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