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
  43 #include <asm/processor.h>
  44 #include <asm/msr.h>
  45 #include <asm/io.h>
  46 #include <asm/uaccess.h>
  47 #include <asm/desc.h>
  48
  49 MODULE_AUTHOR("Qumranet");
  50 MODULE_LICENSE("GPL");
  51
  52 static DEFINE_SPINLOCK(kvm_lock);
  53 static LIST_HEAD(vm_list);
  54
  55 static cpumask_t cpus_hardware_enabled;
  56
  57 struct kvm_x86_ops *kvm_x86_ops;
  58 struct kmem_cache *kvm_vcpu_cache;
  59 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  60
  61 static __read_mostly struct preempt_ops kvm_preempt_ops;
  62
  63 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  64
  65 static struct kvm_stats_debugfs_item {
  66         const char *name;
  67         int offset;
  68         struct dentry *dentry;
  69 } debugfs_entries[] = {
  70         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  71         { "pf_guest", STAT_OFFSET(pf_guest) },
  72         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  73         { "invlpg", STAT_OFFSET(invlpg) },
  74         { "exits", STAT_OFFSET(exits) },
  75         { "io_exits", STAT_OFFSET(io_exits) },
  76         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  77         { "signal_exits", STAT_OFFSET(signal_exits) },
  78         { "irq_window", STAT_OFFSET(irq_window_exits) },
  79         { "halt_exits", STAT_OFFSET(halt_exits) },
  80         { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
  81         { "request_irq", STAT_OFFSET(request_irq_exits) },
  82         { "irq_exits", STAT_OFFSET(irq_exits) },
  83         { "light_exits", STAT_OFFSET(light_exits) },
  84         { "efer_reload", STAT_OFFSET(efer_reload) },
  85         { NULL }
  86 };
  87
  88 static struct dentry *debugfs_dir;
  89
  90 #define MAX_IO_MSRS 256
  91
  92 #define CR0_RESERVED_BITS                                               \
  93         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  94                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  95                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  96 #define CR4_RESERVED_BITS                                               \
  97         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  98                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  99                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
 100                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
 101
 102 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
 103 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
 104
 105 #ifdef CONFIG_X86_64
 106 // LDT or TSS descriptor in the GDT. 16 bytes.
 107 struct segment_descriptor_64 {
 108         struct segment_descriptor s;
 109         u32 base_higher;
 110         u32 pad_zero;
 111 };
 112
 113 #endif
 114
 115 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 116                            unsigned long arg);
 117
 118 unsigned long segment_base(u16 selector)
 119 {
 120         struct descriptor_table gdt;
 121         struct segment_descriptor *d;
 122         unsigned long table_base;
 123         typedef unsigned long ul;
 124         unsigned long v;
 125
 126         if (selector == 0)
 127                 return 0;
 128
 129         asm ("sgdt %0" : "=m"(gdt));
 130         table_base = gdt.base;
 131
 132         if (selector & 4) {           /* from ldt */
 133                 u16 ldt_selector;
 134
 135                 asm ("sldt %0" : "=g"(ldt_selector));
 136                 table_base = segment_base(ldt_selector);
 137         }
 138         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 139         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 140 #ifdef CONFIG_X86_64
 141         if (d->system == 0
 142             && (d->type == 2 || d->type == 9 || d->type == 11))
 143                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 144 #endif
 145         return v;
 146 }
 147 EXPORT_SYMBOL_GPL(segment_base);
 148
 149 static inline int valid_vcpu(int n)
 150 {
 151         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 152 }
 153
 154 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 155 {
 156         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
 157                 return;
 158
 159         vcpu->guest_fpu_loaded = 1;
 160         fx_save(&vcpu->host_fx_image);
 161         fx_restore(&vcpu->guest_fx_image);
 162 }
 163 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
 164
 165 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 166 {
 167         if (!vcpu->guest_fpu_loaded)
 168                 return;
 169
 170         vcpu->guest_fpu_loaded = 0;
 171         fx_save(&vcpu->guest_fx_image);
 172         fx_restore(&vcpu->host_fx_image);
 173 }
 174 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
 175
 176 /*
 177  * Switches to specified vcpu, until a matching vcpu_put()
 178  */
 179 static void vcpu_load(struct kvm_vcpu *vcpu)
 180 {
 181         int cpu;
 182
 183         mutex_lock(&vcpu->mutex);
 184         cpu = get_cpu();
 185         preempt_notifier_register(&vcpu->preempt_notifier);
 186         kvm_x86_ops->vcpu_load(vcpu, cpu);
 187         put_cpu();
 188 }
 189
 190 static void vcpu_put(struct kvm_vcpu *vcpu)
 191 {
 192         preempt_disable();
 193         kvm_x86_ops->vcpu_put(vcpu);
 194         preempt_notifier_unregister(&vcpu->preempt_notifier);
 195         preempt_enable();
 196         mutex_unlock(&vcpu->mutex);
 197 }
 198
 199 static void ack_flush(void *_completed)
 200 {
 201 }
 202
 203 void kvm_flush_remote_tlbs(struct kvm *kvm)
 204 {
 205         int i, cpu;
 206         cpumask_t cpus;
 207         struct kvm_vcpu *vcpu;
 208
 209         cpus_clear(cpus);
 210         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 211                 vcpu = kvm->vcpus[i];
 212                 if (!vcpu)
 213                         continue;
 214                 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
 215                         continue;
 216                 cpu = vcpu->cpu;
 217                 if (cpu != -1 && cpu != raw_smp_processor_id())
 218                         cpu_set(cpu, cpus);
 219         }
 220         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 221 }
 222
 223 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 224 {
 225         struct page *page;
 226         int r;
 227
 228         mutex_init(&vcpu->mutex);
 229         vcpu->cpu = -1;
 230         vcpu->mmu.root_hpa = INVALID_PAGE;
 231         vcpu->kvm = kvm;
 232         vcpu->vcpu_id = id;
 233         if (!irqchip_in_kernel(kvm) || id == 0)
 234                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
 235         else
 236                 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
 237         init_waitqueue_head(&vcpu->wq);
 238
 239         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 240         if (!page) {
 241                 r = -ENOMEM;
 242                 goto fail;
 243         }
 244         vcpu->run = page_address(page);
 245
 246         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 247         if (!page) {
 248                 r = -ENOMEM;
 249                 goto fail_free_run;
 250         }
 251         vcpu->pio_data = page_address(page);
 252
 253         r = kvm_mmu_create(vcpu);
 254         if (r < 0)
 255                 goto fail_free_pio_data;
 256
 257         return 0;
 258
 259 fail_free_pio_data:
 260         free_page((unsigned long)vcpu->pio_data);
 261 fail_free_run:
 262         free_page((unsigned long)vcpu->run);
 263 fail:
 264         return -ENOMEM;
 265 }
 266 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 267
 268 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 269 {
 270         kvm_mmu_destroy(vcpu);
 271         if (vcpu->apic)
 272                 hrtimer_cancel(&vcpu->apic->timer.dev);
 273         kvm_free_apic(vcpu->apic);
 274         free_page((unsigned long)vcpu->pio_data);
 275         free_page((unsigned long)vcpu->run);
 276 }
 277 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 278
 279 static struct kvm *kvm_create_vm(void)
 280 {
 281         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 282
 283         if (!kvm)
 284                 return ERR_PTR(-ENOMEM);
 285
 286         kvm_io_bus_init(&kvm->pio_bus);
 287         mutex_init(&kvm->lock);
 288         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 289         kvm_io_bus_init(&kvm->mmio_bus);
 290         spin_lock(&kvm_lock);
 291         list_add(&kvm->vm_list, &vm_list);
 292         spin_unlock(&kvm_lock);
 293         return kvm;
 294 }
 295
 296 /*
 297  * Free any memory in @free but not in @dont.
 298  */
 299 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 300                                   struct kvm_memory_slot *dont)
 301 {
 302         int i;
 303
 304         if (!dont || free->phys_mem != dont->phys_mem)
 305                 if (free->phys_mem) {
 306                         for (i = 0; i < free->npages; ++i)
 307                                 if (free->phys_mem[i])
 308                                         __free_page(free->phys_mem[i]);
 309                         vfree(free->phys_mem);
 310                 }
 311
 312         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 313                 vfree(free->dirty_bitmap);
 314
 315         free->phys_mem = NULL;
 316         free->npages = 0;
 317         free->dirty_bitmap = NULL;
 318 }
 319
 320 static void kvm_free_physmem(struct kvm *kvm)
 321 {
 322         int i;
 323
 324         for (i = 0; i < kvm->nmemslots; ++i)
 325                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 326 }
 327
 328 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 329 {
 330         int i;
 331
 332         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
 333                 if (vcpu->pio.guest_pages[i]) {
 334                         __free_page(vcpu->pio.guest_pages[i]);
 335                         vcpu->pio.guest_pages[i] = NULL;
 336                 }
 337 }
 338
 339 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 340 {
 341         vcpu_load(vcpu);
 342         kvm_mmu_unload(vcpu);
 343         vcpu_put(vcpu);
 344 }
 345
 346 static void kvm_free_vcpus(struct kvm *kvm)
 347 {
 348         unsigned int i;
 349
 350         /*
 351          * Unpin any mmu pages first.
 352          */
 353         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 354                 if (kvm->vcpus[i])
 355                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
 356         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 357                 if (kvm->vcpus[i]) {
 358                         kvm_x86_ops->vcpu_free(kvm->vcpus[i]);
 359                         kvm->vcpus[i] = NULL;
 360                 }
 361         }
 362
 363 }
 364
 365 static void kvm_destroy_vm(struct kvm *kvm)
 366 {
 367         spin_lock(&kvm_lock);
 368         list_del(&kvm->vm_list);
 369         spin_unlock(&kvm_lock);
 370         kvm_io_bus_destroy(&kvm->pio_bus);
 371         kvm_io_bus_destroy(&kvm->mmio_bus);
 372         kfree(kvm->vpic);
 373         kfree(kvm->vioapic);
 374         kvm_free_vcpus(kvm);
 375         kvm_free_physmem(kvm);
 376         kfree(kvm);
 377 }
 378
 379 static int kvm_vm_release(struct inode *inode, struct file *filp)
 380 {
 381         struct kvm *kvm = filp->private_data;
 382
 383         kvm_destroy_vm(kvm);
 384         return 0;
 385 }
 386
 387 static void inject_gp(struct kvm_vcpu *vcpu)
 388 {
 389         kvm_x86_ops->inject_gp(vcpu, 0);
 390 }
 391
 392 /*
 393  * Load the pae pdptrs.  Return true is they are all valid.
 394  */
 395 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 396 {
 397         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 398         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 399         int i;
 400         u64 *pdpt;
 401         int ret;
 402         struct page *page;
 403         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
 404
 405         mutex_lock(&vcpu->kvm->lock);
 406         page = gfn_to_page(vcpu->kvm, pdpt_gfn);
 407         if (!page) {
 408                 ret = 0;
 409                 goto out;
 410         }
 411
 412         pdpt = kmap_atomic(page, KM_USER0);
 413         memcpy(pdpte, pdpt+offset, sizeof(pdpte));
 414         kunmap_atomic(pdpt, KM_USER0);
 415
 416         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 417                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 418                         ret = 0;
 419                         goto out;
 420                 }
 421         }
 422         ret = 1;
 423
 424         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
 425 out:
 426         mutex_unlock(&vcpu->kvm->lock);
 427
 428         return ret;
 429 }
 430
 431 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 432 {
 433         if (cr0 & CR0_RESERVED_BITS) {
 434                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 435                        cr0, vcpu->cr0);
 436                 inject_gp(vcpu);
 437                 return;
 438         }
 439
 440         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 441                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 442                 inject_gp(vcpu);
 443                 return;
 444         }
 445
 446         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 447                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 448                        "and a clear PE flag\n");
 449                 inject_gp(vcpu);
 450                 return;
 451         }
 452
 453         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 454 #ifdef CONFIG_X86_64
 455                 if ((vcpu->shadow_efer & EFER_LME)) {
 456                         int cs_db, cs_l;
 457
 458                         if (!is_pae(vcpu)) {
 459                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 460                                        "in long mode while PAE is disabled\n");
 461                                 inject_gp(vcpu);
 462                                 return;
 463                         }
 464                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 465                         if (cs_l) {
 466                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 467                                        "in long mode while CS.L == 1\n");
 468                                 inject_gp(vcpu);
 469                                 return;
 470
 471                         }
 472                 } else
 473 #endif
 474                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 475                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 476                                "reserved bits\n");
 477                         inject_gp(vcpu);
 478                         return;
 479                 }
 480
 481         }
 482
 483         kvm_x86_ops->set_cr0(vcpu, cr0);
 484         vcpu->cr0 = cr0;
 485
 486         mutex_lock(&vcpu->kvm->lock);
 487         kvm_mmu_reset_context(vcpu);
 488         mutex_unlock(&vcpu->kvm->lock);
 489         return;
 490 }
 491 EXPORT_SYMBOL_GPL(set_cr0);
 492
 493 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 494 {
 495         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 496 }
 497 EXPORT_SYMBOL_GPL(lmsw);
 498
 499 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 500 {
 501         if (cr4 & CR4_RESERVED_BITS) {
 502                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 503                 inject_gp(vcpu);
 504                 return;
 505         }
 506
 507         if (is_long_mode(vcpu)) {
 508                 if (!(cr4 & X86_CR4_PAE)) {
 509                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 510                                "in long mode\n");
 511                         inject_gp(vcpu);
 512                         return;
 513                 }
 514         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 515                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 516                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 517                 inject_gp(vcpu);
 518                 return;
 519         }
 520
 521         if (cr4 & X86_CR4_VMXE) {
 522                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 523                 inject_gp(vcpu);
 524                 return;
 525         }
 526         kvm_x86_ops->set_cr4(vcpu, cr4);
 527         vcpu->cr4 = cr4;
 528         mutex_lock(&vcpu->kvm->lock);
 529         kvm_mmu_reset_context(vcpu);
 530         mutex_unlock(&vcpu->kvm->lock);
 531 }
 532 EXPORT_SYMBOL_GPL(set_cr4);
 533
 534 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 535 {
 536         if (is_long_mode(vcpu)) {
 537                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 538                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 539                         inject_gp(vcpu);
 540                         return;
 541                 }
 542         } else {
 543                 if (is_pae(vcpu)) {
 544                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 545                                 printk(KERN_DEBUG
 546                                        "set_cr3: #GP, reserved bits\n");
 547                                 inject_gp(vcpu);
 548                                 return;
 549                         }
 550                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 551                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 552                                        "reserved bits\n");
 553                                 inject_gp(vcpu);
 554                                 return;
 555                         }
 556                 } else {
 557                         if (cr3 & CR3_NONPAE_RESERVED_BITS) {
 558                                 printk(KERN_DEBUG
 559                                        "set_cr3: #GP, reserved bits\n");
 560                                 inject_gp(vcpu);
 561                                 return;
 562                         }
 563                 }
 564         }
 565
 566         mutex_lock(&vcpu->kvm->lock);
 567         /*
 568          * Does the new cr3 value map to physical memory? (Note, we
 569          * catch an invalid cr3 even in real-mode, because it would
 570          * cause trouble later on when we turn on paging anyway.)
 571          *
 572          * A real CPU would silently accept an invalid cr3 and would
 573          * attempt to use it - with largely undefined (and often hard
 574          * to debug) behavior on the guest side.
 575          */
 576         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 577                 inject_gp(vcpu);
 578         else {
 579                 vcpu->cr3 = cr3;
 580                 vcpu->mmu.new_cr3(vcpu);
 581         }
 582         mutex_unlock(&vcpu->kvm->lock);
 583 }
 584 EXPORT_SYMBOL_GPL(set_cr3);
 585
 586 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 587 {
 588         if (cr8 & CR8_RESERVED_BITS) {
 589                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 590                 inject_gp(vcpu);
 591                 return;
 592         }
 593         if (irqchip_in_kernel(vcpu->kvm))
 594                 kvm_lapic_set_tpr(vcpu, cr8);
 595         else
 596                 vcpu->cr8 = cr8;
 597 }
 598 EXPORT_SYMBOL_GPL(set_cr8);
 599
 600 unsigned long get_cr8(struct kvm_vcpu *vcpu)
 601 {
 602         if (irqchip_in_kernel(vcpu->kvm))
 603                 return kvm_lapic_get_cr8(vcpu);
 604         else
 605                 return vcpu->cr8;
 606 }
 607 EXPORT_SYMBOL_GPL(get_cr8);
 608
 609 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 610 {
 611         if (irqchip_in_kernel(vcpu->kvm))
 612                 return vcpu->apic_base;
 613         else
 614                 return vcpu->apic_base;
 615 }
 616 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 617
 618 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 619 {
 620         /* TODO: reserve bits check */
 621         if (irqchip_in_kernel(vcpu->kvm))
 622                 kvm_lapic_set_base(vcpu, data);
 623         else
 624                 vcpu->apic_base = data;
 625 }
 626 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 627
 628 void fx_init(struct kvm_vcpu *vcpu)
 629 {
 630         unsigned after_mxcsr_mask;
 631
 632         /* Initialize guest FPU by resetting ours and saving into guest's */
 633         preempt_disable();
 634         fx_save(&vcpu->host_fx_image);
 635         fpu_init();
 636         fx_save(&vcpu->guest_fx_image);
 637         fx_restore(&vcpu->host_fx_image);
 638         preempt_enable();
 639
 640         vcpu->cr0 |= X86_CR0_ET;
 641         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
 642         vcpu->guest_fx_image.mxcsr = 0x1f80;
 643         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
 644                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
 645 }
 646 EXPORT_SYMBOL_GPL(fx_init);
 647
 648 /*
 649  * Allocate some memory and give it an address in the guest physical address
 650  * space.
 651  *
 652  * Discontiguous memory is allowed, mostly for framebuffers.
 653  */
 654 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 655                                           struct kvm_memory_region *mem)
 656 {
 657         int r;
 658         gfn_t base_gfn;
 659         unsigned long npages;
 660         unsigned long i;
 661         struct kvm_memory_slot *memslot;
 662         struct kvm_memory_slot old, new;
 663
 664         r = -EINVAL;
 665         /* General sanity checks */
 666         if (mem->memory_size & (PAGE_SIZE - 1))
 667                 goto out;
 668         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 669                 goto out;
 670         if (mem->slot >= KVM_MEMORY_SLOTS)
 671                 goto out;
 672         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 673                 goto out;
 674
 675         memslot = &kvm->memslots[mem->slot];
 676         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 677         npages = mem->memory_size >> PAGE_SHIFT;
 678
 679         if (!npages)
 680                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 681
 682         mutex_lock(&kvm->lock);
 683
 684         new = old = *memslot;
 685
 686         new.base_gfn = base_gfn;
 687         new.npages = npages;
 688         new.flags = mem->flags;
 689
 690         /* Disallow changing a memory slot's size. */
 691         r = -EINVAL;
 692         if (npages && old.npages && npages != old.npages)
 693                 goto out_unlock;
 694
 695         /* Check for overlaps */
 696         r = -EEXIST;
 697         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 698                 struct kvm_memory_slot *s = &kvm->memslots[i];
 699
 700                 if (s == memslot)
 701                         continue;
 702                 if (!((base_gfn + npages <= s->base_gfn) ||
 703                       (base_gfn >= s->base_gfn + s->npages)))
 704                         goto out_unlock;
 705         }
 706
 707         /* Deallocate if slot is being removed */
 708         if (!npages)
 709                 new.phys_mem = NULL;
 710
 711         /* Free page dirty bitmap if unneeded */
 712         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 713                 new.dirty_bitmap = NULL;
 714
 715         r = -ENOMEM;
 716
 717         /* Allocate if a slot is being created */
 718         if (npages && !new.phys_mem) {
 719                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 720
 721                 if (!new.phys_mem)
 722                         goto out_unlock;
 723
 724                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 725                 for (i = 0; i < npages; ++i) {
 726                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 727                                                      | __GFP_ZERO);
 728                         if (!new.phys_mem[i])
 729                                 goto out_unlock;
 730                         set_page_private(new.phys_mem[i],0);
 731                 }
 732         }
 733
 734         /* Allocate page dirty bitmap if needed */
 735         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 736                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 737
 738                 new.dirty_bitmap = vmalloc(dirty_bytes);
 739                 if (!new.dirty_bitmap)
 740                         goto out_unlock;
 741                 memset(new.dirty_bitmap, 0, dirty_bytes);
 742         }
 743
 744         if (mem->slot >= kvm->nmemslots)
 745                 kvm->nmemslots = mem->slot + 1;
 746
 747         *memslot = new;
 748
 749         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 750         kvm_flush_remote_tlbs(kvm);
 751
 752         mutex_unlock(&kvm->lock);
 753
 754         kvm_free_physmem_slot(&old, &new);
 755         return 0;
 756
 757 out_unlock:
 758         mutex_unlock(&kvm->lock);
 759         kvm_free_physmem_slot(&new, &old);
 760 out:
 761         return r;
 762 }
 763
 764 /*
 765  * Get (and clear) the dirty memory log for a memory slot.
 766  */
 767 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 768                                       struct kvm_dirty_log *log)
 769 {
 770         struct kvm_memory_slot *memslot;
 771         int r, i;
 772         int n;
 773         unsigned long any = 0;
 774
 775         mutex_lock(&kvm->lock);
 776
 777         r = -EINVAL;
 778         if (log->slot >= KVM_MEMORY_SLOTS)
 779                 goto out;
 780
 781         memslot = &kvm->memslots[log->slot];
 782         r = -ENOENT;
 783         if (!memslot->dirty_bitmap)
 784                 goto out;
 785
 786         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 787
 788         for (i = 0; !any && i < n/sizeof(long); ++i)
 789                 any = memslot->dirty_bitmap[i];
 790
 791         r = -EFAULT;
 792         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 793                 goto out;
 794
 795         /* If nothing is dirty, don't bother messing with page tables. */
 796         if (any) {
 797                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 798                 kvm_flush_remote_tlbs(kvm);
 799                 memset(memslot->dirty_bitmap, 0, n);
 800         }
 801
 802         r = 0;
 803
 804 out:
 805         mutex_unlock(&kvm->lock);
 806         return r;
 807 }
 808
 809 /*
 810  * Set a new alias region.  Aliases map a portion of physical memory into
 811  * another portion.  This is useful for memory windows, for example the PC
 812  * VGA region.
 813  */
 814 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 815                                          struct kvm_memory_alias *alias)
 816 {
 817         int r, n;
 818         struct kvm_mem_alias *p;
 819
 820         r = -EINVAL;
 821         /* General sanity checks */
 822         if (alias->memory_size & (PAGE_SIZE - 1))
 823                 goto out;
 824         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 825                 goto out;
 826         if (alias->slot >= KVM_ALIAS_SLOTS)
 827                 goto out;
 828         if (alias->guest_phys_addr + alias->memory_size
 829             < alias->guest_phys_addr)
 830                 goto out;
 831         if (alias->target_phys_addr + alias->memory_size
 832             < alias->target_phys_addr)
 833                 goto out;
 834
 835         mutex_lock(&kvm->lock);
 836
 837         p = &kvm->aliases[alias->slot];
 838         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 839         p->npages = alias->memory_size >> PAGE_SHIFT;
 840         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 841
 842         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 843                 if (kvm->aliases[n - 1].npages)
 844                         break;
 845         kvm->naliases = n;
 846
 847         kvm_mmu_zap_all(kvm);
 848
 849         mutex_unlock(&kvm->lock);
 850
 851         return 0;
 852
 853 out:
 854         return r;
 855 }
 856
 857 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 858 {
 859         int r;
 860
 861         r = 0;
 862         switch (chip->chip_id) {
 863         case KVM_IRQCHIP_PIC_MASTER:
 864                 memcpy (&chip->chip.pic,
 865                         &pic_irqchip(kvm)->pics[0],
 866                         sizeof(struct kvm_pic_state));
 867                 break;
 868         case KVM_IRQCHIP_PIC_SLAVE:
 869                 memcpy (&chip->chip.pic,
 870                         &pic_irqchip(kvm)->pics[1],
 871                         sizeof(struct kvm_pic_state));
 872                 break;
 873         case KVM_IRQCHIP_IOAPIC:
 874                 memcpy (&chip->chip.ioapic,
 875                         ioapic_irqchip(kvm),
 876                         sizeof(struct kvm_ioapic_state));
 877                 break;
 878         default:
 879                 r = -EINVAL;
 880                 break;
 881         }
 882         return r;
 883 }
 884
 885 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 886 {
 887         int r;
 888
 889         r = 0;
 890         switch (chip->chip_id) {
 891         case KVM_IRQCHIP_PIC_MASTER:
 892                 memcpy (&pic_irqchip(kvm)->pics[0],
 893                         &chip->chip.pic,
 894                         sizeof(struct kvm_pic_state));
 895                 break;
 896         case KVM_IRQCHIP_PIC_SLAVE:
 897                 memcpy (&pic_irqchip(kvm)->pics[1],
 898                         &chip->chip.pic,
 899                         sizeof(struct kvm_pic_state));
 900                 break;
 901         case KVM_IRQCHIP_IOAPIC:
 902                 memcpy (ioapic_irqchip(kvm),
 903                         &chip->chip.ioapic,
 904                         sizeof(struct kvm_ioapic_state));
 905                 break;
 906         default:
 907                 r = -EINVAL;
 908                 break;
 909         }
 910         kvm_pic_update_irq(pic_irqchip(kvm));
 911         return r;
 912 }
 913
 914 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 915 {
 916         int i;
 917         struct kvm_mem_alias *alias;
 918
 919         for (i = 0; i < kvm->naliases; ++i) {
 920                 alias = &kvm->aliases[i];
 921                 if (gfn >= alias->base_gfn
 922                     && gfn < alias->base_gfn + alias->npages)
 923                         return alias->target_gfn + gfn - alias->base_gfn;
 924         }
 925         return gfn;
 926 }
 927
 928 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 929 {
 930         int i;
 931
 932         for (i = 0; i < kvm->nmemslots; ++i) {
 933                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 934
 935                 if (gfn >= memslot->base_gfn
 936                     && gfn < memslot->base_gfn + memslot->npages)
 937                         return memslot;
 938         }
 939         return NULL;
 940 }
 941
 942 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 943 {
 944         gfn = unalias_gfn(kvm, gfn);
 945         return __gfn_to_memslot(kvm, gfn);
 946 }
 947
 948 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 949 {
 950         struct kvm_memory_slot *slot;
 951
 952         gfn = unalias_gfn(kvm, gfn);
 953         slot = __gfn_to_memslot(kvm, gfn);
 954         if (!slot)
 955                 return NULL;
 956         return slot->phys_mem[gfn - slot->base_gfn];
 957 }
 958 EXPORT_SYMBOL_GPL(gfn_to_page);
 959
 960 /* WARNING: Does not work on aliased pages. */
 961 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 962 {
 963         struct kvm_memory_slot *memslot;
 964
 965         memslot = __gfn_to_memslot(kvm, gfn);
 966         if (memslot && memslot->dirty_bitmap) {
 967                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 968
 969                 /* avoid RMW */
 970                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 971                         set_bit(rel_gfn, memslot->dirty_bitmap);
 972         }
 973 }
 974
 975 int emulator_read_std(unsigned long addr,
 976                              void *val,
 977                              unsigned int bytes,
 978                              struct kvm_vcpu *vcpu)
 979 {
 980         void *data = val;
 981
 982         while (bytes) {
 983                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 984                 unsigned offset = addr & (PAGE_SIZE-1);
 985                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 986                 unsigned long pfn;
 987                 struct page *page;
 988                 void *page_virt;
 989
 990                 if (gpa == UNMAPPED_GVA)
 991                         return X86EMUL_PROPAGATE_FAULT;
 992                 pfn = gpa >> PAGE_SHIFT;
 993                 page = gfn_to_page(vcpu->kvm, pfn);
 994                 if (!page)
 995                         return X86EMUL_UNHANDLEABLE;
 996                 page_virt = kmap_atomic(page, KM_USER0);
 997
 998                 memcpy(data, page_virt + offset, tocopy);
 999
1000                 kunmap_atomic(page_virt, KM_USER0);
1001
1002                 bytes -= tocopy;
1003                 data += tocopy;
1004                 addr += tocopy;
1005         }
1006
1007         return X86EMUL_CONTINUE;
1008 }
1009 EXPORT_SYMBOL_GPL(emulator_read_std);
1010
1011 static int emulator_write_std(unsigned long addr,
1012                               const void *val,
1013                               unsigned int bytes,
1014                               struct kvm_vcpu *vcpu)
1015 {
1016         pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1017         return X86EMUL_UNHANDLEABLE;
1018 }
1019
1020 /*
1021  * Only apic need an MMIO device hook, so shortcut now..
1022  */
1023 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1024                                                 gpa_t addr)
1025 {
1026         struct kvm_io_device *dev;
1027
1028         if (vcpu->apic) {
1029                 dev = &vcpu->apic->dev;
1030                 if (dev->in_range(dev, addr))
1031                         return dev;
1032         }
1033         return NULL;
1034 }
1035
1036 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1037                                                 gpa_t addr)
1038 {
1039         struct kvm_io_device *dev;
1040
1041         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1042         if (dev == NULL)
1043                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1044         return dev;
1045 }
1046
1047 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1048                                                gpa_t addr)
1049 {
1050         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1051 }
1052
1053 static int emulator_read_emulated(unsigned long addr,
1054                                   void *val,
1055                                   unsigned int bytes,
1056                                   struct kvm_vcpu *vcpu)
1057 {
1058         struct kvm_io_device *mmio_dev;
1059         gpa_t                 gpa;
1060
1061         if (vcpu->mmio_read_completed) {
1062                 memcpy(val, vcpu->mmio_data, bytes);
1063                 vcpu->mmio_read_completed = 0;
1064                 return X86EMUL_CONTINUE;
1065         } else if (emulator_read_std(addr, val, bytes, vcpu)
1066                    == X86EMUL_CONTINUE)
1067                 return X86EMUL_CONTINUE;
1068
1069         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1070         if (gpa == UNMAPPED_GVA)
1071                 return X86EMUL_PROPAGATE_FAULT;
1072
1073         /*
1074          * Is this MMIO handled locally?
1075          */
1076         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1077         if (mmio_dev) {
1078                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1079                 return X86EMUL_CONTINUE;
1080         }
1081
1082         vcpu->mmio_needed = 1;
1083         vcpu->mmio_phys_addr = gpa;
1084         vcpu->mmio_size = bytes;
1085         vcpu->mmio_is_write = 0;
1086
1087         return X86EMUL_UNHANDLEABLE;
1088 }
1089
1090 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1091                                const void *val, int bytes)
1092 {
1093         struct page *page;
1094         void *virt;
1095
1096         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1097                 return 0;
1098         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1099         if (!page)
1100                 return 0;
1101         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1102         virt = kmap_atomic(page, KM_USER0);
1103         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1104         memcpy(virt + offset_in_page(gpa), val, bytes);
1105         kunmap_atomic(virt, KM_USER0);
1106         return 1;
1107 }
1108
1109 static int emulator_write_emulated_onepage(unsigned long addr,
1110                                            const void *val,
1111                                            unsigned int bytes,
1112                                            struct kvm_vcpu *vcpu)
1113 {
1114         struct kvm_io_device *mmio_dev;
1115         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1116
1117         if (gpa == UNMAPPED_GVA) {
1118                 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1119                 return X86EMUL_PROPAGATE_FAULT;
1120         }
1121
1122         if (emulator_write_phys(vcpu, gpa, val, bytes))
1123                 return X86EMUL_CONTINUE;
1124
1125         /*
1126          * Is this MMIO handled locally?
1127          */
1128         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1129         if (mmio_dev) {
1130                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1131                 return X86EMUL_CONTINUE;
1132         }
1133
1134         vcpu->mmio_needed = 1;
1135         vcpu->mmio_phys_addr = gpa;
1136         vcpu->mmio_size = bytes;
1137         vcpu->mmio_is_write = 1;
1138         memcpy(vcpu->mmio_data, val, bytes);
1139
1140         return X86EMUL_CONTINUE;
1141 }
1142
1143 int emulator_write_emulated(unsigned long addr,
1144                                    const void *val,
1145                                    unsigned int bytes,
1146                                    struct kvm_vcpu *vcpu)
1147 {
1148         /* Crossing a page boundary? */
1149         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1150                 int rc, now;
1151
1152                 now = -addr & ~PAGE_MASK;
1153                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1154                 if (rc != X86EMUL_CONTINUE)
1155                         return rc;
1156                 addr += now;
1157                 val += now;
1158                 bytes -= now;
1159         }
1160         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1161 }
1162 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1163
1164 static int emulator_cmpxchg_emulated(unsigned long addr,
1165                                      const void *old,
1166                                      const void *new,
1167                                      unsigned int bytes,
1168                                      struct kvm_vcpu *vcpu)
1169 {
1170         static int reported;
1171
1172         if (!reported) {
1173                 reported = 1;
1174                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1175         }
1176         return emulator_write_emulated(addr, new, bytes, vcpu);
1177 }
1178
1179 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1180 {
1181         return kvm_x86_ops->get_segment_base(vcpu, seg);
1182 }
1183
1184 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1185 {
1186         return X86EMUL_CONTINUE;
1187 }
1188
1189 int emulate_clts(struct kvm_vcpu *vcpu)
1190 {
1191         vcpu->cr0 &= ~X86_CR0_TS;
1192         kvm_x86_ops->set_cr0(vcpu, vcpu->cr0);
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 {
1256         struct x86_emulate_ctxt emulate_ctxt;
1257         int r;
1258         int cs_db, cs_l;
1259
1260         vcpu->mmio_fault_cr2 = cr2;
1261         kvm_x86_ops->cache_regs(vcpu);
1262
1263         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1264
1265         emulate_ctxt.vcpu = vcpu;
1266         emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1267         emulate_ctxt.cr2 = cr2;
1268         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1269                 ? X86EMUL_MODE_REAL : cs_l
1270                 ? X86EMUL_MODE_PROT64 : cs_db
1271                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1272
1273         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1274                 emulate_ctxt.cs_base = 0;
1275                 emulate_ctxt.ds_base = 0;
1276                 emulate_ctxt.es_base = 0;
1277                 emulate_ctxt.ss_base = 0;
1278         } else {
1279                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1280                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1281                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1282                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1283         }
1284
1285         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1286         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1287
1288         vcpu->mmio_is_write = 0;
1289         vcpu->pio.string = 0;
1290         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1291         if (vcpu->pio.string)
1292                 return EMULATE_DO_MMIO;
1293
1294         if ((r || vcpu->mmio_is_write) && run) {
1295                 run->exit_reason = KVM_EXIT_MMIO;
1296                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1297                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1298                 run->mmio.len = vcpu->mmio_size;
1299                 run->mmio.is_write = vcpu->mmio_is_write;
1300         }
1301
1302         if (r) {
1303                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1304                         return EMULATE_DONE;
1305                 if (!vcpu->mmio_needed) {
1306                         kvm_report_emulation_failure(vcpu, "mmio");
1307                         return EMULATE_FAIL;
1308                 }
1309                 return EMULATE_DO_MMIO;
1310         }
1311
1312         kvm_x86_ops->decache_regs(vcpu);
1313         kvm_x86_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1314
1315         if (vcpu->mmio_is_write) {
1316                 vcpu->mmio_needed = 0;
1317                 return EMULATE_DO_MMIO;
1318         }
1319
1320         return EMULATE_DONE;
1321 }
1322 EXPORT_SYMBOL_GPL(emulate_instruction);
1323
1324 /*
1325  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1326  */
1327 static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1328 {
1329         DECLARE_WAITQUEUE(wait, current);
1330
1331         add_wait_queue(&vcpu->wq, &wait);
1332
1333         /*
1334          * We will block until either an interrupt or a signal wakes us up
1335          */
1336         while (!kvm_cpu_has_interrupt(vcpu)
1337                && !signal_pending(current)
1338                && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
1339                && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
1340                 set_current_state(TASK_INTERRUPTIBLE);
1341                 vcpu_put(vcpu);
1342                 schedule();
1343                 vcpu_load(vcpu);
1344         }
1345
1346         __set_current_state(TASK_RUNNING);
1347         remove_wait_queue(&vcpu->wq, &wait);
1348 }
1349
1350 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1351 {
1352         ++vcpu->stat.halt_exits;
1353         if (irqchip_in_kernel(vcpu->kvm)) {
1354                 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1355                 kvm_vcpu_block(vcpu);
1356                 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1357                         return -EINTR;
1358                 return 1;
1359         } else {
1360                 vcpu->run->exit_reason = KVM_EXIT_HLT;
1361                 return 0;
1362         }
1363 }
1364 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1365
1366 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1367 {
1368         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1369
1370         kvm_x86_ops->cache_regs(vcpu);
1371         ret = -KVM_EINVAL;
1372 #ifdef CONFIG_X86_64
1373         if (is_long_mode(vcpu)) {
1374                 nr = vcpu->regs[VCPU_REGS_RAX];
1375                 a0 = vcpu->regs[VCPU_REGS_RDI];
1376                 a1 = vcpu->regs[VCPU_REGS_RSI];
1377                 a2 = vcpu->regs[VCPU_REGS_RDX];
1378                 a3 = vcpu->regs[VCPU_REGS_RCX];
1379                 a4 = vcpu->regs[VCPU_REGS_R8];
1380                 a5 = vcpu->regs[VCPU_REGS_R9];
1381         } else
1382 #endif
1383         {
1384                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1385                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1386                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1387                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1388                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1389                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1390                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1391         }
1392         switch (nr) {
1393         default:
1394                 run->hypercall.nr = nr;
1395                 run->hypercall.args[0] = a0;
1396                 run->hypercall.args[1] = a1;
1397                 run->hypercall.args[2] = a2;
1398                 run->hypercall.args[3] = a3;
1399                 run->hypercall.args[4] = a4;
1400                 run->hypercall.args[5] = a5;
1401                 run->hypercall.ret = ret;
1402                 run->hypercall.longmode = is_long_mode(vcpu);
1403                 kvm_x86_ops->decache_regs(vcpu);
1404                 return 0;
1405         }
1406         vcpu->regs[VCPU_REGS_RAX] = ret;
1407         kvm_x86_ops->decache_regs(vcpu);
1408         return 1;
1409 }
1410 EXPORT_SYMBOL_GPL(kvm_hypercall);
1411
1412 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1413 {
1414         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1415 }
1416
1417 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1418 {
1419         struct descriptor_table dt = { limit, base };
1420
1421         kvm_x86_ops->set_gdt(vcpu, &dt);
1422 }
1423
1424 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1425 {
1426         struct descriptor_table dt = { limit, base };
1427
1428         kvm_x86_ops->set_idt(vcpu, &dt);
1429 }
1430
1431 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1432                    unsigned long *rflags)
1433 {
1434         lmsw(vcpu, msw);
1435         *rflags = kvm_x86_ops->get_rflags(vcpu);
1436 }
1437
1438 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1439 {
1440         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1441         switch (cr) {
1442         case 0:
1443                 return vcpu->cr0;
1444         case 2:
1445                 return vcpu->cr2;
1446         case 3:
1447                 return vcpu->cr3;
1448         case 4:
1449                 return vcpu->cr4;
1450         default:
1451                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1452                 return 0;
1453         }
1454 }
1455
1456 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1457                      unsigned long *rflags)
1458 {
1459         switch (cr) {
1460         case 0:
1461                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1462                 *rflags = kvm_x86_ops->get_rflags(vcpu);
1463                 break;
1464         case 2:
1465                 vcpu->cr2 = val;
1466                 break;
1467         case 3:
1468                 set_cr3(vcpu, val);
1469                 break;
1470         case 4:
1471                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1472                 break;
1473         default:
1474                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1475         }
1476 }
1477
1478 /*
1479  * Register the para guest with the host:
1480  */
1481 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1482 {
1483         struct kvm_vcpu_para_state *para_state;
1484         hpa_t para_state_hpa, hypercall_hpa;
1485         struct page *para_state_page;
1486         unsigned char *hypercall;
1487         gpa_t hypercall_gpa;
1488
1489         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1490         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1491
1492         /*
1493          * Needs to be page aligned:
1494          */
1495         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1496                 goto err_gp;
1497
1498         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1499         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1500         if (is_error_hpa(para_state_hpa))
1501                 goto err_gp;
1502
1503         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1504         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1505         para_state = kmap(para_state_page);
1506
1507         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1508         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1509
1510         para_state->host_version = KVM_PARA_API_VERSION;
1511         /*
1512          * We cannot support guests that try to register themselves
1513          * with a newer API version than the host supports:
1514          */
1515         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1516                 para_state->ret = -KVM_EINVAL;
1517                 goto err_kunmap_skip;
1518         }
1519
1520         hypercall_gpa = para_state->hypercall_gpa;
1521         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1522         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1523         if (is_error_hpa(hypercall_hpa)) {
1524                 para_state->ret = -KVM_EINVAL;
1525                 goto err_kunmap_skip;
1526         }
1527
1528         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1529         vcpu->para_state_page = para_state_page;
1530         vcpu->para_state_gpa = para_state_gpa;
1531         vcpu->hypercall_gpa = hypercall_gpa;
1532
1533         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1534         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1535                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1536         kvm_x86_ops->patch_hypercall(vcpu, hypercall);
1537         kunmap_atomic(hypercall, KM_USER1);
1538
1539         para_state->ret = 0;
1540 err_kunmap_skip:
1541         kunmap(para_state_page);
1542         return 0;
1543 err_gp:
1544         return 1;
1545 }
1546
1547 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1548 {
1549         u64 data;
1550
1551         switch (msr) {
1552         case 0xc0010010: /* SYSCFG */
1553         case 0xc0010015: /* HWCR */
1554         case MSR_IA32_PLATFORM_ID:
1555         case MSR_IA32_P5_MC_ADDR:
1556         case MSR_IA32_P5_MC_TYPE:
1557         case MSR_IA32_MC0_CTL:
1558         case MSR_IA32_MCG_STATUS:
1559         case MSR_IA32_MCG_CAP:
1560         case MSR_IA32_MC0_MISC:
1561         case MSR_IA32_MC0_MISC+4:
1562         case MSR_IA32_MC0_MISC+8:
1563         case MSR_IA32_MC0_MISC+12:
1564         case MSR_IA32_MC0_MISC+16:
1565         case MSR_IA32_UCODE_REV:
1566         case MSR_IA32_PERF_STATUS:
1567         case MSR_IA32_EBL_CR_POWERON:
1568                 /* MTRR registers */
1569         case 0xfe:
1570         case 0x200 ... 0x2ff:
1571                 data = 0;
1572                 break;
1573         case 0xcd: /* fsb frequency */
1574                 data = 3;
1575                 break;
1576         case MSR_IA32_APICBASE:
1577                 data = kvm_get_apic_base(vcpu);
1578                 break;
1579         case MSR_IA32_MISC_ENABLE:
1580                 data = vcpu->ia32_misc_enable_msr;
1581                 break;
1582 #ifdef CONFIG_X86_64
1583         case MSR_EFER:
1584                 data = vcpu->shadow_efer;
1585                 break;
1586 #endif
1587         default:
1588                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1589                 return 1;
1590         }
1591         *pdata = data;
1592         return 0;
1593 }
1594 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1595
1596 /*
1597  * Reads an msr value (of 'msr_index') into 'pdata'.
1598  * Returns 0 on success, non-0 otherwise.
1599  * Assumes vcpu_load() was already called.
1600  */
1601 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1602 {
1603         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1604 }
1605
1606 #ifdef CONFIG_X86_64
1607
1608 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1609 {
1610         if (efer & EFER_RESERVED_BITS) {
1611                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1612                        efer);
1613                 inject_gp(vcpu);
1614                 return;
1615         }
1616
1617         if (is_paging(vcpu)
1618             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1619                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1620                 inject_gp(vcpu);
1621                 return;
1622         }
1623
1624         kvm_x86_ops->set_efer(vcpu, efer);
1625
1626         efer &= ~EFER_LMA;
1627         efer |= vcpu->shadow_efer & EFER_LMA;
1628
1629         vcpu->shadow_efer = efer;
1630 }
1631
1632 #endif
1633
1634 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1635 {
1636         switch (msr) {
1637 #ifdef CONFIG_X86_64
1638         case MSR_EFER:
1639                 set_efer(vcpu, data);
1640                 break;
1641 #endif
1642         case MSR_IA32_MC0_STATUS:
1643                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1644                        __FUNCTION__, data);
1645                 break;
1646         case MSR_IA32_MCG_STATUS:
1647                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1648                         __FUNCTION__, data);
1649                 break;
1650         case MSR_IA32_UCODE_REV:
1651         case MSR_IA32_UCODE_WRITE:
1652         case 0x200 ... 0x2ff: /* MTRRs */
1653                 break;
1654         case MSR_IA32_APICBASE:
1655                 kvm_set_apic_base(vcpu, data);
1656                 break;
1657         case MSR_IA32_MISC_ENABLE:
1658                 vcpu->ia32_misc_enable_msr = data;
1659                 break;
1660         /*
1661          * This is the 'probe whether the host is KVM' logic:
1662          */
1663         case MSR_KVM_API_MAGIC:
1664                 return vcpu_register_para(vcpu, data);
1665
1666         default:
1667                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1668                 return 1;
1669         }
1670         return 0;
1671 }
1672 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1673
1674 /*
1675  * Writes msr value into into the appropriate "register".
1676  * Returns 0 on success, non-0 otherwise.
1677  * Assumes vcpu_load() was already called.
1678  */
1679 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1680 {
1681         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
1682 }
1683
1684 void kvm_resched(struct kvm_vcpu *vcpu)
1685 {
1686         if (!need_resched())
1687                 return;
1688         cond_resched();
1689 }
1690 EXPORT_SYMBOL_GPL(kvm_resched);
1691
1692 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1693 {
1694         int i;
1695         u32 function;
1696         struct kvm_cpuid_entry *e, *best;
1697
1698         kvm_x86_ops->cache_regs(vcpu);
1699         function = vcpu->regs[VCPU_REGS_RAX];
1700         vcpu->regs[VCPU_REGS_RAX] = 0;
1701         vcpu->regs[VCPU_REGS_RBX] = 0;
1702         vcpu->regs[VCPU_REGS_RCX] = 0;
1703         vcpu->regs[VCPU_REGS_RDX] = 0;
1704         best = NULL;
1705         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1706                 e = &vcpu->cpuid_entries[i];
1707                 if (e->function == function) {
1708                         best = e;
1709                         break;
1710                 }
1711                 /*
1712                  * Both basic or both extended?
1713                  */
1714                 if (((e->function ^ function) & 0x80000000) == 0)
1715                         if (!best || e->function > best->function)
1716                                 best = e;
1717         }
1718         if (best) {
1719                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1720                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1721                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1722                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1723         }
1724         kvm_x86_ops->decache_regs(vcpu);
1725         kvm_x86_ops->skip_emulated_instruction(vcpu);
1726 }
1727 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1728
1729 static int pio_copy_data(struct kvm_vcpu *vcpu)
1730 {
1731         void *p = vcpu->pio_data;
1732         void *q;
1733         unsigned bytes;
1734         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1735
1736         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1737                  PAGE_KERNEL);
1738         if (!q) {
1739                 free_pio_guest_pages(vcpu);
1740                 return -ENOMEM;
1741         }
1742         q += vcpu->pio.guest_page_offset;
1743         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1744         if (vcpu->pio.in)
1745                 memcpy(q, p, bytes);
1746         else
1747                 memcpy(p, q, bytes);
1748         q -= vcpu->pio.guest_page_offset;
1749         vunmap(q);
1750         free_pio_guest_pages(vcpu);
1751         return 0;
1752 }
1753
1754 static int complete_pio(struct kvm_vcpu *vcpu)
1755 {
1756         struct kvm_pio_request *io = &vcpu->pio;
1757         long delta;
1758         int r;
1759
1760         kvm_x86_ops->cache_regs(vcpu);
1761
1762         if (!io->string) {
1763                 if (io->in)
1764                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1765                                io->size);
1766         } else {
1767                 if (io->in) {
1768                         r = pio_copy_data(vcpu);
1769                         if (r) {
1770                                 kvm_x86_ops->cache_regs(vcpu);
1771                                 return r;
1772                         }
1773                 }
1774
1775                 delta = 1;
1776                 if (io->rep) {
1777                         delta *= io->cur_count;
1778                         /*
1779                          * The size of the register should really depend on
1780                          * current address size.
1781                          */
1782                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1783                 }
1784                 if (io->down)
1785                         delta = -delta;
1786                 delta *= io->size;
1787                 if (io->in)
1788                         vcpu->regs[VCPU_REGS_RDI] += delta;
1789                 else
1790                         vcpu->regs[VCPU_REGS_RSI] += delta;
1791         }
1792
1793         kvm_x86_ops->decache_regs(vcpu);
1794
1795         io->count -= io->cur_count;
1796         io->cur_count = 0;
1797
1798         return 0;
1799 }
1800
1801 static void kernel_pio(struct kvm_io_device *pio_dev,
1802                        struct kvm_vcpu *vcpu,
1803                        void *pd)
1804 {
1805         /* TODO: String I/O for in kernel device */
1806
1807         mutex_lock(&vcpu->kvm->lock);
1808         if (vcpu->pio.in)
1809                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1810                                   vcpu->pio.size,
1811                                   pd);
1812         else
1813                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1814                                    vcpu->pio.size,
1815                                    pd);
1816         mutex_unlock(&vcpu->kvm->lock);
1817 }
1818
1819 static void pio_string_write(struct kvm_io_device *pio_dev,
1820                              struct kvm_vcpu *vcpu)
1821 {
1822         struct kvm_pio_request *io = &vcpu->pio;
1823         void *pd = vcpu->pio_data;
1824         int i;
1825
1826         mutex_lock(&vcpu->kvm->lock);
1827         for (i = 0; i < io->cur_count; i++) {
1828                 kvm_iodevice_write(pio_dev, io->port,
1829                                    io->size,
1830                                    pd);
1831                 pd += io->size;
1832         }
1833         mutex_unlock(&vcpu->kvm->lock);
1834 }
1835
1836 int kvm_emulate_pio (struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1837                   int size, unsigned port)
1838 {
1839         struct kvm_io_device *pio_dev;
1840
1841         vcpu->run->exit_reason = KVM_EXIT_IO;
1842         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1843         vcpu->run->io.size = vcpu->pio.size = size;
1844         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1845         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1846         vcpu->run->io.port = vcpu->pio.port = port;
1847         vcpu->pio.in = in;
1848         vcpu->pio.string = 0;
1849         vcpu->pio.down = 0;
1850         vcpu->pio.guest_page_offset = 0;
1851         vcpu->pio.rep = 0;
1852
1853         kvm_x86_ops->cache_regs(vcpu);
1854         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1855         kvm_x86_ops->decache_regs(vcpu);
1856
1857         kvm_x86_ops->skip_emulated_instruction(vcpu);
1858
1859         pio_dev = vcpu_find_pio_dev(vcpu, port);
1860         if (pio_dev) {
1861                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1862                 complete_pio(vcpu);
1863                 return 1;
1864         }
1865         return 0;
1866 }
1867 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1868
1869 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1870                   int size, unsigned long count, int down,
1871                   gva_t address, int rep, unsigned port)
1872 {
1873         unsigned now, in_page;
1874         int i, ret = 0;
1875         int nr_pages = 1;
1876         struct page *page;
1877         struct kvm_io_device *pio_dev;
1878
1879         vcpu->run->exit_reason = KVM_EXIT_IO;
1880         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1881         vcpu->run->io.size = vcpu->pio.size = size;
1882         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1883         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1884         vcpu->run->io.port = vcpu->pio.port = port;
1885         vcpu->pio.in = in;
1886         vcpu->pio.string = 1;
1887         vcpu->pio.down = down;
1888         vcpu->pio.guest_page_offset = offset_in_page(address);
1889         vcpu->pio.rep = rep;
1890
1891         if (!count) {
1892                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1893                 return 1;
1894         }
1895
1896         if (!down)
1897                 in_page = PAGE_SIZE - offset_in_page(address);
1898         else
1899                 in_page = offset_in_page(address) + size;
1900         now = min(count, (unsigned long)in_page / size);
1901         if (!now) {
1902                 /*
1903                  * String I/O straddles page boundary.  Pin two guest pages
1904                  * so that we satisfy atomicity constraints.  Do just one
1905                  * transaction to avoid complexity.
1906                  */
1907                 nr_pages = 2;
1908                 now = 1;
1909         }
1910         if (down) {
1911                 /*
1912                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1913                  */
1914                 pr_unimpl(vcpu, "guest string pio down\n");
1915                 inject_gp(vcpu);
1916                 return 1;
1917         }
1918         vcpu->run->io.count = now;
1919         vcpu->pio.cur_count = now;
1920
1921         if (vcpu->pio.cur_count == vcpu->pio.count)
1922                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1923
1924         for (i = 0; i < nr_pages; ++i) {
1925                 mutex_lock(&vcpu->kvm->lock);
1926                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1927                 if (page)
1928                         get_page(page);
1929                 vcpu->pio.guest_pages[i] = page;
1930                 mutex_unlock(&vcpu->kvm->lock);
1931                 if (!page) {
1932                         inject_gp(vcpu);
1933                         free_pio_guest_pages(vcpu);
1934                         return 1;
1935                 }
1936         }
1937
1938         pio_dev = vcpu_find_pio_dev(vcpu, port);
1939         if (!vcpu->pio.in) {
1940                 /* string PIO write */
1941                 ret = pio_copy_data(vcpu);
1942                 if (ret >= 0 && pio_dev) {
1943                         pio_string_write(pio_dev, vcpu);
1944                         complete_pio(vcpu);
1945                         if (vcpu->pio.count == 0)
1946                                 ret = 1;
1947                 }
1948         } else if (pio_dev)
1949                 pr_unimpl(vcpu, "no string pio read support yet, "
1950                        "port %x size %d count %ld\n",
1951                         port, size, count);
1952
1953         return ret;
1954 }
1955 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1956
1957 /*
1958  * Check if userspace requested an interrupt window, and that the
1959  * interrupt window is open.
1960  *
1961  * No need to exit to userspace if we already have an interrupt queued.
1962  */
1963 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1964                                           struct kvm_run *kvm_run)
1965 {
1966         return (!vcpu->irq_summary &&
1967                 kvm_run->request_interrupt_window &&
1968                 vcpu->interrupt_window_open &&
1969                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
1970 }
1971
1972 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1973                               struct kvm_run *kvm_run)
1974 {
1975         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
1976         kvm_run->cr8 = get_cr8(vcpu);
1977         kvm_run->apic_base = kvm_get_apic_base(vcpu);
1978         if (irqchip_in_kernel(vcpu->kvm))
1979                 kvm_run->ready_for_interrupt_injection = 1;
1980         else
1981                 kvm_run->ready_for_interrupt_injection =
1982                                         (vcpu->interrupt_window_open &&
1983                                          vcpu->irq_summary == 0);
1984 }
1985
1986 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1987 {
1988         int r;
1989
1990         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
1991                 printk("vcpu %d received sipi with vector # %x\n",
1992                        vcpu->vcpu_id, vcpu->sipi_vector);
1993                 kvm_lapic_reset(vcpu);
1994                 kvm_x86_ops->vcpu_reset(vcpu);
1995                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
1996         }
1997
1998 preempted:
1999         if (vcpu->guest_debug.enabled)
2000                 kvm_x86_ops->guest_debug_pre(vcpu);
2001
2002 again:
2003         r = kvm_mmu_reload(vcpu);
2004         if (unlikely(r))
2005                 goto out;
2006
2007         preempt_disable();
2008
2009         kvm_x86_ops->prepare_guest_switch(vcpu);
2010         kvm_load_guest_fpu(vcpu);
2011
2012         local_irq_disable();
2013
2014         if (signal_pending(current)) {
2015                 local_irq_enable();
2016                 preempt_enable();
2017                 r = -EINTR;
2018                 kvm_run->exit_reason = KVM_EXIT_INTR;
2019                 ++vcpu->stat.signal_exits;
2020                 goto out;
2021         }
2022
2023         if (irqchip_in_kernel(vcpu->kvm))
2024                 kvm_x86_ops->inject_pending_irq(vcpu);
2025         else if (!vcpu->mmio_read_completed)
2026                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2027
2028         vcpu->guest_mode = 1;
2029         kvm_guest_enter();
2030
2031         if (vcpu->requests)
2032                 if (test_and_clear_bit(KVM_TLB_FLUSH, &vcpu->requests))
2033                         kvm_x86_ops->tlb_flush(vcpu);
2034
2035         kvm_x86_ops->run(vcpu, kvm_run);
2036
2037         vcpu->guest_mode = 0;
2038         local_irq_enable();
2039
2040         ++vcpu->stat.exits;
2041
2042         /*
2043          * We must have an instruction between local_irq_enable() and
2044          * kvm_guest_exit(), so the timer interrupt isn't delayed by
2045          * the interrupt shadow.  The stat.exits increment will do nicely.
2046          * But we need to prevent reordering, hence this barrier():
2047          */
2048         barrier();
2049
2050         kvm_guest_exit();
2051
2052         preempt_enable();
2053
2054         /*
2055          * Profile KVM exit RIPs:
2056          */
2057         if (unlikely(prof_on == KVM_PROFILING)) {
2058                 kvm_x86_ops->cache_regs(vcpu);
2059                 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
2060         }
2061
2062         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2063
2064         if (r > 0) {
2065                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2066                         r = -EINTR;
2067                         kvm_run->exit_reason = KVM_EXIT_INTR;
2068                         ++vcpu->stat.request_irq_exits;
2069                         goto out;
2070                 }
2071                 if (!need_resched()) {
2072                         ++vcpu->stat.light_exits;
2073                         goto again;
2074                 }
2075         }
2076
2077 out:
2078         if (r > 0) {
2079                 kvm_resched(vcpu);
2080                 goto preempted;
2081         }
2082
2083         post_kvm_run_save(vcpu, kvm_run);
2084
2085         return r;
2086 }
2087
2088
2089 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2090 {
2091         int r;
2092         sigset_t sigsaved;
2093
2094         vcpu_load(vcpu);
2095
2096         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2097                 kvm_vcpu_block(vcpu);
2098                 vcpu_put(vcpu);
2099                 return -EAGAIN;
2100         }
2101
2102         if (vcpu->sigset_active)
2103                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2104
2105         /* re-sync apic's tpr */
2106         if (!irqchip_in_kernel(vcpu->kvm))
2107                 set_cr8(vcpu, kvm_run->cr8);
2108
2109         if (vcpu->pio.cur_count) {
2110                 r = complete_pio(vcpu);
2111                 if (r)
2112                         goto out;
2113         }
2114
2115         if (vcpu->mmio_needed) {
2116                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2117                 vcpu->mmio_read_completed = 1;
2118                 vcpu->mmio_needed = 0;
2119                 r = emulate_instruction(vcpu, kvm_run,
2120                                         vcpu->mmio_fault_cr2, 0);
2121                 if (r == EMULATE_DO_MMIO) {
2122                         /*
2123                          * Read-modify-write.  Back to userspace.
2124                          */
2125                         r = 0;
2126                         goto out;
2127                 }
2128         }
2129
2130         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2131                 kvm_x86_ops->cache_regs(vcpu);
2132                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2133                 kvm_x86_ops->decache_regs(vcpu);
2134         }
2135
2136         r = __vcpu_run(vcpu, kvm_run);
2137
2138 out:
2139         if (vcpu->sigset_active)
2140                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2141
2142         vcpu_put(vcpu);
2143         return r;
2144 }
2145
2146 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2147                                    struct kvm_regs *regs)
2148 {
2149         vcpu_load(vcpu);
2150
2151         kvm_x86_ops->cache_regs(vcpu);
2152
2153         regs->rax = vcpu->regs[VCPU_REGS_RAX];
2154         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2155         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2156         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2157         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2158         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2159         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2160         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2161 #ifdef CONFIG_X86_64
2162         regs->r8 = vcpu->regs[VCPU_REGS_R8];
2163         regs->r9 = vcpu->regs[VCPU_REGS_R9];
2164         regs->r10 = vcpu->regs[VCPU_REGS_R10];
2165         regs->r11 = vcpu->regs[VCPU_REGS_R11];
2166         regs->r12 = vcpu->regs[VCPU_REGS_R12];
2167         regs->r13 = vcpu->regs[VCPU_REGS_R13];
2168         regs->r14 = vcpu->regs[VCPU_REGS_R14];
2169         regs->r15 = vcpu->regs[VCPU_REGS_R15];
2170 #endif
2171
2172         regs->rip = vcpu->rip;
2173         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2174
2175         /*
2176          * Don't leak debug flags in case they were set for guest debugging
2177          */
2178         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2179                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2180
2181         vcpu_put(vcpu);
2182
2183         return 0;
2184 }
2185
2186 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2187                                    struct kvm_regs *regs)
2188 {
2189         vcpu_load(vcpu);
2190
2191         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2192         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2193         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2194         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2195         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2196         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2197         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2198         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2199 #ifdef CONFIG_X86_64
2200         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2201         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2202         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2203         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2204         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2205         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2206         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2207         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2208 #endif
2209
2210         vcpu->rip = regs->rip;
2211         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2212
2213         kvm_x86_ops->decache_regs(vcpu);
2214
2215         vcpu_put(vcpu);
2216
2217         return 0;
2218 }
2219
2220 static void get_segment(struct kvm_vcpu *vcpu,
2221                         struct kvm_segment *var, int seg)
2222 {
2223         return kvm_x86_ops->get_segment(vcpu, var, seg);
2224 }
2225
2226 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2227                                     struct kvm_sregs *sregs)
2228 {
2229         struct descriptor_table dt;
2230         int pending_vec;
2231
2232         vcpu_load(vcpu);
2233
2234         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2235         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2236         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2237         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2238         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2239         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2240
2241         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2242         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2243
2244         kvm_x86_ops->get_idt(vcpu, &dt);
2245         sregs->idt.limit = dt.limit;
2246         sregs->idt.base = dt.base;
2247         kvm_x86_ops->get_gdt(vcpu, &dt);
2248         sregs->gdt.limit = dt.limit;
2249         sregs->gdt.base = dt.base;
2250
2251         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2252         sregs->cr0 = vcpu->cr0;
2253         sregs->cr2 = vcpu->cr2;
2254         sregs->cr3 = vcpu->cr3;
2255         sregs->cr4 = vcpu->cr4;
2256         sregs->cr8 = get_cr8(vcpu);
2257         sregs->efer = vcpu->shadow_efer;
2258         sregs->apic_base = kvm_get_apic_base(vcpu);
2259
2260         if (irqchip_in_kernel(vcpu->kvm)) {
2261                 memset(sregs->interrupt_bitmap, 0,
2262                        sizeof sregs->interrupt_bitmap);
2263                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2264                 if (pending_vec >= 0)
2265                         set_bit(pending_vec, (unsigned long *)sregs->interrupt_bitmap);
2266         } else
2267                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2268                        sizeof sregs->interrupt_bitmap);
2269
2270         vcpu_put(vcpu);
2271
2272         return 0;
2273 }
2274
2275 static void set_segment(struct kvm_vcpu *vcpu,
2276                         struct kvm_segment *var, int seg)
2277 {
2278         return kvm_x86_ops->set_segment(vcpu, var, seg);
2279 }
2280
2281 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2282                                     struct kvm_sregs *sregs)
2283 {
2284         int mmu_reset_needed = 0;
2285         int i, pending_vec, max_bits;
2286         struct descriptor_table dt;
2287
2288         vcpu_load(vcpu);
2289
2290         dt.limit = sregs->idt.limit;
2291         dt.base = sregs->idt.base;
2292         kvm_x86_ops->set_idt(vcpu, &dt);
2293         dt.limit = sregs->gdt.limit;
2294         dt.base = sregs->gdt.base;
2295         kvm_x86_ops->set_gdt(vcpu, &dt);
2296
2297         vcpu->cr2 = sregs->cr2;
2298         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2299         vcpu->cr3 = sregs->cr3;
2300
2301         set_cr8(vcpu, sregs->cr8);
2302
2303         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2304 #ifdef CONFIG_X86_64
2305         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2306 #endif
2307         kvm_set_apic_base(vcpu, sregs->apic_base);
2308
2309         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2310
2311         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2312         vcpu->cr0 = sregs->cr0;
2313         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2314
2315         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2316         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2317         if (!is_long_mode(vcpu) && is_pae(vcpu))
2318                 load_pdptrs(vcpu, vcpu->cr3);
2319
2320         if (mmu_reset_needed)
2321                 kvm_mmu_reset_context(vcpu);
2322
2323         if (!irqchip_in_kernel(vcpu->kvm)) {
2324                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2325                        sizeof vcpu->irq_pending);
2326                 vcpu->irq_summary = 0;
2327                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2328                         if (vcpu->irq_pending[i])
2329                                 __set_bit(i, &vcpu->irq_summary);
2330         } else {
2331                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2332                 pending_vec = find_first_bit(
2333                         (const unsigned long *)sregs->interrupt_bitmap,
2334                         max_bits);
2335                 /* Only pending external irq is handled here */
2336                 if (pending_vec < max_bits) {
2337                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2338                         printk("Set back pending irq %d\n", pending_vec);
2339                 }
2340         }
2341
2342         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2343         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2344         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2345         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2346         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2347         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2348
2349         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2350         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2351
2352         vcpu_put(vcpu);
2353
2354         return 0;
2355 }
2356
2357 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2358 {
2359         struct kvm_segment cs;
2360
2361         get_segment(vcpu, &cs, VCPU_SREG_CS);
2362         *db = cs.db;
2363         *l = cs.l;
2364 }
2365 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2366
2367 /*
2368  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2369  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2370  *
2371  * This list is modified at module load time to reflect the
2372  * capabilities of the host cpu.
2373  */
2374 static u32 msrs_to_save[] = {
2375         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2376         MSR_K6_STAR,
2377 #ifdef CONFIG_X86_64
2378         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2379 #endif
2380         MSR_IA32_TIME_STAMP_COUNTER,
2381 };
2382
2383 static unsigned num_msrs_to_save;
2384
2385 static u32 emulated_msrs[] = {
2386         MSR_IA32_MISC_ENABLE,
2387 };
2388
2389 static __init void kvm_init_msr_list(void)
2390 {
2391         u32 dummy[2];
2392         unsigned i, j;
2393
2394         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2395                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2396                         continue;
2397                 if (j < i)
2398                         msrs_to_save[j] = msrs_to_save[i];
2399                 j++;
2400         }
2401         num_msrs_to_save = j;
2402 }
2403
2404 /*
2405  * Adapt set_msr() to msr_io()'s calling convention
2406  */
2407 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2408 {
2409         return kvm_set_msr(vcpu, index, *data);
2410 }
2411
2412 /*
2413  * Read or write a bunch of msrs. All parameters are kernel addresses.
2414  *
2415  * @return number of msrs set successfully.
2416  */
2417 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2418                     struct kvm_msr_entry *entries,
2419                     int (*do_msr)(struct kvm_vcpu *vcpu,
2420                                   unsigned index, u64 *data))
2421 {
2422         int i;
2423
2424         vcpu_load(vcpu);
2425
2426         for (i = 0; i < msrs->nmsrs; ++i)
2427                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2428                         break;
2429
2430         vcpu_put(vcpu);
2431
2432         return i;
2433 }
2434
2435 /*
2436  * Read or write a bunch of msrs. Parameters are user addresses.
2437  *
2438  * @return number of msrs set successfully.
2439  */
2440 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2441                   int (*do_msr)(struct kvm_vcpu *vcpu,
2442                                 unsigned index, u64 *data),
2443                   int writeback)
2444 {
2445         struct kvm_msrs msrs;
2446         struct kvm_msr_entry *entries;
2447         int r, n;
2448         unsigned size;
2449
2450         r = -EFAULT;
2451         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2452                 goto out;
2453
2454         r = -E2BIG;
2455         if (msrs.nmsrs >= MAX_IO_MSRS)
2456                 goto out;
2457
2458         r = -ENOMEM;
2459         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2460         entries = vmalloc(size);
2461         if (!entries)
2462                 goto out;
2463
2464         r = -EFAULT;
2465         if (copy_from_user(entries, user_msrs->entries, size))
2466                 goto out_free;
2467
2468         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2469         if (r < 0)
2470                 goto out_free;
2471
2472         r = -EFAULT;
2473         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2474                 goto out_free;
2475
2476         r = n;
2477
2478 out_free:
2479         vfree(entries);
2480 out:
2481         return r;
2482 }
2483
2484 /*
2485  * Translate a guest virtual address to a guest physical address.
2486  */
2487 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2488                                     struct kvm_translation *tr)
2489 {
2490         unsigned long vaddr = tr->linear_address;
2491         gpa_t gpa;
2492
2493         vcpu_load(vcpu);
2494         mutex_lock(&vcpu->kvm->lock);
2495         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2496         tr->physical_address = gpa;
2497         tr->valid = gpa != UNMAPPED_GVA;
2498         tr->writeable = 1;
2499         tr->usermode = 0;
2500         mutex_unlock(&vcpu->kvm->lock);
2501         vcpu_put(vcpu);
2502
2503         return 0;
2504 }
2505
2506 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2507                                     struct kvm_interrupt *irq)
2508 {
2509         if (irq->irq < 0 || irq->irq >= 256)
2510                 return -EINVAL;
2511         if (irqchip_in_kernel(vcpu->kvm))
2512                 return -ENXIO;
2513         vcpu_load(vcpu);
2514
2515         set_bit(irq->irq, vcpu->irq_pending);
2516         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2517
2518         vcpu_put(vcpu);
2519
2520         return 0;
2521 }
2522
2523 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2524                                       struct kvm_debug_guest *dbg)
2525 {
2526         int r;
2527
2528         vcpu_load(vcpu);
2529
2530         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2531
2532         vcpu_put(vcpu);
2533
2534         return r;
2535 }
2536
2537 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2538                                     unsigned long address,
2539                                     int *type)
2540 {
2541         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2542         unsigned long pgoff;
2543         struct page *page;
2544
2545         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2546         if (pgoff == 0)
2547                 page = virt_to_page(vcpu->run);
2548         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2549                 page = virt_to_page(vcpu->pio_data);
2550         else
2551                 return NOPAGE_SIGBUS;
2552         get_page(page);
2553         if (type != NULL)
2554                 *type = VM_FAULT_MINOR;
2555
2556         return page;
2557 }
2558
2559 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2560         .nopage = kvm_vcpu_nopage,
2561 };
2562
2563 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2564 {
2565         vma->vm_ops = &kvm_vcpu_vm_ops;
2566         return 0;
2567 }
2568
2569 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2570 {
2571         struct kvm_vcpu *vcpu = filp->private_data;
2572
2573         fput(vcpu->kvm->filp);
2574         return 0;
2575 }
2576
2577 static struct file_operations kvm_vcpu_fops = {
2578         .release        = kvm_vcpu_release,
2579         .unlocked_ioctl = kvm_vcpu_ioctl,
2580         .compat_ioctl   = kvm_vcpu_ioctl,
2581         .mmap           = kvm_vcpu_mmap,
2582 };
2583
2584 /*
2585  * Allocates an inode for the vcpu.
2586  */
2587 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2588 {
2589         int fd, r;
2590         struct inode *inode;
2591         struct file *file;
2592
2593         r = anon_inode_getfd(&fd, &inode, &file,
2594                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2595         if (r)
2596                 return r;
2597         atomic_inc(&vcpu->kvm->filp->f_count);
2598         return fd;
2599 }
2600
2601 /*
2602  * Creates some virtual cpus.  Good luck creating more than one.
2603  */
2604 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2605 {
2606         int r;
2607         struct kvm_vcpu *vcpu;
2608
2609         if (!valid_vcpu(n))
2610                 return -EINVAL;
2611
2612         vcpu = kvm_x86_ops->vcpu_create(kvm, n);
2613         if (IS_ERR(vcpu))
2614                 return PTR_ERR(vcpu);
2615
2616         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2617
2618         /* We do fxsave: this must be aligned. */
2619         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2620
2621         vcpu_load(vcpu);
2622         r = kvm_mmu_setup(vcpu);
2623         vcpu_put(vcpu);
2624         if (r < 0)
2625                 goto free_vcpu;
2626
2627         mutex_lock(&kvm->lock);
2628         if (kvm->vcpus[n]) {
2629                 r = -EEXIST;
2630                 mutex_unlock(&kvm->lock);
2631                 goto mmu_unload;
2632         }
2633         kvm->vcpus[n] = vcpu;
2634         mutex_unlock(&kvm->lock);
2635
2636         /* Now it's all set up, let userspace reach it */
2637         r = create_vcpu_fd(vcpu);
2638         if (r < 0)
2639                 goto unlink;
2640         return r;
2641
2642 unlink:
2643         mutex_lock(&kvm->lock);
2644         kvm->vcpus[n] = NULL;
2645         mutex_unlock(&kvm->lock);
2646
2647 mmu_unload:
2648         vcpu_load(vcpu);
2649         kvm_mmu_unload(vcpu);
2650         vcpu_put(vcpu);
2651
2652 free_vcpu:
2653         kvm_x86_ops->vcpu_free(vcpu);
2654         return r;
2655 }
2656
2657 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2658 {
2659         u64 efer;
2660         int i;
2661         struct kvm_cpuid_entry *e, *entry;
2662
2663         rdmsrl(MSR_EFER, efer);
2664         entry = NULL;
2665         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2666                 e = &vcpu->cpuid_entries[i];
2667                 if (e->function == 0x80000001) {
2668                         entry = e;
2669                         break;
2670                 }
2671         }
2672         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2673                 entry->edx &= ~(1 << 20);
2674                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2675         }
2676 }
2677
2678 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2679                                     struct kvm_cpuid *cpuid,
2680                                     struct kvm_cpuid_entry __user *entries)
2681 {
2682         int r;
2683
2684         r = -E2BIG;
2685         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2686                 goto out;
2687         r = -EFAULT;
2688         if (copy_from_user(&vcpu->cpuid_entries, entries,
2689                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2690                 goto out;
2691         vcpu->cpuid_nent = cpuid->nent;
2692         cpuid_fix_nx_cap(vcpu);
2693         return 0;
2694
2695 out:
2696         return r;
2697 }
2698
2699 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2700 {
2701         if (sigset) {
2702                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2703                 vcpu->sigset_active = 1;
2704                 vcpu->sigset = *sigset;
2705         } else
2706                 vcpu->sigset_active = 0;
2707         return 0;
2708 }
2709
2710 /*
2711  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2712  * we have asm/x86/processor.h
2713  */
2714 struct fxsave {
2715         u16     cwd;
2716         u16     swd;
2717         u16     twd;
2718         u16     fop;
2719         u64     rip;
2720         u64     rdp;
2721         u32     mxcsr;
2722         u32     mxcsr_mask;
2723         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2724 #ifdef CONFIG_X86_64
2725         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2726 #else
2727         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2728 #endif
2729 };
2730
2731 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2732 {
2733         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2734
2735         vcpu_load(vcpu);
2736
2737         memcpy(fpu->fpr, fxsave->st_space, 128);
2738         fpu->fcw = fxsave->cwd;
2739         fpu->fsw = fxsave->swd;
2740         fpu->ftwx = fxsave->twd;
2741         fpu->last_opcode = fxsave->fop;
2742         fpu->last_ip = fxsave->rip;
2743         fpu->last_dp = fxsave->rdp;
2744         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2745
2746         vcpu_put(vcpu);
2747
2748         return 0;
2749 }
2750
2751 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2752 {
2753         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2754
2755         vcpu_load(vcpu);
2756
2757         memcpy(fxsave->st_space, fpu->fpr, 128);
2758         fxsave->cwd = fpu->fcw;
2759         fxsave->swd = fpu->fsw;
2760         fxsave->twd = fpu->ftwx;
2761         fxsave->fop = fpu->last_opcode;
2762         fxsave->rip = fpu->last_ip;
2763         fxsave->rdp = fpu->last_dp;
2764         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2765
2766         vcpu_put(vcpu);
2767
2768         return 0;
2769 }
2770
2771 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2772                                     struct kvm_lapic_state *s)
2773 {
2774         vcpu_load(vcpu);
2775         memcpy(s->regs, vcpu->apic->regs, sizeof *s);
2776         vcpu_put(vcpu);
2777
2778         return 0;
2779 }
2780
2781 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2782                                     struct kvm_lapic_state *s)
2783 {
2784         vcpu_load(vcpu);
2785         memcpy(vcpu->apic->regs, s->regs, sizeof *s);
2786         kvm_apic_post_state_restore(vcpu);
2787         vcpu_put(vcpu);
2788
2789         return 0;
2790 }
2791
2792 static long kvm_vcpu_ioctl(struct file *filp,
2793                            unsigned int ioctl, unsigned long arg)
2794 {
2795         struct kvm_vcpu *vcpu = filp->private_data;
2796         void __user *argp = (void __user *)arg;
2797         int r = -EINVAL;
2798
2799         switch (ioctl) {
2800         case KVM_RUN:
2801                 r = -EINVAL;
2802                 if (arg)
2803                         goto out;
2804                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2805                 break;
2806         case KVM_GET_REGS: {
2807                 struct kvm_regs kvm_regs;
2808
2809                 memset(&kvm_regs, 0, sizeof kvm_regs);
2810                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2811                 if (r)
2812                         goto out;
2813                 r = -EFAULT;
2814                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2815                         goto out;
2816                 r = 0;
2817                 break;
2818         }
2819         case KVM_SET_REGS: {
2820                 struct kvm_regs kvm_regs;
2821
2822                 r = -EFAULT;
2823                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2824                         goto out;
2825                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2826                 if (r)
2827                         goto out;
2828                 r = 0;
2829                 break;
2830         }
2831         case KVM_GET_SREGS: {
2832                 struct kvm_sregs kvm_sregs;
2833
2834                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2835                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2836                 if (r)
2837                         goto out;
2838                 r = -EFAULT;
2839                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2840                         goto out;
2841                 r = 0;
2842                 break;
2843         }
2844         case KVM_SET_SREGS: {
2845                 struct kvm_sregs kvm_sregs;
2846
2847                 r = -EFAULT;
2848                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2849                         goto out;
2850                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2851                 if (r)
2852                         goto out;
2853                 r = 0;
2854                 break;
2855         }
2856         case KVM_TRANSLATE: {
2857                 struct kvm_translation tr;
2858
2859                 r = -EFAULT;
2860                 if (copy_from_user(&tr, argp, sizeof tr))
2861                         goto out;
2862                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2863                 if (r)
2864                         goto out;
2865                 r = -EFAULT;
2866                 if (copy_to_user(argp, &tr, sizeof tr))
2867                         goto out;
2868                 r = 0;
2869                 break;
2870         }
2871         case KVM_INTERRUPT: {
2872                 struct kvm_interrupt irq;
2873
2874                 r = -EFAULT;
2875                 if (copy_from_user(&irq, argp, sizeof irq))
2876                         goto out;
2877                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2878                 if (r)
2879                         goto out;
2880                 r = 0;
2881                 break;
2882         }
2883         case KVM_DEBUG_GUEST: {
2884                 struct kvm_debug_guest dbg;
2885
2886                 r = -EFAULT;
2887                 if (copy_from_user(&dbg, argp, sizeof dbg))
2888                         goto out;
2889                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2890                 if (r)
2891                         goto out;
2892                 r = 0;
2893                 break;
2894         }
2895         case KVM_GET_MSRS:
2896                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2897                 break;
2898         case KVM_SET_MSRS:
2899                 r = msr_io(vcpu, argp, do_set_msr, 0);
2900                 break;
2901         case KVM_SET_CPUID: {
2902                 struct kvm_cpuid __user *cpuid_arg = argp;
2903                 struct kvm_cpuid cpuid;
2904
2905                 r = -EFAULT;
2906                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2907                         goto out;
2908                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2909                 if (r)
2910                         goto out;
2911                 break;
2912         }
2913         case KVM_SET_SIGNAL_MASK: {
2914                 struct kvm_signal_mask __user *sigmask_arg = argp;
2915                 struct kvm_signal_mask kvm_sigmask;
2916                 sigset_t sigset, *p;
2917
2918                 p = NULL;
2919                 if (argp) {
2920                         r = -EFAULT;
2921                         if (copy_from_user(&kvm_sigmask, argp,
2922                                            sizeof kvm_sigmask))
2923                                 goto out;
2924                         r = -EINVAL;
2925                         if (kvm_sigmask.len != sizeof sigset)
2926                                 goto out;
2927                         r = -EFAULT;
2928                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2929                                            sizeof sigset))
2930                                 goto out;
2931                         p = &sigset;
2932                 }
2933                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2934                 break;
2935         }
2936         case KVM_GET_FPU: {
2937                 struct kvm_fpu fpu;
2938
2939                 memset(&fpu, 0, sizeof fpu);
2940                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2941                 if (r)
2942                         goto out;
2943                 r = -EFAULT;
2944                 if (copy_to_user(argp, &fpu, sizeof fpu))
2945                         goto out;
2946                 r = 0;
2947                 break;
2948         }
2949         case KVM_SET_FPU: {
2950                 struct kvm_fpu fpu;
2951
2952                 r = -EFAULT;
2953                 if (copy_from_user(&fpu, argp, sizeof fpu))
2954                         goto out;
2955                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2956                 if (r)
2957                         goto out;
2958                 r = 0;
2959                 break;
2960         }
2961         case KVM_GET_LAPIC: {
2962                 struct kvm_lapic_state lapic;
2963
2964                 memset(&lapic, 0, sizeof lapic);
2965                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
2966                 if (r)
2967                         goto out;
2968                 r = -EFAULT;
2969                 if (copy_to_user(argp, &lapic, sizeof lapic))
2970                         goto out;
2971                 r = 0;
2972                 break;
2973         }
2974         case KVM_SET_LAPIC: {
2975                 struct kvm_lapic_state lapic;
2976
2977                 r = -EFAULT;
2978                 if (copy_from_user(&lapic, argp, sizeof lapic))
2979                         goto out;
2980                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
2981                 if (r)
2982                         goto out;
2983                 r = 0;
2984                 break;
2985         }
2986         default:
2987                 ;
2988         }
2989 out:
2990         return r;
2991 }
2992
2993 static long kvm_vm_ioctl(struct file *filp,
2994                            unsigned int ioctl, unsigned long arg)
2995 {
2996         struct kvm *kvm = filp->private_data;
2997         void __user *argp = (void __user *)arg;
2998         int r = -EINVAL;
2999
3000         switch (ioctl) {
3001         case KVM_CREATE_VCPU:
3002                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3003                 if (r < 0)
3004                         goto out;
3005                 break;
3006         case KVM_SET_MEMORY_REGION: {
3007                 struct kvm_memory_region kvm_mem;
3008
3009                 r = -EFAULT;
3010                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
3011                         goto out;
3012                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
3013                 if (r)
3014                         goto out;
3015                 break;
3016         }
3017         case KVM_GET_DIRTY_LOG: {
3018                 struct kvm_dirty_log log;
3019
3020                 r = -EFAULT;
3021                 if (copy_from_user(&log, argp, sizeof log))
3022                         goto out;
3023                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3024                 if (r)
3025                         goto out;
3026                 break;
3027         }
3028         case KVM_SET_MEMORY_ALIAS: {
3029                 struct kvm_memory_alias alias;
3030
3031                 r = -EFAULT;
3032                 if (copy_from_user(&alias, argp, sizeof alias))
3033                         goto out;
3034                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
3035                 if (r)
3036                         goto out;
3037                 break;
3038         }
3039         case KVM_CREATE_IRQCHIP:
3040                 r = -ENOMEM;
3041                 kvm->vpic = kvm_create_pic(kvm);
3042                 if (kvm->vpic) {
3043                         r = kvm_ioapic_init(kvm);
3044                         if (r) {
3045                                 kfree(kvm->vpic);
3046                                 kvm->vpic = NULL;
3047                                 goto out;
3048                         }
3049                 }
3050                 else
3051                         goto out;
3052                 break;
3053         case KVM_IRQ_LINE: {
3054                 struct kvm_irq_level irq_event;
3055
3056                 r = -EFAULT;
3057                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3058                         goto out;
3059                 if (irqchip_in_kernel(kvm)) {
3060                         mutex_lock(&kvm->lock);
3061                         if (irq_event.irq < 16)
3062                                 kvm_pic_set_irq(pic_irqchip(kvm),
3063                                         irq_event.irq,
3064                                         irq_event.level);
3065                         kvm_ioapic_set_irq(kvm->vioapic,
3066                                         irq_event.irq,
3067                                         irq_event.level);
3068                         mutex_unlock(&kvm->lock);
3069                         r = 0;
3070                 }
3071                 break;
3072         }
3073         case KVM_GET_IRQCHIP: {
3074                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3075                 struct kvm_irqchip chip;
3076
3077                 r = -EFAULT;
3078                 if (copy_from_user(&chip, argp, sizeof chip))
3079                         goto out;
3080                 r = -ENXIO;
3081                 if (!irqchip_in_kernel(kvm))
3082                         goto out;
3083                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
3084                 if (r)
3085                         goto out;
3086                 r = -EFAULT;
3087                 if (copy_to_user(argp, &chip, sizeof chip))
3088                         goto out;
3089                 r = 0;
3090                 break;
3091         }
3092         case KVM_SET_IRQCHIP: {
3093                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3094                 struct kvm_irqchip chip;
3095
3096                 r = -EFAULT;
3097                 if (copy_from_user(&chip, argp, sizeof chip))
3098                         goto out;
3099                 r = -ENXIO;
3100                 if (!irqchip_in_kernel(kvm))
3101                         goto out;
3102                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
3103                 if (r)
3104                         goto out;
3105                 r = 0;
3106                 break;
3107         }
3108         default:
3109                 ;
3110         }
3111 out:
3112         return r;
3113 }
3114
3115 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
3116                                   unsigned long address,
3117                                   int *type)
3118 {
3119         struct kvm *kvm = vma->vm_file->private_data;
3120         unsigned long pgoff;
3121         struct page *page;
3122
3123         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
3124         page = gfn_to_page(kvm, pgoff);
3125         if (!page)
3126                 return NOPAGE_SIGBUS;
3127         get_page(page);
3128         if (type != NULL)
3129                 *type = VM_FAULT_MINOR;
3130
3131         return page;
3132 }
3133
3134 static struct vm_operations_struct kvm_vm_vm_ops = {
3135         .nopage = kvm_vm_nopage,
3136 };
3137
3138 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
3139 {
3140         vma->vm_ops = &kvm_vm_vm_ops;
3141         return 0;
3142 }
3143
3144 static struct file_operations kvm_vm_fops = {
3145         .release        = kvm_vm_release,
3146         .unlocked_ioctl = kvm_vm_ioctl,
3147         .compat_ioctl   = kvm_vm_ioctl,
3148         .mmap           = kvm_vm_mmap,
3149 };
3150
3151 static int kvm_dev_ioctl_create_vm(void)
3152 {
3153         int fd, r;
3154         struct inode *inode;
3155         struct file *file;
3156         struct kvm *kvm;
3157
3158         kvm = kvm_create_vm();
3159         if (IS_ERR(kvm))
3160                 return PTR_ERR(kvm);
3161         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
3162         if (r) {
3163                 kvm_destroy_vm(kvm);
3164                 return r;
3165         }
3166
3167         kvm->filp = file;
3168
3169         return fd;
3170 }
3171
3172 static long kvm_dev_ioctl(struct file *filp,
3173                           unsigned int ioctl, unsigned long arg)
3174 {
3175         void __user *argp = (void __user *)arg;
3176         long r = -EINVAL;
3177
3178         switch (ioctl) {
3179         case KVM_GET_API_VERSION:
3180                 r = -EINVAL;
3181                 if (arg)
3182                         goto out;
3183                 r = KVM_API_VERSION;
3184                 break;
3185         case KVM_CREATE_VM:
3186                 r = -EINVAL;
3187                 if (arg)
3188                         goto out;
3189                 r = kvm_dev_ioctl_create_vm();
3190                 break;
3191         case KVM_GET_MSR_INDEX_LIST: {
3192                 struct kvm_msr_list __user *user_msr_list = argp;
3193                 struct kvm_msr_list msr_list;
3194                 unsigned n;
3195
3196                 r = -EFAULT;
3197                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
3198                         goto out;
3199                 n = msr_list.nmsrs;
3200                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
3201                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
3202                         goto out;
3203                 r = -E2BIG;
3204                 if (n < num_msrs_to_save)
3205                         goto out;
3206                 r = -EFAULT;
3207                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3208                                  num_msrs_to_save * sizeof(u32)))
3209                         goto out;
3210                 if (copy_to_user(user_msr_list->indices
3211                                  + num_msrs_to_save * sizeof(u32),
3212                                  &emulated_msrs,
3213                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
3214                         goto out;
3215                 r = 0;
3216                 break;
3217         }
3218         case KVM_CHECK_EXTENSION: {
3219                 int ext = (long)argp;
3220
3221                 switch (ext) {
3222                 case KVM_CAP_IRQCHIP:
3223                 case KVM_CAP_HLT:
3224                         r = 1;
3225                         break;
3226                 default:
3227                         r = 0;
3228                         break;
3229                 }
3230                 break;
3231         }
3232         case KVM_GET_VCPU_MMAP_SIZE:
3233                 r = -EINVAL;
3234                 if (arg)
3235                         goto out;
3236                 r = 2 * PAGE_SIZE;
3237                 break;
3238         default:
3239                 ;
3240         }
3241 out:
3242         return r;
3243 }
3244
3245 static struct file_operations kvm_chardev_ops = {
3246         .unlocked_ioctl = kvm_dev_ioctl,
3247         .compat_ioctl   = kvm_dev_ioctl,
3248 };
3249
3250 static struct miscdevice kvm_dev = {
3251         KVM_MINOR,
3252         "kvm",
3253         &kvm_chardev_ops,
3254 };
3255
3256 /*
3257  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
3258  * cached on it.
3259  */
3260 static void decache_vcpus_on_cpu(int cpu)
3261 {
3262         struct kvm *vm;
3263         struct kvm_vcpu *vcpu;
3264         int i;
3265
3266         spin_lock(&kvm_lock);
3267         list_for_each_entry(vm, &vm_list, vm_list)
3268                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3269                         vcpu = vm->vcpus[i];
3270                         if (!vcpu)
3271                                 continue;
3272                         /*
3273                          * If the vcpu is locked, then it is running on some
3274                          * other cpu and therefore it is not cached on the
3275                          * cpu in question.
3276                          *
3277                          * If it's not locked, check the last cpu it executed
3278                          * on.
3279                          */
3280                         if (mutex_trylock(&vcpu->mutex)) {
3281                                 if (vcpu->cpu == cpu) {
3282                                         kvm_x86_ops->vcpu_decache(vcpu);
3283                                         vcpu->cpu = -1;
3284                                 }
3285                                 mutex_unlock(&vcpu->mutex);
3286                         }
3287                 }
3288         spin_unlock(&kvm_lock);
3289 }
3290
3291 static void hardware_enable(void *junk)
3292 {
3293         int cpu = raw_smp_processor_id();
3294
3295         if (cpu_isset(cpu, cpus_hardware_enabled))
3296                 return;
3297         cpu_set(cpu, cpus_hardware_enabled);
3298         kvm_x86_ops->hardware_enable(NULL);
3299 }
3300
3301 static void hardware_disable(void *junk)
3302 {
3303         int cpu = raw_smp_processor_id();
3304
3305         if (!cpu_isset(cpu, cpus_hardware_enabled))
3306                 return;
3307         cpu_clear(cpu, cpus_hardware_enabled);
3308         decache_vcpus_on_cpu(cpu);
3309         kvm_x86_ops->hardware_disable(NULL);
3310 }
3311
3312 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3313                            void *v)
3314 {
3315         int cpu = (long)v;
3316
3317         switch (val) {
3318         case CPU_DYING:
3319         case CPU_DYING_FROZEN:
3320                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3321                        cpu);
3322                 hardware_disable(NULL);
3323                 break;
3324         case CPU_UP_CANCELED:
3325         case CPU_UP_CANCELED_FROZEN:
3326                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3327                        cpu);
3328                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3329                 break;
3330         case CPU_ONLINE:
3331         case CPU_ONLINE_FROZEN:
3332                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3333                        cpu);
3334                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3335                 break;
3336         }
3337         return NOTIFY_OK;
3338 }
3339
3340 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3341                        void *v)
3342 {
3343         if (val == SYS_RESTART) {
3344                 /*
3345                  * Some (well, at least mine) BIOSes hang on reboot if
3346                  * in vmx root mode.
3347                  */
3348                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3349                 on_each_cpu(hardware_disable, NULL, 0, 1);
3350         }
3351         return NOTIFY_OK;
3352 }
3353
3354 static struct notifier_block kvm_reboot_notifier = {
3355         .notifier_call = kvm_reboot,
3356         .priority = 0,
3357 };
3358
3359 void kvm_io_bus_init(struct kvm_io_bus *bus)
3360 {
3361         memset(bus, 0, sizeof(*bus));
3362 }
3363
3364 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3365 {
3366         int i;
3367
3368         for (i = 0; i < bus->dev_count; i++) {
3369                 struct kvm_io_device *pos = bus->devs[i];
3370
3371                 kvm_iodevice_destructor(pos);
3372         }
3373 }
3374
3375 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3376 {
3377         int i;
3378
3379         for (i = 0; i < bus->dev_count; i++) {
3380                 struct kvm_io_device *pos = bus->devs[i];
3381
3382                 if (pos->in_range(pos, addr))
3383                         return pos;
3384         }
3385
3386         return NULL;
3387 }
3388
3389 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3390 {
3391         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3392
3393         bus->devs[bus->dev_count++] = dev;
3394 }
3395
3396 static struct notifier_block kvm_cpu_notifier = {
3397         .notifier_call = kvm_cpu_hotplug,
3398         .priority = 20, /* must be > scheduler priority */
3399 };
3400
3401 static u64 stat_get(void *_offset)
3402 {
3403         unsigned offset = (long)_offset;
3404         u64 total = 0;
3405         struct kvm *kvm;
3406         struct kvm_vcpu *vcpu;
3407         int i;
3408
3409         spin_lock(&kvm_lock);
3410         list_for_each_entry(kvm, &vm_list, vm_list)
3411                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3412                         vcpu = kvm->vcpus[i];
3413                         if (vcpu)
3414                                 total += *(u32 *)((void *)vcpu + offset);
3415                 }
3416         spin_unlock(&kvm_lock);
3417         return total;
3418 }
3419
3420 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3421
3422 static __init void kvm_init_debug(void)
3423 {
3424         struct kvm_stats_debugfs_item *p;
3425
3426         debugfs_dir = debugfs_create_dir("kvm", NULL);
3427         for (p = debugfs_entries; p->name; ++p)
3428                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3429                                                 (void *)(long)p->offset,
3430                                                 &stat_fops);
3431 }
3432
3433 static void kvm_exit_debug(void)
3434 {
3435         struct kvm_stats_debugfs_item *p;
3436
3437         for (p = debugfs_entries; p->name; ++p)
3438                 debugfs_remove(p->dentry);
3439         debugfs_remove(debugfs_dir);
3440 }
3441
3442 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3443 {
3444         hardware_disable(NULL);
3445         return 0;
3446 }
3447
3448 static int kvm_resume(struct sys_device *dev)
3449 {
3450         hardware_enable(NULL);
3451         return 0;
3452 }
3453
3454 static struct sysdev_class kvm_sysdev_class = {
3455         set_kset_name("kvm"),
3456         .suspend = kvm_suspend,
3457         .resume = kvm_resume,
3458 };
3459
3460 static struct sys_device kvm_sysdev = {
3461         .id = 0,
3462         .cls = &kvm_sysdev_class,
3463 };
3464
3465 hpa_t bad_page_address;
3466
3467 static inline
3468 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3469 {
3470         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3471 }
3472
3473 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3474 {
3475         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3476
3477         kvm_x86_ops->vcpu_load(vcpu, cpu);
3478 }
3479
3480 static void kvm_sched_out(struct preempt_notifier *pn,
3481                           struct task_struct *next)
3482 {
3483         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3484
3485         kvm_x86_ops->vcpu_put(vcpu);
3486 }
3487
3488 int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
3489                   struct module *module)
3490 {
3491         int r;
3492         int cpu;
3493
3494         if (kvm_x86_ops) {
3495                 printk(KERN_ERR "kvm: already loaded the other module\n");
3496                 return -EEXIST;
3497         }
3498
3499         if (!ops->cpu_has_kvm_support()) {
3500                 printk(KERN_ERR "kvm: no hardware support\n");
3501                 return -EOPNOTSUPP;
3502         }
3503         if (ops->disabled_by_bios()) {
3504                 printk(KERN_ERR "kvm: disabled by bios\n");
3505                 return -EOPNOTSUPP;
3506         }
3507
3508         kvm_x86_ops = ops;
3509
3510         r = kvm_x86_ops->hardware_setup();
3511         if (r < 0)
3512                 goto out;
3513
3514         for_each_online_cpu(cpu) {
3515                 smp_call_function_single(cpu,
3516                                 kvm_x86_ops->check_processor_compatibility,
3517                                 &r, 0, 1);
3518                 if (r < 0)
3519                         goto out_free_0;
3520         }
3521
3522         on_each_cpu(hardware_enable, NULL, 0, 1);
3523         r = register_cpu_notifier(&kvm_cpu_notifier);
3524         if (r)
3525                 goto out_free_1;
3526         register_reboot_notifier(&kvm_reboot_notifier);
3527
3528         r = sysdev_class_register(&kvm_sysdev_class);
3529         if (r)
3530                 goto out_free_2;
3531
3532         r = sysdev_register(&kvm_sysdev);
3533         if (r)
3534                 goto out_free_3;
3535
3536         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3537         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3538                                            __alignof__(struct kvm_vcpu), 0, 0);
3539         if (!kvm_vcpu_cache) {
3540                 r = -ENOMEM;
3541                 goto out_free_4;
3542         }
3543
3544         kvm_chardev_ops.owner = module;
3545
3546         r = misc_register(&kvm_dev);
3547         if (r) {
3548                 printk (KERN_ERR "kvm: misc device register failed\n");
3549                 goto out_free;
3550         }
3551
3552         kvm_preempt_ops.sched_in = kvm_sched_in;
3553         kvm_preempt_ops.sched_out = kvm_sched_out;
3554
3555         return r;
3556
3557 out_free:
3558         kmem_cache_destroy(kvm_vcpu_cache);
3559 out_free_4:
3560         sysdev_unregister(&kvm_sysdev);
3561 out_free_3:
3562         sysdev_class_unregister(&kvm_sysdev_class);
3563 out_free_2:
3564         unregister_reboot_notifier(&kvm_reboot_notifier);
3565         unregister_cpu_notifier(&kvm_cpu_notifier);
3566 out_free_1:
3567         on_each_cpu(hardware_disable, NULL, 0, 1);
3568 out_free_0:
3569         kvm_x86_ops->hardware_unsetup();
3570 out:
3571         kvm_x86_ops = NULL;
3572         return r;
3573 }
3574
3575 void kvm_exit_x86(void)
3576 {
3577         misc_deregister(&kvm_dev);
3578         kmem_cache_destroy(kvm_vcpu_cache);
3579         sysdev_unregister(&kvm_sysdev);
3580         sysdev_class_unregister(&kvm_sysdev_class);
3581         unregister_reboot_notifier(&kvm_reboot_notifier);
3582         unregister_cpu_notifier(&kvm_cpu_notifier);
3583         on_each_cpu(hardware_disable, NULL, 0, 1);
3584         kvm_x86_ops->hardware_unsetup();
3585         kvm_x86_ops = NULL;
3586 }
3587
3588 static __init int kvm_init(void)
3589 {
3590         static struct page *bad_page;
3591         int r;
3592
3593         r = kvm_mmu_module_init();
3594         if (r)
3595                 goto out4;
3596
3597         kvm_init_debug();
3598
3599         kvm_init_msr_list();
3600
3601         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3602                 r = -ENOMEM;
3603                 goto out;
3604         }
3605
3606         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3607         memset(__va(bad_page_address), 0, PAGE_SIZE);
3608
3609         return 0;
3610
3611 out:
3612         kvm_exit_debug();
3613         kvm_mmu_module_exit();
3614 out4:
3615         return r;
3616 }
3617
3618 static __exit void kvm_exit(void)
3619 {
3620         kvm_exit_debug();
3621         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3622         kvm_mmu_module_exit();
3623 }
3624
3625 module_init(kvm_init)
3626 module_exit(kvm_exit)
3627
3628 EXPORT_SYMBOL_GPL(kvm_init_x86);
3629 EXPORT_SYMBOL_GPL(kvm_exit_x86);