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