git.oblomov.eu Git - linux-2.6/blob - drivers/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "kvm.h"
  19 #include "x86_emulate.h"
  20 #include "segment_descriptor.h"
  21 #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         kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1192         return X86EMUL_CONTINUE;
1193 }
1194
1195 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1196 {
1197         struct kvm_vcpu *vcpu = ctxt->vcpu;
1198
1199         switch (dr) {
1200         case 0 ... 3:
1201                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1202                 return X86EMUL_CONTINUE;
1203         default:
1204                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1205                 return X86EMUL_UNHANDLEABLE;
1206         }
1207 }
1208
1209 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1210 {
1211         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1212         int exception;
1213
1214         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1215         if (exception) {
1216                 /* FIXME: better handling */
1217                 return X86EMUL_UNHANDLEABLE;
1218         }
1219         return X86EMUL_CONTINUE;
1220 }
1221
1222 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1223 {
1224         static int reported;
1225         u8 opcodes[4];
1226         unsigned long rip = vcpu->rip;
1227         unsigned long rip_linear;
1228
1229         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1230
1231         if (reported)
1232                 return;
1233
1234         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1235
1236         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1237                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1238         reported = 1;
1239 }
1240 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1241
1242 struct x86_emulate_ops emulate_ops = {
1243         .read_std            = emulator_read_std,
1244         .write_std           = emulator_write_std,
1245         .read_emulated       = emulator_read_emulated,
1246         .write_emulated      = emulator_write_emulated,
1247         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1248 };
1249
1250 int emulate_instruction(struct kvm_vcpu *vcpu,
1251                         struct kvm_run *run,
1252                         unsigned long cr2,
1253                         u16 error_code)
1254 {
1255         struct x86_emulate_ctxt emulate_ctxt;
1256         int r;
1257         int cs_db, cs_l;
1258
1259         vcpu->mmio_fault_cr2 = cr2;
1260         kvm_x86_ops->cache_regs(vcpu);
1261
1262         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1263
1264         emulate_ctxt.vcpu = vcpu;
1265         emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1266         emulate_ctxt.cr2 = cr2;
1267         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1268                 ? X86EMUL_MODE_REAL : cs_l
1269                 ? X86EMUL_MODE_PROT64 : cs_db
1270                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1271
1272         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1273                 emulate_ctxt.cs_base = 0;
1274                 emulate_ctxt.ds_base = 0;
1275                 emulate_ctxt.es_base = 0;
1276                 emulate_ctxt.ss_base = 0;
1277         } else {
1278                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1279                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1280                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1281                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1282         }
1283
1284         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1285         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1286
1287         vcpu->mmio_is_write = 0;
1288         vcpu->pio.string = 0;
1289         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1290         if (vcpu->pio.string)
1291                 return EMULATE_DO_MMIO;
1292
1293         if ((r || vcpu->mmio_is_write) && run) {
1294                 run->exit_reason = KVM_EXIT_MMIO;
1295                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1296                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1297                 run->mmio.len = vcpu->mmio_size;
1298                 run->mmio.is_write = vcpu->mmio_is_write;
1299         }
1300
1301         if (r) {
1302                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1303                         return EMULATE_DONE;
1304                 if (!vcpu->mmio_needed) {
1305                         kvm_report_emulation_failure(vcpu, "mmio");
1306                         return EMULATE_FAIL;
1307                 }
1308                 return EMULATE_DO_MMIO;
1309         }
1310
1311         kvm_x86_ops->decache_regs(vcpu);
1312         kvm_x86_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1313
1314         if (vcpu->mmio_is_write) {
1315                 vcpu->mmio_needed = 0;
1316                 return EMULATE_DO_MMIO;
1317         }
1318
1319         return EMULATE_DONE;
1320 }
1321 EXPORT_SYMBOL_GPL(emulate_instruction);
1322
1323 /*
1324  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1325  */
1326 static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1327 {
1328         DECLARE_WAITQUEUE(wait, current);
1329
1330         add_wait_queue(&vcpu->wq, &wait);
1331
1332         /*
1333          * We will block until either an interrupt or a signal wakes us up
1334          */
1335         while (!kvm_cpu_has_interrupt(vcpu)
1336                && !signal_pending(current)
1337                && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
1338                && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
1339                 set_current_state(TASK_INTERRUPTIBLE);
1340                 vcpu_put(vcpu);
1341                 schedule();
1342                 vcpu_load(vcpu);
1343         }
1344
1345         __set_current_state(TASK_RUNNING);
1346         remove_wait_queue(&vcpu->wq, &wait);
1347 }
1348
1349 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1350 {
1351         ++vcpu->stat.halt_exits;
1352         if (irqchip_in_kernel(vcpu->kvm)) {
1353                 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1354                 kvm_vcpu_block(vcpu);
1355                 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1356                         return -EINTR;
1357                 return 1;
1358         } else {
1359                 vcpu->run->exit_reason = KVM_EXIT_HLT;
1360                 return 0;
1361         }
1362 }
1363 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1364
1365 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1366 {
1367         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1368
1369         kvm_x86_ops->cache_regs(vcpu);
1370         ret = -KVM_EINVAL;
1371 #ifdef CONFIG_X86_64
1372         if (is_long_mode(vcpu)) {
1373                 nr = vcpu->regs[VCPU_REGS_RAX];
1374                 a0 = vcpu->regs[VCPU_REGS_RDI];
1375                 a1 = vcpu->regs[VCPU_REGS_RSI];
1376                 a2 = vcpu->regs[VCPU_REGS_RDX];
1377                 a3 = vcpu->regs[VCPU_REGS_RCX];
1378                 a4 = vcpu->regs[VCPU_REGS_R8];
1379                 a5 = vcpu->regs[VCPU_REGS_R9];
1380         } else
1381 #endif
1382         {
1383                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1384                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1385                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1386                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1387                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1388                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1389                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1390         }
1391         switch (nr) {
1392         default:
1393                 run->hypercall.nr = nr;
1394                 run->hypercall.args[0] = a0;
1395                 run->hypercall.args[1] = a1;
1396                 run->hypercall.args[2] = a2;
1397                 run->hypercall.args[3] = a3;
1398                 run->hypercall.args[4] = a4;
1399                 run->hypercall.args[5] = a5;
1400                 run->hypercall.ret = ret;
1401                 run->hypercall.longmode = is_long_mode(vcpu);
1402                 kvm_x86_ops->decache_regs(vcpu);
1403                 return 0;
1404         }
1405         vcpu->regs[VCPU_REGS_RAX] = ret;
1406         kvm_x86_ops->decache_regs(vcpu);
1407         return 1;
1408 }
1409 EXPORT_SYMBOL_GPL(kvm_hypercall);
1410
1411 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1412 {
1413         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1414 }
1415
1416 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1417 {
1418         struct descriptor_table dt = { limit, base };
1419
1420         kvm_x86_ops->set_gdt(vcpu, &dt);
1421 }
1422
1423 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1424 {
1425         struct descriptor_table dt = { limit, base };
1426
1427         kvm_x86_ops->set_idt(vcpu, &dt);
1428 }
1429
1430 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1431                    unsigned long *rflags)
1432 {
1433         lmsw(vcpu, msw);
1434         *rflags = kvm_x86_ops->get_rflags(vcpu);
1435 }
1436
1437 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1438 {
1439         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1440         switch (cr) {
1441         case 0:
1442                 return vcpu->cr0;
1443         case 2:
1444                 return vcpu->cr2;
1445         case 3:
1446                 return vcpu->cr3;
1447         case 4:
1448                 return vcpu->cr4;
1449         default:
1450                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1451                 return 0;
1452         }
1453 }
1454
1455 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1456                      unsigned long *rflags)
1457 {
1458         switch (cr) {
1459         case 0:
1460                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1461                 *rflags = kvm_x86_ops->get_rflags(vcpu);
1462                 break;
1463         case 2:
1464                 vcpu->cr2 = val;
1465                 break;
1466         case 3:
1467                 set_cr3(vcpu, val);
1468                 break;
1469         case 4:
1470                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1471                 break;
1472         default:
1473                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1474         }
1475 }
1476
1477 /*
1478  * Register the para guest with the host:
1479  */
1480 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1481 {
1482         struct kvm_vcpu_para_state *para_state;
1483         hpa_t para_state_hpa, hypercall_hpa;
1484         struct page *para_state_page;
1485         unsigned char *hypercall;
1486         gpa_t hypercall_gpa;
1487
1488         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1489         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1490
1491         /*
1492          * Needs to be page aligned:
1493          */
1494         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1495                 goto err_gp;
1496
1497         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1498         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1499         if (is_error_hpa(para_state_hpa))
1500                 goto err_gp;
1501
1502         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1503         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1504         para_state = kmap(para_state_page);
1505
1506         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1507         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1508
1509         para_state->host_version = KVM_PARA_API_VERSION;
1510         /*
1511          * We cannot support guests that try to register themselves
1512          * with a newer API version than the host supports:
1513          */
1514         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1515                 para_state->ret = -KVM_EINVAL;
1516                 goto err_kunmap_skip;
1517         }
1518
1519         hypercall_gpa = para_state->hypercall_gpa;
1520         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1521         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1522         if (is_error_hpa(hypercall_hpa)) {
1523                 para_state->ret = -KVM_EINVAL;
1524                 goto err_kunmap_skip;
1525         }
1526
1527         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1528         vcpu->para_state_page = para_state_page;
1529         vcpu->para_state_gpa = para_state_gpa;
1530         vcpu->hypercall_gpa = hypercall_gpa;
1531
1532         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1533         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1534                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1535         kvm_x86_ops->patch_hypercall(vcpu, hypercall);
1536         kunmap_atomic(hypercall, KM_USER1);
1537
1538         para_state->ret = 0;
1539 err_kunmap_skip:
1540         kunmap(para_state_page);
1541         return 0;
1542 err_gp:
1543         return 1;
1544 }
1545
1546 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1547 {
1548         u64 data;
1549
1550         switch (msr) {
1551         case 0xc0010010: /* SYSCFG */
1552         case 0xc0010015: /* HWCR */
1553         case MSR_IA32_PLATFORM_ID:
1554         case MSR_IA32_P5_MC_ADDR:
1555         case MSR_IA32_P5_MC_TYPE:
1556         case MSR_IA32_MC0_CTL:
1557         case MSR_IA32_MCG_STATUS:
1558         case MSR_IA32_MCG_CAP:
1559         case MSR_IA32_MC0_MISC:
1560         case MSR_IA32_MC0_MISC+4:
1561         case MSR_IA32_MC0_MISC+8:
1562         case MSR_IA32_MC0_MISC+12:
1563         case MSR_IA32_MC0_MISC+16:
1564         case MSR_IA32_UCODE_REV:
1565         case MSR_IA32_PERF_STATUS:
1566         case MSR_IA32_EBL_CR_POWERON:
1567                 /* MTRR registers */
1568         case 0xfe:
1569         case 0x200 ... 0x2ff:
1570                 data = 0;
1571                 break;
1572         case 0xcd: /* fsb frequency */
1573                 data = 3;
1574                 break;
1575         case MSR_IA32_APICBASE:
1576                 data = kvm_get_apic_base(vcpu);
1577                 break;
1578         case MSR_IA32_MISC_ENABLE:
1579                 data = vcpu->ia32_misc_enable_msr;
1580                 break;
1581 #ifdef CONFIG_X86_64
1582         case MSR_EFER:
1583                 data = vcpu->shadow_efer;
1584                 break;
1585 #endif
1586         default:
1587                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1588                 return 1;
1589         }
1590         *pdata = data;
1591         return 0;
1592 }
1593 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1594
1595 /*
1596  * Reads an msr value (of 'msr_index') into 'pdata'.
1597  * Returns 0 on success, non-0 otherwise.
1598  * Assumes vcpu_load() was already called.
1599  */
1600 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1601 {
1602         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1603 }
1604
1605 #ifdef CONFIG_X86_64
1606
1607 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1608 {
1609         if (efer & EFER_RESERVED_BITS) {
1610                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1611                        efer);
1612                 inject_gp(vcpu);
1613                 return;
1614         }
1615
1616         if (is_paging(vcpu)
1617             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1618                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1619                 inject_gp(vcpu);
1620                 return;
1621         }
1622
1623         kvm_x86_ops->set_efer(vcpu, efer);
1624
1625         efer &= ~EFER_LMA;
1626         efer |= vcpu->shadow_efer & EFER_LMA;
1627
1628         vcpu->shadow_efer = efer;
1629 }
1630
1631 #endif
1632
1633 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1634 {
1635         switch (msr) {
1636 #ifdef CONFIG_X86_64
1637         case MSR_EFER:
1638                 set_efer(vcpu, data);
1639                 break;
1640 #endif
1641         case MSR_IA32_MC0_STATUS:
1642                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1643                        __FUNCTION__, data);
1644                 break;
1645         case MSR_IA32_MCG_STATUS:
1646                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1647                         __FUNCTION__, data);
1648                 break;
1649         case MSR_IA32_UCODE_REV:
1650         case MSR_IA32_UCODE_WRITE:
1651         case 0x200 ... 0x2ff: /* MTRRs */
1652                 break;
1653         case MSR_IA32_APICBASE:
1654                 kvm_set_apic_base(vcpu, data);
1655                 break;
1656         case MSR_IA32_MISC_ENABLE:
1657                 vcpu->ia32_misc_enable_msr = data;
1658                 break;
1659         /*
1660          * This is the 'probe whether the host is KVM' logic:
1661          */
1662         case MSR_KVM_API_MAGIC:
1663                 return vcpu_register_para(vcpu, data);
1664
1665         default:
1666                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1667                 return 1;
1668         }
1669         return 0;
1670 }
1671 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1672
1673 /*
1674  * Writes msr value into into the appropriate "register".
1675  * Returns 0 on success, non-0 otherwise.
1676  * Assumes vcpu_load() was already called.
1677  */
1678 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1679 {
1680         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
1681 }
1682
1683 void kvm_resched(struct kvm_vcpu *vcpu)
1684 {
1685         if (!need_resched())
1686                 return;
1687         cond_resched();
1688 }
1689 EXPORT_SYMBOL_GPL(kvm_resched);
1690
1691 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1692 {
1693         int i;
1694         u32 function;
1695         struct kvm_cpuid_entry *e, *best;
1696
1697         kvm_x86_ops->cache_regs(vcpu);
1698         function = vcpu->regs[VCPU_REGS_RAX];
1699         vcpu->regs[VCPU_REGS_RAX] = 0;
1700         vcpu->regs[VCPU_REGS_RBX] = 0;
1701         vcpu->regs[VCPU_REGS_RCX] = 0;
1702         vcpu->regs[VCPU_REGS_RDX] = 0;
1703         best = NULL;
1704         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1705                 e = &vcpu->cpuid_entries[i];
1706                 if (e->function == function) {
1707                         best = e;
1708                         break;
1709                 }
1710                 /*
1711                  * Both basic or both extended?
1712                  */
1713                 if (((e->function ^ function) & 0x80000000) == 0)
1714                         if (!best || e->function > best->function)
1715                                 best = e;
1716         }
1717         if (best) {
1718                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1719                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1720                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1721                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1722         }
1723         kvm_x86_ops->decache_regs(vcpu);
1724         kvm_x86_ops->skip_emulated_instruction(vcpu);
1725 }
1726 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1727
1728 static int pio_copy_data(struct kvm_vcpu *vcpu)
1729 {
1730         void *p = vcpu->pio_data;
1731         void *q;
1732         unsigned bytes;
1733         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1734
1735         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1736                  PAGE_KERNEL);
1737         if (!q) {
1738                 free_pio_guest_pages(vcpu);
1739                 return -ENOMEM;
1740         }
1741         q += vcpu->pio.guest_page_offset;
1742         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1743         if (vcpu->pio.in)
1744                 memcpy(q, p, bytes);
1745         else
1746                 memcpy(p, q, bytes);
1747         q -= vcpu->pio.guest_page_offset;
1748         vunmap(q);
1749         free_pio_guest_pages(vcpu);
1750         return 0;
1751 }
1752
1753 static int complete_pio(struct kvm_vcpu *vcpu)
1754 {
1755         struct kvm_pio_request *io = &vcpu->pio;
1756         long delta;
1757         int r;
1758
1759         kvm_x86_ops->cache_regs(vcpu);
1760
1761         if (!io->string) {
1762                 if (io->in)
1763                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1764                                io->size);
1765         } else {
1766                 if (io->in) {
1767                         r = pio_copy_data(vcpu);
1768                         if (r) {
1769                                 kvm_x86_ops->cache_regs(vcpu);
1770                                 return r;
1771                         }
1772                 }
1773
1774                 delta = 1;
1775                 if (io->rep) {
1776                         delta *= io->cur_count;
1777                         /*
1778                          * The size of the register should really depend on
1779                          * current address size.
1780                          */
1781                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1782                 }
1783                 if (io->down)
1784                         delta = -delta;
1785                 delta *= io->size;
1786                 if (io->in)
1787                         vcpu->regs[VCPU_REGS_RDI] += delta;
1788                 else
1789                         vcpu->regs[VCPU_REGS_RSI] += delta;
1790         }
1791
1792         kvm_x86_ops->decache_regs(vcpu);
1793
1794         io->count -= io->cur_count;
1795         io->cur_count = 0;
1796
1797         return 0;
1798 }
1799
1800 static void kernel_pio(struct kvm_io_device *pio_dev,
1801                        struct kvm_vcpu *vcpu,
1802                        void *pd)
1803 {
1804         /* TODO: String I/O for in kernel device */
1805
1806         mutex_lock(&vcpu->kvm->lock);
1807         if (vcpu->pio.in)
1808                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1809                                   vcpu->pio.size,
1810                                   pd);
1811         else
1812                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1813                                    vcpu->pio.size,
1814                                    pd);
1815         mutex_unlock(&vcpu->kvm->lock);
1816 }
1817
1818 static void pio_string_write(struct kvm_io_device *pio_dev,
1819                              struct kvm_vcpu *vcpu)
1820 {
1821         struct kvm_pio_request *io = &vcpu->pio;
1822         void *pd = vcpu->pio_data;
1823         int i;
1824
1825         mutex_lock(&vcpu->kvm->lock);
1826         for (i = 0; i < io->cur_count; i++) {
1827                 kvm_iodevice_write(pio_dev, io->port,
1828                                    io->size,
1829                                    pd);
1830                 pd += io->size;
1831         }
1832         mutex_unlock(&vcpu->kvm->lock);
1833 }
1834
1835 int kvm_emulate_pio (struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1836                   int size, unsigned port)
1837 {
1838         struct kvm_io_device *pio_dev;
1839
1840         vcpu->run->exit_reason = KVM_EXIT_IO;
1841         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1842         vcpu->run->io.size = vcpu->pio.size = size;
1843         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1844         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1845         vcpu->run->io.port = vcpu->pio.port = port;
1846         vcpu->pio.in = in;
1847         vcpu->pio.string = 0;
1848         vcpu->pio.down = 0;
1849         vcpu->pio.guest_page_offset = 0;
1850         vcpu->pio.rep = 0;
1851
1852         kvm_x86_ops->cache_regs(vcpu);
1853         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1854         kvm_x86_ops->decache_regs(vcpu);
1855
1856         kvm_x86_ops->skip_emulated_instruction(vcpu);
1857
1858         pio_dev = vcpu_find_pio_dev(vcpu, port);
1859         if (pio_dev) {
1860                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1861                 complete_pio(vcpu);
1862                 return 1;
1863         }
1864         return 0;
1865 }
1866 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1867
1868 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1869                   int size, unsigned long count, int down,
1870                   gva_t address, int rep, unsigned port)
1871 {
1872         unsigned now, in_page;
1873         int i, ret = 0;
1874         int nr_pages = 1;
1875         struct page *page;
1876         struct kvm_io_device *pio_dev;
1877
1878         vcpu->run->exit_reason = KVM_EXIT_IO;
1879         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1880         vcpu->run->io.size = vcpu->pio.size = size;
1881         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1882         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1883         vcpu->run->io.port = vcpu->pio.port = port;
1884         vcpu->pio.in = in;
1885         vcpu->pio.string = 1;
1886         vcpu->pio.down = down;
1887         vcpu->pio.guest_page_offset = offset_in_page(address);
1888         vcpu->pio.rep = rep;
1889
1890         if (!count) {
1891                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1892                 return 1;
1893         }
1894
1895         if (!down)
1896                 in_page = PAGE_SIZE - offset_in_page(address);
1897         else
1898                 in_page = offset_in_page(address) + size;
1899         now = min(count, (unsigned long)in_page / size);
1900         if (!now) {
1901                 /*
1902                  * String I/O straddles page boundary.  Pin two guest pages
1903                  * so that we satisfy atomicity constraints.  Do just one
1904                  * transaction to avoid complexity.
1905                  */
1906                 nr_pages = 2;
1907                 now = 1;
1908         }
1909         if (down) {
1910                 /*
1911                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1912                  */
1913                 pr_unimpl(vcpu, "guest string pio down\n");
1914                 inject_gp(vcpu);
1915                 return 1;
1916         }
1917         vcpu->run->io.count = now;
1918         vcpu->pio.cur_count = now;
1919
1920         if (vcpu->pio.cur_count == vcpu->pio.count)
1921                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1922
1923         for (i = 0; i < nr_pages; ++i) {
1924                 mutex_lock(&vcpu->kvm->lock);
1925                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1926                 if (page)
1927                         get_page(page);
1928                 vcpu->pio.guest_pages[i] = page;
1929                 mutex_unlock(&vcpu->kvm->lock);
1930                 if (!page) {
1931                         inject_gp(vcpu);
1932                         free_pio_guest_pages(vcpu);
1933                         return 1;
1934                 }
1935         }
1936
1937         pio_dev = vcpu_find_pio_dev(vcpu, port);
1938         if (!vcpu->pio.in) {
1939                 /* string PIO write */
1940                 ret = pio_copy_data(vcpu);
1941                 if (ret >= 0 && pio_dev) {
1942                         pio_string_write(pio_dev, vcpu);
1943                         complete_pio(vcpu);
1944                         if (vcpu->pio.count == 0)
1945                                 ret = 1;
1946                 }
1947         } else if (pio_dev)
1948                 pr_unimpl(vcpu, "no string pio read support yet, "
1949                        "port %x size %d count %ld\n",
1950                         port, size, count);
1951
1952         return ret;
1953 }
1954 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1955
1956 /*
1957  * Check if userspace requested an interrupt window, and that the
1958  * interrupt window is open.
1959  *
1960  * No need to exit to userspace if we already have an interrupt queued.
1961  */
1962 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1963                                           struct kvm_run *kvm_run)
1964 {
1965         return (!vcpu->irq_summary &&
1966                 kvm_run->request_interrupt_window &&
1967                 vcpu->interrupt_window_open &&
1968                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
1969 }
1970
1971 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1972                               struct kvm_run *kvm_run)
1973 {
1974         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
1975         kvm_run->cr8 = get_cr8(vcpu);
1976         kvm_run->apic_base = kvm_get_apic_base(vcpu);
1977         if (irqchip_in_kernel(vcpu->kvm))
1978                 kvm_run->ready_for_interrupt_injection = 1;
1979         else
1980                 kvm_run->ready_for_interrupt_injection =
1981                                         (vcpu->interrupt_window_open &&
1982                                          vcpu->irq_summary == 0);
1983 }
1984
1985 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1986 {
1987         int r;
1988
1989         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
1990                 printk("vcpu %d received sipi with vector # %x\n",
1991                        vcpu->vcpu_id, vcpu->sipi_vector);
1992                 kvm_lapic_reset(vcpu);
1993                 kvm_x86_ops->vcpu_reset(vcpu);
1994                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
1995         }
1996
1997 preempted:
1998         if (vcpu->guest_debug.enabled)
1999                 kvm_x86_ops->guest_debug_pre(vcpu);
2000
2001 again:
2002         r = kvm_mmu_reload(vcpu);
2003         if (unlikely(r))
2004                 goto out;
2005
2006         preempt_disable();
2007
2008         kvm_x86_ops->prepare_guest_switch(vcpu);
2009         kvm_load_guest_fpu(vcpu);
2010
2011         local_irq_disable();
2012
2013         if (signal_pending(current)) {
2014                 local_irq_enable();
2015                 preempt_enable();
2016                 r = -EINTR;
2017                 kvm_run->exit_reason = KVM_EXIT_INTR;
2018                 ++vcpu->stat.signal_exits;
2019                 goto out;
2020         }
2021
2022         if (irqchip_in_kernel(vcpu->kvm))
2023                 kvm_x86_ops->inject_pending_irq(vcpu);
2024         else if (!vcpu->mmio_read_completed)
2025                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2026
2027         vcpu->guest_mode = 1;
2028         kvm_guest_enter();
2029
2030         if (vcpu->requests)
2031                 if (test_and_clear_bit(KVM_TLB_FLUSH, &vcpu->requests))
2032                         kvm_x86_ops->tlb_flush(vcpu);
2033
2034         kvm_x86_ops->run(vcpu, kvm_run);
2035
2036         vcpu->guest_mode = 0;
2037         local_irq_enable();
2038
2039         ++vcpu->stat.exits;
2040
2041         /*
2042          * We must have an instruction between local_irq_enable() and
2043          * kvm_guest_exit(), so the timer interrupt isn't delayed by
2044          * the interrupt shadow.  The stat.exits increment will do nicely.
2045          * But we need to prevent reordering, hence this barrier():
2046          */
2047         barrier();
2048
2049         kvm_guest_exit();
2050
2051         preempt_enable();
2052
2053         /*
2054          * Profile KVM exit RIPs:
2055          */
2056         if (unlikely(prof_on == KVM_PROFILING)) {
2057                 kvm_x86_ops->cache_regs(vcpu);
2058                 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
2059         }
2060
2061         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2062
2063         if (r > 0) {
2064                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2065                         r = -EINTR;
2066                         kvm_run->exit_reason = KVM_EXIT_INTR;
2067                         ++vcpu->stat.request_irq_exits;
2068                         goto out;
2069                 }
2070                 if (!need_resched()) {
2071                         ++vcpu->stat.light_exits;
2072                         goto again;
2073                 }
2074         }
2075
2076 out:
2077         if (r > 0) {
2078                 kvm_resched(vcpu);
2079                 goto preempted;
2080         }
2081
2082         post_kvm_run_save(vcpu, kvm_run);
2083
2084         return r;
2085 }
2086
2087
2088 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2089 {
2090         int r;
2091         sigset_t sigsaved;
2092
2093         vcpu_load(vcpu);
2094
2095         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2096                 kvm_vcpu_block(vcpu);
2097                 vcpu_put(vcpu);
2098                 return -EAGAIN;
2099         }
2100
2101         if (vcpu->sigset_active)
2102                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2103
2104         /* re-sync apic's tpr */
2105         if (!irqchip_in_kernel(vcpu->kvm))
2106                 set_cr8(vcpu, kvm_run->cr8);
2107
2108         if (vcpu->pio.cur_count) {
2109                 r = complete_pio(vcpu);
2110                 if (r)
2111                         goto out;
2112         }
2113
2114         if (vcpu->mmio_needed) {
2115                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2116                 vcpu->mmio_read_completed = 1;
2117                 vcpu->mmio_needed = 0;
2118                 r = emulate_instruction(vcpu, kvm_run,
2119                                         vcpu->mmio_fault_cr2, 0);
2120                 if (r == EMULATE_DO_MMIO) {
2121                         /*
2122                          * Read-modify-write.  Back to userspace.
2123                          */
2124                         r = 0;
2125                         goto out;
2126                 }
2127         }
2128
2129         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2130                 kvm_x86_ops->cache_regs(vcpu);
2131                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2132                 kvm_x86_ops->decache_regs(vcpu);
2133         }
2134
2135         r = __vcpu_run(vcpu, kvm_run);
2136
2137 out:
2138         if (vcpu->sigset_active)
2139                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2140
2141         vcpu_put(vcpu);
2142         return r;
2143 }
2144
2145 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2146                                    struct kvm_regs *regs)
2147 {
2148         vcpu_load(vcpu);
2149
2150         kvm_x86_ops->cache_regs(vcpu);
2151
2152         regs->rax = vcpu->regs[VCPU_REGS_RAX];
2153         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2154         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2155         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2156         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2157         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2158         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2159         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2160 #ifdef CONFIG_X86_64
2161         regs->r8 = vcpu->regs[VCPU_REGS_R8];
2162         regs->r9 = vcpu->regs[VCPU_REGS_R9];
2163         regs->r10 = vcpu->regs[VCPU_REGS_R10];
2164         regs->r11 = vcpu->regs[VCPU_REGS_R11];
2165         regs->r12 = vcpu->regs[VCPU_REGS_R12];
2166         regs->r13 = vcpu->regs[VCPU_REGS_R13];
2167         regs->r14 = vcpu->regs[VCPU_REGS_R14];
2168         regs->r15 = vcpu->regs[VCPU_REGS_R15];
2169 #endif
2170
2171         regs->rip = vcpu->rip;
2172         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2173
2174         /*
2175          * Don't leak debug flags in case they were set for guest debugging
2176          */
2177         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2178                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2179
2180         vcpu_put(vcpu);
2181
2182         return 0;
2183 }
2184
2185 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2186                                    struct kvm_regs *regs)
2187 {
2188         vcpu_load(vcpu);
2189
2190         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2191         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2192         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2193         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2194         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2195         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2196         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2197         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2198 #ifdef CONFIG_X86_64
2199         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2200         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2201         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2202         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2203         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2204         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2205         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2206         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2207 #endif
2208
2209         vcpu->rip = regs->rip;
2210         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2211
2212         kvm_x86_ops->decache_regs(vcpu);
2213
2214         vcpu_put(vcpu);
2215
2216         return 0;
2217 }
2218
2219 static void get_segment(struct kvm_vcpu *vcpu,
2220                         struct kvm_segment *var, int seg)
2221 {
2222         return kvm_x86_ops->get_segment(vcpu, var, seg);
2223 }
2224
2225 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2226                                     struct kvm_sregs *sregs)
2227 {
2228         struct descriptor_table dt;
2229         int pending_vec;
2230
2231         vcpu_load(vcpu);
2232
2233         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2234         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2235         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2236         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2237         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2238         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2239
2240         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2241         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2242
2243         kvm_x86_ops->get_idt(vcpu, &dt);
2244         sregs->idt.limit = dt.limit;
2245         sregs->idt.base = dt.base;
2246         kvm_x86_ops->get_gdt(vcpu, &dt);
2247         sregs->gdt.limit = dt.limit;
2248         sregs->gdt.base = dt.base;
2249
2250         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2251         sregs->cr0 = vcpu->cr0;
2252         sregs->cr2 = vcpu->cr2;
2253         sregs->cr3 = vcpu->cr3;
2254         sregs->cr4 = vcpu->cr4;
2255         sregs->cr8 = get_cr8(vcpu);
2256         sregs->efer = vcpu->shadow_efer;
2257         sregs->apic_base = kvm_get_apic_base(vcpu);
2258
2259         if (irqchip_in_kernel(vcpu->kvm)) {
2260                 memset(sregs->interrupt_bitmap, 0,
2261                        sizeof sregs->interrupt_bitmap);
2262                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2263                 if (pending_vec >= 0)
2264                         set_bit(pending_vec, (unsigned long *)sregs->interrupt_bitmap);
2265         } else
2266                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2267                        sizeof sregs->interrupt_bitmap);
2268
2269         vcpu_put(vcpu);
2270
2271         return 0;
2272 }
2273
2274 static void set_segment(struct kvm_vcpu *vcpu,
2275                         struct kvm_segment *var, int seg)
2276 {
2277         return kvm_x86_ops->set_segment(vcpu, var, seg);
2278 }
2279
2280 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2281                                     struct kvm_sregs *sregs)
2282 {
2283         int mmu_reset_needed = 0;
2284         int i, pending_vec, max_bits;
2285         struct descriptor_table dt;
2286
2287         vcpu_load(vcpu);
2288
2289         dt.limit = sregs->idt.limit;
2290         dt.base = sregs->idt.base;
2291         kvm_x86_ops->set_idt(vcpu, &dt);
2292         dt.limit = sregs->gdt.limit;
2293         dt.base = sregs->gdt.base;
2294         kvm_x86_ops->set_gdt(vcpu, &dt);
2295
2296         vcpu->cr2 = sregs->cr2;
2297         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2298         vcpu->cr3 = sregs->cr3;
2299
2300         set_cr8(vcpu, sregs->cr8);
2301
2302         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2303 #ifdef CONFIG_X86_64
2304         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2305 #endif
2306         kvm_set_apic_base(vcpu, sregs->apic_base);
2307
2308         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2309
2310         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2311         vcpu->cr0 = sregs->cr0;
2312         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2313
2314         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2315         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2316         if (!is_long_mode(vcpu) && is_pae(vcpu))
2317                 load_pdptrs(vcpu, vcpu->cr3);
2318
2319         if (mmu_reset_needed)
2320                 kvm_mmu_reset_context(vcpu);
2321
2322         if (!irqchip_in_kernel(vcpu->kvm)) {
2323                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2324                        sizeof vcpu->irq_pending);
2325                 vcpu->irq_summary = 0;
2326                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2327                         if (vcpu->irq_pending[i])
2328                                 __set_bit(i, &vcpu->irq_summary);
2329         } else {
2330                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2331                 pending_vec = find_first_bit(
2332                         (const unsigned long *)sregs->interrupt_bitmap,
2333                         max_bits);
2334                 /* Only pending external irq is handled here */
2335                 if (pending_vec < max_bits) {
2336                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2337                         printk("Set back pending irq %d\n", pending_vec);
2338                 }
2339         }
2340
2341         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2342         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2343         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2344         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2345         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2346         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2347
2348         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2349         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2350
2351         vcpu_put(vcpu);
2352
2353         return 0;
2354 }
2355
2356 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2357 {
2358         struct kvm_segment cs;
2359
2360         get_segment(vcpu, &cs, VCPU_SREG_CS);
2361         *db = cs.db;
2362         *l = cs.l;
2363 }
2364 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2365
2366 /*
2367  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2368  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2369  *
2370  * This list is modified at module load time to reflect the
2371  * capabilities of the host cpu.
2372  */
2373 static u32 msrs_to_save[] = {
2374         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2375         MSR_K6_STAR,
2376 #ifdef CONFIG_X86_64
2377         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2378 #endif
2379         MSR_IA32_TIME_STAMP_COUNTER,
2380 };
2381
2382 static unsigned num_msrs_to_save;
2383
2384 static u32 emulated_msrs[] = {
2385         MSR_IA32_MISC_ENABLE,
2386 };
2387
2388 static __init void kvm_init_msr_list(void)
2389 {
2390         u32 dummy[2];
2391         unsigned i, j;
2392
2393         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2394                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2395                         continue;
2396                 if (j < i)
2397                         msrs_to_save[j] = msrs_to_save[i];
2398                 j++;
2399         }
2400         num_msrs_to_save = j;
2401 }
2402
2403 /*
2404  * Adapt set_msr() to msr_io()'s calling convention
2405  */
2406 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2407 {
2408         return kvm_set_msr(vcpu, index, *data);
2409 }
2410
2411 /*
2412  * Read or write a bunch of msrs. All parameters are kernel addresses.
2413  *
2414  * @return number of msrs set successfully.
2415  */
2416 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2417                     struct kvm_msr_entry *entries,
2418                     int (*do_msr)(struct kvm_vcpu *vcpu,
2419                                   unsigned index, u64 *data))
2420 {
2421         int i;
2422
2423         vcpu_load(vcpu);
2424
2425         for (i = 0; i < msrs->nmsrs; ++i)
2426                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2427                         break;
2428
2429         vcpu_put(vcpu);
2430
2431         return i;
2432 }
2433
2434 /*
2435  * Read or write a bunch of msrs. Parameters are user addresses.
2436  *
2437  * @return number of msrs set successfully.
2438  */
2439 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2440                   int (*do_msr)(struct kvm_vcpu *vcpu,
2441                                 unsigned index, u64 *data),
2442                   int writeback)
2443 {
2444         struct kvm_msrs msrs;
2445         struct kvm_msr_entry *entries;
2446         int r, n;
2447         unsigned size;
2448
2449         r = -EFAULT;
2450         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2451                 goto out;
2452
2453         r = -E2BIG;
2454         if (msrs.nmsrs >= MAX_IO_MSRS)
2455                 goto out;
2456
2457         r = -ENOMEM;
2458         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2459         entries = vmalloc(size);
2460         if (!entries)
2461                 goto out;
2462
2463         r = -EFAULT;
2464         if (copy_from_user(entries, user_msrs->entries, size))
2465                 goto out_free;
2466
2467         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2468         if (r < 0)
2469                 goto out_free;
2470
2471         r = -EFAULT;
2472         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2473                 goto out_free;
2474
2475         r = n;
2476
2477 out_free:
2478         vfree(entries);
2479 out:
2480         return r;
2481 }
2482
2483 /*
2484  * Translate a guest virtual address to a guest physical address.
2485  */
2486 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2487                                     struct kvm_translation *tr)
2488 {
2489         unsigned long vaddr = tr->linear_address;
2490         gpa_t gpa;
2491
2492         vcpu_load(vcpu);
2493         mutex_lock(&vcpu->kvm->lock);
2494         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2495         tr->physical_address = gpa;
2496         tr->valid = gpa != UNMAPPED_GVA;
2497         tr->writeable = 1;
2498         tr->usermode = 0;
2499         mutex_unlock(&vcpu->kvm->lock);
2500         vcpu_put(vcpu);
2501
2502         return 0;
2503 }
2504
2505 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2506                                     struct kvm_interrupt *irq)
2507 {
2508         if (irq->irq < 0 || irq->irq >= 256)
2509                 return -EINVAL;
2510         if (irqchip_in_kernel(vcpu->kvm))
2511                 return -ENXIO;
2512         vcpu_load(vcpu);
2513
2514         set_bit(irq->irq, vcpu->irq_pending);
2515         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2516
2517         vcpu_put(vcpu);
2518
2519         return 0;
2520 }
2521
2522 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2523                                       struct kvm_debug_guest *dbg)
2524 {
2525         int r;
2526
2527         vcpu_load(vcpu);
2528
2529         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2530
2531         vcpu_put(vcpu);
2532
2533         return r;
2534 }
2535
2536 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2537                                     unsigned long address,
2538                                     int *type)
2539 {
2540         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2541         unsigned long pgoff;
2542         struct page *page;
2543
2544         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2545         if (pgoff == 0)
2546                 page = virt_to_page(vcpu->run);
2547         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2548                 page = virt_to_page(vcpu->pio_data);
2549         else
2550                 return NOPAGE_SIGBUS;
2551         get_page(page);
2552         if (type != NULL)
2553                 *type = VM_FAULT_MINOR;
2554
2555         return page;
2556 }
2557
2558 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2559         .nopage = kvm_vcpu_nopage,
2560 };
2561
2562 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2563 {
2564         vma->vm_ops = &kvm_vcpu_vm_ops;
2565         return 0;
2566 }
2567
2568 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2569 {
2570         struct kvm_vcpu *vcpu = filp->private_data;
2571
2572         fput(vcpu->kvm->filp);
2573         return 0;
2574 }
2575
2576 static struct file_operations kvm_vcpu_fops = {
2577         .release        = kvm_vcpu_release,
2578         .unlocked_ioctl = kvm_vcpu_ioctl,
2579         .compat_ioctl   = kvm_vcpu_ioctl,
2580         .mmap           = kvm_vcpu_mmap,
2581 };
2582
2583 /*
2584  * Allocates an inode for the vcpu.
2585  */
2586 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2587 {
2588         int fd, r;
2589         struct inode *inode;
2590         struct file *file;
2591
2592         r = anon_inode_getfd(&fd, &inode, &file,
2593                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2594         if (r)
2595                 return r;
2596         atomic_inc(&vcpu->kvm->filp->f_count);
2597         return fd;
2598 }
2599
2600 /*
2601  * Creates some virtual cpus.  Good luck creating more than one.
2602  */
2603 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2604 {
2605         int r;
2606         struct kvm_vcpu *vcpu;
2607
2608         if (!valid_vcpu(n))
2609                 return -EINVAL;
2610
2611         vcpu = kvm_x86_ops->vcpu_create(kvm, n);
2612         if (IS_ERR(vcpu))
2613                 return PTR_ERR(vcpu);
2614
2615         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2616
2617         /* We do fxsave: this must be aligned. */
2618         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2619
2620         vcpu_load(vcpu);
2621         r = kvm_mmu_setup(vcpu);
2622         vcpu_put(vcpu);
2623         if (r < 0)
2624                 goto free_vcpu;
2625
2626         mutex_lock(&kvm->lock);
2627         if (kvm->vcpus[n]) {
2628                 r = -EEXIST;
2629                 mutex_unlock(&kvm->lock);
2630                 goto mmu_unload;
2631         }
2632         kvm->vcpus[n] = vcpu;
2633         mutex_unlock(&kvm->lock);
2634
2635         /* Now it's all set up, let userspace reach it */
2636         r = create_vcpu_fd(vcpu);
2637         if (r < 0)
2638                 goto unlink;
2639         return r;
2640
2641 unlink:
2642         mutex_lock(&kvm->lock);
2643         kvm->vcpus[n] = NULL;
2644         mutex_unlock(&kvm->lock);
2645
2646 mmu_unload:
2647         vcpu_load(vcpu);
2648         kvm_mmu_unload(vcpu);
2649         vcpu_put(vcpu);
2650
2651 free_vcpu:
2652         kvm_x86_ops->vcpu_free(vcpu);
2653         return r;
2654 }
2655
2656 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2657 {
2658         u64 efer;
2659         int i;
2660         struct kvm_cpuid_entry *e, *entry;
2661
2662         rdmsrl(MSR_EFER, efer);
2663         entry = NULL;
2664         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2665                 e = &vcpu->cpuid_entries[i];
2666                 if (e->function == 0x80000001) {
2667                         entry = e;
2668                         break;
2669                 }
2670         }
2671         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2672                 entry->edx &= ~(1 << 20);
2673                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2674         }
2675 }
2676
2677 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2678                                     struct kvm_cpuid *cpuid,
2679                                     struct kvm_cpuid_entry __user *entries)
2680 {
2681         int r;
2682
2683         r = -E2BIG;
2684         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2685                 goto out;
2686         r = -EFAULT;
2687         if (copy_from_user(&vcpu->cpuid_entries, entries,
2688                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2689                 goto out;
2690         vcpu->cpuid_nent = cpuid->nent;
2691         cpuid_fix_nx_cap(vcpu);
2692         return 0;
2693
2694 out:
2695         return r;
2696 }
2697
2698 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2699 {
2700         if (sigset) {
2701                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2702                 vcpu->sigset_active = 1;
2703                 vcpu->sigset = *sigset;
2704         } else
2705                 vcpu->sigset_active = 0;
2706         return 0;
2707 }
2708
2709 /*
2710  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2711  * we have asm/x86/processor.h
2712  */
2713 struct fxsave {
2714         u16     cwd;
2715         u16     swd;
2716         u16     twd;
2717         u16     fop;
2718         u64     rip;
2719         u64     rdp;
2720         u32     mxcsr;
2721         u32     mxcsr_mask;
2722         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2723 #ifdef CONFIG_X86_64
2724         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2725 #else
2726         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2727 #endif
2728 };
2729
2730 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2731 {
2732         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2733
2734         vcpu_load(vcpu);
2735
2736         memcpy(fpu->fpr, fxsave->st_space, 128);
2737         fpu->fcw = fxsave->cwd;
2738         fpu->fsw = fxsave->swd;
2739         fpu->ftwx = fxsave->twd;
2740         fpu->last_opcode = fxsave->fop;
2741         fpu->last_ip = fxsave->rip;
2742         fpu->last_dp = fxsave->rdp;
2743         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2744
2745         vcpu_put(vcpu);
2746
2747         return 0;
2748 }
2749
2750 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2751 {
2752         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2753
2754         vcpu_load(vcpu);
2755
2756         memcpy(fxsave->st_space, fpu->fpr, 128);
2757         fxsave->cwd = fpu->fcw;
2758         fxsave->swd = fpu->fsw;
2759         fxsave->twd = fpu->ftwx;
2760         fxsave->fop = fpu->last_opcode;
2761         fxsave->rip = fpu->last_ip;
2762         fxsave->rdp = fpu->last_dp;
2763         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2764
2765         vcpu_put(vcpu);
2766
2767         return 0;
2768 }
2769
2770 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2771                                     struct kvm_lapic_state *s)
2772 {
2773         vcpu_load(vcpu);
2774         memcpy(s->regs, vcpu->apic->regs, sizeof *s);
2775         vcpu_put(vcpu);
2776
2777         return 0;
2778 }
2779
2780 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2781                                     struct kvm_lapic_state *s)
2782 {
2783         vcpu_load(vcpu);
2784         memcpy(vcpu->apic->regs, s->regs, sizeof *s);
2785         kvm_apic_post_state_restore(vcpu);
2786         vcpu_put(vcpu);
2787
2788         return 0;
2789 }
2790
2791 static long kvm_vcpu_ioctl(struct file *filp,
2792                            unsigned int ioctl, unsigned long arg)
2793 {
2794         struct kvm_vcpu *vcpu = filp->private_data;
2795         void __user *argp = (void __user *)arg;
2796         int r = -EINVAL;
2797
2798         switch (ioctl) {
2799         case KVM_RUN:
2800                 r = -EINVAL;
2801                 if (arg)
2802                         goto out;
2803                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2804                 break;
2805         case KVM_GET_REGS: {
2806                 struct kvm_regs kvm_regs;
2807
2808                 memset(&kvm_regs, 0, sizeof kvm_regs);
2809                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2810                 if (r)
2811                         goto out;
2812                 r = -EFAULT;
2813                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2814                         goto out;
2815                 r = 0;
2816                 break;
2817         }
2818         case KVM_SET_REGS: {
2819                 struct kvm_regs kvm_regs;
2820
2821                 r = -EFAULT;
2822                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2823                         goto out;
2824                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2825                 if (r)
2826                         goto out;
2827                 r = 0;
2828                 break;
2829         }
2830         case KVM_GET_SREGS: {
2831                 struct kvm_sregs kvm_sregs;
2832
2833                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2834                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2835                 if (r)
2836                         goto out;
2837                 r = -EFAULT;
2838                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2839                         goto out;
2840                 r = 0;
2841                 break;
2842         }
2843         case KVM_SET_SREGS: {
2844                 struct kvm_sregs kvm_sregs;
2845
2846                 r = -EFAULT;
2847                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2848                         goto out;
2849                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2850                 if (r)
2851                         goto out;
2852                 r = 0;
2853                 break;
2854         }
2855         case KVM_TRANSLATE: {
2856                 struct kvm_translation tr;
2857
2858                 r = -EFAULT;
2859                 if (copy_from_user(&tr, argp, sizeof tr))
2860                         goto out;
2861                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2862                 if (r)
2863                         goto out;
2864                 r = -EFAULT;
2865                 if (copy_to_user(argp, &tr, sizeof tr))
2866                         goto out;
2867                 r = 0;
2868                 break;
2869         }
2870         case KVM_INTERRUPT: {
2871                 struct kvm_interrupt irq;
2872
2873                 r = -EFAULT;
2874                 if (copy_from_user(&irq, argp, sizeof irq))
2875                         goto out;
2876                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2877                 if (r)
2878                         goto out;
2879                 r = 0;
2880                 break;
2881         }
2882         case KVM_DEBUG_GUEST: {
2883                 struct kvm_debug_guest dbg;
2884
2885                 r = -EFAULT;
2886                 if (copy_from_user(&dbg, argp, sizeof dbg))
2887                         goto out;
2888                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2889                 if (r)
2890                         goto out;
2891                 r = 0;
2892                 break;
2893         }
2894         case KVM_GET_MSRS:
2895                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2896                 break;
2897         case KVM_SET_MSRS:
2898                 r = msr_io(vcpu, argp, do_set_msr, 0);
2899                 break;
2900         case KVM_SET_CPUID: {
2901                 struct kvm_cpuid __user *cpuid_arg = argp;
2902                 struct kvm_cpuid cpuid;
2903
2904                 r = -EFAULT;
2905                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2906                         goto out;
2907                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2908                 if (r)
2909                         goto out;
2910                 break;
2911         }
2912         case KVM_SET_SIGNAL_MASK: {
2913                 struct kvm_signal_mask __user *sigmask_arg = argp;
2914                 struct kvm_signal_mask kvm_sigmask;
2915                 sigset_t sigset, *p;
2916
2917                 p = NULL;
2918                 if (argp) {
2919                         r = -EFAULT;
2920                         if (copy_from_user(&kvm_sigmask, argp,
2921                                            sizeof kvm_sigmask))
2922                                 goto out;
2923                         r = -EINVAL;
2924                         if (kvm_sigmask.len != sizeof sigset)
2925                                 goto out;
2926                         r = -EFAULT;
2927                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2928                                            sizeof sigset))
2929                                 goto out;
2930                         p = &sigset;
2931                 }
2932                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2933                 break;
2934         }
2935         case KVM_GET_FPU: {
2936                 struct kvm_fpu fpu;
2937
2938                 memset(&fpu, 0, sizeof fpu);
2939                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2940                 if (r)
2941                         goto out;
2942                 r = -EFAULT;
2943                 if (copy_to_user(argp, &fpu, sizeof fpu))
2944                         goto out;
2945                 r = 0;
2946                 break;
2947         }
2948         case KVM_SET_FPU: {
2949                 struct kvm_fpu fpu;
2950
2951                 r = -EFAULT;
2952                 if (copy_from_user(&fpu, argp, sizeof fpu))
2953                         goto out;
2954                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2955                 if (r)
2956                         goto out;
2957                 r = 0;
2958                 break;
2959         }
2960         case KVM_GET_LAPIC: {
2961                 struct kvm_lapic_state lapic;
2962
2963                 memset(&lapic, 0, sizeof lapic);
2964                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
2965                 if (r)
2966                         goto out;
2967                 r = -EFAULT;
2968                 if (copy_to_user(argp, &lapic, sizeof lapic))
2969                         goto out;
2970                 r = 0;
2971                 break;
2972         }
2973         case KVM_SET_LAPIC: {
2974                 struct kvm_lapic_state lapic;
2975
2976                 r = -EFAULT;
2977                 if (copy_from_user(&lapic, argp, sizeof lapic))
2978                         goto out;
2979                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
2980                 if (r)
2981                         goto out;
2982                 r = 0;
2983                 break;
2984         }
2985         default:
2986                 ;
2987         }
2988 out:
2989         return r;
2990 }
2991
2992 static long kvm_vm_ioctl(struct file *filp,
2993                            unsigned int ioctl, unsigned long arg)
2994 {
2995         struct kvm *kvm = filp->private_data;
2996         void __user *argp = (void __user *)arg;
2997         int r = -EINVAL;
2998
2999         switch (ioctl) {
3000         case KVM_CREATE_VCPU:
3001                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3002                 if (r < 0)
3003                         goto out;
3004                 break;
3005         case KVM_SET_MEMORY_REGION: {
3006                 struct kvm_memory_region kvm_mem;
3007
3008                 r = -EFAULT;
3009                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
3010                         goto out;
3011                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
3012                 if (r)
3013                         goto out;
3014                 break;
3015         }
3016         case KVM_GET_DIRTY_LOG: {
3017                 struct kvm_dirty_log log;
3018
3019                 r = -EFAULT;
3020                 if (copy_from_user(&log, argp, sizeof log))
3021                         goto out;
3022                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3023                 if (r)
3024                         goto out;
3025                 break;
3026         }
3027         case KVM_SET_MEMORY_ALIAS: {
3028                 struct kvm_memory_alias alias;
3029
3030                 r = -EFAULT;
3031                 if (copy_from_user(&alias, argp, sizeof alias))
3032                         goto out;
3033                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
3034                 if (r)
3035                         goto out;
3036                 break;
3037         }
3038         case KVM_CREATE_IRQCHIP:
3039                 r = -ENOMEM;
3040                 kvm->vpic = kvm_create_pic(kvm);
3041                 if (kvm->vpic) {
3042                         r = kvm_ioapic_init(kvm);
3043                         if (r) {
3044                                 kfree(kvm->vpic);
3045                                 kvm->vpic = NULL;
3046                                 goto out;
3047                         }
3048                 }
3049                 else
3050                         goto out;
3051                 break;
3052         case KVM_IRQ_LINE: {
3053                 struct kvm_irq_level irq_event;
3054
3055                 r = -EFAULT;
3056                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3057                         goto out;
3058                 if (irqchip_in_kernel(kvm)) {
3059                         mutex_lock(&kvm->lock);
3060                         if (irq_event.irq < 16)
3061                                 kvm_pic_set_irq(pic_irqchip(kvm),
3062                                         irq_event.irq,
3063                                         irq_event.level);
3064                         kvm_ioapic_set_irq(kvm->vioapic,
3065                                         irq_event.irq,
3066                                         irq_event.level);
3067                         mutex_unlock(&kvm->lock);
3068                         r = 0;
3069                 }
3070                 break;
3071         }
3072         case KVM_GET_IRQCHIP: {
3073                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3074                 struct kvm_irqchip chip;
3075
3076                 r = -EFAULT;
3077                 if (copy_from_user(&chip, argp, sizeof chip))
3078                         goto out;
3079                 r = -ENXIO;
3080                 if (!irqchip_in_kernel(kvm))
3081                         goto out;
3082                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
3083                 if (r)
3084                         goto out;
3085                 r = -EFAULT;
3086                 if (copy_to_user(argp, &chip, sizeof chip))
3087                         goto out;
3088                 r = 0;
3089                 break;
3090         }
3091         case KVM_SET_IRQCHIP: {
3092                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3093                 struct kvm_irqchip chip;
3094
3095                 r = -EFAULT;
3096                 if (copy_from_user(&chip, argp, sizeof chip))
3097                         goto out;
3098                 r = -ENXIO;
3099                 if (!irqchip_in_kernel(kvm))
3100                         goto out;
3101                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
3102                 if (r)
3103                         goto out;
3104                 r = 0;
3105                 break;
3106         }
3107         default:
3108                 ;
3109         }
3110 out:
3111         return r;
3112 }
3113
3114 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
3115                                   unsigned long address,
3116                                   int *type)
3117 {
3118         struct kvm *kvm = vma->vm_file->private_data;
3119         unsigned long pgoff;
3120         struct page *page;
3121
3122         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
3123         page = gfn_to_page(kvm, pgoff);
3124         if (!page)
3125                 return NOPAGE_SIGBUS;
3126         get_page(page);
3127         if (type != NULL)
3128                 *type = VM_FAULT_MINOR;
3129
3130         return page;
3131 }
3132
3133 static struct vm_operations_struct kvm_vm_vm_ops = {
3134         .nopage = kvm_vm_nopage,
3135 };
3136
3137 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
3138 {
3139         vma->vm_ops = &kvm_vm_vm_ops;
3140         return 0;
3141 }
3142
3143 static struct file_operations kvm_vm_fops = {
3144         .release        = kvm_vm_release,
3145         .unlocked_ioctl = kvm_vm_ioctl,
3146         .compat_ioctl   = kvm_vm_ioctl,
3147         .mmap           = kvm_vm_mmap,
3148 };
3149
3150 static int kvm_dev_ioctl_create_vm(void)
3151 {
3152         int fd, r;
3153         struct inode *inode;
3154         struct file *file;
3155         struct kvm *kvm;
3156
3157         kvm = kvm_create_vm();
3158         if (IS_ERR(kvm))
3159                 return PTR_ERR(kvm);
3160         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
3161         if (r) {
3162                 kvm_destroy_vm(kvm);
3163                 return r;
3164         }
3165
3166         kvm->filp = file;
3167
3168         return fd;
3169 }
3170
3171 static long kvm_dev_ioctl(struct file *filp,
3172                           unsigned int ioctl, unsigned long arg)
3173 {
3174         void __user *argp = (void __user *)arg;
3175         long r = -EINVAL;
3176
3177         switch (ioctl) {
3178         case KVM_GET_API_VERSION:
3179                 r = -EINVAL;
3180                 if (arg)
3181                         goto out;
3182                 r = KVM_API_VERSION;
3183                 break;
3184         case KVM_CREATE_VM:
3185                 r = -EINVAL;
3186                 if (arg)
3187                         goto out;
3188                 r = kvm_dev_ioctl_create_vm();
3189                 break;
3190         case KVM_GET_MSR_INDEX_LIST: {
3191                 struct kvm_msr_list __user *user_msr_list = argp;
3192                 struct kvm_msr_list msr_list;
3193                 unsigned n;
3194
3195                 r = -EFAULT;
3196                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
3197                         goto out;
3198                 n = msr_list.nmsrs;
3199                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
3200                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
3201                         goto out;
3202                 r = -E2BIG;
3203                 if (n < num_msrs_to_save)
3204                         goto out;
3205                 r = -EFAULT;
3206                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3207                                  num_msrs_to_save * sizeof(u32)))
3208                         goto out;
3209                 if (copy_to_user(user_msr_list->indices
3210                                  + num_msrs_to_save * sizeof(u32),
3211                                  &emulated_msrs,
3212                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
3213                         goto out;
3214                 r = 0;
3215                 break;
3216         }
3217         case KVM_CHECK_EXTENSION: {
3218                 int ext = (long)argp;
3219
3220                 switch (ext) {
3221                 case KVM_CAP_IRQCHIP:
3222                 case KVM_CAP_HLT:
3223                         r = 1;
3224                         break;
3225                 default:
3226                         r = 0;
3227                         break;
3228                 }
3229                 break;
3230         }
3231         case KVM_GET_VCPU_MMAP_SIZE:
3232                 r = -EINVAL;
3233                 if (arg)
3234                         goto out;
3235                 r = 2 * PAGE_SIZE;
3236                 break;
3237         default:
3238                 ;
3239         }
3240 out:
3241         return r;
3242 }
3243
3244 static struct file_operations kvm_chardev_ops = {
3245         .unlocked_ioctl = kvm_dev_ioctl,
3246         .compat_ioctl   = kvm_dev_ioctl,
3247 };
3248
3249 static struct miscdevice kvm_dev = {
3250         KVM_MINOR,
3251         "kvm",
3252         &kvm_chardev_ops,
3253 };
3254
3255 /*
3256  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
3257  * cached on it.
3258  */
3259 static void decache_vcpus_on_cpu(int cpu)
3260 {
3261         struct kvm *vm;
3262         struct kvm_vcpu *vcpu;
3263         int i;
3264
3265         spin_lock(&kvm_lock);
3266         list_for_each_entry(vm, &vm_list, vm_list)
3267                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3268                         vcpu = vm->vcpus[i];
3269                         if (!vcpu)
3270                                 continue;
3271                         /*
3272                          * If the vcpu is locked, then it is running on some
3273                          * other cpu and therefore it is not cached on the
3274                          * cpu in question.
3275                          *
3276                          * If it's not locked, check the last cpu it executed
3277                          * on.
3278                          */
3279                         if (mutex_trylock(&vcpu->mutex)) {
3280                                 if (vcpu->cpu == cpu) {
3281                                         kvm_x86_ops->vcpu_decache(vcpu);
3282                                         vcpu->cpu = -1;
3283                                 }
3284                                 mutex_unlock(&vcpu->mutex);
3285                         }
3286                 }
3287         spin_unlock(&kvm_lock);
3288 }
3289
3290 static void hardware_enable(void *junk)
3291 {
3292         int cpu = raw_smp_processor_id();
3293
3294         if (cpu_isset(cpu, cpus_hardware_enabled))
3295                 return;
3296         cpu_set(cpu, cpus_hardware_enabled);
3297         kvm_x86_ops->hardware_enable(NULL);
3298 }
3299
3300 static void hardware_disable(void *junk)
3301 {
3302         int cpu = raw_smp_processor_id();
3303
3304         if (!cpu_isset(cpu, cpus_hardware_enabled))
3305                 return;
3306         cpu_clear(cpu, cpus_hardware_enabled);
3307         decache_vcpus_on_cpu(cpu);
3308         kvm_x86_ops->hardware_disable(NULL);
3309 }
3310
3311 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3312                            void *v)
3313 {
3314         int cpu = (long)v;
3315
3316         switch (val) {
3317         case CPU_DYING:
3318         case CPU_DYING_FROZEN:
3319                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3320                        cpu);
3321                 hardware_disable(NULL);
3322                 break;
3323         case CPU_UP_CANCELED:
3324         case CPU_UP_CANCELED_FROZEN:
3325                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3326                        cpu);
3327                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3328                 break;
3329         case CPU_ONLINE:
3330         case CPU_ONLINE_FROZEN:
3331                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3332                        cpu);
3333                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3334                 break;
3335         }
3336         return NOTIFY_OK;
3337 }
3338
3339 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3340                        void *v)
3341 {
3342         if (val == SYS_RESTART) {
3343                 /*
3344                  * Some (well, at least mine) BIOSes hang on reboot if
3345                  * in vmx root mode.
3346                  */
3347                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3348                 on_each_cpu(hardware_disable, NULL, 0, 1);
3349         }
3350         return NOTIFY_OK;
3351 }
3352
3353 static struct notifier_block kvm_reboot_notifier = {
3354         .notifier_call = kvm_reboot,
3355         .priority = 0,
3356 };
3357
3358 void kvm_io_bus_init(struct kvm_io_bus *bus)
3359 {
3360         memset(bus, 0, sizeof(*bus));
3361 }
3362
3363 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3364 {
3365         int i;
3366
3367         for (i = 0; i < bus->dev_count; i++) {
3368                 struct kvm_io_device *pos = bus->devs[i];
3369
3370                 kvm_iodevice_destructor(pos);
3371         }
3372 }
3373
3374 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3375 {
3376         int i;
3377
3378         for (i = 0; i < bus->dev_count; i++) {
3379                 struct kvm_io_device *pos = bus->devs[i];
3380
3381                 if (pos->in_range(pos, addr))
3382                         return pos;
3383         }
3384
3385         return NULL;
3386 }
3387
3388 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3389 {
3390         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3391
3392         bus->devs[bus->dev_count++] = dev;
3393 }
3394
3395 static struct notifier_block kvm_cpu_notifier = {
3396         .notifier_call = kvm_cpu_hotplug,
3397         .priority = 20, /* must be > scheduler priority */
3398 };
3399
3400 static u64 stat_get(void *_offset)
3401 {
3402         unsigned offset = (long)_offset;
3403         u64 total = 0;
3404         struct kvm *kvm;
3405         struct kvm_vcpu *vcpu;
3406         int i;
3407
3408         spin_lock(&kvm_lock);
3409         list_for_each_entry(kvm, &vm_list, vm_list)
3410                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3411                         vcpu = kvm->vcpus[i];
3412                         if (vcpu)
3413                                 total += *(u32 *)((void *)vcpu + offset);
3414                 }
3415         spin_unlock(&kvm_lock);
3416         return total;
3417 }
3418
3419 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3420
3421 static __init void kvm_init_debug(void)
3422 {
3423         struct kvm_stats_debugfs_item *p;
3424
3425         debugfs_dir = debugfs_create_dir("kvm", NULL);
3426         for (p = debugfs_entries; p->name; ++p)
3427                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3428                                                 (void *)(long)p->offset,
3429                                                 &stat_fops);
3430 }
3431
3432 static void kvm_exit_debug(void)
3433 {
3434         struct kvm_stats_debugfs_item *p;
3435
3436         for (p = debugfs_entries; p->name; ++p)
3437                 debugfs_remove(p->dentry);
3438         debugfs_remove(debugfs_dir);
3439 }
3440
3441 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3442 {
3443         hardware_disable(NULL);
3444         return 0;
3445 }
3446
3447 static int kvm_resume(struct sys_device *dev)
3448 {
3449         hardware_enable(NULL);
3450         return 0;
3451 }
3452
3453 static struct sysdev_class kvm_sysdev_class = {
3454         set_kset_name("kvm"),
3455         .suspend = kvm_suspend,
3456         .resume = kvm_resume,
3457 };
3458
3459 static struct sys_device kvm_sysdev = {
3460         .id = 0,
3461         .cls = &kvm_sysdev_class,
3462 };
3463
3464 hpa_t bad_page_address;
3465
3466 static inline
3467 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3468 {
3469         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3470 }
3471
3472 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3473 {
3474         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3475
3476         kvm_x86_ops->vcpu_load(vcpu, cpu);
3477 }
3478
3479 static void kvm_sched_out(struct preempt_notifier *pn,
3480                           struct task_struct *next)
3481 {
3482         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3483
3484         kvm_x86_ops->vcpu_put(vcpu);
3485 }
3486
3487 int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
3488                   struct module *module)
3489 {
3490         int r;
3491         int cpu;
3492
3493         if (kvm_x86_ops) {
3494                 printk(KERN_ERR "kvm: already loaded the other module\n");
3495                 return -EEXIST;
3496         }
3497
3498         if (!ops->cpu_has_kvm_support()) {
3499                 printk(KERN_ERR "kvm: no hardware support\n");
3500                 return -EOPNOTSUPP;
3501         }
3502         if (ops->disabled_by_bios()) {
3503                 printk(KERN_ERR "kvm: disabled by bios\n");
3504                 return -EOPNOTSUPP;
3505         }
3506
3507         kvm_x86_ops = ops;
3508
3509         r = kvm_x86_ops->hardware_setup();
3510         if (r < 0)
3511                 goto out;
3512
3513         for_each_online_cpu(cpu) {
3514                 smp_call_function_single(cpu,
3515                                 kvm_x86_ops->check_processor_compatibility,
3516                                 &r, 0, 1);
3517                 if (r < 0)
3518                         goto out_free_0;
3519         }
3520
3521         on_each_cpu(hardware_enable, NULL, 0, 1);
3522         r = register_cpu_notifier(&kvm_cpu_notifier);
3523         if (r)
3524                 goto out_free_1;
3525         register_reboot_notifier(&kvm_reboot_notifier);
3526
3527         r = sysdev_class_register(&kvm_sysdev_class);
3528         if (r)
3529                 goto out_free_2;
3530
3531         r = sysdev_register(&kvm_sysdev);
3532         if (r)
3533                 goto out_free_3;
3534
3535         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3536         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3537                                            __alignof__(struct kvm_vcpu), 0, 0);
3538         if (!kvm_vcpu_cache) {
3539                 r = -ENOMEM;
3540                 goto out_free_4;
3541         }
3542
3543         kvm_chardev_ops.owner = module;
3544
3545         r = misc_register(&kvm_dev);
3546         if (r) {
3547                 printk (KERN_ERR "kvm: misc device register failed\n");
3548                 goto out_free;
3549         }
3550
3551         kvm_preempt_ops.sched_in = kvm_sched_in;
3552         kvm_preempt_ops.sched_out = kvm_sched_out;
3553
3554         return r;
3555
3556 out_free:
3557         kmem_cache_destroy(kvm_vcpu_cache);
3558 out_free_4:
3559         sysdev_unregister(&kvm_sysdev);
3560 out_free_3:
3561         sysdev_class_unregister(&kvm_sysdev_class);
3562 out_free_2:
3563         unregister_reboot_notifier(&kvm_reboot_notifier);
3564         unregister_cpu_notifier(&kvm_cpu_notifier);
3565 out_free_1:
3566         on_each_cpu(hardware_disable, NULL, 0, 1);
3567 out_free_0:
3568         kvm_x86_ops->hardware_unsetup();
3569 out:
3570         kvm_x86_ops = NULL;
3571         return r;
3572 }
3573
3574 void kvm_exit_x86(void)
3575 {
3576         misc_deregister(&kvm_dev);
3577         kmem_cache_destroy(kvm_vcpu_cache);
3578         sysdev_unregister(&kvm_sysdev);
3579         sysdev_class_unregister(&kvm_sysdev_class);
3580         unregister_reboot_notifier(&kvm_reboot_notifier);
3581         unregister_cpu_notifier(&kvm_cpu_notifier);
3582         on_each_cpu(hardware_disable, NULL, 0, 1);
3583         kvm_x86_ops->hardware_unsetup();
3584         kvm_x86_ops = NULL;
3585 }
3586
3587 static __init int kvm_init(void)
3588 {
3589         static struct page *bad_page;
3590         int r;
3591
3592         r = kvm_mmu_module_init();
3593         if (r)
3594                 goto out4;
3595
3596         kvm_init_debug();
3597
3598         kvm_init_msr_list();
3599
3600         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3601                 r = -ENOMEM;
3602                 goto out;
3603         }
3604
3605         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3606         memset(__va(bad_page_address), 0, PAGE_SIZE);
3607
3608         return 0;
3609
3610 out:
3611         kvm_exit_debug();
3612         kvm_mmu_module_exit();
3613 out4:
3614         return r;
3615 }
3616
3617 static __exit void kvm_exit(void)
3618 {
3619         kvm_exit_debug();
3620         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3621         kvm_mmu_module_exit();
3622 }
3623
3624 module_init(kvm_init)
3625 module_exit(kvm_exit)
3626
3627 EXPORT_SYMBOL_GPL(kvm_init_x86);
3628 EXPORT_SYMBOL_GPL(kvm_exit_x86);