drivers/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "kvm.h"
  19 #include "x86_emulate.h"
  20 #include "segment_descriptor.h"
  21 #include "irq.h"
  22
  23 #include <linux/kvm.h>
  24 #include <linux/module.h>
  25 #include <linux/errno.h>
  26 #include <linux/percpu.h>
  27 #include <linux/gfp.h>
  28 #include <linux/mm.h>
  29 #include <linux/miscdevice.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/reboot.h>
  32 #include <linux/debugfs.h>
  33 #include <linux/highmem.h>
  34 #include <linux/file.h>
  35 #include <linux/sysdev.h>
  36 #include <linux/cpu.h>
  37 #include <linux/sched.h>
  38 #include <linux/cpumask.h>
  39 #include <linux/smp.h>
  40 #include <linux/anon_inodes.h>
  41
  42 #include <asm/processor.h>
  43 #include <asm/msr.h>
  44 #include <asm/io.h>
  45 #include <asm/uaccess.h>
  46 #include <asm/desc.h>
  47
  48 MODULE_AUTHOR("Qumranet");
  49 MODULE_LICENSE("GPL");
  50
  51 static DEFINE_SPINLOCK(kvm_lock);
  52 static LIST_HEAD(vm_list);
  53
  54 static cpumask_t cpus_hardware_enabled;
  55
  56 struct kvm_arch_ops *kvm_arch_ops;
  57 struct kmem_cache *kvm_vcpu_cache;
  58 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  59
  60 static __read_mostly struct preempt_ops kvm_preempt_ops;
  61
  62 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  63
  64 static struct kvm_stats_debugfs_item {
  65         const char *name;
  66         int offset;
  67         struct dentry *dentry;
  68 } debugfs_entries[] = {
  69         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  70         { "pf_guest", STAT_OFFSET(pf_guest) },
  71         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  72         { "invlpg", STAT_OFFSET(invlpg) },
  73         { "exits", STAT_OFFSET(exits) },
  74         { "io_exits", STAT_OFFSET(io_exits) },
  75         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  76         { "signal_exits", STAT_OFFSET(signal_exits) },
  77         { "irq_window", STAT_OFFSET(irq_window_exits) },
  78         { "halt_exits", STAT_OFFSET(halt_exits) },
  79         { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
  80         { "request_irq", STAT_OFFSET(request_irq_exits) },
  81         { "irq_exits", STAT_OFFSET(irq_exits) },
  82         { "light_exits", STAT_OFFSET(light_exits) },
  83         { "efer_reload", STAT_OFFSET(efer_reload) },
  84         { NULL }
  85 };
  86
  87 static struct dentry *debugfs_dir;
  88
  89 #define MAX_IO_MSRS 256
  90
  91 #define CR0_RESERVED_BITS                                               \
  92         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  93                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  94                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  95 #define CR4_RESERVED_BITS                                               \
  96         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  97                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  98                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  99                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
 100
 101 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
 102 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
 103
 104 #ifdef CONFIG_X86_64
 105 // LDT or TSS descriptor in the GDT. 16 bytes.
 106 struct segment_descriptor_64 {
 107         struct segment_descriptor s;
 108         u32 base_higher;
 109         u32 pad_zero;
 110 };
 111
 112 #endif
 113
 114 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 115                            unsigned long arg);
 116
 117 unsigned long segment_base(u16 selector)
 118 {
 119         struct descriptor_table gdt;
 120         struct segment_descriptor *d;
 121         unsigned long table_base;
 122         typedef unsigned long ul;
 123         unsigned long v;
 124
 125         if (selector == 0)
 126                 return 0;
 127
 128         asm ("sgdt %0" : "=m"(gdt));
 129         table_base = gdt.base;
 130
 131         if (selector & 4) {           /* from ldt */
 132                 u16 ldt_selector;
 133
 134                 asm ("sldt %0" : "=g"(ldt_selector));
 135                 table_base = segment_base(ldt_selector);
 136         }
 137         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 138         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 139 #ifdef CONFIG_X86_64
 140         if (d->system == 0
 141             && (d->type == 2 || d->type == 9 || d->type == 11))
 142                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 143 #endif
 144         return v;
 145 }
 146 EXPORT_SYMBOL_GPL(segment_base);
 147
 148 static inline int valid_vcpu(int n)
 149 {
 150         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 151 }
 152
 153 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 154 {
 155         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
 156                 return;
 157
 158         vcpu->guest_fpu_loaded = 1;
 159         fx_save(&vcpu->host_fx_image);
 160         fx_restore(&vcpu->guest_fx_image);
 161 }
 162 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
 163
 164 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 165 {
 166         if (!vcpu->guest_fpu_loaded)
 167                 return;
 168
 169         vcpu->guest_fpu_loaded = 0;
 170         fx_save(&vcpu->guest_fx_image);
 171         fx_restore(&vcpu->host_fx_image);
 172 }
 173 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
 174
 175 /*
 176  * Switches to specified vcpu, until a matching vcpu_put()
 177  */
 178 static void vcpu_load(struct kvm_vcpu *vcpu)
 179 {
 180         int cpu;
 181
 182         mutex_lock(&vcpu->mutex);
 183         cpu = get_cpu();
 184         preempt_notifier_register(&vcpu->preempt_notifier);
 185         kvm_arch_ops->vcpu_load(vcpu, cpu);
 186         put_cpu();
 187 }
 188
 189 static void vcpu_put(struct kvm_vcpu *vcpu)
 190 {
 191         preempt_disable();
 192         kvm_arch_ops->vcpu_put(vcpu);
 193         preempt_notifier_unregister(&vcpu->preempt_notifier);
 194         preempt_enable();
 195         mutex_unlock(&vcpu->mutex);
 196 }
 197
 198 static void ack_flush(void *_completed)
 199 {
 200         atomic_t *completed = _completed;
 201
 202         atomic_inc(completed);
 203 }
 204
 205 void kvm_flush_remote_tlbs(struct kvm *kvm)
 206 {
 207         int i, cpu, needed;
 208         cpumask_t cpus;
 209         struct kvm_vcpu *vcpu;
 210         atomic_t completed;
 211
 212         atomic_set(&completed, 0);
 213         cpus_clear(cpus);
 214         needed = 0;
 215         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 216                 vcpu = kvm->vcpus[i];
 217                 if (!vcpu)
 218                         continue;
 219                 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
 220                         continue;
 221                 cpu = vcpu->cpu;
 222                 if (cpu != -1 && cpu != raw_smp_processor_id())
 223                         if (!cpu_isset(cpu, cpus)) {
 224                                 cpu_set(cpu, cpus);
 225                                 ++needed;
 226                         }
 227         }
 228
 229         /*
 230          * We really want smp_call_function_mask() here.  But that's not
 231          * available, so ipi all cpus in parallel and wait for them
 232          * to complete.
 233          */
 234         for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
 235                 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
 236         while (atomic_read(&completed) != needed) {
 237                 cpu_relax();
 238                 barrier();
 239         }
 240 }
 241
 242 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 243 {
 244         struct page *page;
 245         int r;
 246
 247         mutex_init(&vcpu->mutex);
 248         vcpu->cpu = -1;
 249         vcpu->mmu.root_hpa = INVALID_PAGE;
 250         vcpu->kvm = kvm;
 251         vcpu->vcpu_id = id;
 252         if (!irqchip_in_kernel(kvm) || id == 0)
 253                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
 254         else
 255                 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
 256         init_waitqueue_head(&vcpu->wq);
 257
 258         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 259         if (!page) {
 260                 r = -ENOMEM;
 261                 goto fail;
 262         }
 263         vcpu->run = page_address(page);
 264
 265         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 266         if (!page) {
 267                 r = -ENOMEM;
 268                 goto fail_free_run;
 269         }
 270         vcpu->pio_data = page_address(page);
 271
 272         r = kvm_mmu_create(vcpu);
 273         if (r < 0)
 274                 goto fail_free_pio_data;
 275
 276         return 0;
 277
 278 fail_free_pio_data:
 279         free_page((unsigned long)vcpu->pio_data);
 280 fail_free_run:
 281         free_page((unsigned long)vcpu->run);
 282 fail:
 283         return -ENOMEM;
 284 }
 285 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 286
 287 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 288 {
 289         kvm_mmu_destroy(vcpu);
 290         if (vcpu->apic)
 291                 hrtimer_cancel(&vcpu->apic->timer.dev);
 292         kvm_free_apic(vcpu->apic);
 293         free_page((unsigned long)vcpu->pio_data);
 294         free_page((unsigned long)vcpu->run);
 295 }
 296 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 297
 298 static struct kvm *kvm_create_vm(void)
 299 {
 300         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 301
 302         if (!kvm)
 303                 return ERR_PTR(-ENOMEM);
 304
 305         kvm_io_bus_init(&kvm->pio_bus);
 306         mutex_init(&kvm->lock);
 307         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 308         kvm_io_bus_init(&kvm->mmio_bus);
 309         spin_lock(&kvm_lock);
 310         list_add(&kvm->vm_list, &vm_list);
 311         spin_unlock(&kvm_lock);
 312         return kvm;
 313 }
 314
 315 /*
 316  * Free any memory in @free but not in @dont.
 317  */
 318 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 319                                   struct kvm_memory_slot *dont)
 320 {
 321         int i;
 322
 323         if (!dont || free->phys_mem != dont->phys_mem)
 324                 if (free->phys_mem) {
 325                         for (i = 0; i < free->npages; ++i)
 326                                 if (free->phys_mem[i])
 327                                         __free_page(free->phys_mem[i]);
 328                         vfree(free->phys_mem);
 329                 }
 330
 331         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 332                 vfree(free->dirty_bitmap);
 333
 334         free->phys_mem = NULL;
 335         free->npages = 0;
 336         free->dirty_bitmap = NULL;
 337 }
 338
 339 static void kvm_free_physmem(struct kvm *kvm)
 340 {
 341         int i;
 342
 343         for (i = 0; i < kvm->nmemslots; ++i)
 344                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 345 }
 346
 347 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 348 {
 349         int i;
 350
 351         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
 352                 if (vcpu->pio.guest_pages[i]) {
 353                         __free_page(vcpu->pio.guest_pages[i]);
 354                         vcpu->pio.guest_pages[i] = NULL;
 355                 }
 356 }
 357
 358 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 359 {
 360         vcpu_load(vcpu);
 361         kvm_mmu_unload(vcpu);
 362         vcpu_put(vcpu);
 363 }
 364
 365 static void kvm_free_vcpus(struct kvm *kvm)
 366 {
 367         unsigned int i;
 368
 369         /*
 370          * Unpin any mmu pages first.
 371          */
 372         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 373                 if (kvm->vcpus[i])
 374                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
 375         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 376                 if (kvm->vcpus[i]) {
 377                         kvm_arch_ops->vcpu_free(kvm->vcpus[i]);
 378                         kvm->vcpus[i] = NULL;
 379                 }
 380         }
 381
 382 }
 383
 384 static void kvm_destroy_vm(struct kvm *kvm)
 385 {
 386         spin_lock(&kvm_lock);
 387         list_del(&kvm->vm_list);
 388         spin_unlock(&kvm_lock);
 389         kvm_io_bus_destroy(&kvm->pio_bus);
 390         kvm_io_bus_destroy(&kvm->mmio_bus);
 391         kfree(kvm->vpic);
 392         kfree(kvm->vioapic);
 393         kvm_free_vcpus(kvm);
 394         kvm_free_physmem(kvm);
 395         kfree(kvm);
 396 }
 397
 398 static int kvm_vm_release(struct inode *inode, struct file *filp)
 399 {
 400         struct kvm *kvm = filp->private_data;
 401
 402         kvm_destroy_vm(kvm);
 403         return 0;
 404 }
 405
 406 static void inject_gp(struct kvm_vcpu *vcpu)
 407 {
 408         kvm_arch_ops->inject_gp(vcpu, 0);
 409 }
 410
 411 /*
 412  * Load the pae pdptrs.  Return true is they are all valid.
 413  */
 414 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 415 {
 416         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 417         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 418         int i;
 419         u64 *pdpt;
 420         int ret;
 421         struct page *page;
 422         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
 423
 424         mutex_lock(&vcpu->kvm->lock);
 425         page = gfn_to_page(vcpu->kvm, pdpt_gfn);
 426         if (!page) {
 427                 ret = 0;
 428                 goto out;
 429         }
 430
 431         pdpt = kmap_atomic(page, KM_USER0);
 432         memcpy(pdpte, pdpt+offset, sizeof(pdpte));
 433         kunmap_atomic(pdpt, KM_USER0);
 434
 435         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 436                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 437                         ret = 0;
 438                         goto out;
 439                 }
 440         }
 441         ret = 1;
 442
 443         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
 444 out:
 445         mutex_unlock(&vcpu->kvm->lock);
 446
 447         return ret;
 448 }
 449
 450 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 451 {
 452         if (cr0 & CR0_RESERVED_BITS) {
 453                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 454                        cr0, vcpu->cr0);
 455                 inject_gp(vcpu);
 456                 return;
 457         }
 458
 459         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 460                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 461                 inject_gp(vcpu);
 462                 return;
 463         }
 464
 465         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 466                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 467                        "and a clear PE flag\n");
 468                 inject_gp(vcpu);
 469                 return;
 470         }
 471
 472         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 473 #ifdef CONFIG_X86_64
 474                 if ((vcpu->shadow_efer & EFER_LME)) {
 475                         int cs_db, cs_l;
 476
 477                         if (!is_pae(vcpu)) {
 478                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 479                                        "in long mode while PAE is disabled\n");
 480                                 inject_gp(vcpu);
 481                                 return;
 482                         }
 483                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 484                         if (cs_l) {
 485                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 486                                        "in long mode while CS.L == 1\n");
 487                                 inject_gp(vcpu);
 488                                 return;
 489
 490                         }
 491                 } else
 492 #endif
 493                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 494                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 495                                "reserved bits\n");
 496                         inject_gp(vcpu);
 497                         return;
 498                 }
 499
 500         }
 501
 502         kvm_arch_ops->set_cr0(vcpu, cr0);
 503         vcpu->cr0 = cr0;
 504
 505         mutex_lock(&vcpu->kvm->lock);
 506         kvm_mmu_reset_context(vcpu);
 507         mutex_unlock(&vcpu->kvm->lock);
 508         return;
 509 }
 510 EXPORT_SYMBOL_GPL(set_cr0);
 511
 512 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 513 {
 514         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 515 }
 516 EXPORT_SYMBOL_GPL(lmsw);
 517
 518 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 519 {
 520         if (cr4 & CR4_RESERVED_BITS) {
 521                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 522                 inject_gp(vcpu);
 523                 return;
 524         }
 525
 526         if (is_long_mode(vcpu)) {
 527                 if (!(cr4 & X86_CR4_PAE)) {
 528                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 529                                "in long mode\n");
 530                         inject_gp(vcpu);
 531                         return;
 532                 }
 533         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 534                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 535                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 536                 inject_gp(vcpu);
 537                 return;
 538         }
 539
 540         if (cr4 & X86_CR4_VMXE) {
 541                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 542                 inject_gp(vcpu);
 543                 return;
 544         }
 545         kvm_arch_ops->set_cr4(vcpu, cr4);
 546         vcpu->cr4 = cr4;
 547         mutex_lock(&vcpu->kvm->lock);
 548         kvm_mmu_reset_context(vcpu);
 549         mutex_unlock(&vcpu->kvm->lock);
 550 }
 551 EXPORT_SYMBOL_GPL(set_cr4);
 552
 553 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 554 {
 555         if (is_long_mode(vcpu)) {
 556                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 557                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 558                         inject_gp(vcpu);
 559                         return;
 560                 }
 561         } else {
 562                 if (is_pae(vcpu)) {
 563                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 564                                 printk(KERN_DEBUG
 565                                        "set_cr3: #GP, reserved bits\n");
 566                                 inject_gp(vcpu);
 567                                 return;
 568                         }
 569                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 570                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 571                                        "reserved bits\n");
 572                                 inject_gp(vcpu);
 573                                 return;
 574                         }
 575                 } else {
 576                         if (cr3 & CR3_NONPAE_RESERVED_BITS) {
 577                                 printk(KERN_DEBUG
 578                                        "set_cr3: #GP, reserved bits\n");
 579                                 inject_gp(vcpu);
 580                                 return;
 581                         }
 582                 }
 583         }
 584
 585         mutex_lock(&vcpu->kvm->lock);
 586         /*
 587          * Does the new cr3 value map to physical memory? (Note, we
 588          * catch an invalid cr3 even in real-mode, because it would
 589          * cause trouble later on when we turn on paging anyway.)
 590          *
 591          * A real CPU would silently accept an invalid cr3 and would
 592          * attempt to use it - with largely undefined (and often hard
 593          * to debug) behavior on the guest side.
 594          */
 595         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 596                 inject_gp(vcpu);
 597         else {
 598                 vcpu->cr3 = cr3;
 599                 vcpu->mmu.new_cr3(vcpu);
 600         }
 601         mutex_unlock(&vcpu->kvm->lock);
 602 }
 603 EXPORT_SYMBOL_GPL(set_cr3);
 604
 605 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 606 {
 607         if (cr8 & CR8_RESERVED_BITS) {
 608                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 609                 inject_gp(vcpu);
 610                 return;
 611         }
 612         if (irqchip_in_kernel(vcpu->kvm))
 613                 kvm_lapic_set_tpr(vcpu, cr8);
 614         else
 615                 vcpu->cr8 = cr8;
 616 }
 617 EXPORT_SYMBOL_GPL(set_cr8);
 618
 619 unsigned long get_cr8(struct kvm_vcpu *vcpu)
 620 {
 621         if (irqchip_in_kernel(vcpu->kvm))
 622                 return kvm_lapic_get_cr8(vcpu);
 623         else
 624                 return vcpu->cr8;
 625 }
 626 EXPORT_SYMBOL_GPL(get_cr8);
 627
 628 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 629 {
 630         if (irqchip_in_kernel(vcpu->kvm))
 631                 return vcpu->apic_base;
 632         else
 633                 return vcpu->apic_base;
 634 }
 635 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 636
 637 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 638 {
 639         /* TODO: reserve bits check */
 640         if (irqchip_in_kernel(vcpu->kvm))
 641                 kvm_lapic_set_base(vcpu, data);
 642         else
 643                 vcpu->apic_base = data;
 644 }
 645 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 646
 647 void fx_init(struct kvm_vcpu *vcpu)
 648 {
 649         unsigned after_mxcsr_mask;
 650
 651         /* Initialize guest FPU by resetting ours and saving into guest's */
 652         preempt_disable();
 653         fx_save(&vcpu->host_fx_image);
 654         fpu_init();
 655         fx_save(&vcpu->guest_fx_image);
 656         fx_restore(&vcpu->host_fx_image);
 657         preempt_enable();
 658
 659         vcpu->cr0 |= X86_CR0_ET;
 660         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
 661         vcpu->guest_fx_image.mxcsr = 0x1f80;
 662         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
 663                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
 664 }
 665 EXPORT_SYMBOL_GPL(fx_init);
 666
 667 /*
 668  * Allocate some memory and give it an address in the guest physical address
 669  * space.
 670  *
 671  * Discontiguous memory is allowed, mostly for framebuffers.
 672  */
 673 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 674                                           struct kvm_memory_region *mem)
 675 {
 676         int r;
 677         gfn_t base_gfn;
 678         unsigned long npages;
 679         unsigned long i;
 680         struct kvm_memory_slot *memslot;
 681         struct kvm_memory_slot old, new;
 682         int memory_config_version;
 683
 684         r = -EINVAL;
 685         /* General sanity checks */
 686         if (mem->memory_size & (PAGE_SIZE - 1))
 687                 goto out;
 688         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 689                 goto out;
 690         if (mem->slot >= KVM_MEMORY_SLOTS)
 691                 goto out;
 692         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 693                 goto out;
 694
 695         memslot = &kvm->memslots[mem->slot];
 696         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 697         npages = mem->memory_size >> PAGE_SHIFT;
 698
 699         if (!npages)
 700                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 701
 702 raced:
 703         mutex_lock(&kvm->lock);
 704
 705         memory_config_version = kvm->memory_config_version;
 706         new = old = *memslot;
 707
 708         new.base_gfn = base_gfn;
 709         new.npages = npages;
 710         new.flags = mem->flags;
 711
 712         /* Disallow changing a memory slot's size. */
 713         r = -EINVAL;
 714         if (npages && old.npages && npages != old.npages)
 715                 goto out_unlock;
 716
 717         /* Check for overlaps */
 718         r = -EEXIST;
 719         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 720                 struct kvm_memory_slot *s = &kvm->memslots[i];
 721
 722                 if (s == memslot)
 723                         continue;
 724                 if (!((base_gfn + npages <= s->base_gfn) ||
 725                       (base_gfn >= s->base_gfn + s->npages)))
 726                         goto out_unlock;
 727         }
 728         /*
 729          * Do memory allocations outside lock.  memory_config_version will
 730          * detect any races.
 731          */
 732         mutex_unlock(&kvm->lock);
 733
 734         /* Deallocate if slot is being removed */
 735         if (!npages)
 736                 new.phys_mem = NULL;
 737
 738         /* Free page dirty bitmap if unneeded */
 739         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 740                 new.dirty_bitmap = NULL;
 741
 742         r = -ENOMEM;
 743
 744         /* Allocate if a slot is being created */
 745         if (npages && !new.phys_mem) {
 746                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 747
 748                 if (!new.phys_mem)
 749                         goto out_free;
 750
 751                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 752                 for (i = 0; i < npages; ++i) {
 753                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 754                                                      | __GFP_ZERO);
 755                         if (!new.phys_mem[i])
 756                                 goto out_free;
 757                         set_page_private(new.phys_mem[i],0);
 758                 }
 759         }
 760
 761         /* Allocate page dirty bitmap if needed */
 762         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 763                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 764
 765                 new.dirty_bitmap = vmalloc(dirty_bytes);
 766                 if (!new.dirty_bitmap)
 767                         goto out_free;
 768                 memset(new.dirty_bitmap, 0, dirty_bytes);
 769         }
 770
 771         mutex_lock(&kvm->lock);
 772
 773         if (memory_config_version != kvm->memory_config_version) {
 774                 mutex_unlock(&kvm->lock);
 775                 kvm_free_physmem_slot(&new, &old);
 776                 goto raced;
 777         }
 778
 779         r = -EAGAIN;
 780         if (kvm->busy)
 781                 goto out_unlock;
 782
 783         if (mem->slot >= kvm->nmemslots)
 784                 kvm->nmemslots = mem->slot + 1;
 785
 786         *memslot = new;
 787         ++kvm->memory_config_version;
 788
 789         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 790         kvm_flush_remote_tlbs(kvm);
 791
 792         mutex_unlock(&kvm->lock);
 793
 794         kvm_free_physmem_slot(&old, &new);
 795         return 0;
 796
 797 out_unlock:
 798         mutex_unlock(&kvm->lock);
 799 out_free:
 800         kvm_free_physmem_slot(&new, &old);
 801 out:
 802         return r;
 803 }
 804
 805 /*
 806  * Get (and clear) the dirty memory log for a memory slot.
 807  */
 808 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 809                                       struct kvm_dirty_log *log)
 810 {
 811         struct kvm_memory_slot *memslot;
 812         int r, i;
 813         int n;
 814         unsigned long any = 0;
 815
 816         mutex_lock(&kvm->lock);
 817
 818         /*
 819          * Prevent changes to guest memory configuration even while the lock
 820          * is not taken.
 821          */
 822         ++kvm->busy;
 823         mutex_unlock(&kvm->lock);
 824         r = -EINVAL;
 825         if (log->slot >= KVM_MEMORY_SLOTS)
 826                 goto out;
 827
 828         memslot = &kvm->memslots[log->slot];
 829         r = -ENOENT;
 830         if (!memslot->dirty_bitmap)
 831                 goto out;
 832
 833         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 834
 835         for (i = 0; !any && i < n/sizeof(long); ++i)
 836                 any = memslot->dirty_bitmap[i];
 837
 838         r = -EFAULT;
 839         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 840                 goto out;
 841
 842         /* If nothing is dirty, don't bother messing with page tables. */
 843         if (any) {
 844                 mutex_lock(&kvm->lock);
 845                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 846                 kvm_flush_remote_tlbs(kvm);
 847                 memset(memslot->dirty_bitmap, 0, n);
 848                 mutex_unlock(&kvm->lock);
 849         }
 850
 851         r = 0;
 852
 853 out:
 854         mutex_lock(&kvm->lock);
 855         --kvm->busy;
 856         mutex_unlock(&kvm->lock);
 857         return r;
 858 }
 859
 860 /*
 861  * Set a new alias region.  Aliases map a portion of physical memory into
 862  * another portion.  This is useful for memory windows, for example the PC
 863  * VGA region.
 864  */
 865 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 866                                          struct kvm_memory_alias *alias)
 867 {
 868         int r, n;
 869         struct kvm_mem_alias *p;
 870
 871         r = -EINVAL;
 872         /* General sanity checks */
 873         if (alias->memory_size & (PAGE_SIZE - 1))
 874                 goto out;
 875         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 876                 goto out;
 877         if (alias->slot >= KVM_ALIAS_SLOTS)
 878                 goto out;
 879         if (alias->guest_phys_addr + alias->memory_size
 880             < alias->guest_phys_addr)
 881                 goto out;
 882         if (alias->target_phys_addr + alias->memory_size
 883             < alias->target_phys_addr)
 884                 goto out;
 885
 886         mutex_lock(&kvm->lock);
 887
 888         p = &kvm->aliases[alias->slot];
 889         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 890         p->npages = alias->memory_size >> PAGE_SHIFT;
 891         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 892
 893         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 894                 if (kvm->aliases[n - 1].npages)
 895                         break;
 896         kvm->naliases = n;
 897
 898         kvm_mmu_zap_all(kvm);
 899
 900         mutex_unlock(&kvm->lock);
 901
 902         return 0;
 903
 904 out:
 905         return r;
 906 }
 907
 908 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 909 {
 910         int r;
 911
 912         r = 0;
 913         switch (chip->chip_id) {
 914         case KVM_IRQCHIP_PIC_MASTER:
 915                 memcpy (&chip->chip.pic,
 916                         &pic_irqchip(kvm)->pics[0],
 917                         sizeof(struct kvm_pic_state));
 918                 break;
 919         case KVM_IRQCHIP_PIC_SLAVE:
 920                 memcpy (&chip->chip.pic,
 921                         &pic_irqchip(kvm)->pics[1],
 922                         sizeof(struct kvm_pic_state));
 923                 break;
 924         case KVM_IRQCHIP_IOAPIC:
 925                 memcpy (&chip->chip.ioapic,
 926                         ioapic_irqchip(kvm),
 927                         sizeof(struct kvm_ioapic_state));
 928                 break;
 929         default:
 930                 r = -EINVAL;
 931                 break;
 932         }
 933         return r;
 934 }
 935
 936 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 937 {
 938         int r;
 939
 940         r = 0;
 941         switch (chip->chip_id) {
 942         case KVM_IRQCHIP_PIC_MASTER:
 943                 memcpy (&pic_irqchip(kvm)->pics[0],
 944                         &chip->chip.pic,
 945                         sizeof(struct kvm_pic_state));
 946                 break;
 947         case KVM_IRQCHIP_PIC_SLAVE:
 948                 memcpy (&pic_irqchip(kvm)->pics[1],
 949                         &chip->chip.pic,
 950                         sizeof(struct kvm_pic_state));
 951                 break;
 952         case KVM_IRQCHIP_IOAPIC:
 953                 memcpy (ioapic_irqchip(kvm),
 954                         &chip->chip.ioapic,
 955                         sizeof(struct kvm_ioapic_state));
 956                 break;
 957         default:
 958                 r = -EINVAL;
 959                 break;
 960         }
 961         kvm_pic_update_irq(pic_irqchip(kvm));
 962         return r;
 963 }
 964
 965 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 966 {
 967         int i;
 968         struct kvm_mem_alias *alias;
 969
 970         for (i = 0; i < kvm->naliases; ++i) {
 971                 alias = &kvm->aliases[i];
 972                 if (gfn >= alias->base_gfn
 973                     && gfn < alias->base_gfn + alias->npages)
 974                         return alias->target_gfn + gfn - alias->base_gfn;
 975         }
 976         return gfn;
 977 }
 978
 979 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 980 {
 981         int i;
 982
 983         for (i = 0; i < kvm->nmemslots; ++i) {
 984                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 985
 986                 if (gfn >= memslot->base_gfn
 987                     && gfn < memslot->base_gfn + memslot->npages)
 988                         return memslot;
 989         }
 990         return NULL;
 991 }
 992
 993 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 994 {
 995         gfn = unalias_gfn(kvm, gfn);
 996         return __gfn_to_memslot(kvm, gfn);
 997 }
 998
 999 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1000 {
1001         struct kvm_memory_slot *slot;
1002
1003         gfn = unalias_gfn(kvm, gfn);
1004         slot = __gfn_to_memslot(kvm, gfn);
1005         if (!slot)
1006                 return NULL;
1007         return slot->phys_mem[gfn - slot->base_gfn];
1008 }
1009 EXPORT_SYMBOL_GPL(gfn_to_page);
1010
1011 /* WARNING: Does not work on aliased pages. */
1012 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1013 {
1014         struct kvm_memory_slot *memslot;
1015
1016         memslot = __gfn_to_memslot(kvm, gfn);
1017         if (memslot && memslot->dirty_bitmap) {
1018                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1019
1020                 /* avoid RMW */
1021                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1022                         set_bit(rel_gfn, memslot->dirty_bitmap);
1023         }
1024 }
1025
1026 int emulator_read_std(unsigned long addr,
1027                              void *val,
1028                              unsigned int bytes,
1029                              struct kvm_vcpu *vcpu)
1030 {
1031         void *data = val;
1032
1033         while (bytes) {
1034                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1035                 unsigned offset = addr & (PAGE_SIZE-1);
1036                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1037                 unsigned long pfn;
1038                 struct page *page;
1039                 void *page_virt;
1040
1041                 if (gpa == UNMAPPED_GVA)
1042                         return X86EMUL_PROPAGATE_FAULT;
1043                 pfn = gpa >> PAGE_SHIFT;
1044                 page = gfn_to_page(vcpu->kvm, pfn);
1045                 if (!page)
1046                         return X86EMUL_UNHANDLEABLE;
1047                 page_virt = kmap_atomic(page, KM_USER0);
1048
1049                 memcpy(data, page_virt + offset, tocopy);
1050
1051                 kunmap_atomic(page_virt, KM_USER0);
1052
1053                 bytes -= tocopy;
1054                 data += tocopy;
1055                 addr += tocopy;
1056         }
1057
1058         return X86EMUL_CONTINUE;
1059 }
1060 EXPORT_SYMBOL_GPL(emulator_read_std);
1061
1062 static int emulator_write_std(unsigned long addr,
1063                               const void *val,
1064                               unsigned int bytes,
1065                               struct kvm_vcpu *vcpu)
1066 {
1067         pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1068         return X86EMUL_UNHANDLEABLE;
1069 }
1070
1071 /*
1072  * Only apic need an MMIO device hook, so shortcut now..
1073  */
1074 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1075                                                 gpa_t addr)
1076 {
1077         struct kvm_io_device *dev;
1078
1079         if (vcpu->apic) {
1080                 dev = &vcpu->apic->dev;
1081                 if (dev->in_range(dev, addr))
1082                         return dev;
1083         }
1084         return NULL;
1085 }
1086
1087 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1088                                                 gpa_t addr)
1089 {
1090         struct kvm_io_device *dev;
1091
1092         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1093         if (dev == NULL)
1094                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1095         return dev;
1096 }
1097
1098 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1099                                                gpa_t addr)
1100 {
1101         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1102 }
1103
1104 static int emulator_read_emulated(unsigned long addr,
1105                                   void *val,
1106                                   unsigned int bytes,
1107                                   struct kvm_vcpu *vcpu)
1108 {
1109         struct kvm_io_device *mmio_dev;
1110         gpa_t                 gpa;
1111
1112         if (vcpu->mmio_read_completed) {
1113                 memcpy(val, vcpu->mmio_data, bytes);
1114                 vcpu->mmio_read_completed = 0;
1115                 return X86EMUL_CONTINUE;
1116         } else if (emulator_read_std(addr, val, bytes, vcpu)
1117                    == X86EMUL_CONTINUE)
1118                 return X86EMUL_CONTINUE;
1119
1120         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1121         if (gpa == UNMAPPED_GVA)
1122                 return X86EMUL_PROPAGATE_FAULT;
1123
1124         /*
1125          * Is this MMIO handled locally?
1126          */
1127         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1128         if (mmio_dev) {
1129                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1130                 return X86EMUL_CONTINUE;
1131         }
1132
1133         vcpu->mmio_needed = 1;
1134         vcpu->mmio_phys_addr = gpa;
1135         vcpu->mmio_size = bytes;
1136         vcpu->mmio_is_write = 0;
1137
1138         return X86EMUL_UNHANDLEABLE;
1139 }
1140
1141 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1142                                const void *val, int bytes)
1143 {
1144         struct page *page;
1145         void *virt;
1146
1147         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1148                 return 0;
1149         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1150         if (!page)
1151                 return 0;
1152         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1153         virt = kmap_atomic(page, KM_USER0);
1154         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1155         memcpy(virt + offset_in_page(gpa), val, bytes);
1156         kunmap_atomic(virt, KM_USER0);
1157         return 1;
1158 }
1159
1160 static int emulator_write_emulated_onepage(unsigned long addr,
1161                                            const void *val,
1162                                            unsigned int bytes,
1163                                            struct kvm_vcpu *vcpu)
1164 {
1165         struct kvm_io_device *mmio_dev;
1166         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1167
1168         if (gpa == UNMAPPED_GVA) {
1169                 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1170                 return X86EMUL_PROPAGATE_FAULT;
1171         }
1172
1173         if (emulator_write_phys(vcpu, gpa, val, bytes))
1174                 return X86EMUL_CONTINUE;
1175
1176         /*
1177          * Is this MMIO handled locally?
1178          */
1179         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1180         if (mmio_dev) {
1181                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1182                 return X86EMUL_CONTINUE;
1183         }
1184
1185         vcpu->mmio_needed = 1;
1186         vcpu->mmio_phys_addr = gpa;
1187         vcpu->mmio_size = bytes;
1188         vcpu->mmio_is_write = 1;
1189         memcpy(vcpu->mmio_data, val, bytes);
1190
1191         return X86EMUL_CONTINUE;
1192 }
1193
1194 int emulator_write_emulated(unsigned long addr,
1195                                    const void *val,
1196                                    unsigned int bytes,
1197                                    struct kvm_vcpu *vcpu)
1198 {
1199         /* Crossing a page boundary? */
1200         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1201                 int rc, now;
1202
1203                 now = -addr & ~PAGE_MASK;
1204                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1205                 if (rc != X86EMUL_CONTINUE)
1206                         return rc;
1207                 addr += now;
1208                 val += now;
1209                 bytes -= now;
1210         }
1211         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1212 }
1213 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1214
1215 static int emulator_cmpxchg_emulated(unsigned long addr,
1216                                      const void *old,
1217                                      const void *new,
1218                                      unsigned int bytes,
1219                                      struct kvm_vcpu *vcpu)
1220 {
1221         static int reported;
1222
1223         if (!reported) {
1224                 reported = 1;
1225                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1226         }
1227         return emulator_write_emulated(addr, new, bytes, vcpu);
1228 }
1229
1230 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1231 {
1232         return kvm_arch_ops->get_segment_base(vcpu, seg);
1233 }
1234
1235 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1236 {
1237         return X86EMUL_CONTINUE;
1238 }
1239
1240 int emulate_clts(struct kvm_vcpu *vcpu)
1241 {
1242         vcpu->cr0 &= ~X86_CR0_TS;
1243         kvm_arch_ops->set_cr0(vcpu, vcpu->cr0);
1244         return X86EMUL_CONTINUE;
1245 }
1246
1247 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1248 {
1249         struct kvm_vcpu *vcpu = ctxt->vcpu;
1250
1251         switch (dr) {
1252         case 0 ... 3:
1253                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1254                 return X86EMUL_CONTINUE;
1255         default:
1256                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1257                 return X86EMUL_UNHANDLEABLE;
1258         }
1259 }
1260
1261 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1262 {
1263         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1264         int exception;
1265
1266         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1267         if (exception) {
1268                 /* FIXME: better handling */
1269                 return X86EMUL_UNHANDLEABLE;
1270         }
1271         return X86EMUL_CONTINUE;
1272 }
1273
1274 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1275 {
1276         static int reported;
1277         u8 opcodes[4];
1278         unsigned long rip = ctxt->vcpu->rip;
1279         unsigned long rip_linear;
1280
1281         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1282
1283         if (reported)
1284                 return;
1285
1286         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt->vcpu);
1287
1288         printk(KERN_ERR "emulation failed but !mmio_needed?"
1289                " rip %lx %02x %02x %02x %02x\n",
1290                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1291         reported = 1;
1292 }
1293
1294 struct x86_emulate_ops emulate_ops = {
1295         .read_std            = emulator_read_std,
1296         .write_std           = emulator_write_std,
1297         .read_emulated       = emulator_read_emulated,
1298         .write_emulated      = emulator_write_emulated,
1299         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1300 };
1301
1302 int emulate_instruction(struct kvm_vcpu *vcpu,
1303                         struct kvm_run *run,
1304                         unsigned long cr2,
1305                         u16 error_code)
1306 {
1307         struct x86_emulate_ctxt emulate_ctxt;
1308         int r;
1309         int cs_db, cs_l;
1310
1311         vcpu->mmio_fault_cr2 = cr2;
1312         kvm_arch_ops->cache_regs(vcpu);
1313
1314         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1315
1316         emulate_ctxt.vcpu = vcpu;
1317         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1318         emulate_ctxt.cr2 = cr2;
1319         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1320                 ? X86EMUL_MODE_REAL : cs_l
1321                 ? X86EMUL_MODE_PROT64 : cs_db
1322                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1323
1324         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1325                 emulate_ctxt.cs_base = 0;
1326                 emulate_ctxt.ds_base = 0;
1327                 emulate_ctxt.es_base = 0;
1328                 emulate_ctxt.ss_base = 0;
1329         } else {
1330                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1331                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1332                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1333                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1334         }
1335
1336         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1337         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1338
1339         vcpu->mmio_is_write = 0;
1340         vcpu->pio.string = 0;
1341         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1342         if (vcpu->pio.string)
1343                 return EMULATE_DO_MMIO;
1344
1345         if ((r || vcpu->mmio_is_write) && run) {
1346                 run->exit_reason = KVM_EXIT_MMIO;
1347                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1348                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1349                 run->mmio.len = vcpu->mmio_size;
1350                 run->mmio.is_write = vcpu->mmio_is_write;
1351         }
1352
1353         if (r) {
1354                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1355                         return EMULATE_DONE;
1356                 if (!vcpu->mmio_needed) {
1357                         report_emulation_failure(&emulate_ctxt);
1358                         return EMULATE_FAIL;
1359                 }
1360                 return EMULATE_DO_MMIO;
1361         }
1362
1363         kvm_arch_ops->decache_regs(vcpu);
1364         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1365
1366         if (vcpu->mmio_is_write) {
1367                 vcpu->mmio_needed = 0;
1368                 return EMULATE_DO_MMIO;
1369         }
1370
1371         return EMULATE_DONE;
1372 }
1373 EXPORT_SYMBOL_GPL(emulate_instruction);
1374
1375 /*
1376  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1377  */
1378 static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1379 {
1380         DECLARE_WAITQUEUE(wait, current);
1381
1382         add_wait_queue(&vcpu->wq, &wait);
1383
1384         /*
1385          * We will block until either an interrupt or a signal wakes us up
1386          */
1387         while (!kvm_cpu_has_interrupt(vcpu)
1388                && !signal_pending(current)
1389                && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
1390                && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
1391                 set_current_state(TASK_INTERRUPTIBLE);
1392                 vcpu_put(vcpu);
1393                 schedule();
1394                 vcpu_load(vcpu);
1395         }
1396
1397         __set_current_state(TASK_RUNNING);
1398         remove_wait_queue(&vcpu->wq, &wait);
1399 }
1400
1401 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1402 {
1403         ++vcpu->stat.halt_exits;
1404         if (irqchip_in_kernel(vcpu->kvm)) {
1405                 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1406                 kvm_vcpu_block(vcpu);
1407                 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1408                         return -EINTR;
1409                 return 1;
1410         } else {
1411                 vcpu->run->exit_reason = KVM_EXIT_HLT;
1412                 return 0;
1413         }
1414 }
1415 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1416
1417 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1418 {
1419         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1420
1421         kvm_arch_ops->cache_regs(vcpu);
1422         ret = -KVM_EINVAL;
1423 #ifdef CONFIG_X86_64
1424         if (is_long_mode(vcpu)) {
1425                 nr = vcpu->regs[VCPU_REGS_RAX];
1426                 a0 = vcpu->regs[VCPU_REGS_RDI];
1427                 a1 = vcpu->regs[VCPU_REGS_RSI];
1428                 a2 = vcpu->regs[VCPU_REGS_RDX];
1429                 a3 = vcpu->regs[VCPU_REGS_RCX];
1430                 a4 = vcpu->regs[VCPU_REGS_R8];
1431                 a5 = vcpu->regs[VCPU_REGS_R9];
1432         } else
1433 #endif
1434         {
1435                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1436                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1437                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1438                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1439                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1440                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1441                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1442         }
1443         switch (nr) {
1444         default:
1445                 run->hypercall.nr = nr;
1446                 run->hypercall.args[0] = a0;
1447                 run->hypercall.args[1] = a1;
1448                 run->hypercall.args[2] = a2;
1449                 run->hypercall.args[3] = a3;
1450                 run->hypercall.args[4] = a4;
1451                 run->hypercall.args[5] = a5;
1452                 run->hypercall.ret = ret;
1453                 run->hypercall.longmode = is_long_mode(vcpu);
1454                 kvm_arch_ops->decache_regs(vcpu);
1455                 return 0;
1456         }
1457         vcpu->regs[VCPU_REGS_RAX] = ret;
1458         kvm_arch_ops->decache_regs(vcpu);
1459         return 1;
1460 }
1461 EXPORT_SYMBOL_GPL(kvm_hypercall);
1462
1463 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1464 {
1465         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1466 }
1467
1468 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1469 {
1470         struct descriptor_table dt = { limit, base };
1471
1472         kvm_arch_ops->set_gdt(vcpu, &dt);
1473 }
1474
1475 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1476 {
1477         struct descriptor_table dt = { limit, base };
1478
1479         kvm_arch_ops->set_idt(vcpu, &dt);
1480 }
1481
1482 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1483                    unsigned long *rflags)
1484 {
1485         lmsw(vcpu, msw);
1486         *rflags = kvm_arch_ops->get_rflags(vcpu);
1487 }
1488
1489 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1490 {
1491         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1492         switch (cr) {
1493         case 0:
1494                 return vcpu->cr0;
1495         case 2:
1496                 return vcpu->cr2;
1497         case 3:
1498                 return vcpu->cr3;
1499         case 4:
1500                 return vcpu->cr4;
1501         default:
1502                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1503                 return 0;
1504         }
1505 }
1506
1507 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1508                      unsigned long *rflags)
1509 {
1510         switch (cr) {
1511         case 0:
1512                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1513                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1514                 break;
1515         case 2:
1516                 vcpu->cr2 = val;
1517                 break;
1518         case 3:
1519                 set_cr3(vcpu, val);
1520                 break;
1521         case 4:
1522                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1523                 break;
1524         default:
1525                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1526         }
1527 }
1528
1529 /*
1530  * Register the para guest with the host:
1531  */
1532 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1533 {
1534         struct kvm_vcpu_para_state *para_state;
1535         hpa_t para_state_hpa, hypercall_hpa;
1536         struct page *para_state_page;
1537         unsigned char *hypercall;
1538         gpa_t hypercall_gpa;
1539
1540         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1541         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1542
1543         /*
1544          * Needs to be page aligned:
1545          */
1546         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1547                 goto err_gp;
1548
1549         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1550         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1551         if (is_error_hpa(para_state_hpa))
1552                 goto err_gp;
1553
1554         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1555         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1556         para_state = kmap(para_state_page);
1557
1558         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1559         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1560
1561         para_state->host_version = KVM_PARA_API_VERSION;
1562         /*
1563          * We cannot support guests that try to register themselves
1564          * with a newer API version than the host supports:
1565          */
1566         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1567                 para_state->ret = -KVM_EINVAL;
1568                 goto err_kunmap_skip;
1569         }
1570
1571         hypercall_gpa = para_state->hypercall_gpa;
1572         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1573         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1574         if (is_error_hpa(hypercall_hpa)) {
1575                 para_state->ret = -KVM_EINVAL;
1576                 goto err_kunmap_skip;
1577         }
1578
1579         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1580         vcpu->para_state_page = para_state_page;
1581         vcpu->para_state_gpa = para_state_gpa;
1582         vcpu->hypercall_gpa = hypercall_gpa;
1583
1584         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1585         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1586                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1587         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1588         kunmap_atomic(hypercall, KM_USER1);
1589
1590         para_state->ret = 0;
1591 err_kunmap_skip:
1592         kunmap(para_state_page);
1593         return 0;
1594 err_gp:
1595         return 1;
1596 }
1597
1598 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1599 {
1600         u64 data;
1601
1602         switch (msr) {
1603         case 0xc0010010: /* SYSCFG */
1604         case 0xc0010015: /* HWCR */
1605         case MSR_IA32_PLATFORM_ID:
1606         case MSR_IA32_P5_MC_ADDR:
1607         case MSR_IA32_P5_MC_TYPE:
1608         case MSR_IA32_MC0_CTL:
1609         case MSR_IA32_MCG_STATUS:
1610         case MSR_IA32_MCG_CAP:
1611         case MSR_IA32_MC0_MISC:
1612         case MSR_IA32_MC0_MISC+4:
1613         case MSR_IA32_MC0_MISC+8:
1614         case MSR_IA32_MC0_MISC+12:
1615         case MSR_IA32_MC0_MISC+16:
1616         case MSR_IA32_UCODE_REV:
1617         case MSR_IA32_PERF_STATUS:
1618         case MSR_IA32_EBL_CR_POWERON:
1619                 /* MTRR registers */
1620         case 0xfe:
1621         case 0x200 ... 0x2ff:
1622                 data = 0;
1623                 break;
1624         case 0xcd: /* fsb frequency */
1625                 data = 3;
1626                 break;
1627         case MSR_IA32_APICBASE:
1628                 data = kvm_get_apic_base(vcpu);
1629                 break;
1630         case MSR_IA32_MISC_ENABLE:
1631                 data = vcpu->ia32_misc_enable_msr;
1632                 break;
1633 #ifdef CONFIG_X86_64
1634         case MSR_EFER:
1635                 data = vcpu->shadow_efer;
1636                 break;
1637 #endif
1638         default:
1639                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1640                 return 1;
1641         }
1642         *pdata = data;
1643         return 0;
1644 }
1645 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1646
1647 /*
1648  * Reads an msr value (of 'msr_index') into 'pdata'.
1649  * Returns 0 on success, non-0 otherwise.
1650  * Assumes vcpu_load() was already called.
1651  */
1652 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1653 {
1654         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1655 }
1656
1657 #ifdef CONFIG_X86_64
1658
1659 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1660 {
1661         if (efer & EFER_RESERVED_BITS) {
1662                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1663                        efer);
1664                 inject_gp(vcpu);
1665                 return;
1666         }
1667
1668         if (is_paging(vcpu)
1669             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1670                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1671                 inject_gp(vcpu);
1672                 return;
1673         }
1674
1675         kvm_arch_ops->set_efer(vcpu, efer);
1676
1677         efer &= ~EFER_LMA;
1678         efer |= vcpu->shadow_efer & EFER_LMA;
1679
1680         vcpu->shadow_efer = efer;
1681 }
1682
1683 #endif
1684
1685 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1686 {
1687         switch (msr) {
1688 #ifdef CONFIG_X86_64
1689         case MSR_EFER:
1690                 set_efer(vcpu, data);
1691                 break;
1692 #endif
1693         case MSR_IA32_MC0_STATUS:
1694                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1695                        __FUNCTION__, data);
1696                 break;
1697         case MSR_IA32_MCG_STATUS:
1698                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1699                         __FUNCTION__, data);
1700                 break;
1701         case MSR_IA32_UCODE_REV:
1702         case MSR_IA32_UCODE_WRITE:
1703         case 0x200 ... 0x2ff: /* MTRRs */
1704                 break;
1705         case MSR_IA32_APICBASE:
1706                 kvm_set_apic_base(vcpu, data);
1707                 break;
1708         case MSR_IA32_MISC_ENABLE:
1709                 vcpu->ia32_misc_enable_msr = data;
1710                 break;
1711         /*
1712          * This is the 'probe whether the host is KVM' logic:
1713          */
1714         case MSR_KVM_API_MAGIC:
1715                 return vcpu_register_para(vcpu, data);
1716
1717         default:
1718                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1719                 return 1;
1720         }
1721         return 0;
1722 }
1723 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1724
1725 /*
1726  * Writes msr value into into the appropriate "register".
1727  * Returns 0 on success, non-0 otherwise.
1728  * Assumes vcpu_load() was already called.
1729  */
1730 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1731 {
1732         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1733 }
1734
1735 void kvm_resched(struct kvm_vcpu *vcpu)
1736 {
1737         if (!need_resched())
1738                 return;
1739         cond_resched();
1740 }
1741 EXPORT_SYMBOL_GPL(kvm_resched);
1742
1743 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1744 {
1745         int i;
1746         u32 function;
1747         struct kvm_cpuid_entry *e, *best;
1748
1749         kvm_arch_ops->cache_regs(vcpu);
1750         function = vcpu->regs[VCPU_REGS_RAX];
1751         vcpu->regs[VCPU_REGS_RAX] = 0;
1752         vcpu->regs[VCPU_REGS_RBX] = 0;
1753         vcpu->regs[VCPU_REGS_RCX] = 0;
1754         vcpu->regs[VCPU_REGS_RDX] = 0;
1755         best = NULL;
1756         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1757                 e = &vcpu->cpuid_entries[i];
1758                 if (e->function == function) {
1759                         best = e;
1760                         break;
1761                 }
1762                 /*
1763                  * Both basic or both extended?
1764                  */
1765                 if (((e->function ^ function) & 0x80000000) == 0)
1766                         if (!best || e->function > best->function)
1767                                 best = e;
1768         }
1769         if (best) {
1770                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1771                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1772                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1773                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1774         }
1775         kvm_arch_ops->decache_regs(vcpu);
1776         kvm_arch_ops->skip_emulated_instruction(vcpu);
1777 }
1778 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1779
1780 static int pio_copy_data(struct kvm_vcpu *vcpu)
1781 {
1782         void *p = vcpu->pio_data;
1783         void *q;
1784         unsigned bytes;
1785         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1786
1787         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1788                  PAGE_KERNEL);
1789         if (!q) {
1790                 free_pio_guest_pages(vcpu);
1791                 return -ENOMEM;
1792         }
1793         q += vcpu->pio.guest_page_offset;
1794         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1795         if (vcpu->pio.in)
1796                 memcpy(q, p, bytes);
1797         else
1798                 memcpy(p, q, bytes);
1799         q -= vcpu->pio.guest_page_offset;
1800         vunmap(q);
1801         free_pio_guest_pages(vcpu);
1802         return 0;
1803 }
1804
1805 static int complete_pio(struct kvm_vcpu *vcpu)
1806 {
1807         struct kvm_pio_request *io = &vcpu->pio;
1808         long delta;
1809         int r;
1810
1811         kvm_arch_ops->cache_regs(vcpu);
1812
1813         if (!io->string) {
1814                 if (io->in)
1815                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1816                                io->size);
1817         } else {
1818                 if (io->in) {
1819                         r = pio_copy_data(vcpu);
1820                         if (r) {
1821                                 kvm_arch_ops->cache_regs(vcpu);
1822                                 return r;
1823                         }
1824                 }
1825
1826                 delta = 1;
1827                 if (io->rep) {
1828                         delta *= io->cur_count;
1829                         /*
1830                          * The size of the register should really depend on
1831                          * current address size.
1832                          */
1833                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1834                 }
1835                 if (io->down)
1836                         delta = -delta;
1837                 delta *= io->size;
1838                 if (io->in)
1839                         vcpu->regs[VCPU_REGS_RDI] += delta;
1840                 else
1841                         vcpu->regs[VCPU_REGS_RSI] += delta;
1842         }
1843
1844         kvm_arch_ops->decache_regs(vcpu);
1845
1846         io->count -= io->cur_count;
1847         io->cur_count = 0;
1848
1849         if (!io->count)
1850                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1851         return 0;
1852 }
1853
1854 static void kernel_pio(struct kvm_io_device *pio_dev,
1855                        struct kvm_vcpu *vcpu,
1856                        void *pd)
1857 {
1858         /* TODO: String I/O for in kernel device */
1859
1860         mutex_lock(&vcpu->kvm->lock);
1861         if (vcpu->pio.in)
1862                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1863                                   vcpu->pio.size,
1864                                   pd);
1865         else
1866                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1867                                    vcpu->pio.size,
1868                                    pd);
1869         mutex_unlock(&vcpu->kvm->lock);
1870 }
1871
1872 static void pio_string_write(struct kvm_io_device *pio_dev,
1873                              struct kvm_vcpu *vcpu)
1874 {
1875         struct kvm_pio_request *io = &vcpu->pio;
1876         void *pd = vcpu->pio_data;
1877         int i;
1878
1879         mutex_lock(&vcpu->kvm->lock);
1880         for (i = 0; i < io->cur_count; i++) {
1881                 kvm_iodevice_write(pio_dev, io->port,
1882                                    io->size,
1883                                    pd);
1884                 pd += io->size;
1885         }
1886         mutex_unlock(&vcpu->kvm->lock);
1887 }
1888
1889 int kvm_emulate_pio (struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1890                   int size, unsigned port)
1891 {
1892         struct kvm_io_device *pio_dev;
1893
1894         vcpu->run->exit_reason = KVM_EXIT_IO;
1895         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1896         vcpu->run->io.size = vcpu->pio.size = size;
1897         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1898         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1899         vcpu->run->io.port = vcpu->pio.port = port;
1900         vcpu->pio.in = in;
1901         vcpu->pio.string = 0;
1902         vcpu->pio.down = 0;
1903         vcpu->pio.guest_page_offset = 0;
1904         vcpu->pio.rep = 0;
1905
1906         kvm_arch_ops->cache_regs(vcpu);
1907         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1908         kvm_arch_ops->decache_regs(vcpu);
1909
1910         pio_dev = vcpu_find_pio_dev(vcpu, port);
1911         if (pio_dev) {
1912                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1913                 complete_pio(vcpu);
1914                 return 1;
1915         }
1916         return 0;
1917 }
1918 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1919
1920 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1921                   int size, unsigned long count, int down,
1922                   gva_t address, int rep, unsigned port)
1923 {
1924         unsigned now, in_page;
1925         int i, ret = 0;
1926         int nr_pages = 1;
1927         struct page *page;
1928         struct kvm_io_device *pio_dev;
1929
1930         vcpu->run->exit_reason = KVM_EXIT_IO;
1931         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1932         vcpu->run->io.size = vcpu->pio.size = size;
1933         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1934         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1935         vcpu->run->io.port = vcpu->pio.port = port;
1936         vcpu->pio.in = in;
1937         vcpu->pio.string = 1;
1938         vcpu->pio.down = down;
1939         vcpu->pio.guest_page_offset = offset_in_page(address);
1940         vcpu->pio.rep = rep;
1941
1942         if (!count) {
1943                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1944                 return 1;
1945         }
1946
1947         if (!down)
1948                 in_page = PAGE_SIZE - offset_in_page(address);
1949         else
1950                 in_page = offset_in_page(address) + size;
1951         now = min(count, (unsigned long)in_page / size);
1952         if (!now) {
1953                 /*
1954                  * String I/O straddles page boundary.  Pin two guest pages
1955                  * so that we satisfy atomicity constraints.  Do just one
1956                  * transaction to avoid complexity.
1957                  */
1958                 nr_pages = 2;
1959                 now = 1;
1960         }
1961         if (down) {
1962                 /*
1963                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1964                  */
1965                 pr_unimpl(vcpu, "guest string pio down\n");
1966                 inject_gp(vcpu);
1967                 return 1;
1968         }
1969         vcpu->run->io.count = now;
1970         vcpu->pio.cur_count = now;
1971
1972         for (i = 0; i < nr_pages; ++i) {
1973                 mutex_lock(&vcpu->kvm->lock);
1974                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1975                 if (page)
1976                         get_page(page);
1977                 vcpu->pio.guest_pages[i] = page;
1978                 mutex_unlock(&vcpu->kvm->lock);
1979                 if (!page) {
1980                         inject_gp(vcpu);
1981                         free_pio_guest_pages(vcpu);
1982                         return 1;
1983                 }
1984         }
1985
1986         pio_dev = vcpu_find_pio_dev(vcpu, port);
1987         if (!vcpu->pio.in) {
1988                 /* string PIO write */
1989                 ret = pio_copy_data(vcpu);
1990                 if (ret >= 0 && pio_dev) {
1991                         pio_string_write(pio_dev, vcpu);
1992                         complete_pio(vcpu);
1993                         if (vcpu->pio.count == 0)
1994                                 ret = 1;
1995                 }
1996         } else if (pio_dev)
1997                 pr_unimpl(vcpu, "no string pio read support yet, "
1998                        "port %x size %d count %ld\n",
1999                         port, size, count);
2000
2001         return ret;
2002 }
2003 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2004
2005 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2006 {
2007         int r;
2008         sigset_t sigsaved;
2009
2010         vcpu_load(vcpu);
2011
2012         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2013                 kvm_vcpu_block(vcpu);
2014                 vcpu_put(vcpu);
2015                 return -EAGAIN;
2016         }
2017
2018         if (vcpu->sigset_active)
2019                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2020
2021         /* re-sync apic's tpr */
2022         if (!irqchip_in_kernel(vcpu->kvm))
2023                 set_cr8(vcpu, kvm_run->cr8);
2024
2025         if (vcpu->pio.cur_count) {
2026                 r = complete_pio(vcpu);
2027                 if (r)
2028                         goto out;
2029         }
2030
2031         if (vcpu->mmio_needed) {
2032                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2033                 vcpu->mmio_read_completed = 1;
2034                 vcpu->mmio_needed = 0;
2035                 r = emulate_instruction(vcpu, kvm_run,
2036                                         vcpu->mmio_fault_cr2, 0);
2037                 if (r == EMULATE_DO_MMIO) {
2038                         /*
2039                          * Read-modify-write.  Back to userspace.
2040                          */
2041                         r = 0;
2042                         goto out;
2043                 }
2044         }
2045
2046         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2047                 kvm_arch_ops->cache_regs(vcpu);
2048                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2049                 kvm_arch_ops->decache_regs(vcpu);
2050         }
2051
2052         r = kvm_arch_ops->run(vcpu, kvm_run);
2053
2054 out:
2055         if (vcpu->sigset_active)
2056                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2057
2058         vcpu_put(vcpu);
2059         return r;
2060 }
2061
2062 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2063                                    struct kvm_regs *regs)
2064 {
2065         vcpu_load(vcpu);
2066
2067         kvm_arch_ops->cache_regs(vcpu);
2068
2069         regs->rax = vcpu->regs[VCPU_REGS_RAX];
2070         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2071         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2072         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2073         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2074         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2075         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2076         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2077 #ifdef CONFIG_X86_64
2078         regs->r8 = vcpu->regs[VCPU_REGS_R8];
2079         regs->r9 = vcpu->regs[VCPU_REGS_R9];
2080         regs->r10 = vcpu->regs[VCPU_REGS_R10];
2081         regs->r11 = vcpu->regs[VCPU_REGS_R11];
2082         regs->r12 = vcpu->regs[VCPU_REGS_R12];
2083         regs->r13 = vcpu->regs[VCPU_REGS_R13];
2084         regs->r14 = vcpu->regs[VCPU_REGS_R14];
2085         regs->r15 = vcpu->regs[VCPU_REGS_R15];
2086 #endif
2087
2088         regs->rip = vcpu->rip;
2089         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
2090
2091         /*
2092          * Don't leak debug flags in case they were set for guest debugging
2093          */
2094         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2095                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2096
2097         vcpu_put(vcpu);
2098
2099         return 0;
2100 }
2101
2102 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2103                                    struct kvm_regs *regs)
2104 {
2105         vcpu_load(vcpu);
2106
2107         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2108         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2109         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2110         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2111         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2112         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2113         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2114         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2115 #ifdef CONFIG_X86_64
2116         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2117         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2118         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2119         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2120         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2121         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2122         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2123         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2124 #endif
2125
2126         vcpu->rip = regs->rip;
2127         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
2128
2129         kvm_arch_ops->decache_regs(vcpu);
2130
2131         vcpu_put(vcpu);
2132
2133         return 0;
2134 }
2135
2136 static void get_segment(struct kvm_vcpu *vcpu,
2137                         struct kvm_segment *var, int seg)
2138 {
2139         return kvm_arch_ops->get_segment(vcpu, var, seg);
2140 }
2141
2142 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2143                                     struct kvm_sregs *sregs)
2144 {
2145         struct descriptor_table dt;
2146         int pending_vec;
2147
2148         vcpu_load(vcpu);
2149
2150         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2151         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2152         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2153         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2154         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2155         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2156
2157         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2158         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2159
2160         kvm_arch_ops->get_idt(vcpu, &dt);
2161         sregs->idt.limit = dt.limit;
2162         sregs->idt.base = dt.base;
2163         kvm_arch_ops->get_gdt(vcpu, &dt);
2164         sregs->gdt.limit = dt.limit;
2165         sregs->gdt.base = dt.base;
2166
2167         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2168         sregs->cr0 = vcpu->cr0;
2169         sregs->cr2 = vcpu->cr2;
2170         sregs->cr3 = vcpu->cr3;
2171         sregs->cr4 = vcpu->cr4;
2172         sregs->cr8 = get_cr8(vcpu);
2173         sregs->efer = vcpu->shadow_efer;
2174         sregs->apic_base = kvm_get_apic_base(vcpu);
2175
2176         if (irqchip_in_kernel(vcpu->kvm)) {
2177                 memset(sregs->interrupt_bitmap, 0,
2178                        sizeof sregs->interrupt_bitmap);
2179                 pending_vec = kvm_arch_ops->get_irq(vcpu);
2180                 if (pending_vec >= 0)
2181                         set_bit(pending_vec, (unsigned long *)sregs->interrupt_bitmap);
2182         } else
2183                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2184                        sizeof sregs->interrupt_bitmap);
2185
2186         vcpu_put(vcpu);
2187
2188         return 0;
2189 }
2190
2191 static void set_segment(struct kvm_vcpu *vcpu,
2192                         struct kvm_segment *var, int seg)
2193 {
2194         return kvm_arch_ops->set_segment(vcpu, var, seg);
2195 }
2196
2197 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2198                                     struct kvm_sregs *sregs)
2199 {
2200         int mmu_reset_needed = 0;
2201         int i, pending_vec, max_bits;
2202         struct descriptor_table dt;
2203
2204         vcpu_load(vcpu);
2205
2206         dt.limit = sregs->idt.limit;
2207         dt.base = sregs->idt.base;
2208         kvm_arch_ops->set_idt(vcpu, &dt);
2209         dt.limit = sregs->gdt.limit;
2210         dt.base = sregs->gdt.base;
2211         kvm_arch_ops->set_gdt(vcpu, &dt);
2212
2213         vcpu->cr2 = sregs->cr2;
2214         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2215         vcpu->cr3 = sregs->cr3;
2216
2217         set_cr8(vcpu, sregs->cr8);
2218
2219         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2220 #ifdef CONFIG_X86_64
2221         kvm_arch_ops->set_efer(vcpu, sregs->efer);
2222 #endif
2223         kvm_set_apic_base(vcpu, sregs->apic_base);
2224
2225         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2226
2227         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2228         vcpu->cr0 = sregs->cr0;
2229         kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2230
2231         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2232         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2233         if (!is_long_mode(vcpu) && is_pae(vcpu))
2234                 load_pdptrs(vcpu, vcpu->cr3);
2235
2236         if (mmu_reset_needed)
2237                 kvm_mmu_reset_context(vcpu);
2238
2239         if (!irqchip_in_kernel(vcpu->kvm)) {
2240                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2241                        sizeof vcpu->irq_pending);
2242                 vcpu->irq_summary = 0;
2243                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2244                         if (vcpu->irq_pending[i])
2245                                 __set_bit(i, &vcpu->irq_summary);
2246         } else {
2247                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2248                 pending_vec = find_first_bit(
2249                         (const unsigned long *)sregs->interrupt_bitmap,
2250                         max_bits);
2251                 /* Only pending external irq is handled here */
2252                 if (pending_vec < max_bits) {
2253                         kvm_arch_ops->set_irq(vcpu, pending_vec);
2254                         printk("Set back pending irq %d\n", pending_vec);
2255                 }
2256         }
2257
2258         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2259         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2260         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2261         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2262         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2263         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2264
2265         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2266         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2267
2268         vcpu_put(vcpu);
2269
2270         return 0;
2271 }
2272
2273 /*
2274  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2275  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2276  *
2277  * This list is modified at module load time to reflect the
2278  * capabilities of the host cpu.
2279  */
2280 static u32 msrs_to_save[] = {
2281         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2282         MSR_K6_STAR,
2283 #ifdef CONFIG_X86_64
2284         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2285 #endif
2286         MSR_IA32_TIME_STAMP_COUNTER,
2287 };
2288
2289 static unsigned num_msrs_to_save;
2290
2291 static u32 emulated_msrs[] = {
2292         MSR_IA32_MISC_ENABLE,
2293 };
2294
2295 static __init void kvm_init_msr_list(void)
2296 {
2297         u32 dummy[2];
2298         unsigned i, j;
2299
2300         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2301                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2302                         continue;
2303                 if (j < i)
2304                         msrs_to_save[j] = msrs_to_save[i];
2305                 j++;
2306         }
2307         num_msrs_to_save = j;
2308 }
2309
2310 /*
2311  * Adapt set_msr() to msr_io()'s calling convention
2312  */
2313 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2314 {
2315         return kvm_set_msr(vcpu, index, *data);
2316 }
2317
2318 /*
2319  * Read or write a bunch of msrs. All parameters are kernel addresses.
2320  *
2321  * @return number of msrs set successfully.
2322  */
2323 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2324                     struct kvm_msr_entry *entries,
2325                     int (*do_msr)(struct kvm_vcpu *vcpu,
2326                                   unsigned index, u64 *data))
2327 {
2328         int i;
2329
2330         vcpu_load(vcpu);
2331
2332         for (i = 0; i < msrs->nmsrs; ++i)
2333                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2334                         break;
2335
2336         vcpu_put(vcpu);
2337
2338         return i;
2339 }
2340
2341 /*
2342  * Read or write a bunch of msrs. Parameters are user addresses.
2343  *
2344  * @return number of msrs set successfully.
2345  */
2346 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2347                   int (*do_msr)(struct kvm_vcpu *vcpu,
2348                                 unsigned index, u64 *data),
2349                   int writeback)
2350 {
2351         struct kvm_msrs msrs;
2352         struct kvm_msr_entry *entries;
2353         int r, n;
2354         unsigned size;
2355
2356         r = -EFAULT;
2357         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2358                 goto out;
2359
2360         r = -E2BIG;
2361         if (msrs.nmsrs >= MAX_IO_MSRS)
2362                 goto out;
2363
2364         r = -ENOMEM;
2365         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2366         entries = vmalloc(size);
2367         if (!entries)
2368                 goto out;
2369
2370         r = -EFAULT;
2371         if (copy_from_user(entries, user_msrs->entries, size))
2372                 goto out_free;
2373
2374         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2375         if (r < 0)
2376                 goto out_free;
2377
2378         r = -EFAULT;
2379         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2380                 goto out_free;
2381
2382         r = n;
2383
2384 out_free:
2385         vfree(entries);
2386 out:
2387         return r;
2388 }
2389
2390 /*
2391  * Translate a guest virtual address to a guest physical address.
2392  */
2393 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2394                                     struct kvm_translation *tr)
2395 {
2396         unsigned long vaddr = tr->linear_address;
2397         gpa_t gpa;
2398
2399         vcpu_load(vcpu);
2400         mutex_lock(&vcpu->kvm->lock);
2401         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2402         tr->physical_address = gpa;
2403         tr->valid = gpa != UNMAPPED_GVA;
2404         tr->writeable = 1;
2405         tr->usermode = 0;
2406         mutex_unlock(&vcpu->kvm->lock);
2407         vcpu_put(vcpu);
2408
2409         return 0;
2410 }
2411
2412 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2413                                     struct kvm_interrupt *irq)
2414 {
2415         if (irq->irq < 0 || irq->irq >= 256)
2416                 return -EINVAL;
2417         if (irqchip_in_kernel(vcpu->kvm))
2418                 return -ENXIO;
2419         vcpu_load(vcpu);
2420
2421         set_bit(irq->irq, vcpu->irq_pending);
2422         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2423
2424         vcpu_put(vcpu);
2425
2426         return 0;
2427 }
2428
2429 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2430                                       struct kvm_debug_guest *dbg)
2431 {
2432         int r;
2433
2434         vcpu_load(vcpu);
2435
2436         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2437
2438         vcpu_put(vcpu);
2439
2440         return r;
2441 }
2442
2443 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2444                                     unsigned long address,
2445                                     int *type)
2446 {
2447         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2448         unsigned long pgoff;
2449         struct page *page;
2450
2451         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2452         if (pgoff == 0)
2453                 page = virt_to_page(vcpu->run);
2454         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2455                 page = virt_to_page(vcpu->pio_data);
2456         else
2457                 return NOPAGE_SIGBUS;
2458         get_page(page);
2459         if (type != NULL)
2460                 *type = VM_FAULT_MINOR;
2461
2462         return page;
2463 }
2464
2465 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2466         .nopage = kvm_vcpu_nopage,
2467 };
2468
2469 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2470 {
2471         vma->vm_ops = &kvm_vcpu_vm_ops;
2472         return 0;
2473 }
2474
2475 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2476 {
2477         struct kvm_vcpu *vcpu = filp->private_data;
2478
2479         fput(vcpu->kvm->filp);
2480         return 0;
2481 }
2482
2483 static struct file_operations kvm_vcpu_fops = {
2484         .release        = kvm_vcpu_release,
2485         .unlocked_ioctl = kvm_vcpu_ioctl,
2486         .compat_ioctl   = kvm_vcpu_ioctl,
2487         .mmap           = kvm_vcpu_mmap,
2488 };
2489
2490 /*
2491  * Allocates an inode for the vcpu.
2492  */
2493 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2494 {
2495         int fd, r;
2496         struct inode *inode;
2497         struct file *file;
2498
2499         r = anon_inode_getfd(&fd, &inode, &file,
2500                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2501         if (r)
2502                 return r;
2503         atomic_inc(&vcpu->kvm->filp->f_count);
2504         return fd;
2505 }
2506
2507 /*
2508  * Creates some virtual cpus.  Good luck creating more than one.
2509  */
2510 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2511 {
2512         int r;
2513         struct kvm_vcpu *vcpu;
2514
2515         if (!valid_vcpu(n))
2516                 return -EINVAL;
2517
2518         vcpu = kvm_arch_ops->vcpu_create(kvm, n);
2519         if (IS_ERR(vcpu))
2520                 return PTR_ERR(vcpu);
2521
2522         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2523
2524         /* We do fxsave: this must be aligned. */
2525         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2526
2527         vcpu_load(vcpu);
2528         r = kvm_mmu_setup(vcpu);
2529         vcpu_put(vcpu);
2530         if (r < 0)
2531                 goto free_vcpu;
2532
2533         mutex_lock(&kvm->lock);
2534         if (kvm->vcpus[n]) {
2535                 r = -EEXIST;
2536                 mutex_unlock(&kvm->lock);
2537                 goto mmu_unload;
2538         }
2539         kvm->vcpus[n] = vcpu;
2540         mutex_unlock(&kvm->lock);
2541
2542         /* Now it's all set up, let userspace reach it */
2543         r = create_vcpu_fd(vcpu);
2544         if (r < 0)
2545                 goto unlink;
2546         return r;
2547
2548 unlink:
2549         mutex_lock(&kvm->lock);
2550         kvm->vcpus[n] = NULL;
2551         mutex_unlock(&kvm->lock);
2552
2553 mmu_unload:
2554         vcpu_load(vcpu);
2555         kvm_mmu_unload(vcpu);
2556         vcpu_put(vcpu);
2557
2558 free_vcpu:
2559         kvm_arch_ops->vcpu_free(vcpu);
2560         return r;
2561 }
2562
2563 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2564 {
2565         u64 efer;
2566         int i;
2567         struct kvm_cpuid_entry *e, *entry;
2568
2569         rdmsrl(MSR_EFER, efer);
2570         entry = NULL;
2571         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2572                 e = &vcpu->cpuid_entries[i];
2573                 if (e->function == 0x80000001) {
2574                         entry = e;
2575                         break;
2576                 }
2577         }
2578         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2579                 entry->edx &= ~(1 << 20);
2580                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2581         }
2582 }
2583
2584 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2585                                     struct kvm_cpuid *cpuid,
2586                                     struct kvm_cpuid_entry __user *entries)
2587 {
2588         int r;
2589
2590         r = -E2BIG;
2591         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2592                 goto out;
2593         r = -EFAULT;
2594         if (copy_from_user(&vcpu->cpuid_entries, entries,
2595                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2596                 goto out;
2597         vcpu->cpuid_nent = cpuid->nent;
2598         cpuid_fix_nx_cap(vcpu);
2599         return 0;
2600
2601 out:
2602         return r;
2603 }
2604
2605 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2606 {
2607         if (sigset) {
2608                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2609                 vcpu->sigset_active = 1;
2610                 vcpu->sigset = *sigset;
2611         } else
2612                 vcpu->sigset_active = 0;
2613         return 0;
2614 }
2615
2616 /*
2617  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2618  * we have asm/x86/processor.h
2619  */
2620 struct fxsave {
2621         u16     cwd;
2622         u16     swd;
2623         u16     twd;
2624         u16     fop;
2625         u64     rip;
2626         u64     rdp;
2627         u32     mxcsr;
2628         u32     mxcsr_mask;
2629         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2630 #ifdef CONFIG_X86_64
2631         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2632 #else
2633         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2634 #endif
2635 };
2636
2637 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2638 {
2639         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2640
2641         vcpu_load(vcpu);
2642
2643         memcpy(fpu->fpr, fxsave->st_space, 128);
2644         fpu->fcw = fxsave->cwd;
2645         fpu->fsw = fxsave->swd;
2646         fpu->ftwx = fxsave->twd;
2647         fpu->last_opcode = fxsave->fop;
2648         fpu->last_ip = fxsave->rip;
2649         fpu->last_dp = fxsave->rdp;
2650         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2651
2652         vcpu_put(vcpu);
2653
2654         return 0;
2655 }
2656
2657 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2658 {
2659         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2660
2661         vcpu_load(vcpu);
2662
2663         memcpy(fxsave->st_space, fpu->fpr, 128);
2664         fxsave->cwd = fpu->fcw;
2665         fxsave->swd = fpu->fsw;
2666         fxsave->twd = fpu->ftwx;
2667         fxsave->fop = fpu->last_opcode;
2668         fxsave->rip = fpu->last_ip;
2669         fxsave->rdp = fpu->last_dp;
2670         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2671
2672         vcpu_put(vcpu);
2673
2674         return 0;
2675 }
2676
2677 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2678                                     struct kvm_lapic_state *s)
2679 {
2680         vcpu_load(vcpu);
2681         memcpy(s->regs, vcpu->apic->regs, sizeof *s);
2682         vcpu_put(vcpu);
2683
2684         return 0;
2685 }
2686
2687 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2688                                     struct kvm_lapic_state *s)
2689 {
2690         vcpu_load(vcpu);
2691         memcpy(vcpu->apic->regs, s->regs, sizeof *s);
2692         kvm_apic_post_state_restore(vcpu);
2693         vcpu_put(vcpu);
2694
2695         return 0;
2696 }
2697
2698 static long kvm_vcpu_ioctl(struct file *filp,
2699                            unsigned int ioctl, unsigned long arg)
2700 {
2701         struct kvm_vcpu *vcpu = filp->private_data;
2702         void __user *argp = (void __user *)arg;
2703         int r = -EINVAL;
2704
2705         switch (ioctl) {
2706         case KVM_RUN:
2707                 r = -EINVAL;
2708                 if (arg)
2709                         goto out;
2710                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2711                 break;
2712         case KVM_GET_REGS: {
2713                 struct kvm_regs kvm_regs;
2714
2715                 memset(&kvm_regs, 0, sizeof kvm_regs);
2716                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2717                 if (r)
2718                         goto out;
2719                 r = -EFAULT;
2720                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2721                         goto out;
2722                 r = 0;
2723                 break;
2724         }
2725         case KVM_SET_REGS: {
2726                 struct kvm_regs kvm_regs;
2727
2728                 r = -EFAULT;
2729                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2730                         goto out;
2731                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2732                 if (r)
2733                         goto out;
2734                 r = 0;
2735                 break;
2736         }
2737         case KVM_GET_SREGS: {
2738                 struct kvm_sregs kvm_sregs;
2739
2740                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2741                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2742                 if (r)
2743                         goto out;
2744                 r = -EFAULT;
2745                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2746                         goto out;
2747                 r = 0;
2748                 break;
2749         }
2750         case KVM_SET_SREGS: {
2751                 struct kvm_sregs kvm_sregs;
2752
2753                 r = -EFAULT;
2754                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2755                         goto out;
2756                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2757                 if (r)
2758                         goto out;
2759                 r = 0;
2760                 break;
2761         }
2762         case KVM_TRANSLATE: {
2763                 struct kvm_translation tr;
2764
2765                 r = -EFAULT;
2766                 if (copy_from_user(&tr, argp, sizeof tr))
2767                         goto out;
2768                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2769                 if (r)
2770                         goto out;
2771                 r = -EFAULT;
2772                 if (copy_to_user(argp, &tr, sizeof tr))
2773                         goto out;
2774                 r = 0;
2775                 break;
2776         }
2777         case KVM_INTERRUPT: {
2778                 struct kvm_interrupt irq;
2779
2780                 r = -EFAULT;
2781                 if (copy_from_user(&irq, argp, sizeof irq))
2782                         goto out;
2783                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2784                 if (r)
2785                         goto out;
2786                 r = 0;
2787                 break;
2788         }
2789         case KVM_DEBUG_GUEST: {
2790                 struct kvm_debug_guest dbg;
2791
2792                 r = -EFAULT;
2793                 if (copy_from_user(&dbg, argp, sizeof dbg))
2794                         goto out;
2795                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2796                 if (r)
2797                         goto out;
2798                 r = 0;
2799                 break;
2800         }
2801         case KVM_GET_MSRS:
2802                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2803                 break;
2804         case KVM_SET_MSRS:
2805                 r = msr_io(vcpu, argp, do_set_msr, 0);
2806                 break;
2807         case KVM_SET_CPUID: {
2808                 struct kvm_cpuid __user *cpuid_arg = argp;
2809                 struct kvm_cpuid cpuid;
2810
2811                 r = -EFAULT;
2812                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2813                         goto out;
2814                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2815                 if (r)
2816                         goto out;
2817                 break;
2818         }
2819         case KVM_SET_SIGNAL_MASK: {
2820                 struct kvm_signal_mask __user *sigmask_arg = argp;
2821                 struct kvm_signal_mask kvm_sigmask;
2822                 sigset_t sigset, *p;
2823
2824                 p = NULL;
2825                 if (argp) {
2826                         r = -EFAULT;
2827                         if (copy_from_user(&kvm_sigmask, argp,
2828                                            sizeof kvm_sigmask))
2829                                 goto out;
2830                         r = -EINVAL;
2831                         if (kvm_sigmask.len != sizeof sigset)
2832                                 goto out;
2833                         r = -EFAULT;
2834                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2835                                            sizeof sigset))
2836                                 goto out;
2837                         p = &sigset;
2838                 }
2839                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2840                 break;
2841         }
2842         case KVM_GET_FPU: {
2843                 struct kvm_fpu fpu;
2844
2845                 memset(&fpu, 0, sizeof fpu);
2846                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2847                 if (r)
2848                         goto out;
2849                 r = -EFAULT;
2850                 if (copy_to_user(argp, &fpu, sizeof fpu))
2851                         goto out;
2852                 r = 0;
2853                 break;
2854         }
2855         case KVM_SET_FPU: {
2856                 struct kvm_fpu fpu;
2857
2858                 r = -EFAULT;
2859                 if (copy_from_user(&fpu, argp, sizeof fpu))
2860                         goto out;
2861                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2862                 if (r)
2863                         goto out;
2864                 r = 0;
2865                 break;
2866         }
2867         case KVM_GET_LAPIC: {
2868                 struct kvm_lapic_state lapic;
2869
2870                 memset(&lapic, 0, sizeof lapic);
2871                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
2872                 if (r)
2873                         goto out;
2874                 r = -EFAULT;
2875                 if (copy_to_user(argp, &lapic, sizeof lapic))
2876                         goto out;
2877                 r = 0;
2878                 break;
2879         }
2880         case KVM_SET_LAPIC: {
2881                 struct kvm_lapic_state lapic;
2882
2883                 r = -EFAULT;
2884                 if (copy_from_user(&lapic, argp, sizeof lapic))
2885                         goto out;
2886                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
2887                 if (r)
2888                         goto out;
2889                 r = 0;
2890                 break;
2891         }
2892         default:
2893                 ;
2894         }
2895 out:
2896         return r;
2897 }
2898
2899 static long kvm_vm_ioctl(struct file *filp,
2900                            unsigned int ioctl, unsigned long arg)
2901 {
2902         struct kvm *kvm = filp->private_data;
2903         void __user *argp = (void __user *)arg;
2904         int r = -EINVAL;
2905
2906         switch (ioctl) {
2907         case KVM_CREATE_VCPU:
2908                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2909                 if (r < 0)
2910                         goto out;
2911                 break;
2912         case KVM_SET_MEMORY_REGION: {
2913                 struct kvm_memory_region kvm_mem;
2914
2915                 r = -EFAULT;
2916                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2917                         goto out;
2918                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2919                 if (r)
2920                         goto out;
2921                 break;
2922         }
2923         case KVM_GET_DIRTY_LOG: {
2924                 struct kvm_dirty_log log;
2925
2926                 r = -EFAULT;
2927                 if (copy_from_user(&log, argp, sizeof log))
2928                         goto out;
2929                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2930                 if (r)
2931                         goto out;
2932                 break;
2933         }
2934         case KVM_SET_MEMORY_ALIAS: {
2935                 struct kvm_memory_alias alias;
2936
2937                 r = -EFAULT;
2938                 if (copy_from_user(&alias, argp, sizeof alias))
2939                         goto out;
2940                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2941                 if (r)
2942                         goto out;
2943                 break;
2944         }
2945         case KVM_CREATE_IRQCHIP:
2946                 r = -ENOMEM;
2947                 kvm->vpic = kvm_create_pic(kvm);
2948                 if (kvm->vpic) {
2949                         r = kvm_ioapic_init(kvm);
2950                         if (r) {
2951                                 kfree(kvm->vpic);
2952                                 kvm->vpic = NULL;
2953                                 goto out;
2954                         }
2955                 }
2956                 else
2957                         goto out;
2958                 break;
2959         case KVM_IRQ_LINE: {
2960                 struct kvm_irq_level irq_event;
2961
2962                 r = -EFAULT;
2963                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2964                         goto out;
2965                 if (irqchip_in_kernel(kvm)) {
2966                         mutex_lock(&kvm->lock);
2967                         if (irq_event.irq < 16)
2968                                 kvm_pic_set_irq(pic_irqchip(kvm),
2969                                         irq_event.irq,
2970                                         irq_event.level);
2971                         kvm_ioapic_set_irq(kvm->vioapic,
2972                                         irq_event.irq,
2973                                         irq_event.level);
2974                         mutex_unlock(&kvm->lock);
2975                         r = 0;
2976                 }
2977                 break;
2978         }
2979         case KVM_GET_IRQCHIP: {
2980                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2981                 struct kvm_irqchip chip;
2982
2983                 r = -EFAULT;
2984                 if (copy_from_user(&chip, argp, sizeof chip))
2985                         goto out;
2986                 r = -ENXIO;
2987                 if (!irqchip_in_kernel(kvm))
2988                         goto out;
2989                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
2990                 if (r)
2991                         goto out;
2992                 r = -EFAULT;
2993                 if (copy_to_user(argp, &chip, sizeof chip))
2994                         goto out;
2995                 r = 0;
2996                 break;
2997         }
2998         case KVM_SET_IRQCHIP: {
2999                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3000                 struct kvm_irqchip chip;
3001
3002                 r = -EFAULT;
3003                 if (copy_from_user(&chip, argp, sizeof chip))
3004                         goto out;
3005                 r = -ENXIO;
3006                 if (!irqchip_in_kernel(kvm))
3007                         goto out;
3008                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
3009                 if (r)
3010                         goto out;
3011                 r = 0;
3012                 break;
3013         }
3014         default:
3015                 ;
3016         }
3017 out:
3018         return r;
3019 }
3020
3021 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
3022                                   unsigned long address,
3023                                   int *type)
3024 {
3025         struct kvm *kvm = vma->vm_file->private_data;
3026         unsigned long pgoff;
3027         struct page *page;
3028
3029         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
3030         page = gfn_to_page(kvm, pgoff);
3031         if (!page)
3032                 return NOPAGE_SIGBUS;
3033         get_page(page);
3034         if (type != NULL)
3035                 *type = VM_FAULT_MINOR;
3036
3037         return page;
3038 }
3039
3040 static struct vm_operations_struct kvm_vm_vm_ops = {
3041         .nopage = kvm_vm_nopage,
3042 };
3043
3044 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
3045 {
3046         vma->vm_ops = &kvm_vm_vm_ops;
3047         return 0;
3048 }
3049
3050 static struct file_operations kvm_vm_fops = {
3051         .release        = kvm_vm_release,
3052         .unlocked_ioctl = kvm_vm_ioctl,
3053         .compat_ioctl   = kvm_vm_ioctl,
3054         .mmap           = kvm_vm_mmap,
3055 };
3056
3057 static int kvm_dev_ioctl_create_vm(void)
3058 {
3059         int fd, r;
3060         struct inode *inode;
3061         struct file *file;
3062         struct kvm *kvm;
3063
3064         kvm = kvm_create_vm();
3065         if (IS_ERR(kvm))
3066                 return PTR_ERR(kvm);
3067         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
3068         if (r) {
3069                 kvm_destroy_vm(kvm);
3070                 return r;
3071         }
3072
3073         kvm->filp = file;
3074
3075         return fd;
3076 }
3077
3078 static long kvm_dev_ioctl(struct file *filp,
3079                           unsigned int ioctl, unsigned long arg)
3080 {
3081         void __user *argp = (void __user *)arg;
3082         long r = -EINVAL;
3083
3084         switch (ioctl) {
3085         case KVM_GET_API_VERSION:
3086                 r = -EINVAL;
3087                 if (arg)
3088                         goto out;
3089                 r = KVM_API_VERSION;
3090                 break;
3091         case KVM_CREATE_VM:
3092                 r = -EINVAL;
3093                 if (arg)
3094                         goto out;
3095                 r = kvm_dev_ioctl_create_vm();
3096                 break;
3097         case KVM_GET_MSR_INDEX_LIST: {
3098                 struct kvm_msr_list __user *user_msr_list = argp;
3099                 struct kvm_msr_list msr_list;
3100                 unsigned n;
3101
3102                 r = -EFAULT;
3103                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
3104                         goto out;
3105                 n = msr_list.nmsrs;
3106                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
3107                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
3108                         goto out;
3109                 r = -E2BIG;
3110                 if (n < num_msrs_to_save)
3111                         goto out;
3112                 r = -EFAULT;
3113                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3114                                  num_msrs_to_save * sizeof(u32)))
3115                         goto out;
3116                 if (copy_to_user(user_msr_list->indices
3117                                  + num_msrs_to_save * sizeof(u32),
3118                                  &emulated_msrs,
3119                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
3120                         goto out;
3121                 r = 0;
3122                 break;
3123         }
3124         case KVM_CHECK_EXTENSION: {
3125                 int ext = (long)argp;
3126
3127                 switch (ext) {
3128                 case KVM_CAP_IRQCHIP:
3129                 case KVM_CAP_HLT:
3130                         r = 1;
3131                         break;
3132                 default:
3133                         r = 0;
3134                         break;
3135                 }
3136                 break;
3137         }
3138         case KVM_GET_VCPU_MMAP_SIZE:
3139                 r = -EINVAL;
3140                 if (arg)
3141                         goto out;
3142                 r = 2 * PAGE_SIZE;
3143                 break;
3144         default:
3145                 ;
3146         }
3147 out:
3148         return r;
3149 }
3150
3151 static struct file_operations kvm_chardev_ops = {
3152         .unlocked_ioctl = kvm_dev_ioctl,
3153         .compat_ioctl   = kvm_dev_ioctl,
3154 };
3155
3156 static struct miscdevice kvm_dev = {
3157         KVM_MINOR,
3158         "kvm",
3159         &kvm_chardev_ops,
3160 };
3161
3162 /*
3163  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
3164  * cached on it.
3165  */
3166 static void decache_vcpus_on_cpu(int cpu)
3167 {
3168         struct kvm *vm;
3169         struct kvm_vcpu *vcpu;
3170         int i;
3171
3172         spin_lock(&kvm_lock);
3173         list_for_each_entry(vm, &vm_list, vm_list)
3174                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3175                         vcpu = vm->vcpus[i];
3176                         if (!vcpu)
3177                                 continue;
3178                         /*
3179                          * If the vcpu is locked, then it is running on some
3180                          * other cpu and therefore it is not cached on the
3181                          * cpu in question.
3182                          *
3183                          * If it's not locked, check the last cpu it executed
3184                          * on.
3185                          */
3186                         if (mutex_trylock(&vcpu->mutex)) {
3187                                 if (vcpu->cpu == cpu) {
3188                                         kvm_arch_ops->vcpu_decache(vcpu);
3189                                         vcpu->cpu = -1;
3190                                 }
3191                                 mutex_unlock(&vcpu->mutex);
3192                         }
3193                 }
3194         spin_unlock(&kvm_lock);
3195 }
3196
3197 static void hardware_enable(void *junk)
3198 {
3199         int cpu = raw_smp_processor_id();
3200
3201         if (cpu_isset(cpu, cpus_hardware_enabled))
3202                 return;
3203         cpu_set(cpu, cpus_hardware_enabled);
3204         kvm_arch_ops->hardware_enable(NULL);
3205 }
3206
3207 static void hardware_disable(void *junk)
3208 {
3209         int cpu = raw_smp_processor_id();
3210
3211         if (!cpu_isset(cpu, cpus_hardware_enabled))
3212                 return;
3213         cpu_clear(cpu, cpus_hardware_enabled);
3214         decache_vcpus_on_cpu(cpu);
3215         kvm_arch_ops->hardware_disable(NULL);
3216 }
3217
3218 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3219                            void *v)
3220 {
3221         int cpu = (long)v;
3222
3223         switch (val) {
3224         case CPU_DYING:
3225         case CPU_DYING_FROZEN:
3226                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3227                        cpu);
3228                 hardware_disable(NULL);
3229                 break;
3230         case CPU_UP_CANCELED:
3231         case CPU_UP_CANCELED_FROZEN:
3232                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3233                        cpu);
3234                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3235                 break;
3236         case CPU_ONLINE:
3237         case CPU_ONLINE_FROZEN:
3238                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3239                        cpu);
3240                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3241                 break;
3242         }
3243         return NOTIFY_OK;
3244 }
3245
3246 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3247                        void *v)
3248 {
3249         if (val == SYS_RESTART) {
3250                 /*
3251                  * Some (well, at least mine) BIOSes hang on reboot if
3252                  * in vmx root mode.
3253                  */
3254                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3255                 on_each_cpu(hardware_disable, NULL, 0, 1);
3256         }
3257         return NOTIFY_OK;
3258 }
3259
3260 static struct notifier_block kvm_reboot_notifier = {
3261         .notifier_call = kvm_reboot,
3262         .priority = 0,
3263 };
3264
3265 void kvm_io_bus_init(struct kvm_io_bus *bus)
3266 {
3267         memset(bus, 0, sizeof(*bus));
3268 }
3269
3270 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3271 {
3272         int i;
3273
3274         for (i = 0; i < bus->dev_count; i++) {
3275                 struct kvm_io_device *pos = bus->devs[i];
3276
3277                 kvm_iodevice_destructor(pos);
3278         }
3279 }
3280
3281 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3282 {
3283         int i;
3284
3285         for (i = 0; i < bus->dev_count; i++) {
3286                 struct kvm_io_device *pos = bus->devs[i];
3287
3288                 if (pos->in_range(pos, addr))
3289                         return pos;
3290         }
3291
3292         return NULL;
3293 }
3294
3295 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3296 {
3297         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3298
3299         bus->devs[bus->dev_count++] = dev;
3300 }
3301
3302 static struct notifier_block kvm_cpu_notifier = {
3303         .notifier_call = kvm_cpu_hotplug,
3304         .priority = 20, /* must be > scheduler priority */
3305 };
3306
3307 static u64 stat_get(void *_offset)
3308 {
3309         unsigned offset = (long)_offset;
3310         u64 total = 0;
3311         struct kvm *kvm;
3312         struct kvm_vcpu *vcpu;
3313         int i;
3314
3315         spin_lock(&kvm_lock);
3316         list_for_each_entry(kvm, &vm_list, vm_list)
3317                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3318                         vcpu = kvm->vcpus[i];
3319                         if (vcpu)
3320                                 total += *(u32 *)((void *)vcpu + offset);
3321                 }
3322         spin_unlock(&kvm_lock);
3323         return total;
3324 }
3325
3326 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3327
3328 static __init void kvm_init_debug(void)
3329 {
3330         struct kvm_stats_debugfs_item *p;
3331
3332         debugfs_dir = debugfs_create_dir("kvm", NULL);
3333         for (p = debugfs_entries; p->name; ++p)
3334                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3335                                                 (void *)(long)p->offset,
3336                                                 &stat_fops);
3337 }
3338
3339 static void kvm_exit_debug(void)
3340 {
3341         struct kvm_stats_debugfs_item *p;
3342
3343         for (p = debugfs_entries; p->name; ++p)
3344                 debugfs_remove(p->dentry);
3345         debugfs_remove(debugfs_dir);
3346 }
3347
3348 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3349 {
3350         hardware_disable(NULL);
3351         return 0;
3352 }
3353
3354 static int kvm_resume(struct sys_device *dev)
3355 {
3356         hardware_enable(NULL);
3357         return 0;
3358 }
3359
3360 static struct sysdev_class kvm_sysdev_class = {
3361         set_kset_name("kvm"),
3362         .suspend = kvm_suspend,
3363         .resume = kvm_resume,
3364 };
3365
3366 static struct sys_device kvm_sysdev = {
3367         .id = 0,
3368         .cls = &kvm_sysdev_class,
3369 };
3370
3371 hpa_t bad_page_address;
3372
3373 static inline
3374 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3375 {
3376         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3377 }
3378
3379 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3380 {
3381         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3382
3383         kvm_arch_ops->vcpu_load(vcpu, cpu);
3384 }
3385
3386 static void kvm_sched_out(struct preempt_notifier *pn,
3387                           struct task_struct *next)
3388 {
3389         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3390
3391         kvm_arch_ops->vcpu_put(vcpu);
3392 }
3393
3394 int kvm_init_arch(struct kvm_arch_ops *ops, unsigned int vcpu_size,
3395                   struct module *module)
3396 {
3397         int r;
3398         int cpu;
3399
3400         if (kvm_arch_ops) {
3401                 printk(KERN_ERR "kvm: already loaded the other module\n");
3402                 return -EEXIST;
3403         }
3404
3405         if (!ops->cpu_has_kvm_support()) {
3406                 printk(KERN_ERR "kvm: no hardware support\n");
3407                 return -EOPNOTSUPP;
3408         }
3409         if (ops->disabled_by_bios()) {
3410                 printk(KERN_ERR "kvm: disabled by bios\n");
3411                 return -EOPNOTSUPP;
3412         }
3413
3414         kvm_arch_ops = ops;
3415
3416         r = kvm_arch_ops->hardware_setup();
3417         if (r < 0)
3418                 goto out;
3419
3420         for_each_online_cpu(cpu) {
3421                 smp_call_function_single(cpu,
3422                                 kvm_arch_ops->check_processor_compatibility,
3423                                 &r, 0, 1);
3424                 if (r < 0)
3425                         goto out_free_0;
3426         }
3427
3428         on_each_cpu(hardware_enable, NULL, 0, 1);
3429         r = register_cpu_notifier(&kvm_cpu_notifier);
3430         if (r)
3431                 goto out_free_1;
3432         register_reboot_notifier(&kvm_reboot_notifier);
3433
3434         r = sysdev_class_register(&kvm_sysdev_class);
3435         if (r)
3436                 goto out_free_2;
3437
3438         r = sysdev_register(&kvm_sysdev);
3439         if (r)
3440                 goto out_free_3;
3441
3442         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3443         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3444                                            __alignof__(struct kvm_vcpu), 0, 0);
3445         if (!kvm_vcpu_cache) {
3446                 r = -ENOMEM;
3447                 goto out_free_4;
3448         }
3449
3450         kvm_chardev_ops.owner = module;
3451
3452         r = misc_register(&kvm_dev);
3453         if (r) {
3454                 printk (KERN_ERR "kvm: misc device register failed\n");
3455                 goto out_free;
3456         }
3457
3458         kvm_preempt_ops.sched_in = kvm_sched_in;
3459         kvm_preempt_ops.sched_out = kvm_sched_out;
3460
3461         return r;
3462
3463 out_free:
3464         kmem_cache_destroy(kvm_vcpu_cache);
3465 out_free_4:
3466         sysdev_unregister(&kvm_sysdev);
3467 out_free_3:
3468         sysdev_class_unregister(&kvm_sysdev_class);
3469 out_free_2:
3470         unregister_reboot_notifier(&kvm_reboot_notifier);
3471         unregister_cpu_notifier(&kvm_cpu_notifier);
3472 out_free_1:
3473         on_each_cpu(hardware_disable, NULL, 0, 1);
3474 out_free_0:
3475         kvm_arch_ops->hardware_unsetup();
3476 out:
3477         kvm_arch_ops = NULL;
3478         return r;
3479 }
3480
3481 void kvm_exit_arch(void)
3482 {
3483         misc_deregister(&kvm_dev);
3484         kmem_cache_destroy(kvm_vcpu_cache);
3485         sysdev_unregister(&kvm_sysdev);
3486         sysdev_class_unregister(&kvm_sysdev_class);
3487         unregister_reboot_notifier(&kvm_reboot_notifier);
3488         unregister_cpu_notifier(&kvm_cpu_notifier);
3489         on_each_cpu(hardware_disable, NULL, 0, 1);
3490         kvm_arch_ops->hardware_unsetup();
3491         kvm_arch_ops = NULL;
3492 }
3493
3494 static __init int kvm_init(void)
3495 {
3496         static struct page *bad_page;
3497         int r;
3498
3499         r = kvm_mmu_module_init();
3500         if (r)
3501                 goto out4;
3502
3503         kvm_init_debug();
3504
3505         kvm_init_msr_list();
3506
3507         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3508                 r = -ENOMEM;
3509                 goto out;
3510         }
3511
3512         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3513         memset(__va(bad_page_address), 0, PAGE_SIZE);
3514
3515         return 0;
3516
3517 out:
3518         kvm_exit_debug();
3519         kvm_mmu_module_exit();
3520 out4:
3521         return r;
3522 }
3523
3524 static __exit void kvm_exit(void)
3525 {
3526         kvm_exit_debug();
3527         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3528         kvm_mmu_module_exit();
3529 }
3530
3531 module_init(kvm_init)
3532 module_exit(kvm_exit)
3533
3534 EXPORT_SYMBOL_GPL(kvm_init_arch);
3535 EXPORT_SYMBOL_GPL(kvm_exit_arch);