SAFE public projects git trees. - safe/jmp/linux-2.6/blob - drivers/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "kvm.h"
  19 #include "x86_emulate.h"
  20 #include "segment_descriptor.h"
  21 #include "irq.h"
  22
  23 #include <linux/kvm.h>
  24 #include <linux/module.h>
  25 #include <linux/errno.h>
  26 #include <linux/percpu.h>
  27 #include <linux/gfp.h>
  28 #include <linux/mm.h>
  29 #include <linux/miscdevice.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/reboot.h>
  32 #include <linux/debugfs.h>
  33 #include <linux/highmem.h>
  34 #include <linux/file.h>
  35 #include <linux/sysdev.h>
  36 #include <linux/cpu.h>
  37 #include <linux/sched.h>
  38 #include <linux/cpumask.h>
  39 #include <linux/smp.h>
  40 #include <linux/anon_inodes.h>
  41 #include <linux/profile.h>
  42 #include <linux/kvm_para.h>
  43
  44 #include <asm/processor.h>
  45 #include <asm/msr.h>
  46 #include <asm/io.h>
  47 #include <asm/uaccess.h>
  48 #include <asm/desc.h>
  49
  50 MODULE_AUTHOR("Qumranet");
  51 MODULE_LICENSE("GPL");
  52
  53 static DEFINE_SPINLOCK(kvm_lock);
  54 static LIST_HEAD(vm_list);
  55
  56 static cpumask_t cpus_hardware_enabled;
  57
  58 struct kvm_x86_ops *kvm_x86_ops;
  59 struct kmem_cache *kvm_vcpu_cache;
  60 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  61
  62 static __read_mostly struct preempt_ops kvm_preempt_ops;
  63
  64 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  65
  66 static struct kvm_stats_debugfs_item {
  67         const char *name;
  68         int offset;
  69         struct dentry *dentry;
  70 } debugfs_entries[] = {
  71         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  72         { "pf_guest", STAT_OFFSET(pf_guest) },
  73         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  74         { "invlpg", STAT_OFFSET(invlpg) },
  75         { "exits", STAT_OFFSET(exits) },
  76         { "io_exits", STAT_OFFSET(io_exits) },
  77         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  78         { "signal_exits", STAT_OFFSET(signal_exits) },
  79         { "irq_window", STAT_OFFSET(irq_window_exits) },
  80         { "halt_exits", STAT_OFFSET(halt_exits) },
  81         { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
  82         { "request_irq", STAT_OFFSET(request_irq_exits) },
  83         { "irq_exits", STAT_OFFSET(irq_exits) },
  84         { "light_exits", STAT_OFFSET(light_exits) },
  85         { "efer_reload", STAT_OFFSET(efer_reload) },
  86         { NULL }
  87 };
  88
  89 static struct dentry *debugfs_dir;
  90
  91 #define MAX_IO_MSRS 256
  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 #ifdef CONFIG_X86_64
 107 // LDT or TSS descriptor in the GDT. 16 bytes.
 108 struct segment_descriptor_64 {
 109         struct segment_descriptor s;
 110         u32 base_higher;
 111         u32 pad_zero;
 112 };
 113
 114 #endif
 115
 116 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 117                            unsigned long arg);
 118
 119 unsigned long segment_base(u16 selector)
 120 {
 121         struct descriptor_table gdt;
 122         struct segment_descriptor *d;
 123         unsigned long table_base;
 124         typedef unsigned long ul;
 125         unsigned long v;
 126
 127         if (selector == 0)
 128                 return 0;
 129
 130         asm ("sgdt %0" : "=m"(gdt));
 131         table_base = gdt.base;
 132
 133         if (selector & 4) {           /* from ldt */
 134                 u16 ldt_selector;
 135
 136                 asm ("sldt %0" : "=g"(ldt_selector));
 137                 table_base = segment_base(ldt_selector);
 138         }
 139         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 140         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 141 #ifdef CONFIG_X86_64
 142         if (d->system == 0
 143             && (d->type == 2 || d->type == 9 || d->type == 11))
 144                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 145 #endif
 146         return v;
 147 }
 148 EXPORT_SYMBOL_GPL(segment_base);
 149
 150 static inline int valid_vcpu(int n)
 151 {
 152         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 153 }
 154
 155 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 156 {
 157         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
 158                 return;
 159
 160         vcpu->guest_fpu_loaded = 1;
 161         fx_save(&vcpu->host_fx_image);
 162         fx_restore(&vcpu->guest_fx_image);
 163 }
 164 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
 165
 166 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 167 {
 168         if (!vcpu->guest_fpu_loaded)
 169                 return;
 170
 171         vcpu->guest_fpu_loaded = 0;
 172         fx_save(&vcpu->guest_fx_image);
 173         fx_restore(&vcpu->host_fx_image);
 174 }
 175 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
 176
 177 /*
 178  * Switches to specified vcpu, until a matching vcpu_put()
 179  */
 180 static void vcpu_load(struct kvm_vcpu *vcpu)
 181 {
 182         int cpu;
 183
 184         mutex_lock(&vcpu->mutex);
 185         cpu = get_cpu();
 186         preempt_notifier_register(&vcpu->preempt_notifier);
 187         kvm_x86_ops->vcpu_load(vcpu, cpu);
 188         put_cpu();
 189 }
 190
 191 static void vcpu_put(struct kvm_vcpu *vcpu)
 192 {
 193         preempt_disable();
 194         kvm_x86_ops->vcpu_put(vcpu);
 195         preempt_notifier_unregister(&vcpu->preempt_notifier);
 196         preempt_enable();
 197         mutex_unlock(&vcpu->mutex);
 198 }
 199
 200 static void ack_flush(void *_completed)
 201 {
 202 }
 203
 204 void kvm_flush_remote_tlbs(struct kvm *kvm)
 205 {
 206         int i, cpu;
 207         cpumask_t cpus;
 208         struct kvm_vcpu *vcpu;
 209
 210         cpus_clear(cpus);
 211         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 212                 vcpu = kvm->vcpus[i];
 213                 if (!vcpu)
 214                         continue;
 215                 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
 216                         continue;
 217                 cpu = vcpu->cpu;
 218                 if (cpu != -1 && cpu != raw_smp_processor_id())
 219                         cpu_set(cpu, cpus);
 220         }
 221         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 222 }
 223
 224 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 225 {
 226         struct page *page;
 227         int r;
 228
 229         mutex_init(&vcpu->mutex);
 230         vcpu->cpu = -1;
 231         vcpu->mmu.root_hpa = INVALID_PAGE;
 232         vcpu->kvm = kvm;
 233         vcpu->vcpu_id = id;
 234         if (!irqchip_in_kernel(kvm) || id == 0)
 235                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
 236         else
 237                 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
 238         init_waitqueue_head(&vcpu->wq);
 239
 240         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 241         if (!page) {
 242                 r = -ENOMEM;
 243                 goto fail;
 244         }
 245         vcpu->run = page_address(page);
 246
 247         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 248         if (!page) {
 249                 r = -ENOMEM;
 250                 goto fail_free_run;
 251         }
 252         vcpu->pio_data = page_address(page);
 253
 254         r = kvm_mmu_create(vcpu);
 255         if (r < 0)
 256                 goto fail_free_pio_data;
 257
 258         return 0;
 259
 260 fail_free_pio_data:
 261         free_page((unsigned long)vcpu->pio_data);
 262 fail_free_run:
 263         free_page((unsigned long)vcpu->run);
 264 fail:
 265         return -ENOMEM;
 266 }
 267 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 268
 269 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 270 {
 271         kvm_mmu_destroy(vcpu);
 272         if (vcpu->apic)
 273                 hrtimer_cancel(&vcpu->apic->timer.dev);
 274         kvm_free_apic(vcpu->apic);
 275         free_page((unsigned long)vcpu->pio_data);
 276         free_page((unsigned long)vcpu->run);
 277 }
 278 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 279
 280 static struct kvm *kvm_create_vm(void)
 281 {
 282         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 283
 284         if (!kvm)
 285                 return ERR_PTR(-ENOMEM);
 286
 287         kvm_io_bus_init(&kvm->pio_bus);
 288         mutex_init(&kvm->lock);
 289         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 290         kvm_io_bus_init(&kvm->mmio_bus);
 291         spin_lock(&kvm_lock);
 292         list_add(&kvm->vm_list, &vm_list);
 293         spin_unlock(&kvm_lock);
 294         return kvm;
 295 }
 296
 297 /*
 298  * Free any memory in @free but not in @dont.
 299  */
 300 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 301                                   struct kvm_memory_slot *dont)
 302 {
 303         int i;
 304
 305         if (!dont || free->phys_mem != dont->phys_mem)
 306                 if (free->phys_mem) {
 307                         for (i = 0; i < free->npages; ++i)
 308                                 if (free->phys_mem[i])
 309                                         __free_page(free->phys_mem[i]);
 310                         vfree(free->phys_mem);
 311                 }
 312
 313         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 314                 vfree(free->dirty_bitmap);
 315
 316         free->phys_mem = NULL;
 317         free->npages = 0;
 318         free->dirty_bitmap = NULL;
 319 }
 320
 321 static void kvm_free_physmem(struct kvm *kvm)
 322 {
 323         int i;
 324
 325         for (i = 0; i < kvm->nmemslots; ++i)
 326                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 327 }
 328
 329 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 330 {
 331         int i;
 332
 333         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
 334                 if (vcpu->pio.guest_pages[i]) {
 335                         __free_page(vcpu->pio.guest_pages[i]);
 336                         vcpu->pio.guest_pages[i] = NULL;
 337                 }
 338 }
 339
 340 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 341 {
 342         vcpu_load(vcpu);
 343         kvm_mmu_unload(vcpu);
 344         vcpu_put(vcpu);
 345 }
 346
 347 static void kvm_free_vcpus(struct kvm *kvm)
 348 {
 349         unsigned int i;
 350
 351         /*
 352          * Unpin any mmu pages first.
 353          */
 354         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 355                 if (kvm->vcpus[i])
 356                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
 357         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 358                 if (kvm->vcpus[i]) {
 359                         kvm_x86_ops->vcpu_free(kvm->vcpus[i]);
 360                         kvm->vcpus[i] = NULL;
 361                 }
 362         }
 363
 364 }
 365
 366 static void kvm_destroy_vm(struct kvm *kvm)
 367 {
 368         spin_lock(&kvm_lock);
 369         list_del(&kvm->vm_list);
 370         spin_unlock(&kvm_lock);
 371         kvm_io_bus_destroy(&kvm->pio_bus);
 372         kvm_io_bus_destroy(&kvm->mmio_bus);
 373         kfree(kvm->vpic);
 374         kfree(kvm->vioapic);
 375         kvm_free_vcpus(kvm);
 376         kvm_free_physmem(kvm);
 377         kfree(kvm);
 378 }
 379
 380 static int kvm_vm_release(struct inode *inode, struct file *filp)
 381 {
 382         struct kvm *kvm = filp->private_data;
 383
 384         kvm_destroy_vm(kvm);
 385         return 0;
 386 }
 387
 388 static void inject_gp(struct kvm_vcpu *vcpu)
 389 {
 390         kvm_x86_ops->inject_gp(vcpu, 0);
 391 }
 392
 393 /*
 394  * Load the pae pdptrs.  Return true is they are all valid.
 395  */
 396 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 397 {
 398         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 399         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 400         int i;
 401         u64 *pdpt;
 402         int ret;
 403         struct page *page;
 404         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
 405
 406         mutex_lock(&vcpu->kvm->lock);
 407         page = gfn_to_page(vcpu->kvm, pdpt_gfn);
 408         if (!page) {
 409                 ret = 0;
 410                 goto out;
 411         }
 412
 413         pdpt = kmap_atomic(page, KM_USER0);
 414         memcpy(pdpte, pdpt+offset, sizeof(pdpte));
 415         kunmap_atomic(pdpt, KM_USER0);
 416
 417         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 418                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 419                         ret = 0;
 420                         goto out;
 421                 }
 422         }
 423         ret = 1;
 424
 425         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
 426 out:
 427         mutex_unlock(&vcpu->kvm->lock);
 428
 429         return ret;
 430 }
 431
 432 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 433 {
 434         if (cr0 & CR0_RESERVED_BITS) {
 435                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 436                        cr0, vcpu->cr0);
 437                 inject_gp(vcpu);
 438                 return;
 439         }
 440
 441         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 442                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 443                 inject_gp(vcpu);
 444                 return;
 445         }
 446
 447         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 448                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 449                        "and a clear PE flag\n");
 450                 inject_gp(vcpu);
 451                 return;
 452         }
 453
 454         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 455 #ifdef CONFIG_X86_64
 456                 if ((vcpu->shadow_efer & EFER_LME)) {
 457                         int cs_db, cs_l;
 458
 459                         if (!is_pae(vcpu)) {
 460                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 461                                        "in long mode while PAE is disabled\n");
 462                                 inject_gp(vcpu);
 463                                 return;
 464                         }
 465                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 466                         if (cs_l) {
 467                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 468                                        "in long mode while CS.L == 1\n");
 469                                 inject_gp(vcpu);
 470                                 return;
 471
 472                         }
 473                 } else
 474 #endif
 475                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 476                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 477                                "reserved bits\n");
 478                         inject_gp(vcpu);
 479                         return;
 480                 }
 481
 482         }
 483
 484         kvm_x86_ops->set_cr0(vcpu, cr0);
 485         vcpu->cr0 = cr0;
 486
 487         mutex_lock(&vcpu->kvm->lock);
 488         kvm_mmu_reset_context(vcpu);
 489         mutex_unlock(&vcpu->kvm->lock);
 490         return;
 491 }
 492 EXPORT_SYMBOL_GPL(set_cr0);
 493
 494 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 495 {
 496         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 497 }
 498 EXPORT_SYMBOL_GPL(lmsw);
 499
 500 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 501 {
 502         if (cr4 & CR4_RESERVED_BITS) {
 503                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 504                 inject_gp(vcpu);
 505                 return;
 506         }
 507
 508         if (is_long_mode(vcpu)) {
 509                 if (!(cr4 & X86_CR4_PAE)) {
 510                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 511                                "in long mode\n");
 512                         inject_gp(vcpu);
 513                         return;
 514                 }
 515         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 516                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 517                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 518                 inject_gp(vcpu);
 519                 return;
 520         }
 521
 522         if (cr4 & X86_CR4_VMXE) {
 523                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 524                 inject_gp(vcpu);
 525                 return;
 526         }
 527         kvm_x86_ops->set_cr4(vcpu, cr4);
 528         vcpu->cr4 = cr4;
 529         mutex_lock(&vcpu->kvm->lock);
 530         kvm_mmu_reset_context(vcpu);
 531         mutex_unlock(&vcpu->kvm->lock);
 532 }
 533 EXPORT_SYMBOL_GPL(set_cr4);
 534
 535 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 536 {
 537         if (is_long_mode(vcpu)) {
 538                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 539                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 540                         inject_gp(vcpu);
 541                         return;
 542                 }
 543         } else {
 544                 if (is_pae(vcpu)) {
 545                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 546                                 printk(KERN_DEBUG
 547                                        "set_cr3: #GP, reserved bits\n");
 548                                 inject_gp(vcpu);
 549                                 return;
 550                         }
 551                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 552                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 553                                        "reserved bits\n");
 554                                 inject_gp(vcpu);
 555                                 return;
 556                         }
 557                 } else {
 558                         if (cr3 & CR3_NONPAE_RESERVED_BITS) {
 559                                 printk(KERN_DEBUG
 560                                        "set_cr3: #GP, reserved bits\n");
 561                                 inject_gp(vcpu);
 562                                 return;
 563                         }
 564                 }
 565         }
 566
 567         mutex_lock(&vcpu->kvm->lock);
 568         /*
 569          * Does the new cr3 value map to physical memory? (Note, we
 570          * catch an invalid cr3 even in real-mode, because it would
 571          * cause trouble later on when we turn on paging anyway.)
 572          *
 573          * A real CPU would silently accept an invalid cr3 and would
 574          * attempt to use it - with largely undefined (and often hard
 575          * to debug) behavior on the guest side.
 576          */
 577         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 578                 inject_gp(vcpu);
 579         else {
 580                 vcpu->cr3 = cr3;
 581                 vcpu->mmu.new_cr3(vcpu);
 582         }
 583         mutex_unlock(&vcpu->kvm->lock);
 584 }
 585 EXPORT_SYMBOL_GPL(set_cr3);
 586
 587 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 588 {
 589         if (cr8 & CR8_RESERVED_BITS) {
 590                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 591                 inject_gp(vcpu);
 592                 return;
 593         }
 594         if (irqchip_in_kernel(vcpu->kvm))
 595                 kvm_lapic_set_tpr(vcpu, cr8);
 596         else
 597                 vcpu->cr8 = cr8;
 598 }
 599 EXPORT_SYMBOL_GPL(set_cr8);
 600
 601 unsigned long get_cr8(struct kvm_vcpu *vcpu)
 602 {
 603         if (irqchip_in_kernel(vcpu->kvm))
 604                 return kvm_lapic_get_cr8(vcpu);
 605         else
 606                 return vcpu->cr8;
 607 }
 608 EXPORT_SYMBOL_GPL(get_cr8);
 609
 610 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 611 {
 612         if (irqchip_in_kernel(vcpu->kvm))
 613                 return vcpu->apic_base;
 614         else
 615                 return vcpu->apic_base;
 616 }
 617 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 618
 619 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 620 {
 621         /* TODO: reserve bits check */
 622         if (irqchip_in_kernel(vcpu->kvm))
 623                 kvm_lapic_set_base(vcpu, data);
 624         else
 625                 vcpu->apic_base = data;
 626 }
 627 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 628
 629 void fx_init(struct kvm_vcpu *vcpu)
 630 {
 631         unsigned after_mxcsr_mask;
 632
 633         /* Initialize guest FPU by resetting ours and saving into guest's */
 634         preempt_disable();
 635         fx_save(&vcpu->host_fx_image);
 636         fpu_init();
 637         fx_save(&vcpu->guest_fx_image);
 638         fx_restore(&vcpu->host_fx_image);
 639         preempt_enable();
 640
 641         vcpu->cr0 |= X86_CR0_ET;
 642         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
 643         vcpu->guest_fx_image.mxcsr = 0x1f80;
 644         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
 645                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
 646 }
 647 EXPORT_SYMBOL_GPL(fx_init);
 648
 649 /*
 650  * Allocate some memory and give it an address in the guest physical address
 651  * space.
 652  *
 653  * Discontiguous memory is allowed, mostly for framebuffers.
 654  */
 655 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 656                                           struct kvm_memory_region *mem)
 657 {
 658         int r;
 659         gfn_t base_gfn;
 660         unsigned long npages;
 661         unsigned long i;
 662         struct kvm_memory_slot *memslot;
 663         struct kvm_memory_slot old, new;
 664
 665         r = -EINVAL;
 666         /* General sanity checks */
 667         if (mem->memory_size & (PAGE_SIZE - 1))
 668                 goto out;
 669         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 670                 goto out;
 671         if (mem->slot >= KVM_MEMORY_SLOTS)
 672                 goto out;
 673         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 674                 goto out;
 675
 676         memslot = &kvm->memslots[mem->slot];
 677         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 678         npages = mem->memory_size >> PAGE_SHIFT;
 679
 680         if (!npages)
 681                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 682
 683         mutex_lock(&kvm->lock);
 684
 685         new = old = *memslot;
 686
 687         new.base_gfn = base_gfn;
 688         new.npages = npages;
 689         new.flags = mem->flags;
 690
 691         /* Disallow changing a memory slot's size. */
 692         r = -EINVAL;
 693         if (npages && old.npages && npages != old.npages)
 694                 goto out_unlock;
 695
 696         /* Check for overlaps */
 697         r = -EEXIST;
 698         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 699                 struct kvm_memory_slot *s = &kvm->memslots[i];
 700
 701                 if (s == memslot)
 702                         continue;
 703                 if (!((base_gfn + npages <= s->base_gfn) ||
 704                       (base_gfn >= s->base_gfn + s->npages)))
 705                         goto out_unlock;
 706         }
 707
 708         /* Deallocate if slot is being removed */
 709         if (!npages)
 710                 new.phys_mem = NULL;
 711
 712         /* Free page dirty bitmap if unneeded */
 713         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 714                 new.dirty_bitmap = NULL;
 715
 716         r = -ENOMEM;
 717
 718         /* Allocate if a slot is being created */
 719         if (npages && !new.phys_mem) {
 720                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 721
 722                 if (!new.phys_mem)
 723                         goto out_unlock;
 724
 725                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 726                 for (i = 0; i < npages; ++i) {
 727                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 728                                                      | __GFP_ZERO);
 729                         if (!new.phys_mem[i])
 730                                 goto out_unlock;
 731                         set_page_private(new.phys_mem[i],0);
 732                 }
 733         }
 734
 735         /* Allocate page dirty bitmap if needed */
 736         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 737                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 738
 739                 new.dirty_bitmap = vmalloc(dirty_bytes);
 740                 if (!new.dirty_bitmap)
 741                         goto out_unlock;
 742                 memset(new.dirty_bitmap, 0, dirty_bytes);
 743         }
 744
 745         if (mem->slot >= kvm->nmemslots)
 746                 kvm->nmemslots = mem->slot + 1;
 747
 748         *memslot = new;
 749
 750         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 751         kvm_flush_remote_tlbs(kvm);
 752
 753         mutex_unlock(&kvm->lock);
 754
 755         kvm_free_physmem_slot(&old, &new);
 756         return 0;
 757
 758 out_unlock:
 759         mutex_unlock(&kvm->lock);
 760         kvm_free_physmem_slot(&new, &old);
 761 out:
 762         return r;
 763 }
 764
 765 /*
 766  * Get (and clear) the dirty memory log for a memory slot.
 767  */
 768 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 769                                       struct kvm_dirty_log *log)
 770 {
 771         struct kvm_memory_slot *memslot;
 772         int r, i;
 773         int n;
 774         unsigned long any = 0;
 775
 776         mutex_lock(&kvm->lock);
 777
 778         r = -EINVAL;
 779         if (log->slot >= KVM_MEMORY_SLOTS)
 780                 goto out;
 781
 782         memslot = &kvm->memslots[log->slot];
 783         r = -ENOENT;
 784         if (!memslot->dirty_bitmap)
 785                 goto out;
 786
 787         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 788
 789         for (i = 0; !any && i < n/sizeof(long); ++i)
 790                 any = memslot->dirty_bitmap[i];
 791
 792         r = -EFAULT;
 793         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 794                 goto out;
 795
 796         /* If nothing is dirty, don't bother messing with page tables. */
 797         if (any) {
 798                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 799                 kvm_flush_remote_tlbs(kvm);
 800                 memset(memslot->dirty_bitmap, 0, n);
 801         }
 802
 803         r = 0;
 804
 805 out:
 806         mutex_unlock(&kvm->lock);
 807         return r;
 808 }
 809
 810 /*
 811  * Set a new alias region.  Aliases map a portion of physical memory into
 812  * another portion.  This is useful for memory windows, for example the PC
 813  * VGA region.
 814  */
 815 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 816                                          struct kvm_memory_alias *alias)
 817 {
 818         int r, n;
 819         struct kvm_mem_alias *p;
 820
 821         r = -EINVAL;
 822         /* General sanity checks */
 823         if (alias->memory_size & (PAGE_SIZE - 1))
 824                 goto out;
 825         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 826                 goto out;
 827         if (alias->slot >= KVM_ALIAS_SLOTS)
 828                 goto out;
 829         if (alias->guest_phys_addr + alias->memory_size
 830             < alias->guest_phys_addr)
 831                 goto out;
 832         if (alias->target_phys_addr + alias->memory_size
 833             < alias->target_phys_addr)
 834                 goto out;
 835
 836         mutex_lock(&kvm->lock);
 837
 838         p = &kvm->aliases[alias->slot];
 839         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 840         p->npages = alias->memory_size >> PAGE_SHIFT;
 841         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 842
 843         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 844                 if (kvm->aliases[n - 1].npages)
 845                         break;
 846         kvm->naliases = n;
 847
 848         kvm_mmu_zap_all(kvm);
 849
 850         mutex_unlock(&kvm->lock);
 851
 852         return 0;
 853
 854 out:
 855         return r;
 856 }
 857
 858 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 859 {
 860         int r;
 861
 862         r = 0;
 863         switch (chip->chip_id) {
 864         case KVM_IRQCHIP_PIC_MASTER:
 865                 memcpy (&chip->chip.pic,
 866                         &pic_irqchip(kvm)->pics[0],
 867                         sizeof(struct kvm_pic_state));
 868                 break;
 869         case KVM_IRQCHIP_PIC_SLAVE:
 870                 memcpy (&chip->chip.pic,
 871                         &pic_irqchip(kvm)->pics[1],
 872                         sizeof(struct kvm_pic_state));
 873                 break;
 874         case KVM_IRQCHIP_IOAPIC:
 875                 memcpy (&chip->chip.ioapic,
 876                         ioapic_irqchip(kvm),
 877                         sizeof(struct kvm_ioapic_state));
 878                 break;
 879         default:
 880                 r = -EINVAL;
 881                 break;
 882         }
 883         return r;
 884 }
 885
 886 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 887 {
 888         int r;
 889
 890         r = 0;
 891         switch (chip->chip_id) {
 892         case KVM_IRQCHIP_PIC_MASTER:
 893                 memcpy (&pic_irqchip(kvm)->pics[0],
 894                         &chip->chip.pic,
 895                         sizeof(struct kvm_pic_state));
 896                 break;
 897         case KVM_IRQCHIP_PIC_SLAVE:
 898                 memcpy (&pic_irqchip(kvm)->pics[1],
 899                         &chip->chip.pic,
 900                         sizeof(struct kvm_pic_state));
 901                 break;
 902         case KVM_IRQCHIP_IOAPIC:
 903                 memcpy (ioapic_irqchip(kvm),
 904                         &chip->chip.ioapic,
 905                         sizeof(struct kvm_ioapic_state));
 906                 break;
 907         default:
 908                 r = -EINVAL;
 909                 break;
 910         }
 911         kvm_pic_update_irq(pic_irqchip(kvm));
 912         return r;
 913 }
 914
 915 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 916 {
 917         int i;
 918         struct kvm_mem_alias *alias;
 919
 920         for (i = 0; i < kvm->naliases; ++i) {
 921                 alias = &kvm->aliases[i];
 922                 if (gfn >= alias->base_gfn
 923                     && gfn < alias->base_gfn + alias->npages)
 924                         return alias->target_gfn + gfn - alias->base_gfn;
 925         }
 926         return gfn;
 927 }
 928
 929 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 930 {
 931         int i;
 932
 933         for (i = 0; i < kvm->nmemslots; ++i) {
 934                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 935
 936                 if (gfn >= memslot->base_gfn
 937                     && gfn < memslot->base_gfn + memslot->npages)
 938                         return memslot;
 939         }
 940         return NULL;
 941 }
 942
 943 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 944 {
 945         gfn = unalias_gfn(kvm, gfn);
 946         return __gfn_to_memslot(kvm, gfn);
 947 }
 948
 949 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 950 {
 951         struct kvm_memory_slot *slot;
 952
 953         gfn = unalias_gfn(kvm, gfn);
 954         slot = __gfn_to_memslot(kvm, gfn);
 955         if (!slot)
 956                 return NULL;
 957         return slot->phys_mem[gfn - slot->base_gfn];
 958 }
 959 EXPORT_SYMBOL_GPL(gfn_to_page);
 960
 961 /* WARNING: Does not work on aliased pages. */
 962 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 963 {
 964         struct kvm_memory_slot *memslot;
 965
 966         memslot = __gfn_to_memslot(kvm, gfn);
 967         if (memslot && memslot->dirty_bitmap) {
 968                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 969
 970                 /* avoid RMW */
 971                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 972                         set_bit(rel_gfn, memslot->dirty_bitmap);
 973         }
 974 }
 975
 976 int emulator_read_std(unsigned long addr,
 977                              void *val,
 978                              unsigned int bytes,
 979                              struct kvm_vcpu *vcpu)
 980 {
 981         void *data = val;
 982
 983         while (bytes) {
 984                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 985                 unsigned offset = addr & (PAGE_SIZE-1);
 986                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 987                 unsigned long pfn;
 988                 struct page *page;
 989                 void *page_virt;
 990
 991                 if (gpa == UNMAPPED_GVA)
 992                         return X86EMUL_PROPAGATE_FAULT;
 993                 pfn = gpa >> PAGE_SHIFT;
 994                 page = gfn_to_page(vcpu->kvm, pfn);
 995                 if (!page)
 996                         return X86EMUL_UNHANDLEABLE;
 997                 page_virt = kmap_atomic(page, KM_USER0);
 998
 999                 memcpy(data, page_virt + offset, tocopy);
1000
1001                 kunmap_atomic(page_virt, KM_USER0);
1002
1003                 bytes -= tocopy;
1004                 data += tocopy;
1005                 addr += tocopy;
1006         }
1007
1008         return X86EMUL_CONTINUE;
1009 }
1010 EXPORT_SYMBOL_GPL(emulator_read_std);
1011
1012 static int emulator_write_std(unsigned long addr,
1013                               const void *val,
1014                               unsigned int bytes,
1015                               struct kvm_vcpu *vcpu)
1016 {
1017         pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1018         return X86EMUL_UNHANDLEABLE;
1019 }
1020
1021 /*
1022  * Only apic need an MMIO device hook, so shortcut now..
1023  */
1024 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1025                                                 gpa_t addr)
1026 {
1027         struct kvm_io_device *dev;
1028
1029         if (vcpu->apic) {
1030                 dev = &vcpu->apic->dev;
1031                 if (dev->in_range(dev, addr))
1032                         return dev;
1033         }
1034         return NULL;
1035 }
1036
1037 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1038                                                 gpa_t addr)
1039 {
1040         struct kvm_io_device *dev;
1041
1042         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1043         if (dev == NULL)
1044                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1045         return dev;
1046 }
1047
1048 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1049                                                gpa_t addr)
1050 {
1051         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1052 }
1053
1054 static int emulator_read_emulated(unsigned long addr,
1055                                   void *val,
1056                                   unsigned int bytes,
1057                                   struct kvm_vcpu *vcpu)
1058 {
1059         struct kvm_io_device *mmio_dev;
1060         gpa_t                 gpa;
1061
1062         if (vcpu->mmio_read_completed) {
1063                 memcpy(val, vcpu->mmio_data, bytes);
1064                 vcpu->mmio_read_completed = 0;
1065                 return X86EMUL_CONTINUE;
1066         } else if (emulator_read_std(addr, val, bytes, vcpu)
1067                    == X86EMUL_CONTINUE)
1068                 return X86EMUL_CONTINUE;
1069
1070         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1071         if (gpa == UNMAPPED_GVA)
1072                 return X86EMUL_PROPAGATE_FAULT;
1073
1074         /*
1075          * Is this MMIO handled locally?
1076          */
1077         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1078         if (mmio_dev) {
1079                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1080                 return X86EMUL_CONTINUE;
1081         }
1082
1083         vcpu->mmio_needed = 1;
1084         vcpu->mmio_phys_addr = gpa;
1085         vcpu->mmio_size = bytes;
1086         vcpu->mmio_is_write = 0;
1087
1088         return X86EMUL_UNHANDLEABLE;
1089 }
1090
1091 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1092                                const void *val, int bytes)
1093 {
1094         struct page *page;
1095         void *virt;
1096
1097         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1098                 return 0;
1099         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1100         if (!page)
1101                 return 0;
1102         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1103         virt = kmap_atomic(page, KM_USER0);
1104         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1105         memcpy(virt + offset_in_page(gpa), val, bytes);
1106         kunmap_atomic(virt, KM_USER0);
1107         return 1;
1108 }
1109
1110 static int emulator_write_emulated_onepage(unsigned long addr,
1111                                            const void *val,
1112                                            unsigned int bytes,
1113                                            struct kvm_vcpu *vcpu)
1114 {
1115         struct kvm_io_device *mmio_dev;
1116         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1117
1118         if (gpa == UNMAPPED_GVA) {
1119                 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1120                 return X86EMUL_PROPAGATE_FAULT;
1121         }
1122
1123         if (emulator_write_phys(vcpu, gpa, val, bytes))
1124                 return X86EMUL_CONTINUE;
1125
1126         /*
1127          * Is this MMIO handled locally?
1128          */
1129         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1130         if (mmio_dev) {
1131                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1132                 return X86EMUL_CONTINUE;
1133         }
1134
1135         vcpu->mmio_needed = 1;
1136         vcpu->mmio_phys_addr = gpa;
1137         vcpu->mmio_size = bytes;
1138         vcpu->mmio_is_write = 1;
1139         memcpy(vcpu->mmio_data, val, bytes);
1140
1141         return X86EMUL_CONTINUE;
1142 }
1143
1144 int emulator_write_emulated(unsigned long addr,
1145                                    const void *val,
1146                                    unsigned int bytes,
1147                                    struct kvm_vcpu *vcpu)
1148 {
1149         /* Crossing a page boundary? */
1150         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1151                 int rc, now;
1152
1153                 now = -addr & ~PAGE_MASK;
1154                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1155                 if (rc != X86EMUL_CONTINUE)
1156                         return rc;
1157                 addr += now;
1158                 val += now;
1159                 bytes -= now;
1160         }
1161         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1162 }
1163 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1164
1165 static int emulator_cmpxchg_emulated(unsigned long addr,
1166                                      const void *old,
1167                                      const void *new,
1168                                      unsigned int bytes,
1169                                      struct kvm_vcpu *vcpu)
1170 {
1171         static int reported;
1172
1173         if (!reported) {
1174                 reported = 1;
1175                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1176         }
1177         return emulator_write_emulated(addr, new, bytes, vcpu);
1178 }
1179
1180 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1181 {
1182         return kvm_x86_ops->get_segment_base(vcpu, seg);
1183 }
1184
1185 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1186 {
1187         return X86EMUL_CONTINUE;
1188 }
1189
1190 int emulate_clts(struct kvm_vcpu *vcpu)
1191 {
1192         kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1193         return X86EMUL_CONTINUE;
1194 }
1195
1196 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1197 {
1198         struct kvm_vcpu *vcpu = ctxt->vcpu;
1199
1200         switch (dr) {
1201         case 0 ... 3:
1202                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1203                 return X86EMUL_CONTINUE;
1204         default:
1205                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1206                 return X86EMUL_UNHANDLEABLE;
1207         }
1208 }
1209
1210 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1211 {
1212         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1213         int exception;
1214
1215         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1216         if (exception) {
1217                 /* FIXME: better handling */
1218                 return X86EMUL_UNHANDLEABLE;
1219         }
1220         return X86EMUL_CONTINUE;
1221 }
1222
1223 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1224 {
1225         static int reported;
1226         u8 opcodes[4];
1227         unsigned long rip = vcpu->rip;
1228         unsigned long rip_linear;
1229
1230         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1231
1232         if (reported)
1233                 return;
1234
1235         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1236
1237         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1238                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1239         reported = 1;
1240 }
1241 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1242
1243 struct x86_emulate_ops emulate_ops = {
1244         .read_std            = emulator_read_std,
1245         .write_std           = emulator_write_std,
1246         .read_emulated       = emulator_read_emulated,
1247         .write_emulated      = emulator_write_emulated,
1248         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1249 };
1250
1251 int emulate_instruction(struct kvm_vcpu *vcpu,
1252                         struct kvm_run *run,
1253                         unsigned long cr2,
1254                         u16 error_code,
1255                         int no_decode)
1256 {
1257         int r = 0;
1258
1259         vcpu->mmio_fault_cr2 = cr2;
1260         kvm_x86_ops->cache_regs(vcpu);
1261
1262         vcpu->mmio_is_write = 0;
1263         vcpu->pio.string = 0;
1264
1265         if (!no_decode) {
1266                 int cs_db, cs_l;
1267                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1268
1269                 vcpu->emulate_ctxt.vcpu = vcpu;
1270                 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1271                 vcpu->emulate_ctxt.cr2 = cr2;
1272                 vcpu->emulate_ctxt.mode =
1273                         (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1274                         ? X86EMUL_MODE_REAL : cs_l
1275                         ? X86EMUL_MODE_PROT64 : cs_db
1276                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1277
1278                 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1279                         vcpu->emulate_ctxt.cs_base = 0;
1280                         vcpu->emulate_ctxt.ds_base = 0;
1281                         vcpu->emulate_ctxt.es_base = 0;
1282                         vcpu->emulate_ctxt.ss_base = 0;
1283                 } else {
1284                         vcpu->emulate_ctxt.cs_base =
1285                                         get_segment_base(vcpu, VCPU_SREG_CS);
1286                         vcpu->emulate_ctxt.ds_base =
1287                                         get_segment_base(vcpu, VCPU_SREG_DS);
1288                         vcpu->emulate_ctxt.es_base =
1289                                         get_segment_base(vcpu, VCPU_SREG_ES);
1290                         vcpu->emulate_ctxt.ss_base =
1291                                         get_segment_base(vcpu, VCPU_SREG_SS);
1292                 }
1293
1294                 vcpu->emulate_ctxt.gs_base =
1295                                         get_segment_base(vcpu, VCPU_SREG_GS);
1296                 vcpu->emulate_ctxt.fs_base =
1297                                         get_segment_base(vcpu, VCPU_SREG_FS);
1298
1299                 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
1300         }
1301
1302         if (r == 0)
1303                 r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1304
1305         if (vcpu->pio.string)
1306                 return EMULATE_DO_MMIO;
1307
1308         if ((r || vcpu->mmio_is_write) && run) {
1309                 run->exit_reason = KVM_EXIT_MMIO;
1310                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1311                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1312                 run->mmio.len = vcpu->mmio_size;
1313                 run->mmio.is_write = vcpu->mmio_is_write;
1314         }
1315
1316         if (r) {
1317                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1318                         return EMULATE_DONE;
1319                 if (!vcpu->mmio_needed) {
1320                         kvm_report_emulation_failure(vcpu, "mmio");
1321                         return EMULATE_FAIL;
1322                 }
1323                 return EMULATE_DO_MMIO;
1324         }
1325
1326         kvm_x86_ops->decache_regs(vcpu);
1327         kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1328
1329         if (vcpu->mmio_is_write) {
1330                 vcpu->mmio_needed = 0;
1331                 return EMULATE_DO_MMIO;
1332         }
1333
1334         return EMULATE_DONE;
1335 }
1336 EXPORT_SYMBOL_GPL(emulate_instruction);
1337
1338 /*
1339  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1340  */
1341 static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1342 {
1343         DECLARE_WAITQUEUE(wait, current);
1344
1345         add_wait_queue(&vcpu->wq, &wait);
1346
1347         /*
1348          * We will block until either an interrupt or a signal wakes us up
1349          */
1350         while (!kvm_cpu_has_interrupt(vcpu)
1351                && !signal_pending(current)
1352                && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
1353                && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
1354                 set_current_state(TASK_INTERRUPTIBLE);
1355                 vcpu_put(vcpu);
1356                 schedule();
1357                 vcpu_load(vcpu);
1358         }
1359
1360         __set_current_state(TASK_RUNNING);
1361         remove_wait_queue(&vcpu->wq, &wait);
1362 }
1363
1364 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1365 {
1366         ++vcpu->stat.halt_exits;
1367         if (irqchip_in_kernel(vcpu->kvm)) {
1368                 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1369                 kvm_vcpu_block(vcpu);
1370                 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1371                         return -EINTR;
1372                 return 1;
1373         } else {
1374                 vcpu->run->exit_reason = KVM_EXIT_HLT;
1375                 return 0;
1376         }
1377 }
1378 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1379
1380 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
1381 {
1382         unsigned long nr, a0, a1, a2, a3, ret;
1383
1384         kvm_x86_ops->cache_regs(vcpu);
1385
1386         nr = vcpu->regs[VCPU_REGS_RAX];
1387         a0 = vcpu->regs[VCPU_REGS_RBX];
1388         a1 = vcpu->regs[VCPU_REGS_RCX];
1389         a2 = vcpu->regs[VCPU_REGS_RDX];
1390         a3 = vcpu->regs[VCPU_REGS_RSI];
1391
1392         if (!is_long_mode(vcpu)) {
1393                 nr &= 0xFFFFFFFF;
1394                 a0 &= 0xFFFFFFFF;
1395                 a1 &= 0xFFFFFFFF;
1396                 a2 &= 0xFFFFFFFF;
1397                 a3 &= 0xFFFFFFFF;
1398         }
1399
1400         switch (nr) {
1401         default:
1402                 ret = -KVM_ENOSYS;
1403                 break;
1404         }
1405         vcpu->regs[VCPU_REGS_RAX] = ret;
1406         kvm_x86_ops->decache_regs(vcpu);
1407         return 0;
1408 }
1409 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1410
1411 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1412 {
1413         char instruction[3];
1414         int ret = 0;
1415
1416         mutex_lock(&vcpu->kvm->lock);
1417
1418         /*
1419          * Blow out the MMU to ensure that no other VCPU has an active mapping
1420          * to ensure that the updated hypercall appears atomically across all
1421          * VCPUs.
1422          */
1423         kvm_mmu_zap_all(vcpu->kvm);
1424
1425         kvm_x86_ops->cache_regs(vcpu);
1426         kvm_x86_ops->patch_hypercall(vcpu, instruction);
1427         if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1428             != X86EMUL_CONTINUE)
1429                 ret = -EFAULT;
1430
1431         mutex_unlock(&vcpu->kvm->lock);
1432
1433         return ret;
1434 }
1435
1436 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1437 {
1438         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1439 }
1440
1441 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1442 {
1443         struct descriptor_table dt = { limit, base };
1444
1445         kvm_x86_ops->set_gdt(vcpu, &dt);
1446 }
1447
1448 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1449 {
1450         struct descriptor_table dt = { limit, base };
1451
1452         kvm_x86_ops->set_idt(vcpu, &dt);
1453 }
1454
1455 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1456                    unsigned long *rflags)
1457 {
1458         lmsw(vcpu, msw);
1459         *rflags = kvm_x86_ops->get_rflags(vcpu);
1460 }
1461
1462 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1463 {
1464         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1465         switch (cr) {
1466         case 0:
1467                 return vcpu->cr0;
1468         case 2:
1469                 return vcpu->cr2;
1470         case 3:
1471                 return vcpu->cr3;
1472         case 4:
1473                 return vcpu->cr4;
1474         default:
1475                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1476                 return 0;
1477         }
1478 }
1479
1480 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1481                      unsigned long *rflags)
1482 {
1483         switch (cr) {
1484         case 0:
1485                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1486                 *rflags = kvm_x86_ops->get_rflags(vcpu);
1487                 break;
1488         case 2:
1489                 vcpu->cr2 = val;
1490                 break;
1491         case 3:
1492                 set_cr3(vcpu, val);
1493                 break;
1494         case 4:
1495                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1496                 break;
1497         default:
1498                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1499         }
1500 }
1501
1502 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1503 {
1504         u64 data;
1505
1506         switch (msr) {
1507         case 0xc0010010: /* SYSCFG */
1508         case 0xc0010015: /* HWCR */
1509         case MSR_IA32_PLATFORM_ID:
1510         case MSR_IA32_P5_MC_ADDR:
1511         case MSR_IA32_P5_MC_TYPE:
1512         case MSR_IA32_MC0_CTL:
1513         case MSR_IA32_MCG_STATUS:
1514         case MSR_IA32_MCG_CAP:
1515         case MSR_IA32_MC0_MISC:
1516         case MSR_IA32_MC0_MISC+4:
1517         case MSR_IA32_MC0_MISC+8:
1518         case MSR_IA32_MC0_MISC+12:
1519         case MSR_IA32_MC0_MISC+16:
1520         case MSR_IA32_UCODE_REV:
1521         case MSR_IA32_PERF_STATUS:
1522         case MSR_IA32_EBL_CR_POWERON:
1523                 /* MTRR registers */
1524         case 0xfe:
1525         case 0x200 ... 0x2ff:
1526                 data = 0;
1527                 break;
1528         case 0xcd: /* fsb frequency */
1529                 data = 3;
1530                 break;
1531         case MSR_IA32_APICBASE:
1532                 data = kvm_get_apic_base(vcpu);
1533                 break;
1534         case MSR_IA32_MISC_ENABLE:
1535                 data = vcpu->ia32_misc_enable_msr;
1536                 break;
1537 #ifdef CONFIG_X86_64
1538         case MSR_EFER:
1539                 data = vcpu->shadow_efer;
1540                 break;
1541 #endif
1542         default:
1543                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1544                 return 1;
1545         }
1546         *pdata = data;
1547         return 0;
1548 }
1549 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1550
1551 /*
1552  * Reads an msr value (of 'msr_index') into 'pdata'.
1553  * Returns 0 on success, non-0 otherwise.
1554  * Assumes vcpu_load() was already called.
1555  */
1556 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1557 {
1558         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1559 }
1560
1561 #ifdef CONFIG_X86_64
1562
1563 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1564 {
1565         if (efer & EFER_RESERVED_BITS) {
1566                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1567                        efer);
1568                 inject_gp(vcpu);
1569                 return;
1570         }
1571
1572         if (is_paging(vcpu)
1573             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1574                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1575                 inject_gp(vcpu);
1576                 return;
1577         }
1578
1579         kvm_x86_ops->set_efer(vcpu, efer);
1580
1581         efer &= ~EFER_LMA;
1582         efer |= vcpu->shadow_efer & EFER_LMA;
1583
1584         vcpu->shadow_efer = efer;
1585 }
1586
1587 #endif
1588
1589 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1590 {
1591         switch (msr) {
1592 #ifdef CONFIG_X86_64
1593         case MSR_EFER:
1594                 set_efer(vcpu, data);
1595                 break;
1596 #endif
1597         case MSR_IA32_MC0_STATUS:
1598                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1599                        __FUNCTION__, data);
1600                 break;
1601         case MSR_IA32_MCG_STATUS:
1602                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1603                         __FUNCTION__, data);
1604                 break;
1605         case MSR_IA32_UCODE_REV:
1606         case MSR_IA32_UCODE_WRITE:
1607         case 0x200 ... 0x2ff: /* MTRRs */
1608                 break;
1609         case MSR_IA32_APICBASE:
1610                 kvm_set_apic_base(vcpu, data);
1611                 break;
1612         case MSR_IA32_MISC_ENABLE:
1613                 vcpu->ia32_misc_enable_msr = data;
1614                 break;
1615         default:
1616                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1617                 return 1;
1618         }
1619         return 0;
1620 }
1621 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1622
1623 /*
1624  * Writes msr value into into the appropriate "register".
1625  * Returns 0 on success, non-0 otherwise.
1626  * Assumes vcpu_load() was already called.
1627  */
1628 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1629 {
1630         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
1631 }
1632
1633 void kvm_resched(struct kvm_vcpu *vcpu)
1634 {
1635         if (!need_resched())
1636                 return;
1637         cond_resched();
1638 }
1639 EXPORT_SYMBOL_GPL(kvm_resched);
1640
1641 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1642 {
1643         int i;
1644         u32 function;
1645         struct kvm_cpuid_entry *e, *best;
1646
1647         kvm_x86_ops->cache_regs(vcpu);
1648         function = vcpu->regs[VCPU_REGS_RAX];
1649         vcpu->regs[VCPU_REGS_RAX] = 0;
1650         vcpu->regs[VCPU_REGS_RBX] = 0;
1651         vcpu->regs[VCPU_REGS_RCX] = 0;
1652         vcpu->regs[VCPU_REGS_RDX] = 0;
1653         best = NULL;
1654         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1655                 e = &vcpu->cpuid_entries[i];
1656                 if (e->function == function) {
1657                         best = e;
1658                         break;
1659                 }
1660                 /*
1661                  * Both basic or both extended?
1662                  */
1663                 if (((e->function ^ function) & 0x80000000) == 0)
1664                         if (!best || e->function > best->function)
1665                                 best = e;
1666         }
1667         if (best) {
1668                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1669                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1670                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1671                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1672         }
1673         kvm_x86_ops->decache_regs(vcpu);
1674         kvm_x86_ops->skip_emulated_instruction(vcpu);
1675 }
1676 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1677
1678 static int pio_copy_data(struct kvm_vcpu *vcpu)
1679 {
1680         void *p = vcpu->pio_data;
1681         void *q;
1682         unsigned bytes;
1683         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1684
1685         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1686                  PAGE_KERNEL);
1687         if (!q) {
1688                 free_pio_guest_pages(vcpu);
1689                 return -ENOMEM;
1690         }
1691         q += vcpu->pio.guest_page_offset;
1692         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1693         if (vcpu->pio.in)
1694                 memcpy(q, p, bytes);
1695         else
1696                 memcpy(p, q, bytes);
1697         q -= vcpu->pio.guest_page_offset;
1698         vunmap(q);
1699         free_pio_guest_pages(vcpu);
1700         return 0;
1701 }
1702
1703 static int complete_pio(struct kvm_vcpu *vcpu)
1704 {
1705         struct kvm_pio_request *io = &vcpu->pio;
1706         long delta;
1707         int r;
1708
1709         kvm_x86_ops->cache_regs(vcpu);
1710
1711         if (!io->string) {
1712                 if (io->in)
1713                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1714                                io->size);
1715         } else {
1716                 if (io->in) {
1717                         r = pio_copy_data(vcpu);
1718                         if (r) {
1719                                 kvm_x86_ops->cache_regs(vcpu);
1720                                 return r;
1721                         }
1722                 }
1723
1724                 delta = 1;
1725                 if (io->rep) {
1726                         delta *= io->cur_count;
1727                         /*
1728                          * The size of the register should really depend on
1729                          * current address size.
1730                          */
1731                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1732                 }
1733                 if (io->down)
1734                         delta = -delta;
1735                 delta *= io->size;
1736                 if (io->in)
1737                         vcpu->regs[VCPU_REGS_RDI] += delta;
1738                 else
1739                         vcpu->regs[VCPU_REGS_RSI] += delta;
1740         }
1741
1742         kvm_x86_ops->decache_regs(vcpu);
1743
1744         io->count -= io->cur_count;
1745         io->cur_count = 0;
1746
1747         return 0;
1748 }
1749
1750 static void kernel_pio(struct kvm_io_device *pio_dev,
1751                        struct kvm_vcpu *vcpu,
1752                        void *pd)
1753 {
1754         /* TODO: String I/O for in kernel device */
1755
1756         mutex_lock(&vcpu->kvm->lock);
1757         if (vcpu->pio.in)
1758                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1759                                   vcpu->pio.size,
1760                                   pd);
1761         else
1762                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1763                                    vcpu->pio.size,
1764                                    pd);
1765         mutex_unlock(&vcpu->kvm->lock);
1766 }
1767
1768 static void pio_string_write(struct kvm_io_device *pio_dev,
1769                              struct kvm_vcpu *vcpu)
1770 {
1771         struct kvm_pio_request *io = &vcpu->pio;
1772         void *pd = vcpu->pio_data;
1773         int i;
1774
1775         mutex_lock(&vcpu->kvm->lock);
1776         for (i = 0; i < io->cur_count; i++) {
1777                 kvm_iodevice_write(pio_dev, io->port,
1778                                    io->size,
1779                                    pd);
1780                 pd += io->size;
1781         }
1782         mutex_unlock(&vcpu->kvm->lock);
1783 }
1784
1785 int kvm_emulate_pio (struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1786                   int size, unsigned port)
1787 {
1788         struct kvm_io_device *pio_dev;
1789
1790         vcpu->run->exit_reason = KVM_EXIT_IO;
1791         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1792         vcpu->run->io.size = vcpu->pio.size = size;
1793         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1794         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1795         vcpu->run->io.port = vcpu->pio.port = port;
1796         vcpu->pio.in = in;
1797         vcpu->pio.string = 0;
1798         vcpu->pio.down = 0;
1799         vcpu->pio.guest_page_offset = 0;
1800         vcpu->pio.rep = 0;
1801
1802         kvm_x86_ops->cache_regs(vcpu);
1803         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1804         kvm_x86_ops->decache_regs(vcpu);
1805
1806         kvm_x86_ops->skip_emulated_instruction(vcpu);
1807
1808         pio_dev = vcpu_find_pio_dev(vcpu, port);
1809         if (pio_dev) {
1810                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1811                 complete_pio(vcpu);
1812                 return 1;
1813         }
1814         return 0;
1815 }
1816 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1817
1818 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1819                   int size, unsigned long count, int down,
1820                   gva_t address, int rep, unsigned port)
1821 {
1822         unsigned now, in_page;
1823         int i, ret = 0;
1824         int nr_pages = 1;
1825         struct page *page;
1826         struct kvm_io_device *pio_dev;
1827
1828         vcpu->run->exit_reason = KVM_EXIT_IO;
1829         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1830         vcpu->run->io.size = vcpu->pio.size = size;
1831         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1832         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1833         vcpu->run->io.port = vcpu->pio.port = port;
1834         vcpu->pio.in = in;
1835         vcpu->pio.string = 1;
1836         vcpu->pio.down = down;
1837         vcpu->pio.guest_page_offset = offset_in_page(address);
1838         vcpu->pio.rep = rep;
1839
1840         if (!count) {
1841                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1842                 return 1;
1843         }
1844
1845         if (!down)
1846                 in_page = PAGE_SIZE - offset_in_page(address);
1847         else
1848                 in_page = offset_in_page(address) + size;
1849         now = min(count, (unsigned long)in_page / size);
1850         if (!now) {
1851                 /*
1852                  * String I/O straddles page boundary.  Pin two guest pages
1853                  * so that we satisfy atomicity constraints.  Do just one
1854                  * transaction to avoid complexity.
1855                  */
1856                 nr_pages = 2;
1857                 now = 1;
1858         }
1859         if (down) {
1860                 /*
1861                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1862                  */
1863                 pr_unimpl(vcpu, "guest string pio down\n");
1864                 inject_gp(vcpu);
1865                 return 1;
1866         }
1867         vcpu->run->io.count = now;
1868         vcpu->pio.cur_count = now;
1869
1870         if (vcpu->pio.cur_count == vcpu->pio.count)
1871                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1872
1873         for (i = 0; i < nr_pages; ++i) {
1874                 mutex_lock(&vcpu->kvm->lock);
1875                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1876                 if (page)
1877                         get_page(page);
1878                 vcpu->pio.guest_pages[i] = page;
1879                 mutex_unlock(&vcpu->kvm->lock);
1880                 if (!page) {
1881                         inject_gp(vcpu);
1882                         free_pio_guest_pages(vcpu);
1883                         return 1;
1884                 }
1885         }
1886
1887         pio_dev = vcpu_find_pio_dev(vcpu, port);
1888         if (!vcpu->pio.in) {
1889                 /* string PIO write */
1890                 ret = pio_copy_data(vcpu);
1891                 if (ret >= 0 && pio_dev) {
1892                         pio_string_write(pio_dev, vcpu);
1893                         complete_pio(vcpu);
1894                         if (vcpu->pio.count == 0)
1895                                 ret = 1;
1896                 }
1897         } else if (pio_dev)
1898                 pr_unimpl(vcpu, "no string pio read support yet, "
1899                        "port %x size %d count %ld\n",
1900                         port, size, count);
1901
1902         return ret;
1903 }
1904 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1905
1906 /*
1907  * Check if userspace requested an interrupt window, and that the
1908  * interrupt window is open.
1909  *
1910  * No need to exit to userspace if we already have an interrupt queued.
1911  */
1912 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1913                                           struct kvm_run *kvm_run)
1914 {
1915         return (!vcpu->irq_summary &&
1916                 kvm_run->request_interrupt_window &&
1917                 vcpu->interrupt_window_open &&
1918                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
1919 }
1920
1921 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1922                               struct kvm_run *kvm_run)
1923 {
1924         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
1925         kvm_run->cr8 = get_cr8(vcpu);
1926         kvm_run->apic_base = kvm_get_apic_base(vcpu);
1927         if (irqchip_in_kernel(vcpu->kvm))
1928                 kvm_run->ready_for_interrupt_injection = 1;
1929         else
1930                 kvm_run->ready_for_interrupt_injection =
1931                                         (vcpu->interrupt_window_open &&
1932                                          vcpu->irq_summary == 0);
1933 }
1934
1935 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1936 {
1937         int r;
1938
1939         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
1940                 printk("vcpu %d received sipi with vector # %x\n",
1941                        vcpu->vcpu_id, vcpu->sipi_vector);
1942                 kvm_lapic_reset(vcpu);
1943                 kvm_x86_ops->vcpu_reset(vcpu);
1944                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
1945         }
1946
1947 preempted:
1948         if (vcpu->guest_debug.enabled)
1949                 kvm_x86_ops->guest_debug_pre(vcpu);
1950
1951 again:
1952         r = kvm_mmu_reload(vcpu);
1953         if (unlikely(r))
1954                 goto out;
1955
1956         preempt_disable();
1957
1958         kvm_x86_ops->prepare_guest_switch(vcpu);
1959         kvm_load_guest_fpu(vcpu);
1960
1961         local_irq_disable();
1962
1963         if (signal_pending(current)) {
1964                 local_irq_enable();
1965                 preempt_enable();
1966                 r = -EINTR;
1967                 kvm_run->exit_reason = KVM_EXIT_INTR;
1968                 ++vcpu->stat.signal_exits;
1969                 goto out;
1970         }
1971
1972         if (irqchip_in_kernel(vcpu->kvm))
1973                 kvm_x86_ops->inject_pending_irq(vcpu);
1974         else if (!vcpu->mmio_read_completed)
1975                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
1976
1977         vcpu->guest_mode = 1;
1978         kvm_guest_enter();
1979
1980         if (vcpu->requests)
1981                 if (test_and_clear_bit(KVM_TLB_FLUSH, &vcpu->requests))
1982                         kvm_x86_ops->tlb_flush(vcpu);
1983
1984         kvm_x86_ops->run(vcpu, kvm_run);
1985
1986         vcpu->guest_mode = 0;
1987         local_irq_enable();
1988
1989         ++vcpu->stat.exits;
1990
1991         /*
1992          * We must have an instruction between local_irq_enable() and
1993          * kvm_guest_exit(), so the timer interrupt isn't delayed by
1994          * the interrupt shadow.  The stat.exits increment will do nicely.
1995          * But we need to prevent reordering, hence this barrier():
1996          */
1997         barrier();
1998
1999         kvm_guest_exit();
2000
2001         preempt_enable();
2002
2003         /*
2004          * Profile KVM exit RIPs:
2005          */
2006         if (unlikely(prof_on == KVM_PROFILING)) {
2007                 kvm_x86_ops->cache_regs(vcpu);
2008                 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
2009         }
2010
2011         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2012
2013         if (r > 0) {
2014                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2015                         r = -EINTR;
2016                         kvm_run->exit_reason = KVM_EXIT_INTR;
2017                         ++vcpu->stat.request_irq_exits;
2018                         goto out;
2019                 }
2020                 if (!need_resched()) {
2021                         ++vcpu->stat.light_exits;
2022                         goto again;
2023                 }
2024         }
2025
2026 out:
2027         if (r > 0) {
2028                 kvm_resched(vcpu);
2029                 goto preempted;
2030         }
2031
2032         post_kvm_run_save(vcpu, kvm_run);
2033
2034         return r;
2035 }
2036
2037
2038 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2039 {
2040         int r;
2041         sigset_t sigsaved;
2042
2043         vcpu_load(vcpu);
2044
2045         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2046                 kvm_vcpu_block(vcpu);
2047                 vcpu_put(vcpu);
2048                 return -EAGAIN;
2049         }
2050
2051         if (vcpu->sigset_active)
2052                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2053
2054         /* re-sync apic's tpr */
2055         if (!irqchip_in_kernel(vcpu->kvm))
2056                 set_cr8(vcpu, kvm_run->cr8);
2057
2058         if (vcpu->pio.cur_count) {
2059                 r = complete_pio(vcpu);
2060                 if (r)
2061                         goto out;
2062         }
2063
2064         if (vcpu->mmio_needed) {
2065                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2066                 vcpu->mmio_read_completed = 1;
2067                 vcpu->mmio_needed = 0;
2068                 r = emulate_instruction(vcpu, kvm_run,
2069                                         vcpu->mmio_fault_cr2, 0, 1);
2070                 if (r == EMULATE_DO_MMIO) {
2071                         /*
2072                          * Read-modify-write.  Back to userspace.
2073                          */
2074                         r = 0;
2075                         goto out;
2076                 }
2077         }
2078
2079         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2080                 kvm_x86_ops->cache_regs(vcpu);
2081                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2082                 kvm_x86_ops->decache_regs(vcpu);
2083         }
2084
2085         r = __vcpu_run(vcpu, kvm_run);
2086
2087 out:
2088         if (vcpu->sigset_active)
2089                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2090
2091         vcpu_put(vcpu);
2092         return r;
2093 }
2094
2095 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2096                                    struct kvm_regs *regs)
2097 {
2098         vcpu_load(vcpu);
2099
2100         kvm_x86_ops->cache_regs(vcpu);
2101
2102         regs->rax = vcpu->regs[VCPU_REGS_RAX];
2103         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2104         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2105         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2106         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2107         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2108         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2109         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2110 #ifdef CONFIG_X86_64
2111         regs->r8 = vcpu->regs[VCPU_REGS_R8];
2112         regs->r9 = vcpu->regs[VCPU_REGS_R9];
2113         regs->r10 = vcpu->regs[VCPU_REGS_R10];
2114         regs->r11 = vcpu->regs[VCPU_REGS_R11];
2115         regs->r12 = vcpu->regs[VCPU_REGS_R12];
2116         regs->r13 = vcpu->regs[VCPU_REGS_R13];
2117         regs->r14 = vcpu->regs[VCPU_REGS_R14];
2118         regs->r15 = vcpu->regs[VCPU_REGS_R15];
2119 #endif
2120
2121         regs->rip = vcpu->rip;
2122         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2123
2124         /*
2125          * Don't leak debug flags in case they were set for guest debugging
2126          */
2127         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2128                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2129
2130         vcpu_put(vcpu);
2131
2132         return 0;
2133 }
2134
2135 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2136                                    struct kvm_regs *regs)
2137 {
2138         vcpu_load(vcpu);
2139
2140         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2141         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2142         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2143         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2144         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2145         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2146         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2147         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2148 #ifdef CONFIG_X86_64
2149         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2150         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2151         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2152         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2153         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2154         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2155         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2156         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2157 #endif
2158
2159         vcpu->rip = regs->rip;
2160         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2161
2162         kvm_x86_ops->decache_regs(vcpu);
2163
2164         vcpu_put(vcpu);
2165
2166         return 0;
2167 }
2168
2169 static void get_segment(struct kvm_vcpu *vcpu,
2170                         struct kvm_segment *var, int seg)
2171 {
2172         return kvm_x86_ops->get_segment(vcpu, var, seg);
2173 }
2174
2175 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2176                                     struct kvm_sregs *sregs)
2177 {
2178         struct descriptor_table dt;
2179         int pending_vec;
2180
2181         vcpu_load(vcpu);
2182
2183         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2184         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2185         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2186         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2187         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2188         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2189
2190         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2191         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2192
2193         kvm_x86_ops->get_idt(vcpu, &dt);
2194         sregs->idt.limit = dt.limit;
2195         sregs->idt.base = dt.base;
2196         kvm_x86_ops->get_gdt(vcpu, &dt);
2197         sregs->gdt.limit = dt.limit;
2198         sregs->gdt.base = dt.base;
2199
2200         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2201         sregs->cr0 = vcpu->cr0;
2202         sregs->cr2 = vcpu->cr2;
2203         sregs->cr3 = vcpu->cr3;
2204         sregs->cr4 = vcpu->cr4;
2205         sregs->cr8 = get_cr8(vcpu);
2206         sregs->efer = vcpu->shadow_efer;
2207         sregs->apic_base = kvm_get_apic_base(vcpu);
2208
2209         if (irqchip_in_kernel(vcpu->kvm)) {
2210                 memset(sregs->interrupt_bitmap, 0,
2211                        sizeof sregs->interrupt_bitmap);
2212                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2213                 if (pending_vec >= 0)
2214                         set_bit(pending_vec, (unsigned long *)sregs->interrupt_bitmap);
2215         } else
2216                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2217                        sizeof sregs->interrupt_bitmap);
2218
2219         vcpu_put(vcpu);
2220
2221         return 0;
2222 }
2223
2224 static void set_segment(struct kvm_vcpu *vcpu,
2225                         struct kvm_segment *var, int seg)
2226 {
2227         return kvm_x86_ops->set_segment(vcpu, var, seg);
2228 }
2229
2230 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2231                                     struct kvm_sregs *sregs)
2232 {
2233         int mmu_reset_needed = 0;
2234         int i, pending_vec, max_bits;
2235         struct descriptor_table dt;
2236
2237         vcpu_load(vcpu);
2238
2239         dt.limit = sregs->idt.limit;
2240         dt.base = sregs->idt.base;
2241         kvm_x86_ops->set_idt(vcpu, &dt);
2242         dt.limit = sregs->gdt.limit;
2243         dt.base = sregs->gdt.base;
2244         kvm_x86_ops->set_gdt(vcpu, &dt);
2245
2246         vcpu->cr2 = sregs->cr2;
2247         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2248         vcpu->cr3 = sregs->cr3;
2249
2250         set_cr8(vcpu, sregs->cr8);
2251
2252         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2253 #ifdef CONFIG_X86_64
2254         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2255 #endif
2256         kvm_set_apic_base(vcpu, sregs->apic_base);
2257
2258         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2259
2260         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2261         vcpu->cr0 = sregs->cr0;
2262         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2263
2264         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2265         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2266         if (!is_long_mode(vcpu) && is_pae(vcpu))
2267                 load_pdptrs(vcpu, vcpu->cr3);
2268
2269         if (mmu_reset_needed)
2270                 kvm_mmu_reset_context(vcpu);
2271
2272         if (!irqchip_in_kernel(vcpu->kvm)) {
2273                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2274                        sizeof vcpu->irq_pending);
2275                 vcpu->irq_summary = 0;
2276                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2277                         if (vcpu->irq_pending[i])
2278                                 __set_bit(i, &vcpu->irq_summary);
2279         } else {
2280                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2281                 pending_vec = find_first_bit(
2282                         (const unsigned long *)sregs->interrupt_bitmap,
2283                         max_bits);
2284                 /* Only pending external irq is handled here */
2285                 if (pending_vec < max_bits) {
2286                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2287                         printk("Set back pending irq %d\n", pending_vec);
2288                 }
2289         }
2290
2291         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2292         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2293         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2294         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2295         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2296         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2297
2298         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2299         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2300
2301         vcpu_put(vcpu);
2302
2303         return 0;
2304 }
2305
2306 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2307 {
2308         struct kvm_segment cs;
2309
2310         get_segment(vcpu, &cs, VCPU_SREG_CS);
2311         *db = cs.db;
2312         *l = cs.l;
2313 }
2314 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2315
2316 /*
2317  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2318  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2319  *
2320  * This list is modified at module load time to reflect the
2321  * capabilities of the host cpu.
2322  */
2323 static u32 msrs_to_save[] = {
2324         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2325         MSR_K6_STAR,
2326 #ifdef CONFIG_X86_64
2327         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2328 #endif
2329         MSR_IA32_TIME_STAMP_COUNTER,
2330 };
2331
2332 static unsigned num_msrs_to_save;
2333
2334 static u32 emulated_msrs[] = {
2335         MSR_IA32_MISC_ENABLE,
2336 };
2337
2338 static __init void kvm_init_msr_list(void)
2339 {
2340         u32 dummy[2];
2341         unsigned i, j;
2342
2343         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2344                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2345                         continue;
2346                 if (j < i)
2347                         msrs_to_save[j] = msrs_to_save[i];
2348                 j++;
2349         }
2350         num_msrs_to_save = j;
2351 }
2352
2353 /*
2354  * Adapt set_msr() to msr_io()'s calling convention
2355  */
2356 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2357 {
2358         return kvm_set_msr(vcpu, index, *data);
2359 }
2360
2361 /*
2362  * Read or write a bunch of msrs. All parameters are kernel addresses.
2363  *
2364  * @return number of msrs set successfully.
2365  */
2366 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2367                     struct kvm_msr_entry *entries,
2368                     int (*do_msr)(struct kvm_vcpu *vcpu,
2369                                   unsigned index, u64 *data))
2370 {
2371         int i;
2372
2373         vcpu_load(vcpu);
2374
2375         for (i = 0; i < msrs->nmsrs; ++i)
2376                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2377                         break;
2378
2379         vcpu_put(vcpu);
2380
2381         return i;
2382 }
2383
2384 /*
2385  * Read or write a bunch of msrs. Parameters are user addresses.
2386  *
2387  * @return number of msrs set successfully.
2388  */
2389 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2390                   int (*do_msr)(struct kvm_vcpu *vcpu,
2391                                 unsigned index, u64 *data),
2392                   int writeback)
2393 {
2394         struct kvm_msrs msrs;
2395         struct kvm_msr_entry *entries;
2396         int r, n;
2397         unsigned size;
2398
2399         r = -EFAULT;
2400         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2401                 goto out;
2402
2403         r = -E2BIG;
2404         if (msrs.nmsrs >= MAX_IO_MSRS)
2405                 goto out;
2406
2407         r = -ENOMEM;
2408         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2409         entries = vmalloc(size);
2410         if (!entries)
2411                 goto out;
2412
2413         r = -EFAULT;
2414         if (copy_from_user(entries, user_msrs->entries, size))
2415                 goto out_free;
2416
2417         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2418         if (r < 0)
2419                 goto out_free;
2420
2421         r = -EFAULT;
2422         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2423                 goto out_free;
2424
2425         r = n;
2426
2427 out_free:
2428         vfree(entries);
2429 out:
2430         return r;
2431 }
2432
2433 /*
2434  * Translate a guest virtual address to a guest physical address.
2435  */
2436 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2437                                     struct kvm_translation *tr)
2438 {
2439         unsigned long vaddr = tr->linear_address;
2440         gpa_t gpa;
2441
2442         vcpu_load(vcpu);
2443         mutex_lock(&vcpu->kvm->lock);
2444         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2445         tr->physical_address = gpa;
2446         tr->valid = gpa != UNMAPPED_GVA;
2447         tr->writeable = 1;
2448         tr->usermode = 0;
2449         mutex_unlock(&vcpu->kvm->lock);
2450         vcpu_put(vcpu);
2451
2452         return 0;
2453 }
2454
2455 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2456                                     struct kvm_interrupt *irq)
2457 {
2458         if (irq->irq < 0 || irq->irq >= 256)
2459                 return -EINVAL;
2460         if (irqchip_in_kernel(vcpu->kvm))
2461                 return -ENXIO;
2462         vcpu_load(vcpu);
2463
2464         set_bit(irq->irq, vcpu->irq_pending);
2465         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2466
2467         vcpu_put(vcpu);
2468
2469         return 0;
2470 }
2471
2472 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2473                                       struct kvm_debug_guest *dbg)
2474 {
2475         int r;
2476
2477         vcpu_load(vcpu);
2478
2479         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2480
2481         vcpu_put(vcpu);
2482
2483         return r;
2484 }
2485
2486 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2487                                     unsigned long address,
2488                                     int *type)
2489 {
2490         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2491         unsigned long pgoff;
2492         struct page *page;
2493
2494         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2495         if (pgoff == 0)
2496                 page = virt_to_page(vcpu->run);
2497         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2498                 page = virt_to_page(vcpu->pio_data);
2499         else
2500                 return NOPAGE_SIGBUS;
2501         get_page(page);
2502         if (type != NULL)
2503                 *type = VM_FAULT_MINOR;
2504
2505         return page;
2506 }
2507
2508 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2509         .nopage = kvm_vcpu_nopage,
2510 };
2511
2512 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2513 {
2514         vma->vm_ops = &kvm_vcpu_vm_ops;
2515         return 0;
2516 }
2517
2518 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2519 {
2520         struct kvm_vcpu *vcpu = filp->private_data;
2521
2522         fput(vcpu->kvm->filp);
2523         return 0;
2524 }
2525
2526 static struct file_operations kvm_vcpu_fops = {
2527         .release        = kvm_vcpu_release,
2528         .unlocked_ioctl = kvm_vcpu_ioctl,
2529         .compat_ioctl   = kvm_vcpu_ioctl,
2530         .mmap           = kvm_vcpu_mmap,
2531 };
2532
2533 /*
2534  * Allocates an inode for the vcpu.
2535  */
2536 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2537 {
2538         int fd, r;
2539         struct inode *inode;
2540         struct file *file;
2541
2542         r = anon_inode_getfd(&fd, &inode, &file,
2543                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2544         if (r)
2545                 return r;
2546         atomic_inc(&vcpu->kvm->filp->f_count);
2547         return fd;
2548 }
2549
2550 /*
2551  * Creates some virtual cpus.  Good luck creating more than one.
2552  */
2553 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2554 {
2555         int r;
2556         struct kvm_vcpu *vcpu;
2557
2558         if (!valid_vcpu(n))
2559                 return -EINVAL;
2560
2561         vcpu = kvm_x86_ops->vcpu_create(kvm, n);
2562         if (IS_ERR(vcpu))
2563                 return PTR_ERR(vcpu);
2564
2565         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2566
2567         /* We do fxsave: this must be aligned. */
2568         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2569
2570         vcpu_load(vcpu);
2571         r = kvm_mmu_setup(vcpu);
2572         vcpu_put(vcpu);
2573         if (r < 0)
2574                 goto free_vcpu;
2575
2576         mutex_lock(&kvm->lock);
2577         if (kvm->vcpus[n]) {
2578                 r = -EEXIST;
2579                 mutex_unlock(&kvm->lock);
2580                 goto mmu_unload;
2581         }
2582         kvm->vcpus[n] = vcpu;
2583         mutex_unlock(&kvm->lock);
2584
2585         /* Now it's all set up, let userspace reach it */
2586         r = create_vcpu_fd(vcpu);
2587         if (r < 0)
2588                 goto unlink;
2589         return r;
2590
2591 unlink:
2592         mutex_lock(&kvm->lock);
2593         kvm->vcpus[n] = NULL;
2594         mutex_unlock(&kvm->lock);
2595
2596 mmu_unload:
2597         vcpu_load(vcpu);
2598         kvm_mmu_unload(vcpu);
2599         vcpu_put(vcpu);
2600
2601 free_vcpu:
2602         kvm_x86_ops->vcpu_free(vcpu);
2603         return r;
2604 }
2605
2606 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2607 {
2608         u64 efer;
2609         int i;
2610         struct kvm_cpuid_entry *e, *entry;
2611
2612         rdmsrl(MSR_EFER, efer);
2613         entry = NULL;
2614         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2615                 e = &vcpu->cpuid_entries[i];
2616                 if (e->function == 0x80000001) {
2617                         entry = e;
2618                         break;
2619                 }
2620         }
2621         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2622                 entry->edx &= ~(1 << 20);
2623                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2624         }
2625 }
2626
2627 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2628                                     struct kvm_cpuid *cpuid,
2629                                     struct kvm_cpuid_entry __user *entries)
2630 {
2631         int r;
2632
2633         r = -E2BIG;
2634         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2635                 goto out;
2636         r = -EFAULT;
2637         if (copy_from_user(&vcpu->cpuid_entries, entries,
2638                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2639                 goto out;
2640         vcpu->cpuid_nent = cpuid->nent;
2641         cpuid_fix_nx_cap(vcpu);
2642         return 0;
2643
2644 out:
2645         return r;
2646 }
2647
2648 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2649 {
2650         if (sigset) {
2651                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2652                 vcpu->sigset_active = 1;
2653                 vcpu->sigset = *sigset;
2654         } else
2655                 vcpu->sigset_active = 0;
2656         return 0;
2657 }
2658
2659 /*
2660  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2661  * we have asm/x86/processor.h
2662  */
2663 struct fxsave {
2664         u16     cwd;
2665         u16     swd;
2666         u16     twd;
2667         u16     fop;
2668         u64     rip;
2669         u64     rdp;
2670         u32     mxcsr;
2671         u32     mxcsr_mask;
2672         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2673 #ifdef CONFIG_X86_64
2674         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2675 #else
2676         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2677 #endif
2678 };
2679
2680 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2681 {
2682         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2683
2684         vcpu_load(vcpu);
2685
2686         memcpy(fpu->fpr, fxsave->st_space, 128);
2687         fpu->fcw = fxsave->cwd;
2688         fpu->fsw = fxsave->swd;
2689         fpu->ftwx = fxsave->twd;
2690         fpu->last_opcode = fxsave->fop;
2691         fpu->last_ip = fxsave->rip;
2692         fpu->last_dp = fxsave->rdp;
2693         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2694
2695         vcpu_put(vcpu);
2696
2697         return 0;
2698 }
2699
2700 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2701 {
2702         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2703
2704         vcpu_load(vcpu);
2705
2706         memcpy(fxsave->st_space, fpu->fpr, 128);
2707         fxsave->cwd = fpu->fcw;
2708         fxsave->swd = fpu->fsw;
2709         fxsave->twd = fpu->ftwx;
2710         fxsave->fop = fpu->last_opcode;
2711         fxsave->rip = fpu->last_ip;
2712         fxsave->rdp = fpu->last_dp;
2713         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2714
2715         vcpu_put(vcpu);
2716
2717         return 0;
2718 }
2719
2720 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2721                                     struct kvm_lapic_state *s)
2722 {
2723         vcpu_load(vcpu);
2724         memcpy(s->regs, vcpu->apic->regs, sizeof *s);
2725         vcpu_put(vcpu);
2726
2727         return 0;
2728 }
2729
2730 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2731                                     struct kvm_lapic_state *s)
2732 {
2733         vcpu_load(vcpu);
2734         memcpy(vcpu->apic->regs, s->regs, sizeof *s);
2735         kvm_apic_post_state_restore(vcpu);
2736         vcpu_put(vcpu);
2737
2738         return 0;
2739 }
2740
2741 static long kvm_vcpu_ioctl(struct file *filp,
2742                            unsigned int ioctl, unsigned long arg)
2743 {
2744         struct kvm_vcpu *vcpu = filp->private_data;
2745         void __user *argp = (void __user *)arg;
2746         int r = -EINVAL;
2747
2748         switch (ioctl) {
2749         case KVM_RUN:
2750                 r = -EINVAL;
2751                 if (arg)
2752                         goto out;
2753                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2754                 break;
2755         case KVM_GET_REGS: {
2756                 struct kvm_regs kvm_regs;
2757
2758                 memset(&kvm_regs, 0, sizeof kvm_regs);
2759                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2760                 if (r)
2761                         goto out;
2762                 r = -EFAULT;
2763                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2764                         goto out;
2765                 r = 0;
2766                 break;
2767         }
2768         case KVM_SET_REGS: {
2769                 struct kvm_regs kvm_regs;
2770
2771                 r = -EFAULT;
2772                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2773                         goto out;
2774                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2775                 if (r)
2776                         goto out;
2777                 r = 0;
2778                 break;
2779         }
2780         case KVM_GET_SREGS: {
2781                 struct kvm_sregs kvm_sregs;
2782
2783                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2784                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2785                 if (r)
2786                         goto out;
2787                 r = -EFAULT;
2788                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2789                         goto out;
2790                 r = 0;
2791                 break;
2792         }
2793         case KVM_SET_SREGS: {
2794                 struct kvm_sregs kvm_sregs;
2795
2796                 r = -EFAULT;
2797                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2798                         goto out;
2799                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2800                 if (r)
2801                         goto out;
2802                 r = 0;
2803                 break;
2804         }
2805         case KVM_TRANSLATE: {
2806                 struct kvm_translation tr;
2807
2808                 r = -EFAULT;
2809                 if (copy_from_user(&tr, argp, sizeof tr))
2810                         goto out;
2811                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2812                 if (r)
2813                         goto out;
2814                 r = -EFAULT;
2815                 if (copy_to_user(argp, &tr, sizeof tr))
2816                         goto out;
2817                 r = 0;
2818                 break;
2819         }
2820         case KVM_INTERRUPT: {
2821                 struct kvm_interrupt irq;
2822
2823                 r = -EFAULT;
2824                 if (copy_from_user(&irq, argp, sizeof irq))
2825                         goto out;
2826                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2827                 if (r)
2828                         goto out;
2829                 r = 0;
2830                 break;
2831         }
2832         case KVM_DEBUG_GUEST: {
2833                 struct kvm_debug_guest dbg;
2834
2835                 r = -EFAULT;
2836                 if (copy_from_user(&dbg, argp, sizeof dbg))
2837                         goto out;
2838                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2839                 if (r)
2840                         goto out;
2841                 r = 0;
2842                 break;
2843         }
2844         case KVM_GET_MSRS:
2845                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2846                 break;
2847         case KVM_SET_MSRS:
2848                 r = msr_io(vcpu, argp, do_set_msr, 0);
2849                 break;
2850         case KVM_SET_CPUID: {
2851                 struct kvm_cpuid __user *cpuid_arg = argp;
2852                 struct kvm_cpuid cpuid;
2853
2854                 r = -EFAULT;
2855                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2856                         goto out;
2857                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2858                 if (r)
2859                         goto out;
2860                 break;
2861         }
2862         case KVM_SET_SIGNAL_MASK: {
2863                 struct kvm_signal_mask __user *sigmask_arg = argp;
2864                 struct kvm_signal_mask kvm_sigmask;
2865                 sigset_t sigset, *p;
2866
2867                 p = NULL;
2868                 if (argp) {
2869                         r = -EFAULT;
2870                         if (copy_from_user(&kvm_sigmask, argp,
2871                                            sizeof kvm_sigmask))
2872                                 goto out;
2873                         r = -EINVAL;
2874                         if (kvm_sigmask.len != sizeof sigset)
2875                                 goto out;
2876                         r = -EFAULT;
2877                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2878                                            sizeof sigset))
2879                                 goto out;
2880                         p = &sigset;
2881                 }
2882                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2883                 break;
2884         }
2885         case KVM_GET_FPU: {
2886                 struct kvm_fpu fpu;
2887
2888                 memset(&fpu, 0, sizeof fpu);
2889                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2890                 if (r)
2891                         goto out;
2892                 r = -EFAULT;
2893                 if (copy_to_user(argp, &fpu, sizeof fpu))
2894                         goto out;
2895                 r = 0;
2896                 break;
2897         }
2898         case KVM_SET_FPU: {
2899                 struct kvm_fpu fpu;
2900
2901                 r = -EFAULT;
2902                 if (copy_from_user(&fpu, argp, sizeof fpu))
2903                         goto out;
2904                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2905                 if (r)
2906                         goto out;
2907                 r = 0;
2908                 break;
2909         }
2910         case KVM_GET_LAPIC: {
2911                 struct kvm_lapic_state lapic;
2912
2913                 memset(&lapic, 0, sizeof lapic);
2914                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
2915                 if (r)
2916                         goto out;
2917                 r = -EFAULT;
2918                 if (copy_to_user(argp, &lapic, sizeof lapic))
2919                         goto out;
2920                 r = 0;
2921                 break;
2922         }
2923         case KVM_SET_LAPIC: {
2924                 struct kvm_lapic_state lapic;
2925
2926                 r = -EFAULT;
2927                 if (copy_from_user(&lapic, argp, sizeof lapic))
2928                         goto out;
2929                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
2930                 if (r)
2931                         goto out;
2932                 r = 0;
2933                 break;
2934         }
2935         default:
2936                 ;
2937         }
2938 out:
2939         return r;
2940 }
2941
2942 static long kvm_vm_ioctl(struct file *filp,
2943                            unsigned int ioctl, unsigned long arg)
2944 {
2945         struct kvm *kvm = filp->private_data;
2946         void __user *argp = (void __user *)arg;
2947         int r = -EINVAL;
2948
2949         switch (ioctl) {
2950         case KVM_CREATE_VCPU:
2951                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2952                 if (r < 0)
2953                         goto out;
2954                 break;
2955         case KVM_SET_MEMORY_REGION: {
2956                 struct kvm_memory_region kvm_mem;
2957
2958                 r = -EFAULT;
2959                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2960                         goto out;
2961                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2962                 if (r)
2963                         goto out;
2964                 break;
2965         }
2966         case KVM_GET_DIRTY_LOG: {
2967                 struct kvm_dirty_log log;
2968
2969                 r = -EFAULT;
2970                 if (copy_from_user(&log, argp, sizeof log))
2971                         goto out;
2972                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2973                 if (r)
2974                         goto out;
2975                 break;
2976         }
2977         case KVM_SET_MEMORY_ALIAS: {
2978                 struct kvm_memory_alias alias;
2979
2980                 r = -EFAULT;
2981                 if (copy_from_user(&alias, argp, sizeof alias))
2982                         goto out;
2983                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2984                 if (r)
2985                         goto out;
2986                 break;
2987         }
2988         case KVM_CREATE_IRQCHIP:
2989                 r = -ENOMEM;
2990                 kvm->vpic = kvm_create_pic(kvm);
2991                 if (kvm->vpic) {
2992                         r = kvm_ioapic_init(kvm);
2993                         if (r) {
2994                                 kfree(kvm->vpic);
2995                                 kvm->vpic = NULL;
2996                                 goto out;
2997                         }
2998                 }
2999                 else
3000                         goto out;
3001                 break;
3002         case KVM_IRQ_LINE: {
3003                 struct kvm_irq_level irq_event;
3004
3005                 r = -EFAULT;
3006                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3007                         goto out;
3008                 if (irqchip_in_kernel(kvm)) {
3009                         mutex_lock(&kvm->lock);
3010                         if (irq_event.irq < 16)
3011                                 kvm_pic_set_irq(pic_irqchip(kvm),
3012                                         irq_event.irq,
3013                                         irq_event.level);
3014                         kvm_ioapic_set_irq(kvm->vioapic,
3015                                         irq_event.irq,
3016                                         irq_event.level);
3017                         mutex_unlock(&kvm->lock);
3018                         r = 0;
3019                 }
3020                 break;
3021         }
3022         case KVM_GET_IRQCHIP: {
3023                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3024                 struct kvm_irqchip chip;
3025
3026                 r = -EFAULT;
3027                 if (copy_from_user(&chip, argp, sizeof chip))
3028                         goto out;
3029                 r = -ENXIO;
3030                 if (!irqchip_in_kernel(kvm))
3031                         goto out;
3032                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
3033                 if (r)
3034                         goto out;
3035                 r = -EFAULT;
3036                 if (copy_to_user(argp, &chip, sizeof chip))
3037                         goto out;
3038                 r = 0;
3039                 break;
3040         }
3041         case KVM_SET_IRQCHIP: {
3042                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3043                 struct kvm_irqchip chip;
3044
3045                 r = -EFAULT;
3046                 if (copy_from_user(&chip, argp, sizeof chip))
3047                         goto out;
3048                 r = -ENXIO;
3049                 if (!irqchip_in_kernel(kvm))
3050                         goto out;
3051                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
3052                 if (r)
3053                         goto out;
3054                 r = 0;
3055                 break;
3056         }
3057         default:
3058                 ;
3059         }
3060 out:
3061         return r;
3062 }
3063
3064 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
3065                                   unsigned long address,
3066                                   int *type)
3067 {
3068         struct kvm *kvm = vma->vm_file->private_data;
3069         unsigned long pgoff;
3070         struct page *page;
3071
3072         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
3073         page = gfn_to_page(kvm, pgoff);
3074         if (!page)
3075                 return NOPAGE_SIGBUS;
3076         get_page(page);
3077         if (type != NULL)
3078                 *type = VM_FAULT_MINOR;
3079
3080         return page;
3081 }
3082
3083 static struct vm_operations_struct kvm_vm_vm_ops = {
3084         .nopage = kvm_vm_nopage,
3085 };
3086
3087 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
3088 {
3089         vma->vm_ops = &kvm_vm_vm_ops;
3090         return 0;
3091 }
3092
3093 static struct file_operations kvm_vm_fops = {
3094         .release        = kvm_vm_release,
3095         .unlocked_ioctl = kvm_vm_ioctl,
3096         .compat_ioctl   = kvm_vm_ioctl,
3097         .mmap           = kvm_vm_mmap,
3098 };
3099
3100 static int kvm_dev_ioctl_create_vm(void)
3101 {
3102         int fd, r;
3103         struct inode *inode;
3104         struct file *file;
3105         struct kvm *kvm;
3106
3107         kvm = kvm_create_vm();
3108         if (IS_ERR(kvm))
3109                 return PTR_ERR(kvm);
3110         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
3111         if (r) {
3112                 kvm_destroy_vm(kvm);
3113                 return r;
3114         }
3115
3116         kvm->filp = file;
3117
3118         return fd;
3119 }
3120
3121 static long kvm_dev_ioctl(struct file *filp,
3122                           unsigned int ioctl, unsigned long arg)
3123 {
3124         void __user *argp = (void __user *)arg;
3125         long r = -EINVAL;
3126
3127         switch (ioctl) {
3128         case KVM_GET_API_VERSION:
3129                 r = -EINVAL;
3130                 if (arg)
3131                         goto out;
3132                 r = KVM_API_VERSION;
3133                 break;
3134         case KVM_CREATE_VM:
3135                 r = -EINVAL;
3136                 if (arg)
3137                         goto out;
3138                 r = kvm_dev_ioctl_create_vm();
3139                 break;
3140         case KVM_GET_MSR_INDEX_LIST: {
3141                 struct kvm_msr_list __user *user_msr_list = argp;
3142                 struct kvm_msr_list msr_list;
3143                 unsigned n;
3144
3145                 r = -EFAULT;
3146                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
3147                         goto out;
3148                 n = msr_list.nmsrs;
3149                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
3150                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
3151                         goto out;
3152                 r = -E2BIG;
3153                 if (n < num_msrs_to_save)
3154                         goto out;
3155                 r = -EFAULT;
3156                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3157                                  num_msrs_to_save * sizeof(u32)))
3158                         goto out;
3159                 if (copy_to_user(user_msr_list->indices
3160                                  + num_msrs_to_save * sizeof(u32),
3161                                  &emulated_msrs,
3162                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
3163                         goto out;
3164                 r = 0;
3165                 break;
3166         }
3167         case KVM_CHECK_EXTENSION: {
3168                 int ext = (long)argp;
3169
3170                 switch (ext) {
3171                 case KVM_CAP_IRQCHIP:
3172                 case KVM_CAP_HLT:
3173                         r = 1;
3174                         break;
3175                 default:
3176                         r = 0;
3177                         break;
3178                 }
3179                 break;
3180         }
3181         case KVM_GET_VCPU_MMAP_SIZE:
3182                 r = -EINVAL;
3183                 if (arg)
3184                         goto out;
3185                 r = 2 * PAGE_SIZE;
3186                 break;
3187         default:
3188                 ;
3189         }
3190 out:
3191         return r;
3192 }
3193
3194 static struct file_operations kvm_chardev_ops = {
3195         .unlocked_ioctl = kvm_dev_ioctl,
3196         .compat_ioctl   = kvm_dev_ioctl,
3197 };
3198
3199 static struct miscdevice kvm_dev = {
3200         KVM_MINOR,
3201         "kvm",
3202         &kvm_chardev_ops,
3203 };
3204
3205 /*
3206  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
3207  * cached on it.
3208  */
3209 static void decache_vcpus_on_cpu(int cpu)
3210 {
3211         struct kvm *vm;
3212         struct kvm_vcpu *vcpu;
3213         int i;
3214
3215         spin_lock(&kvm_lock);
3216         list_for_each_entry(vm, &vm_list, vm_list)
3217                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3218                         vcpu = vm->vcpus[i];
3219                         if (!vcpu)
3220                                 continue;
3221                         /*
3222                          * If the vcpu is locked, then it is running on some
3223                          * other cpu and therefore it is not cached on the
3224                          * cpu in question.
3225                          *
3226                          * If it's not locked, check the last cpu it executed
3227                          * on.
3228                          */
3229                         if (mutex_trylock(&vcpu->mutex)) {
3230                                 if (vcpu->cpu == cpu) {
3231                                         kvm_x86_ops->vcpu_decache(vcpu);
3232                                         vcpu->cpu = -1;
3233                                 }
3234                                 mutex_unlock(&vcpu->mutex);
3235                         }
3236                 }
3237         spin_unlock(&kvm_lock);
3238 }
3239
3240 static void hardware_enable(void *junk)
3241 {
3242         int cpu = raw_smp_processor_id();
3243
3244         if (cpu_isset(cpu, cpus_hardware_enabled))
3245                 return;
3246         cpu_set(cpu, cpus_hardware_enabled);
3247         kvm_x86_ops->hardware_enable(NULL);
3248 }
3249
3250 static void hardware_disable(void *junk)
3251 {
3252         int cpu = raw_smp_processor_id();
3253
3254         if (!cpu_isset(cpu, cpus_hardware_enabled))
3255                 return;
3256         cpu_clear(cpu, cpus_hardware_enabled);
3257         decache_vcpus_on_cpu(cpu);
3258         kvm_x86_ops->hardware_disable(NULL);
3259 }
3260
3261 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3262                            void *v)
3263 {
3264         int cpu = (long)v;
3265
3266         switch (val) {
3267         case CPU_DYING:
3268         case CPU_DYING_FROZEN:
3269                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3270                        cpu);
3271                 hardware_disable(NULL);
3272                 break;
3273         case CPU_UP_CANCELED:
3274         case CPU_UP_CANCELED_FROZEN:
3275                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3276                        cpu);
3277                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3278                 break;
3279         case CPU_ONLINE:
3280         case CPU_ONLINE_FROZEN:
3281                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3282                        cpu);
3283                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3284                 break;
3285         }
3286         return NOTIFY_OK;
3287 }
3288
3289 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3290                        void *v)
3291 {
3292         if (val == SYS_RESTART) {
3293                 /*
3294                  * Some (well, at least mine) BIOSes hang on reboot if
3295                  * in vmx root mode.
3296                  */
3297                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3298                 on_each_cpu(hardware_disable, NULL, 0, 1);
3299         }
3300         return NOTIFY_OK;
3301 }
3302
3303 static struct notifier_block kvm_reboot_notifier = {
3304         .notifier_call = kvm_reboot,
3305         .priority = 0,
3306 };
3307
3308 void kvm_io_bus_init(struct kvm_io_bus *bus)
3309 {
3310         memset(bus, 0, sizeof(*bus));
3311 }
3312
3313 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3314 {
3315         int i;
3316
3317         for (i = 0; i < bus->dev_count; i++) {
3318                 struct kvm_io_device *pos = bus->devs[i];
3319
3320                 kvm_iodevice_destructor(pos);
3321         }
3322 }
3323
3324 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3325 {
3326         int i;
3327
3328         for (i = 0; i < bus->dev_count; i++) {
3329                 struct kvm_io_device *pos = bus->devs[i];
3330
3331                 if (pos->in_range(pos, addr))
3332                         return pos;
3333         }
3334
3335         return NULL;
3336 }
3337
3338 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3339 {
3340         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3341
3342         bus->devs[bus->dev_count++] = dev;
3343 }
3344
3345 static struct notifier_block kvm_cpu_notifier = {
3346         .notifier_call = kvm_cpu_hotplug,
3347         .priority = 20, /* must be > scheduler priority */
3348 };
3349
3350 static u64 stat_get(void *_offset)
3351 {
3352         unsigned offset = (long)_offset;
3353         u64 total = 0;
3354         struct kvm *kvm;
3355         struct kvm_vcpu *vcpu;
3356         int i;
3357
3358         spin_lock(&kvm_lock);
3359         list_for_each_entry(kvm, &vm_list, vm_list)
3360                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3361                         vcpu = kvm->vcpus[i];
3362                         if (vcpu)
3363                                 total += *(u32 *)((void *)vcpu + offset);
3364                 }
3365         spin_unlock(&kvm_lock);
3366         return total;
3367 }
3368
3369 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3370
3371 static __init void kvm_init_debug(void)
3372 {
3373         struct kvm_stats_debugfs_item *p;
3374
3375         debugfs_dir = debugfs_create_dir("kvm", NULL);
3376         for (p = debugfs_entries; p->name; ++p)
3377                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3378                                                 (void *)(long)p->offset,
3379                                                 &stat_fops);
3380 }
3381
3382 static void kvm_exit_debug(void)
3383 {
3384         struct kvm_stats_debugfs_item *p;
3385
3386         for (p = debugfs_entries; p->name; ++p)
3387                 debugfs_remove(p->dentry);
3388         debugfs_remove(debugfs_dir);
3389 }
3390
3391 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3392 {
3393         hardware_disable(NULL);
3394         return 0;
3395 }
3396
3397 static int kvm_resume(struct sys_device *dev)
3398 {
3399         hardware_enable(NULL);
3400         return 0;
3401 }
3402
3403 static struct sysdev_class kvm_sysdev_class = {
3404         .name = "kvm",
3405         .suspend = kvm_suspend,
3406         .resume = kvm_resume,
3407 };
3408
3409 static struct sys_device kvm_sysdev = {
3410         .id = 0,
3411         .cls = &kvm_sysdev_class,
3412 };
3413
3414 hpa_t bad_page_address;
3415
3416 static inline
3417 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3418 {
3419         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3420 }
3421
3422 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3423 {
3424         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3425
3426         kvm_x86_ops->vcpu_load(vcpu, cpu);
3427 }
3428
3429 static void kvm_sched_out(struct preempt_notifier *pn,
3430                           struct task_struct *next)
3431 {
3432         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3433
3434         kvm_x86_ops->vcpu_put(vcpu);
3435 }
3436
3437 int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
3438                   struct module *module)
3439 {
3440         int r;
3441         int cpu;
3442
3443         if (kvm_x86_ops) {
3444                 printk(KERN_ERR "kvm: already loaded the other module\n");
3445                 return -EEXIST;
3446         }
3447
3448         if (!ops->cpu_has_kvm_support()) {
3449                 printk(KERN_ERR "kvm: no hardware support\n");
3450                 return -EOPNOTSUPP;
3451         }
3452         if (ops->disabled_by_bios()) {
3453                 printk(KERN_ERR "kvm: disabled by bios\n");
3454                 return -EOPNOTSUPP;
3455         }
3456
3457         kvm_x86_ops = ops;
3458
3459         r = kvm_x86_ops->hardware_setup();
3460         if (r < 0)
3461                 goto out;
3462
3463         for_each_online_cpu(cpu) {
3464                 smp_call_function_single(cpu,
3465                                 kvm_x86_ops->check_processor_compatibility,
3466                                 &r, 0, 1);
3467                 if (r < 0)
3468                         goto out_free_0;
3469         }
3470
3471         on_each_cpu(hardware_enable, NULL, 0, 1);
3472         r = register_cpu_notifier(&kvm_cpu_notifier);
3473         if (r)
3474                 goto out_free_1;
3475         register_reboot_notifier(&kvm_reboot_notifier);
3476
3477         r = sysdev_class_register(&kvm_sysdev_class);
3478         if (r)
3479                 goto out_free_2;
3480
3481         r = sysdev_register(&kvm_sysdev);
3482         if (r)
3483                 goto out_free_3;
3484
3485         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3486         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3487                                            __alignof__(struct kvm_vcpu), 0, 0);
3488         if (!kvm_vcpu_cache) {
3489                 r = -ENOMEM;
3490                 goto out_free_4;
3491         }
3492
3493         kvm_chardev_ops.owner = module;
3494
3495         r = misc_register(&kvm_dev);
3496         if (r) {
3497                 printk (KERN_ERR "kvm: misc device register failed\n");
3498                 goto out_free;
3499         }
3500
3501         kvm_preempt_ops.sched_in = kvm_sched_in;
3502         kvm_preempt_ops.sched_out = kvm_sched_out;
3503
3504         return r;
3505
3506 out_free:
3507         kmem_cache_destroy(kvm_vcpu_cache);
3508 out_free_4:
3509         sysdev_unregister(&kvm_sysdev);
3510 out_free_3:
3511         sysdev_class_unregister(&kvm_sysdev_class);
3512 out_free_2:
3513         unregister_reboot_notifier(&kvm_reboot_notifier);
3514         unregister_cpu_notifier(&kvm_cpu_notifier);
3515 out_free_1:
3516         on_each_cpu(hardware_disable, NULL, 0, 1);
3517 out_free_0:
3518         kvm_x86_ops->hardware_unsetup();
3519 out:
3520         kvm_x86_ops = NULL;
3521         return r;
3522 }
3523
3524 void kvm_exit_x86(void)
3525 {
3526         misc_deregister(&kvm_dev);
3527         kmem_cache_destroy(kvm_vcpu_cache);
3528         sysdev_unregister(&kvm_sysdev);
3529         sysdev_class_unregister(&kvm_sysdev_class);
3530         unregister_reboot_notifier(&kvm_reboot_notifier);
3531         unregister_cpu_notifier(&kvm_cpu_notifier);
3532         on_each_cpu(hardware_disable, NULL, 0, 1);
3533         kvm_x86_ops->hardware_unsetup();
3534         kvm_x86_ops = NULL;
3535 }
3536
3537 static __init int kvm_init(void)
3538 {
3539         static struct page *bad_page;
3540         int r;
3541
3542         r = kvm_mmu_module_init();
3543         if (r)
3544                 goto out4;
3545
3546         kvm_init_debug();
3547
3548         kvm_init_msr_list();
3549
3550         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3551                 r = -ENOMEM;
3552                 goto out;
3553         }
3554
3555         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3556         memset(__va(bad_page_address), 0, PAGE_SIZE);
3557
3558         return 0;
3559
3560 out:
3561         kvm_exit_debug();
3562         kvm_mmu_module_exit();
3563 out4:
3564         return r;
3565 }
3566
3567 static __exit void kvm_exit(void)
3568 {
3569         kvm_exit_debug();
3570         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3571         kvm_mmu_module_exit();
3572 }
3573
3574 module_init(kvm_init)
3575 module_exit(kvm_exit)
3576
3577 EXPORT_SYMBOL_GPL(kvm_init_x86);
3578 EXPORT_SYMBOL_GPL(kvm_exit_x86);