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

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "kvm.h"
  19 #include "x86_emulate.h"
  20 #include "segment_descriptor.h"
  21 #include "irq.h"
  22
  23 #include <linux/kvm.h>
  24 #include <linux/module.h>
  25 #include <linux/errno.h>
  26 #include <linux/percpu.h>
  27 #include <linux/gfp.h>
  28 #include <linux/mm.h>
  29 #include <linux/miscdevice.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/reboot.h>
  32 #include <linux/debugfs.h>
  33 #include <linux/highmem.h>
  34 #include <linux/file.h>
  35 #include <linux/sysdev.h>
  36 #include <linux/cpu.h>
  37 #include <linux/sched.h>
  38 #include <linux/cpumask.h>
  39 #include <linux/smp.h>
  40 #include <linux/anon_inodes.h>
  41 #include <linux/profile.h>
  42 #include <linux/kvm_para.h>
  43
  44 #include <asm/processor.h>
  45 #include <asm/msr.h>
  46 #include <asm/io.h>
  47 #include <asm/uaccess.h>
  48 #include <asm/desc.h>
  49
  50 MODULE_AUTHOR("Qumranet");
  51 MODULE_LICENSE("GPL");
  52
  53 static DEFINE_SPINLOCK(kvm_lock);
  54 static LIST_HEAD(vm_list);
  55
  56 static cpumask_t cpus_hardware_enabled;
  57
  58 struct kvm_x86_ops *kvm_x86_ops;
  59 struct kmem_cache *kvm_vcpu_cache;
  60 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  61
  62 static __read_mostly struct preempt_ops kvm_preempt_ops;
  63
  64 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  65
  66 static struct kvm_stats_debugfs_item {
  67         const char *name;
  68         int offset;
  69         struct dentry *dentry;
  70 } debugfs_entries[] = {
  71         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  72         { "pf_guest", STAT_OFFSET(pf_guest) },
  73         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  74         { "invlpg", STAT_OFFSET(invlpg) },
  75         { "exits", STAT_OFFSET(exits) },
  76         { "io_exits", STAT_OFFSET(io_exits) },
  77         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  78         { "signal_exits", STAT_OFFSET(signal_exits) },
  79         { "irq_window", STAT_OFFSET(irq_window_exits) },
  80         { "halt_exits", STAT_OFFSET(halt_exits) },
  81         { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
  82         { "request_irq", STAT_OFFSET(request_irq_exits) },
  83         { "irq_exits", STAT_OFFSET(irq_exits) },
  84         { "light_exits", STAT_OFFSET(light_exits) },
  85         { "efer_reload", STAT_OFFSET(efer_reload) },
  86         { NULL }
  87 };
  88
  89 static struct dentry *debugfs_dir;
  90
  91 #define MAX_IO_MSRS 256
  92
  93 #define CR0_RESERVED_BITS                                               \
  94         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  95                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  96                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  97 #define CR4_RESERVED_BITS                                               \
  98         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  99                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
 100                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
 101                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
 102
 103 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
 104 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
 105
 106 #ifdef CONFIG_X86_64
 107 // LDT or TSS descriptor in the GDT. 16 bytes.
 108 struct segment_descriptor_64 {
 109         struct segment_descriptor s;
 110         u32 base_higher;
 111         u32 pad_zero;
 112 };
 113
 114 #endif
 115
 116 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 117                            unsigned long arg);
 118
 119 unsigned long segment_base(u16 selector)
 120 {
 121         struct descriptor_table gdt;
 122         struct segment_descriptor *d;
 123         unsigned long table_base;
 124         typedef unsigned long ul;
 125         unsigned long v;
 126
 127         if (selector == 0)
 128                 return 0;
 129
 130         asm ("sgdt %0" : "=m"(gdt));
 131         table_base = gdt.base;
 132
 133         if (selector & 4) {           /* from ldt */
 134                 u16 ldt_selector;
 135
 136                 asm ("sldt %0" : "=g"(ldt_selector));
 137                 table_base = segment_base(ldt_selector);
 138         }
 139         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 140         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 141 #ifdef CONFIG_X86_64
 142         if (d->system == 0
 143             && (d->type == 2 || d->type == 9 || d->type == 11))
 144                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 145 #endif
 146         return v;
 147 }
 148 EXPORT_SYMBOL_GPL(segment_base);
 149
 150 static inline int valid_vcpu(int n)
 151 {
 152         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 153 }
 154
 155 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 156 {
 157         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
 158                 return;
 159
 160         vcpu->guest_fpu_loaded = 1;
 161         fx_save(&vcpu->host_fx_image);
 162         fx_restore(&vcpu->guest_fx_image);
 163 }
 164 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
 165
 166 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 167 {
 168         if (!vcpu->guest_fpu_loaded)
 169                 return;
 170
 171         vcpu->guest_fpu_loaded = 0;
 172         fx_save(&vcpu->guest_fx_image);
 173         fx_restore(&vcpu->host_fx_image);
 174 }
 175 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
 176
 177 /*
 178  * Switches to specified vcpu, until a matching vcpu_put()
 179  */
 180 static void vcpu_load(struct kvm_vcpu *vcpu)
 181 {
 182         int cpu;
 183
 184         mutex_lock(&vcpu->mutex);
 185         cpu = get_cpu();
 186         preempt_notifier_register(&vcpu->preempt_notifier);
 187         kvm_x86_ops->vcpu_load(vcpu, cpu);
 188         put_cpu();
 189 }
 190
 191 static void vcpu_put(struct kvm_vcpu *vcpu)
 192 {
 193         preempt_disable();
 194         kvm_x86_ops->vcpu_put(vcpu);
 195         preempt_notifier_unregister(&vcpu->preempt_notifier);
 196         preempt_enable();
 197         mutex_unlock(&vcpu->mutex);
 198 }
 199
 200 static void ack_flush(void *_completed)
 201 {
 202 }
 203
 204 void kvm_flush_remote_tlbs(struct kvm *kvm)
 205 {
 206         int i, cpu;
 207         cpumask_t cpus;
 208         struct kvm_vcpu *vcpu;
 209
 210         cpus_clear(cpus);
 211         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 212                 vcpu = kvm->vcpus[i];
 213                 if (!vcpu)
 214                         continue;
 215                 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
 216                         continue;
 217                 cpu = vcpu->cpu;
 218                 if (cpu != -1 && cpu != raw_smp_processor_id())
 219                         cpu_set(cpu, cpus);
 220         }
 221         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 222 }
 223
 224 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 225 {
 226         struct page *page;
 227         int r;
 228
 229         mutex_init(&vcpu->mutex);
 230         vcpu->cpu = -1;
 231         vcpu->mmu.root_hpa = INVALID_PAGE;
 232         vcpu->kvm = kvm;
 233         vcpu->vcpu_id = id;
 234         if (!irqchip_in_kernel(kvm) || id == 0)
 235                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
 236         else
 237                 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
 238         init_waitqueue_head(&vcpu->wq);
 239
 240         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 241         if (!page) {
 242                 r = -ENOMEM;
 243                 goto fail;
 244         }
 245         vcpu->run = page_address(page);
 246
 247         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 248         if (!page) {
 249                 r = -ENOMEM;
 250                 goto fail_free_run;
 251         }
 252         vcpu->pio_data = page_address(page);
 253
 254         r = kvm_mmu_create(vcpu);
 255         if (r < 0)
 256                 goto fail_free_pio_data;
 257
 258         if (irqchip_in_kernel(kvm)) {
 259                 r = kvm_create_lapic(vcpu);
 260                 if (r < 0)
 261                         goto fail_mmu_destroy;
 262         }
 263
 264         return 0;
 265
 266 fail_mmu_destroy:
 267         kvm_mmu_destroy(vcpu);
 268 fail_free_pio_data:
 269         free_page((unsigned long)vcpu->pio_data);
 270 fail_free_run:
 271         free_page((unsigned long)vcpu->run);
 272 fail:
 273         return r;
 274 }
 275 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 276
 277 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 278 {
 279         kvm_free_lapic(vcpu);
 280         kvm_mmu_destroy(vcpu);
 281         free_page((unsigned long)vcpu->pio_data);
 282         free_page((unsigned long)vcpu->run);
 283 }
 284 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 285
 286 static struct kvm *kvm_create_vm(void)
 287 {
 288         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 289
 290         if (!kvm)
 291                 return ERR_PTR(-ENOMEM);
 292
 293         kvm_io_bus_init(&kvm->pio_bus);
 294         mutex_init(&kvm->lock);
 295         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 296         kvm_io_bus_init(&kvm->mmio_bus);
 297         spin_lock(&kvm_lock);
 298         list_add(&kvm->vm_list, &vm_list);
 299         spin_unlock(&kvm_lock);
 300         return kvm;
 301 }
 302
 303 /*
 304  * Free any memory in @free but not in @dont.
 305  */
 306 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 307                                   struct kvm_memory_slot *dont)
 308 {
 309         int i;
 310
 311         if (!dont || free->phys_mem != dont->phys_mem)
 312                 if (free->phys_mem) {
 313                         for (i = 0; i < free->npages; ++i)
 314                                 if (free->phys_mem[i])
 315                                         __free_page(free->phys_mem[i]);
 316                         vfree(free->phys_mem);
 317                 }
 318         if (!dont || free->rmap != dont->rmap)
 319                 vfree(free->rmap);
 320
 321         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 322                 vfree(free->dirty_bitmap);
 323
 324         free->phys_mem = NULL;
 325         free->npages = 0;
 326         free->dirty_bitmap = NULL;
 327 }
 328
 329 static void kvm_free_physmem(struct kvm *kvm)
 330 {
 331         int i;
 332
 333         for (i = 0; i < kvm->nmemslots; ++i)
 334                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 335 }
 336
 337 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 338 {
 339         int i;
 340
 341         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
 342                 if (vcpu->pio.guest_pages[i]) {
 343                         __free_page(vcpu->pio.guest_pages[i]);
 344                         vcpu->pio.guest_pages[i] = NULL;
 345                 }
 346 }
 347
 348 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 349 {
 350         vcpu_load(vcpu);
 351         kvm_mmu_unload(vcpu);
 352         vcpu_put(vcpu);
 353 }
 354
 355 static void kvm_free_vcpus(struct kvm *kvm)
 356 {
 357         unsigned int i;
 358
 359         /*
 360          * Unpin any mmu pages first.
 361          */
 362         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 363                 if (kvm->vcpus[i])
 364                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
 365         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 366                 if (kvm->vcpus[i]) {
 367                         kvm_x86_ops->vcpu_free(kvm->vcpus[i]);
 368                         kvm->vcpus[i] = NULL;
 369                 }
 370         }
 371
 372 }
 373
 374 static void kvm_destroy_vm(struct kvm *kvm)
 375 {
 376         spin_lock(&kvm_lock);
 377         list_del(&kvm->vm_list);
 378         spin_unlock(&kvm_lock);
 379         kvm_io_bus_destroy(&kvm->pio_bus);
 380         kvm_io_bus_destroy(&kvm->mmio_bus);
 381         kfree(kvm->vpic);
 382         kfree(kvm->vioapic);
 383         kvm_free_vcpus(kvm);
 384         kvm_free_physmem(kvm);
 385         kfree(kvm);
 386 }
 387
 388 static int kvm_vm_release(struct inode *inode, struct file *filp)
 389 {
 390         struct kvm *kvm = filp->private_data;
 391
 392         kvm_destroy_vm(kvm);
 393         return 0;
 394 }
 395
 396 static void inject_gp(struct kvm_vcpu *vcpu)
 397 {
 398         kvm_x86_ops->inject_gp(vcpu, 0);
 399 }
 400
 401 /*
 402  * Load the pae pdptrs.  Return true is they are all valid.
 403  */
 404 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 405 {
 406         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 407         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 408         int i;
 409         int ret;
 410         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
 411
 412         mutex_lock(&vcpu->kvm->lock);
 413         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 414                                   offset * sizeof(u64), sizeof(pdpte));
 415         if (ret < 0) {
 416                 ret = 0;
 417                 goto out;
 418         }
 419         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 420                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 421                         ret = 0;
 422                         goto out;
 423                 }
 424         }
 425         ret = 1;
 426
 427         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
 428 out:
 429         mutex_unlock(&vcpu->kvm->lock);
 430
 431         return ret;
 432 }
 433
 434 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 435 {
 436         if (cr0 & CR0_RESERVED_BITS) {
 437                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 438                        cr0, vcpu->cr0);
 439                 inject_gp(vcpu);
 440                 return;
 441         }
 442
 443         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 444                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 445                 inject_gp(vcpu);
 446                 return;
 447         }
 448
 449         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 450                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 451                        "and a clear PE flag\n");
 452                 inject_gp(vcpu);
 453                 return;
 454         }
 455
 456         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 457 #ifdef CONFIG_X86_64
 458                 if ((vcpu->shadow_efer & EFER_LME)) {
 459                         int cs_db, cs_l;
 460
 461                         if (!is_pae(vcpu)) {
 462                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 463                                        "in long mode while PAE is disabled\n");
 464                                 inject_gp(vcpu);
 465                                 return;
 466                         }
 467                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 468                         if (cs_l) {
 469                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 470                                        "in long mode while CS.L == 1\n");
 471                                 inject_gp(vcpu);
 472                                 return;
 473
 474                         }
 475                 } else
 476 #endif
 477                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 478                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 479                                "reserved bits\n");
 480                         inject_gp(vcpu);
 481                         return;
 482                 }
 483
 484         }
 485
 486         kvm_x86_ops->set_cr0(vcpu, cr0);
 487         vcpu->cr0 = cr0;
 488
 489         mutex_lock(&vcpu->kvm->lock);
 490         kvm_mmu_reset_context(vcpu);
 491         mutex_unlock(&vcpu->kvm->lock);
 492         return;
 493 }
 494 EXPORT_SYMBOL_GPL(set_cr0);
 495
 496 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 497 {
 498         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 499 }
 500 EXPORT_SYMBOL_GPL(lmsw);
 501
 502 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 503 {
 504         if (cr4 & CR4_RESERVED_BITS) {
 505                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 506                 inject_gp(vcpu);
 507                 return;
 508         }
 509
 510         if (is_long_mode(vcpu)) {
 511                 if (!(cr4 & X86_CR4_PAE)) {
 512                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 513                                "in long mode\n");
 514                         inject_gp(vcpu);
 515                         return;
 516                 }
 517         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 518                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 519                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 520                 inject_gp(vcpu);
 521                 return;
 522         }
 523
 524         if (cr4 & X86_CR4_VMXE) {
 525                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 526                 inject_gp(vcpu);
 527                 return;
 528         }
 529         kvm_x86_ops->set_cr4(vcpu, cr4);
 530         vcpu->cr4 = cr4;
 531         mutex_lock(&vcpu->kvm->lock);
 532         kvm_mmu_reset_context(vcpu);
 533         mutex_unlock(&vcpu->kvm->lock);
 534 }
 535 EXPORT_SYMBOL_GPL(set_cr4);
 536
 537 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 538 {
 539         if (is_long_mode(vcpu)) {
 540                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 541                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 542                         inject_gp(vcpu);
 543                         return;
 544                 }
 545         } else {
 546                 if (is_pae(vcpu)) {
 547                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 548                                 printk(KERN_DEBUG
 549                                        "set_cr3: #GP, reserved bits\n");
 550                                 inject_gp(vcpu);
 551                                 return;
 552                         }
 553                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 554                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 555                                        "reserved bits\n");
 556                                 inject_gp(vcpu);
 557                                 return;
 558                         }
 559                 }
 560                 /*
 561                  * We don't check reserved bits in nonpae mode, because
 562                  * this isn't enforced, and VMware depends on this.
 563                  */
 564         }
 565
 566         mutex_lock(&vcpu->kvm->lock);
 567         /*
 568          * Does the new cr3 value map to physical memory? (Note, we
 569          * catch an invalid cr3 even in real-mode, because it would
 570          * cause trouble later on when we turn on paging anyway.)
 571          *
 572          * A real CPU would silently accept an invalid cr3 and would
 573          * attempt to use it - with largely undefined (and often hard
 574          * to debug) behavior on the guest side.
 575          */
 576         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 577                 inject_gp(vcpu);
 578         else {
 579                 vcpu->cr3 = cr3;
 580                 vcpu->mmu.new_cr3(vcpu);
 581         }
 582         mutex_unlock(&vcpu->kvm->lock);
 583 }
 584 EXPORT_SYMBOL_GPL(set_cr3);
 585
 586 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 587 {
 588         if (cr8 & CR8_RESERVED_BITS) {
 589                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 590                 inject_gp(vcpu);
 591                 return;
 592         }
 593         if (irqchip_in_kernel(vcpu->kvm))
 594                 kvm_lapic_set_tpr(vcpu, cr8);
 595         else
 596                 vcpu->cr8 = cr8;
 597 }
 598 EXPORT_SYMBOL_GPL(set_cr8);
 599
 600 unsigned long get_cr8(struct kvm_vcpu *vcpu)
 601 {
 602         if (irqchip_in_kernel(vcpu->kvm))
 603                 return kvm_lapic_get_cr8(vcpu);
 604         else
 605                 return vcpu->cr8;
 606 }
 607 EXPORT_SYMBOL_GPL(get_cr8);
 608
 609 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 610 {
 611         if (irqchip_in_kernel(vcpu->kvm))
 612                 return vcpu->apic_base;
 613         else
 614                 return vcpu->apic_base;
 615 }
 616 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 617
 618 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 619 {
 620         /* TODO: reserve bits check */
 621         if (irqchip_in_kernel(vcpu->kvm))
 622                 kvm_lapic_set_base(vcpu, data);
 623         else
 624                 vcpu->apic_base = data;
 625 }
 626 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 627
 628 void fx_init(struct kvm_vcpu *vcpu)
 629 {
 630         unsigned after_mxcsr_mask;
 631
 632         /* Initialize guest FPU by resetting ours and saving into guest's */
 633         preempt_disable();
 634         fx_save(&vcpu->host_fx_image);
 635         fpu_init();
 636         fx_save(&vcpu->guest_fx_image);
 637         fx_restore(&vcpu->host_fx_image);
 638         preempt_enable();
 639
 640         vcpu->cr0 |= X86_CR0_ET;
 641         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
 642         vcpu->guest_fx_image.mxcsr = 0x1f80;
 643         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
 644                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
 645 }
 646 EXPORT_SYMBOL_GPL(fx_init);
 647
 648 /*
 649  * Allocate some memory and give it an address in the guest physical address
 650  * space.
 651  *
 652  * Discontiguous memory is allowed, mostly for framebuffers.
 653  */
 654 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 655                                           struct kvm_memory_region *mem)
 656 {
 657         int r;
 658         gfn_t base_gfn;
 659         unsigned long npages;
 660         unsigned long i;
 661         struct kvm_memory_slot *memslot;
 662         struct kvm_memory_slot old, new;
 663
 664         r = -EINVAL;
 665         /* General sanity checks */
 666         if (mem->memory_size & (PAGE_SIZE - 1))
 667                 goto out;
 668         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 669                 goto out;
 670         if (mem->slot >= KVM_MEMORY_SLOTS)
 671                 goto out;
 672         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 673                 goto out;
 674
 675         memslot = &kvm->memslots[mem->slot];
 676         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 677         npages = mem->memory_size >> PAGE_SHIFT;
 678
 679         if (!npages)
 680                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 681
 682         mutex_lock(&kvm->lock);
 683
 684         new = old = *memslot;
 685
 686         new.base_gfn = base_gfn;
 687         new.npages = npages;
 688         new.flags = mem->flags;
 689
 690         /* Disallow changing a memory slot's size. */
 691         r = -EINVAL;
 692         if (npages && old.npages && npages != old.npages)
 693                 goto out_unlock;
 694
 695         /* Check for overlaps */
 696         r = -EEXIST;
 697         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 698                 struct kvm_memory_slot *s = &kvm->memslots[i];
 699
 700                 if (s == memslot)
 701                         continue;
 702                 if (!((base_gfn + npages <= s->base_gfn) ||
 703                       (base_gfn >= s->base_gfn + s->npages)))
 704                         goto out_unlock;
 705         }
 706
 707         /* Deallocate if slot is being removed */
 708         if (!npages)
 709                 new.phys_mem = NULL;
 710
 711         /* Free page dirty bitmap if unneeded */
 712         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 713                 new.dirty_bitmap = NULL;
 714
 715         r = -ENOMEM;
 716
 717         /* Allocate if a slot is being created */
 718         if (npages && !new.phys_mem) {
 719                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 720
 721                 if (!new.phys_mem)
 722                         goto out_unlock;
 723
 724                 new.rmap = vmalloc(npages * sizeof(struct page*));
 725
 726                 if (!new.rmap)
 727                         goto out_unlock;
 728
 729                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 730                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 731                 for (i = 0; i < npages; ++i) {
 732                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 733                                                      | __GFP_ZERO);
 734                         if (!new.phys_mem[i])
 735                                 goto out_unlock;
 736                 }
 737         }
 738
 739         /* Allocate page dirty bitmap if needed */
 740         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 741                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 742
 743                 new.dirty_bitmap = vmalloc(dirty_bytes);
 744                 if (!new.dirty_bitmap)
 745                         goto out_unlock;
 746                 memset(new.dirty_bitmap, 0, dirty_bytes);
 747         }
 748
 749         if (mem->slot >= kvm->nmemslots)
 750                 kvm->nmemslots = mem->slot + 1;
 751
 752         if (!kvm->n_requested_mmu_pages) {
 753                 unsigned int n_pages;
 754
 755                 if (npages) {
 756                         n_pages = npages * KVM_PERMILLE_MMU_PAGES / 1000;
 757                         kvm_mmu_change_mmu_pages(kvm, kvm->n_alloc_mmu_pages +
 758                                                  n_pages);
 759                 } else {
 760                         unsigned int nr_mmu_pages;
 761
 762                         n_pages = old.npages * KVM_PERMILLE_MMU_PAGES / 1000;
 763                         nr_mmu_pages = kvm->n_alloc_mmu_pages - n_pages;
 764                         nr_mmu_pages = max(nr_mmu_pages,
 765                                         (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
 766                         kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
 767                 }
 768         }
 769
 770         *memslot = new;
 771
 772         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 773         kvm_flush_remote_tlbs(kvm);
 774
 775         mutex_unlock(&kvm->lock);
 776
 777         kvm_free_physmem_slot(&old, &new);
 778         return 0;
 779
 780 out_unlock:
 781         mutex_unlock(&kvm->lock);
 782         kvm_free_physmem_slot(&new, &old);
 783 out:
 784         return r;
 785 }
 786
 787 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
 788                                           u32 kvm_nr_mmu_pages)
 789 {
 790         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
 791                 return -EINVAL;
 792
 793         mutex_lock(&kvm->lock);
 794
 795         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
 796         kvm->n_requested_mmu_pages = kvm_nr_mmu_pages;
 797
 798         mutex_unlock(&kvm->lock);
 799         return 0;
 800 }
 801
 802 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
 803 {
 804         return kvm->n_alloc_mmu_pages;
 805 }
 806
 807 /*
 808  * Get (and clear) the dirty memory log for a memory slot.
 809  */
 810 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 811                                       struct kvm_dirty_log *log)
 812 {
 813         struct kvm_memory_slot *memslot;
 814         int r, i;
 815         int n;
 816         unsigned long any = 0;
 817
 818         mutex_lock(&kvm->lock);
 819
 820         r = -EINVAL;
 821         if (log->slot >= KVM_MEMORY_SLOTS)
 822                 goto out;
 823
 824         memslot = &kvm->memslots[log->slot];
 825         r = -ENOENT;
 826         if (!memslot->dirty_bitmap)
 827                 goto out;
 828
 829         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 830
 831         for (i = 0; !any && i < n/sizeof(long); ++i)
 832                 any = memslot->dirty_bitmap[i];
 833
 834         r = -EFAULT;
 835         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 836                 goto out;
 837
 838         /* If nothing is dirty, don't bother messing with page tables. */
 839         if (any) {
 840                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 841                 kvm_flush_remote_tlbs(kvm);
 842                 memset(memslot->dirty_bitmap, 0, n);
 843         }
 844
 845         r = 0;
 846
 847 out:
 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 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 static int next_segment(unsigned long len, int offset)
1004 {
1005         if (len > PAGE_SIZE - offset)
1006                 return PAGE_SIZE - offset;
1007         else
1008                 return len;
1009 }
1010
1011 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1012                         int len)
1013 {
1014         void *page_virt;
1015         struct page *page;
1016
1017         page = gfn_to_page(kvm, gfn);
1018         if (!page)
1019                 return -EFAULT;
1020         page_virt = kmap_atomic(page, KM_USER0);
1021
1022         memcpy(data, page_virt + offset, len);
1023
1024         kunmap_atomic(page_virt, KM_USER0);
1025         return 0;
1026 }
1027 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1028
1029 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1030 {
1031         gfn_t gfn = gpa >> PAGE_SHIFT;
1032         int seg;
1033         int offset = offset_in_page(gpa);
1034         int ret;
1035
1036         while ((seg = next_segment(len, offset)) != 0) {
1037                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1038                 if (ret < 0)
1039                         return ret;
1040                 offset = 0;
1041                 len -= seg;
1042                 data += seg;
1043                 ++gfn;
1044         }
1045         return 0;
1046 }
1047 EXPORT_SYMBOL_GPL(kvm_read_guest);
1048
1049 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1050                          int offset, int len)
1051 {
1052         void *page_virt;
1053         struct page *page;
1054
1055         page = gfn_to_page(kvm, gfn);
1056         if (!page)
1057                 return -EFAULT;
1058         page_virt = kmap_atomic(page, KM_USER0);
1059
1060         memcpy(page_virt + offset, data, len);
1061
1062         kunmap_atomic(page_virt, KM_USER0);
1063         mark_page_dirty(kvm, gfn);
1064         return 0;
1065 }
1066 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1067
1068 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1069                     unsigned long len)
1070 {
1071         gfn_t gfn = gpa >> PAGE_SHIFT;
1072         int seg;
1073         int offset = offset_in_page(gpa);
1074         int ret;
1075
1076         while ((seg = next_segment(len, offset)) != 0) {
1077                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1078                 if (ret < 0)
1079                         return ret;
1080                 offset = 0;
1081                 len -= seg;
1082                 data += seg;
1083                 ++gfn;
1084         }
1085         return 0;
1086 }
1087
1088 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1089 {
1090         void *page_virt;
1091         struct page *page;
1092
1093         page = gfn_to_page(kvm, gfn);
1094         if (!page)
1095                 return -EFAULT;
1096         page_virt = kmap_atomic(page, KM_USER0);
1097
1098         memset(page_virt + offset, 0, len);
1099
1100         kunmap_atomic(page_virt, KM_USER0);
1101         return 0;
1102 }
1103 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1104
1105 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1106 {
1107         gfn_t gfn = gpa >> PAGE_SHIFT;
1108         int seg;
1109         int offset = offset_in_page(gpa);
1110         int ret;
1111
1112         while ((seg = next_segment(len, offset)) != 0) {
1113                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1114                 if (ret < 0)
1115                         return ret;
1116                 offset = 0;
1117                 len -= seg;
1118                 ++gfn;
1119         }
1120         return 0;
1121 }
1122 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1123
1124 /* WARNING: Does not work on aliased pages. */
1125 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1126 {
1127         struct kvm_memory_slot *memslot;
1128
1129         memslot = __gfn_to_memslot(kvm, gfn);
1130         if (memslot && memslot->dirty_bitmap) {
1131                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1132
1133                 /* avoid RMW */
1134                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1135                         set_bit(rel_gfn, memslot->dirty_bitmap);
1136         }
1137 }
1138
1139 int emulator_read_std(unsigned long addr,
1140                              void *val,
1141                              unsigned int bytes,
1142                              struct kvm_vcpu *vcpu)
1143 {
1144         void *data = val;
1145
1146         while (bytes) {
1147                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1148                 unsigned offset = addr & (PAGE_SIZE-1);
1149                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1150                 int ret;
1151
1152                 if (gpa == UNMAPPED_GVA)
1153                         return X86EMUL_PROPAGATE_FAULT;
1154                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1155                 if (ret < 0)
1156                         return X86EMUL_UNHANDLEABLE;
1157
1158                 bytes -= tocopy;
1159                 data += tocopy;
1160                 addr += tocopy;
1161         }
1162
1163         return X86EMUL_CONTINUE;
1164 }
1165 EXPORT_SYMBOL_GPL(emulator_read_std);
1166
1167 static int emulator_write_std(unsigned long addr,
1168                               const void *val,
1169                               unsigned int bytes,
1170                               struct kvm_vcpu *vcpu)
1171 {
1172         pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1173         return X86EMUL_UNHANDLEABLE;
1174 }
1175
1176 /*
1177  * Only apic need an MMIO device hook, so shortcut now..
1178  */
1179 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1180                                                 gpa_t addr)
1181 {
1182         struct kvm_io_device *dev;
1183
1184         if (vcpu->apic) {
1185                 dev = &vcpu->apic->dev;
1186                 if (dev->in_range(dev, addr))
1187                         return dev;
1188         }
1189         return NULL;
1190 }
1191
1192 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1193                                                 gpa_t addr)
1194 {
1195         struct kvm_io_device *dev;
1196
1197         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1198         if (dev == NULL)
1199                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1200         return dev;
1201 }
1202
1203 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1204                                                gpa_t addr)
1205 {
1206         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1207 }
1208
1209 static int emulator_read_emulated(unsigned long addr,
1210                                   void *val,
1211                                   unsigned int bytes,
1212                                   struct kvm_vcpu *vcpu)
1213 {
1214         struct kvm_io_device *mmio_dev;
1215         gpa_t                 gpa;
1216
1217         if (vcpu->mmio_read_completed) {
1218                 memcpy(val, vcpu->mmio_data, bytes);
1219                 vcpu->mmio_read_completed = 0;
1220                 return X86EMUL_CONTINUE;
1221         } else if (emulator_read_std(addr, val, bytes, vcpu)
1222                    == X86EMUL_CONTINUE)
1223                 return X86EMUL_CONTINUE;
1224
1225         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1226         if (gpa == UNMAPPED_GVA)
1227                 return X86EMUL_PROPAGATE_FAULT;
1228
1229         /*
1230          * Is this MMIO handled locally?
1231          */
1232         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1233         if (mmio_dev) {
1234                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1235                 return X86EMUL_CONTINUE;
1236         }
1237
1238         vcpu->mmio_needed = 1;
1239         vcpu->mmio_phys_addr = gpa;
1240         vcpu->mmio_size = bytes;
1241         vcpu->mmio_is_write = 0;
1242
1243         return X86EMUL_UNHANDLEABLE;
1244 }
1245
1246 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1247                                const void *val, int bytes)
1248 {
1249         int ret;
1250
1251         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1252         if (ret < 0)
1253                 return 0;
1254         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1255         return 1;
1256 }
1257
1258 static int emulator_write_emulated_onepage(unsigned long addr,
1259                                            const void *val,
1260                                            unsigned int bytes,
1261                                            struct kvm_vcpu *vcpu)
1262 {
1263         struct kvm_io_device *mmio_dev;
1264         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1265
1266         if (gpa == UNMAPPED_GVA) {
1267                 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1268                 return X86EMUL_PROPAGATE_FAULT;
1269         }
1270
1271         if (emulator_write_phys(vcpu, gpa, val, bytes))
1272                 return X86EMUL_CONTINUE;
1273
1274         /*
1275          * Is this MMIO handled locally?
1276          */
1277         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1278         if (mmio_dev) {
1279                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1280                 return X86EMUL_CONTINUE;
1281         }
1282
1283         vcpu->mmio_needed = 1;
1284         vcpu->mmio_phys_addr = gpa;
1285         vcpu->mmio_size = bytes;
1286         vcpu->mmio_is_write = 1;
1287         memcpy(vcpu->mmio_data, val, bytes);
1288
1289         return X86EMUL_CONTINUE;
1290 }
1291
1292 int emulator_write_emulated(unsigned long addr,
1293                                    const void *val,
1294                                    unsigned int bytes,
1295                                    struct kvm_vcpu *vcpu)
1296 {
1297         /* Crossing a page boundary? */
1298         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1299                 int rc, now;
1300
1301                 now = -addr & ~PAGE_MASK;
1302                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1303                 if (rc != X86EMUL_CONTINUE)
1304                         return rc;
1305                 addr += now;
1306                 val += now;
1307                 bytes -= now;
1308         }
1309         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1310 }
1311 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1312
1313 static int emulator_cmpxchg_emulated(unsigned long addr,
1314                                      const void *old,
1315                                      const void *new,
1316                                      unsigned int bytes,
1317                                      struct kvm_vcpu *vcpu)
1318 {
1319         static int reported;
1320
1321         if (!reported) {
1322                 reported = 1;
1323                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1324         }
1325         return emulator_write_emulated(addr, new, bytes, vcpu);
1326 }
1327
1328 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1329 {
1330         return kvm_x86_ops->get_segment_base(vcpu, seg);
1331 }
1332
1333 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1334 {
1335         return X86EMUL_CONTINUE;
1336 }
1337
1338 int emulate_clts(struct kvm_vcpu *vcpu)
1339 {
1340         kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1341         return X86EMUL_CONTINUE;
1342 }
1343
1344 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1345 {
1346         struct kvm_vcpu *vcpu = ctxt->vcpu;
1347
1348         switch (dr) {
1349         case 0 ... 3:
1350                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1351                 return X86EMUL_CONTINUE;
1352         default:
1353                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1354                 return X86EMUL_UNHANDLEABLE;
1355         }
1356 }
1357
1358 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1359 {
1360         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1361         int exception;
1362
1363         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1364         if (exception) {
1365                 /* FIXME: better handling */
1366                 return X86EMUL_UNHANDLEABLE;
1367         }
1368         return X86EMUL_CONTINUE;
1369 }
1370
1371 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1372 {
1373         static int reported;
1374         u8 opcodes[4];
1375         unsigned long rip = vcpu->rip;
1376         unsigned long rip_linear;
1377
1378         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1379
1380         if (reported)
1381                 return;
1382
1383         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1384
1385         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1386                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1387         reported = 1;
1388 }
1389 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1390
1391 struct x86_emulate_ops emulate_ops = {
1392         .read_std            = emulator_read_std,
1393         .write_std           = emulator_write_std,
1394         .read_emulated       = emulator_read_emulated,
1395         .write_emulated      = emulator_write_emulated,
1396         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1397 };
1398
1399 int emulate_instruction(struct kvm_vcpu *vcpu,
1400                         struct kvm_run *run,
1401                         unsigned long cr2,
1402                         u16 error_code,
1403                         int no_decode)
1404 {
1405         int r;
1406
1407         vcpu->mmio_fault_cr2 = cr2;
1408         kvm_x86_ops->cache_regs(vcpu);
1409
1410         vcpu->mmio_is_write = 0;
1411         vcpu->pio.string = 0;
1412
1413         if (!no_decode) {
1414                 int cs_db, cs_l;
1415                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1416
1417                 vcpu->emulate_ctxt.vcpu = vcpu;
1418                 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1419                 vcpu->emulate_ctxt.cr2 = cr2;
1420                 vcpu->emulate_ctxt.mode =
1421                         (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1422                         ? X86EMUL_MODE_REAL : cs_l
1423                         ? X86EMUL_MODE_PROT64 : cs_db
1424                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1425
1426                 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1427                         vcpu->emulate_ctxt.cs_base = 0;
1428                         vcpu->emulate_ctxt.ds_base = 0;
1429                         vcpu->emulate_ctxt.es_base = 0;
1430                         vcpu->emulate_ctxt.ss_base = 0;
1431                 } else {
1432                         vcpu->emulate_ctxt.cs_base =
1433                                         get_segment_base(vcpu, VCPU_SREG_CS);
1434                         vcpu->emulate_ctxt.ds_base =
1435                                         get_segment_base(vcpu, VCPU_SREG_DS);
1436                         vcpu->emulate_ctxt.es_base =
1437                                         get_segment_base(vcpu, VCPU_SREG_ES);
1438                         vcpu->emulate_ctxt.ss_base =
1439                                         get_segment_base(vcpu, VCPU_SREG_SS);
1440                 }
1441
1442                 vcpu->emulate_ctxt.gs_base =
1443                                         get_segment_base(vcpu, VCPU_SREG_GS);
1444                 vcpu->emulate_ctxt.fs_base =
1445                                         get_segment_base(vcpu, VCPU_SREG_FS);
1446
1447                 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
1448                 if (r)  {
1449                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1450                                 return EMULATE_DONE;
1451                         return EMULATE_FAIL;
1452                 }
1453         }
1454
1455         r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1456
1457         if (vcpu->pio.string)
1458                 return EMULATE_DO_MMIO;
1459
1460         if ((r || vcpu->mmio_is_write) && run) {
1461                 run->exit_reason = KVM_EXIT_MMIO;
1462                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1463                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1464                 run->mmio.len = vcpu->mmio_size;
1465                 run->mmio.is_write = vcpu->mmio_is_write;
1466         }
1467
1468         if (r) {
1469                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1470                         return EMULATE_DONE;
1471                 if (!vcpu->mmio_needed) {
1472                         kvm_report_emulation_failure(vcpu, "mmio");
1473                         return EMULATE_FAIL;
1474                 }
1475                 return EMULATE_DO_MMIO;
1476         }
1477
1478         kvm_x86_ops->decache_regs(vcpu);
1479         kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1480
1481         if (vcpu->mmio_is_write) {
1482                 vcpu->mmio_needed = 0;
1483                 return EMULATE_DO_MMIO;
1484         }
1485
1486         return EMULATE_DONE;
1487 }
1488 EXPORT_SYMBOL_GPL(emulate_instruction);
1489
1490 /*
1491  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1492  */
1493 static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1494 {
1495         DECLARE_WAITQUEUE(wait, current);
1496
1497         add_wait_queue(&vcpu->wq, &wait);
1498
1499         /*
1500          * We will block until either an interrupt or a signal wakes us up
1501          */
1502         while (!kvm_cpu_has_interrupt(vcpu)
1503                && !signal_pending(current)
1504                && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
1505                && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
1506                 set_current_state(TASK_INTERRUPTIBLE);
1507                 vcpu_put(vcpu);
1508                 schedule();
1509                 vcpu_load(vcpu);
1510         }
1511
1512         __set_current_state(TASK_RUNNING);
1513         remove_wait_queue(&vcpu->wq, &wait);
1514 }
1515
1516 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1517 {
1518         ++vcpu->stat.halt_exits;
1519         if (irqchip_in_kernel(vcpu->kvm)) {
1520                 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1521                 kvm_vcpu_block(vcpu);
1522                 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1523                         return -EINTR;
1524                 return 1;
1525         } else {
1526                 vcpu->run->exit_reason = KVM_EXIT_HLT;
1527                 return 0;
1528         }
1529 }
1530 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1531
1532 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
1533 {
1534         unsigned long nr, a0, a1, a2, a3, ret;
1535
1536         kvm_x86_ops->cache_regs(vcpu);
1537
1538         nr = vcpu->regs[VCPU_REGS_RAX];
1539         a0 = vcpu->regs[VCPU_REGS_RBX];
1540         a1 = vcpu->regs[VCPU_REGS_RCX];
1541         a2 = vcpu->regs[VCPU_REGS_RDX];
1542         a3 = vcpu->regs[VCPU_REGS_RSI];
1543
1544         if (!is_long_mode(vcpu)) {
1545                 nr &= 0xFFFFFFFF;
1546                 a0 &= 0xFFFFFFFF;
1547                 a1 &= 0xFFFFFFFF;
1548                 a2 &= 0xFFFFFFFF;
1549                 a3 &= 0xFFFFFFFF;
1550         }
1551
1552         switch (nr) {
1553         default:
1554                 ret = -KVM_ENOSYS;
1555                 break;
1556         }
1557         vcpu->regs[VCPU_REGS_RAX] = ret;
1558         kvm_x86_ops->decache_regs(vcpu);
1559         return 0;
1560 }
1561 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1562
1563 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1564 {
1565         char instruction[3];
1566         int ret = 0;
1567
1568         mutex_lock(&vcpu->kvm->lock);
1569
1570         /*
1571          * Blow out the MMU to ensure that no other VCPU has an active mapping
1572          * to ensure that the updated hypercall appears atomically across all
1573          * VCPUs.
1574          */
1575         kvm_mmu_zap_all(vcpu->kvm);
1576
1577         kvm_x86_ops->cache_regs(vcpu);
1578         kvm_x86_ops->patch_hypercall(vcpu, instruction);
1579         if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1580             != X86EMUL_CONTINUE)
1581                 ret = -EFAULT;
1582
1583         mutex_unlock(&vcpu->kvm->lock);
1584
1585         return ret;
1586 }
1587
1588 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1589 {
1590         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1591 }
1592
1593 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1594 {
1595         struct descriptor_table dt = { limit, base };
1596
1597         kvm_x86_ops->set_gdt(vcpu, &dt);
1598 }
1599
1600 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1601 {
1602         struct descriptor_table dt = { limit, base };
1603
1604         kvm_x86_ops->set_idt(vcpu, &dt);
1605 }
1606
1607 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1608                    unsigned long *rflags)
1609 {
1610         lmsw(vcpu, msw);
1611         *rflags = kvm_x86_ops->get_rflags(vcpu);
1612 }
1613
1614 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1615 {
1616         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1617         switch (cr) {
1618         case 0:
1619                 return vcpu->cr0;
1620         case 2:
1621                 return vcpu->cr2;
1622         case 3:
1623                 return vcpu->cr3;
1624         case 4:
1625                 return vcpu->cr4;
1626         default:
1627                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1628                 return 0;
1629         }
1630 }
1631
1632 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1633                      unsigned long *rflags)
1634 {
1635         switch (cr) {
1636         case 0:
1637                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1638                 *rflags = kvm_x86_ops->get_rflags(vcpu);
1639                 break;
1640         case 2:
1641                 vcpu->cr2 = val;
1642                 break;
1643         case 3:
1644                 set_cr3(vcpu, val);
1645                 break;
1646         case 4:
1647                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1648                 break;
1649         default:
1650                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1651         }
1652 }
1653
1654 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1655 {
1656         u64 data;
1657
1658         switch (msr) {
1659         case 0xc0010010: /* SYSCFG */
1660         case 0xc0010015: /* HWCR */
1661         case MSR_IA32_PLATFORM_ID:
1662         case MSR_IA32_P5_MC_ADDR:
1663         case MSR_IA32_P5_MC_TYPE:
1664         case MSR_IA32_MC0_CTL:
1665         case MSR_IA32_MCG_STATUS:
1666         case MSR_IA32_MCG_CAP:
1667         case MSR_IA32_MC0_MISC:
1668         case MSR_IA32_MC0_MISC+4:
1669         case MSR_IA32_MC0_MISC+8:
1670         case MSR_IA32_MC0_MISC+12:
1671         case MSR_IA32_MC0_MISC+16:
1672         case MSR_IA32_UCODE_REV:
1673         case MSR_IA32_PERF_STATUS:
1674         case MSR_IA32_EBL_CR_POWERON:
1675                 /* MTRR registers */
1676         case 0xfe:
1677         case 0x200 ... 0x2ff:
1678                 data = 0;
1679                 break;
1680         case 0xcd: /* fsb frequency */
1681                 data = 3;
1682                 break;
1683         case MSR_IA32_APICBASE:
1684                 data = kvm_get_apic_base(vcpu);
1685                 break;
1686         case MSR_IA32_MISC_ENABLE:
1687                 data = vcpu->ia32_misc_enable_msr;
1688                 break;
1689 #ifdef CONFIG_X86_64
1690         case MSR_EFER:
1691                 data = vcpu->shadow_efer;
1692                 break;
1693 #endif
1694         default:
1695                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1696                 return 1;
1697         }
1698         *pdata = data;
1699         return 0;
1700 }
1701 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1702
1703 /*
1704  * Reads an msr value (of 'msr_index') into 'pdata'.
1705  * Returns 0 on success, non-0 otherwise.
1706  * Assumes vcpu_load() was already called.
1707  */
1708 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1709 {
1710         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1711 }
1712
1713 #ifdef CONFIG_X86_64
1714
1715 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1716 {
1717         if (efer & EFER_RESERVED_BITS) {
1718                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1719                        efer);
1720                 inject_gp(vcpu);
1721                 return;
1722         }
1723
1724         if (is_paging(vcpu)
1725             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1726                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1727                 inject_gp(vcpu);
1728                 return;
1729         }
1730
1731         kvm_x86_ops->set_efer(vcpu, efer);
1732
1733         efer &= ~EFER_LMA;
1734         efer |= vcpu->shadow_efer & EFER_LMA;
1735
1736         vcpu->shadow_efer = efer;
1737 }
1738
1739 #endif
1740
1741 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1742 {
1743         switch (msr) {
1744 #ifdef CONFIG_X86_64
1745         case MSR_EFER:
1746                 set_efer(vcpu, data);
1747                 break;
1748 #endif
1749         case MSR_IA32_MC0_STATUS:
1750                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1751                        __FUNCTION__, data);
1752                 break;
1753         case MSR_IA32_MCG_STATUS:
1754                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1755                         __FUNCTION__, data);
1756                 break;
1757         case MSR_IA32_UCODE_REV:
1758         case MSR_IA32_UCODE_WRITE:
1759         case 0x200 ... 0x2ff: /* MTRRs */
1760                 break;
1761         case MSR_IA32_APICBASE:
1762                 kvm_set_apic_base(vcpu, data);
1763                 break;
1764         case MSR_IA32_MISC_ENABLE:
1765                 vcpu->ia32_misc_enable_msr = data;
1766                 break;
1767         default:
1768                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1769                 return 1;
1770         }
1771         return 0;
1772 }
1773 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1774
1775 /*
1776  * Writes msr value into into the appropriate "register".
1777  * Returns 0 on success, non-0 otherwise.
1778  * Assumes vcpu_load() was already called.
1779  */
1780 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1781 {
1782         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
1783 }
1784
1785 void kvm_resched(struct kvm_vcpu *vcpu)
1786 {
1787         if (!need_resched())
1788                 return;
1789         cond_resched();
1790 }
1791 EXPORT_SYMBOL_GPL(kvm_resched);
1792
1793 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1794 {
1795         int i;
1796         u32 function;
1797         struct kvm_cpuid_entry *e, *best;
1798
1799         kvm_x86_ops->cache_regs(vcpu);
1800         function = vcpu->regs[VCPU_REGS_RAX];
1801         vcpu->regs[VCPU_REGS_RAX] = 0;
1802         vcpu->regs[VCPU_REGS_RBX] = 0;
1803         vcpu->regs[VCPU_REGS_RCX] = 0;
1804         vcpu->regs[VCPU_REGS_RDX] = 0;
1805         best = NULL;
1806         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1807                 e = &vcpu->cpuid_entries[i];
1808                 if (e->function == function) {
1809                         best = e;
1810                         break;
1811                 }
1812                 /*
1813                  * Both basic or both extended?
1814                  */
1815                 if (((e->function ^ function) & 0x80000000) == 0)
1816                         if (!best || e->function > best->function)
1817                                 best = e;
1818         }
1819         if (best) {
1820                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1821                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1822                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1823                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1824         }
1825         kvm_x86_ops->decache_regs(vcpu);
1826         kvm_x86_ops->skip_emulated_instruction(vcpu);
1827 }
1828 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1829
1830 static int pio_copy_data(struct kvm_vcpu *vcpu)
1831 {
1832         void *p = vcpu->pio_data;
1833         void *q;
1834         unsigned bytes;
1835         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1836
1837         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1838                  PAGE_KERNEL);
1839         if (!q) {
1840                 free_pio_guest_pages(vcpu);
1841                 return -ENOMEM;
1842         }
1843         q += vcpu->pio.guest_page_offset;
1844         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1845         if (vcpu->pio.in)
1846                 memcpy(q, p, bytes);
1847         else
1848                 memcpy(p, q, bytes);
1849         q -= vcpu->pio.guest_page_offset;
1850         vunmap(q);
1851         free_pio_guest_pages(vcpu);
1852         return 0;
1853 }
1854
1855 static int complete_pio(struct kvm_vcpu *vcpu)
1856 {
1857         struct kvm_pio_request *io = &vcpu->pio;
1858         long delta;
1859         int r;
1860
1861         kvm_x86_ops->cache_regs(vcpu);
1862
1863         if (!io->string) {
1864                 if (io->in)
1865                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1866                                io->size);
1867         } else {
1868                 if (io->in) {
1869                         r = pio_copy_data(vcpu);
1870                         if (r) {
1871                                 kvm_x86_ops->cache_regs(vcpu);
1872                                 return r;
1873                         }
1874                 }
1875
1876                 delta = 1;
1877                 if (io->rep) {
1878                         delta *= io->cur_count;
1879                         /*
1880                          * The size of the register should really depend on
1881                          * current address size.
1882                          */
1883                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1884                 }
1885                 if (io->down)
1886                         delta = -delta;
1887                 delta *= io->size;
1888                 if (io->in)
1889                         vcpu->regs[VCPU_REGS_RDI] += delta;
1890                 else
1891                         vcpu->regs[VCPU_REGS_RSI] += delta;
1892         }
1893
1894         kvm_x86_ops->decache_regs(vcpu);
1895
1896         io->count -= io->cur_count;
1897         io->cur_count = 0;
1898
1899         return 0;
1900 }
1901
1902 static void kernel_pio(struct kvm_io_device *pio_dev,
1903                        struct kvm_vcpu *vcpu,
1904                        void *pd)
1905 {
1906         /* TODO: String I/O for in kernel device */
1907
1908         mutex_lock(&vcpu->kvm->lock);
1909         if (vcpu->pio.in)
1910                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1911                                   vcpu->pio.size,
1912                                   pd);
1913         else
1914                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1915                                    vcpu->pio.size,
1916                                    pd);
1917         mutex_unlock(&vcpu->kvm->lock);
1918 }
1919
1920 static void pio_string_write(struct kvm_io_device *pio_dev,
1921                              struct kvm_vcpu *vcpu)
1922 {
1923         struct kvm_pio_request *io = &vcpu->pio;
1924         void *pd = vcpu->pio_data;
1925         int i;
1926
1927         mutex_lock(&vcpu->kvm->lock);
1928         for (i = 0; i < io->cur_count; i++) {
1929                 kvm_iodevice_write(pio_dev, io->port,
1930                                    io->size,
1931                                    pd);
1932                 pd += io->size;
1933         }
1934         mutex_unlock(&vcpu->kvm->lock);
1935 }
1936
1937 int kvm_emulate_pio (struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1938                   int size, unsigned port)
1939 {
1940         struct kvm_io_device *pio_dev;
1941
1942         vcpu->run->exit_reason = KVM_EXIT_IO;
1943         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1944         vcpu->run->io.size = vcpu->pio.size = size;
1945         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1946         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1947         vcpu->run->io.port = vcpu->pio.port = port;
1948         vcpu->pio.in = in;
1949         vcpu->pio.string = 0;
1950         vcpu->pio.down = 0;
1951         vcpu->pio.guest_page_offset = 0;
1952         vcpu->pio.rep = 0;
1953
1954         kvm_x86_ops->cache_regs(vcpu);
1955         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1956         kvm_x86_ops->decache_regs(vcpu);
1957
1958         kvm_x86_ops->skip_emulated_instruction(vcpu);
1959
1960         pio_dev = vcpu_find_pio_dev(vcpu, port);
1961         if (pio_dev) {
1962                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1963                 complete_pio(vcpu);
1964                 return 1;
1965         }
1966         return 0;
1967 }
1968 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1969
1970 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1971                   int size, unsigned long count, int down,
1972                   gva_t address, int rep, unsigned port)
1973 {
1974         unsigned now, in_page;
1975         int i, ret = 0;
1976         int nr_pages = 1;
1977         struct page *page;
1978         struct kvm_io_device *pio_dev;
1979
1980         vcpu->run->exit_reason = KVM_EXIT_IO;
1981         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1982         vcpu->run->io.size = vcpu->pio.size = size;
1983         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1984         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1985         vcpu->run->io.port = vcpu->pio.port = port;
1986         vcpu->pio.in = in;
1987         vcpu->pio.string = 1;
1988         vcpu->pio.down = down;
1989         vcpu->pio.guest_page_offset = offset_in_page(address);
1990         vcpu->pio.rep = rep;
1991
1992         if (!count) {
1993                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1994                 return 1;
1995         }
1996
1997         if (!down)
1998                 in_page = PAGE_SIZE - offset_in_page(address);
1999         else
2000                 in_page = offset_in_page(address) + size;
2001         now = min(count, (unsigned long)in_page / size);
2002         if (!now) {
2003                 /*
2004                  * String I/O straddles page boundary.  Pin two guest pages
2005                  * so that we satisfy atomicity constraints.  Do just one
2006                  * transaction to avoid complexity.
2007                  */
2008                 nr_pages = 2;
2009                 now = 1;
2010         }
2011         if (down) {
2012                 /*
2013                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2014                  */
2015                 pr_unimpl(vcpu, "guest string pio down\n");
2016                 inject_gp(vcpu);
2017                 return 1;
2018         }
2019         vcpu->run->io.count = now;
2020         vcpu->pio.cur_count = now;
2021
2022         if (vcpu->pio.cur_count == vcpu->pio.count)
2023                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2024
2025         for (i = 0; i < nr_pages; ++i) {
2026                 mutex_lock(&vcpu->kvm->lock);
2027                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2028                 if (page)
2029                         get_page(page);
2030                 vcpu->pio.guest_pages[i] = page;
2031                 mutex_unlock(&vcpu->kvm->lock);
2032                 if (!page) {
2033                         inject_gp(vcpu);
2034                         free_pio_guest_pages(vcpu);
2035                         return 1;
2036                 }
2037         }
2038
2039         pio_dev = vcpu_find_pio_dev(vcpu, port);
2040         if (!vcpu->pio.in) {
2041                 /* string PIO write */
2042                 ret = pio_copy_data(vcpu);
2043                 if (ret >= 0 && pio_dev) {
2044                         pio_string_write(pio_dev, vcpu);
2045                         complete_pio(vcpu);
2046                         if (vcpu->pio.count == 0)
2047                                 ret = 1;
2048                 }
2049         } else if (pio_dev)
2050                 pr_unimpl(vcpu, "no string pio read support yet, "
2051                        "port %x size %d count %ld\n",
2052                         port, size, count);
2053
2054         return ret;
2055 }
2056 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2057
2058 /*
2059  * Check if userspace requested an interrupt window, and that the
2060  * interrupt window is open.
2061  *
2062  * No need to exit to userspace if we already have an interrupt queued.
2063  */
2064 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2065                                           struct kvm_run *kvm_run)
2066 {
2067         return (!vcpu->irq_summary &&
2068                 kvm_run->request_interrupt_window &&
2069                 vcpu->interrupt_window_open &&
2070                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2071 }
2072
2073 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2074                               struct kvm_run *kvm_run)
2075 {
2076         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2077         kvm_run->cr8 = get_cr8(vcpu);
2078         kvm_run->apic_base = kvm_get_apic_base(vcpu);
2079         if (irqchip_in_kernel(vcpu->kvm))
2080                 kvm_run->ready_for_interrupt_injection = 1;
2081         else
2082                 kvm_run->ready_for_interrupt_injection =
2083                                         (vcpu->interrupt_window_open &&
2084                                          vcpu->irq_summary == 0);
2085 }
2086
2087 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2088 {
2089         int r;
2090
2091         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
2092                 printk("vcpu %d received sipi with vector # %x\n",
2093                        vcpu->vcpu_id, vcpu->sipi_vector);
2094                 kvm_lapic_reset(vcpu);
2095                 kvm_x86_ops->vcpu_reset(vcpu);
2096                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2097         }
2098
2099 preempted:
2100         if (vcpu->guest_debug.enabled)
2101                 kvm_x86_ops->guest_debug_pre(vcpu);
2102
2103 again:
2104         r = kvm_mmu_reload(vcpu);
2105         if (unlikely(r))
2106                 goto out;
2107
2108         preempt_disable();
2109
2110         kvm_x86_ops->prepare_guest_switch(vcpu);
2111         kvm_load_guest_fpu(vcpu);
2112
2113         local_irq_disable();
2114
2115         if (signal_pending(current)) {
2116                 local_irq_enable();
2117                 preempt_enable();
2118                 r = -EINTR;
2119                 kvm_run->exit_reason = KVM_EXIT_INTR;
2120                 ++vcpu->stat.signal_exits;
2121                 goto out;
2122         }
2123
2124         if (irqchip_in_kernel(vcpu->kvm))
2125                 kvm_x86_ops->inject_pending_irq(vcpu);
2126         else if (!vcpu->mmio_read_completed)
2127                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2128
2129         vcpu->guest_mode = 1;
2130         kvm_guest_enter();
2131
2132         if (vcpu->requests)
2133                 if (test_and_clear_bit(KVM_TLB_FLUSH, &vcpu->requests))
2134                         kvm_x86_ops->tlb_flush(vcpu);
2135
2136         kvm_x86_ops->run(vcpu, kvm_run);
2137
2138         vcpu->guest_mode = 0;
2139         local_irq_enable();
2140
2141         ++vcpu->stat.exits;
2142
2143         /*
2144          * We must have an instruction between local_irq_enable() and
2145          * kvm_guest_exit(), so the timer interrupt isn't delayed by
2146          * the interrupt shadow.  The stat.exits increment will do nicely.
2147          * But we need to prevent reordering, hence this barrier():
2148          */
2149         barrier();
2150
2151         kvm_guest_exit();
2152
2153         preempt_enable();
2154
2155         /*
2156          * Profile KVM exit RIPs:
2157          */
2158         if (unlikely(prof_on == KVM_PROFILING)) {
2159                 kvm_x86_ops->cache_regs(vcpu);
2160                 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
2161         }
2162
2163         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2164
2165         if (r > 0) {
2166                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2167                         r = -EINTR;
2168                         kvm_run->exit_reason = KVM_EXIT_INTR;
2169                         ++vcpu->stat.request_irq_exits;
2170                         goto out;
2171                 }
2172                 if (!need_resched()) {
2173                         ++vcpu->stat.light_exits;
2174                         goto again;
2175                 }
2176         }
2177
2178 out:
2179         if (r > 0) {
2180                 kvm_resched(vcpu);
2181                 goto preempted;
2182         }
2183
2184         post_kvm_run_save(vcpu, kvm_run);
2185
2186         return r;
2187 }
2188
2189
2190 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2191 {
2192         int r;
2193         sigset_t sigsaved;
2194
2195         vcpu_load(vcpu);
2196
2197         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2198                 kvm_vcpu_block(vcpu);
2199                 vcpu_put(vcpu);
2200                 return -EAGAIN;
2201         }
2202
2203         if (vcpu->sigset_active)
2204                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2205
2206         /* re-sync apic's tpr */
2207         if (!irqchip_in_kernel(vcpu->kvm))
2208                 set_cr8(vcpu, kvm_run->cr8);
2209
2210         if (vcpu->pio.cur_count) {
2211                 r = complete_pio(vcpu);
2212                 if (r)
2213                         goto out;
2214         }
2215
2216         if (vcpu->mmio_needed) {
2217                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2218                 vcpu->mmio_read_completed = 1;
2219                 vcpu->mmio_needed = 0;
2220                 r = emulate_instruction(vcpu, kvm_run,
2221                                         vcpu->mmio_fault_cr2, 0, 1);
2222                 if (r == EMULATE_DO_MMIO) {
2223                         /*
2224                          * Read-modify-write.  Back to userspace.
2225                          */
2226                         r = 0;
2227                         goto out;
2228                 }
2229         }
2230
2231         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2232                 kvm_x86_ops->cache_regs(vcpu);
2233                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2234                 kvm_x86_ops->decache_regs(vcpu);
2235         }
2236
2237         r = __vcpu_run(vcpu, kvm_run);
2238
2239 out:
2240         if (vcpu->sigset_active)
2241                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2242
2243         vcpu_put(vcpu);
2244         return r;
2245 }
2246
2247 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2248                                    struct kvm_regs *regs)
2249 {
2250         vcpu_load(vcpu);
2251
2252         kvm_x86_ops->cache_regs(vcpu);
2253
2254         regs->rax = vcpu->regs[VCPU_REGS_RAX];
2255         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2256         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2257         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2258         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2259         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2260         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2261         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2262 #ifdef CONFIG_X86_64
2263         regs->r8 = vcpu->regs[VCPU_REGS_R8];
2264         regs->r9 = vcpu->regs[VCPU_REGS_R9];
2265         regs->r10 = vcpu->regs[VCPU_REGS_R10];
2266         regs->r11 = vcpu->regs[VCPU_REGS_R11];
2267         regs->r12 = vcpu->regs[VCPU_REGS_R12];
2268         regs->r13 = vcpu->regs[VCPU_REGS_R13];
2269         regs->r14 = vcpu->regs[VCPU_REGS_R14];
2270         regs->r15 = vcpu->regs[VCPU_REGS_R15];
2271 #endif
2272
2273         regs->rip = vcpu->rip;
2274         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2275
2276         /*
2277          * Don't leak debug flags in case they were set for guest debugging
2278          */
2279         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2280                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2281
2282         vcpu_put(vcpu);
2283
2284         return 0;
2285 }
2286
2287 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2288                                    struct kvm_regs *regs)
2289 {
2290         vcpu_load(vcpu);
2291
2292         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2293         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2294         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2295         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2296         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2297         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2298         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2299         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2300 #ifdef CONFIG_X86_64
2301         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2302         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2303         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2304         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2305         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2306         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2307         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2308         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2309 #endif
2310
2311         vcpu->rip = regs->rip;
2312         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2313
2314         kvm_x86_ops->decache_regs(vcpu);
2315
2316         vcpu_put(vcpu);
2317
2318         return 0;
2319 }
2320
2321 static void get_segment(struct kvm_vcpu *vcpu,
2322                         struct kvm_segment *var, int seg)
2323 {
2324         return kvm_x86_ops->get_segment(vcpu, var, seg);
2325 }
2326
2327 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2328                                     struct kvm_sregs *sregs)
2329 {
2330         struct descriptor_table dt;
2331         int pending_vec;
2332
2333         vcpu_load(vcpu);
2334
2335         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2336         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2337         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2338         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2339         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2340         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2341
2342         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2343         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2344
2345         kvm_x86_ops->get_idt(vcpu, &dt);
2346         sregs->idt.limit = dt.limit;
2347         sregs->idt.base = dt.base;
2348         kvm_x86_ops->get_gdt(vcpu, &dt);
2349         sregs->gdt.limit = dt.limit;
2350         sregs->gdt.base = dt.base;
2351
2352         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2353         sregs->cr0 = vcpu->cr0;
2354         sregs->cr2 = vcpu->cr2;
2355         sregs->cr3 = vcpu->cr3;
2356         sregs->cr4 = vcpu->cr4;
2357         sregs->cr8 = get_cr8(vcpu);
2358         sregs->efer = vcpu->shadow_efer;
2359         sregs->apic_base = kvm_get_apic_base(vcpu);
2360
2361         if (irqchip_in_kernel(vcpu->kvm)) {
2362                 memset(sregs->interrupt_bitmap, 0,
2363                        sizeof sregs->interrupt_bitmap);
2364                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2365                 if (pending_vec >= 0)
2366                         set_bit(pending_vec, (unsigned long *)sregs->interrupt_bitmap);
2367         } else
2368                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2369                        sizeof sregs->interrupt_bitmap);
2370
2371         vcpu_put(vcpu);
2372
2373         return 0;
2374 }
2375
2376 static void set_segment(struct kvm_vcpu *vcpu,
2377                         struct kvm_segment *var, int seg)
2378 {
2379         return kvm_x86_ops->set_segment(vcpu, var, seg);
2380 }
2381
2382 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2383                                     struct kvm_sregs *sregs)
2384 {
2385         int mmu_reset_needed = 0;
2386         int i, pending_vec, max_bits;
2387         struct descriptor_table dt;
2388
2389         vcpu_load(vcpu);
2390
2391         dt.limit = sregs->idt.limit;
2392         dt.base = sregs->idt.base;
2393         kvm_x86_ops->set_idt(vcpu, &dt);
2394         dt.limit = sregs->gdt.limit;
2395         dt.base = sregs->gdt.base;
2396         kvm_x86_ops->set_gdt(vcpu, &dt);
2397
2398         vcpu->cr2 = sregs->cr2;
2399         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2400         vcpu->cr3 = sregs->cr3;
2401
2402         set_cr8(vcpu, sregs->cr8);
2403
2404         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2405 #ifdef CONFIG_X86_64
2406         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2407 #endif
2408         kvm_set_apic_base(vcpu, sregs->apic_base);
2409
2410         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2411
2412         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2413         vcpu->cr0 = sregs->cr0;
2414         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2415
2416         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2417         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2418         if (!is_long_mode(vcpu) && is_pae(vcpu))
2419                 load_pdptrs(vcpu, vcpu->cr3);
2420
2421         if (mmu_reset_needed)
2422                 kvm_mmu_reset_context(vcpu);
2423
2424         if (!irqchip_in_kernel(vcpu->kvm)) {
2425                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2426                        sizeof vcpu->irq_pending);
2427                 vcpu->irq_summary = 0;
2428                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2429                         if (vcpu->irq_pending[i])
2430                                 __set_bit(i, &vcpu->irq_summary);
2431         } else {
2432                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2433                 pending_vec = find_first_bit(
2434                         (const unsigned long *)sregs->interrupt_bitmap,
2435                         max_bits);
2436                 /* Only pending external irq is handled here */
2437                 if (pending_vec < max_bits) {
2438                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2439                         printk("Set back pending irq %d\n", pending_vec);
2440                 }
2441         }
2442
2443         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2444         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2445         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2446         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2447         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2448         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2449
2450         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2451         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2452
2453         vcpu_put(vcpu);
2454
2455         return 0;
2456 }
2457
2458 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2459 {
2460         struct kvm_segment cs;
2461
2462         get_segment(vcpu, &cs, VCPU_SREG_CS);
2463         *db = cs.db;
2464         *l = cs.l;
2465 }
2466 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2467
2468 /*
2469  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2470  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2471  *
2472  * This list is modified at module load time to reflect the
2473  * capabilities of the host cpu.
2474  */
2475 static u32 msrs_to_save[] = {
2476         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2477         MSR_K6_STAR,
2478 #ifdef CONFIG_X86_64
2479         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2480 #endif
2481         MSR_IA32_TIME_STAMP_COUNTER,
2482 };
2483
2484 static unsigned num_msrs_to_save;
2485
2486 static u32 emulated_msrs[] = {
2487         MSR_IA32_MISC_ENABLE,
2488 };
2489
2490 static __init void kvm_init_msr_list(void)
2491 {
2492         u32 dummy[2];
2493         unsigned i, j;
2494
2495         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2496                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2497                         continue;
2498                 if (j < i)
2499                         msrs_to_save[j] = msrs_to_save[i];
2500                 j++;
2501         }
2502         num_msrs_to_save = j;
2503 }
2504
2505 /*
2506  * Adapt set_msr() to msr_io()'s calling convention
2507  */
2508 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2509 {
2510         return kvm_set_msr(vcpu, index, *data);
2511 }
2512
2513 /*
2514  * Read or write a bunch of msrs. All parameters are kernel addresses.
2515  *
2516  * @return number of msrs set successfully.
2517  */
2518 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2519                     struct kvm_msr_entry *entries,
2520                     int (*do_msr)(struct kvm_vcpu *vcpu,
2521                                   unsigned index, u64 *data))
2522 {
2523         int i;
2524
2525         vcpu_load(vcpu);
2526
2527         for (i = 0; i < msrs->nmsrs; ++i)
2528                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2529                         break;
2530
2531         vcpu_put(vcpu);
2532
2533         return i;
2534 }
2535
2536 /*
2537  * Read or write a bunch of msrs. Parameters are user addresses.
2538  *
2539  * @return number of msrs set successfully.
2540  */
2541 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2542                   int (*do_msr)(struct kvm_vcpu *vcpu,
2543                                 unsigned index, u64 *data),
2544                   int writeback)
2545 {
2546         struct kvm_msrs msrs;
2547         struct kvm_msr_entry *entries;
2548         int r, n;
2549         unsigned size;
2550
2551         r = -EFAULT;
2552         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2553                 goto out;
2554
2555         r = -E2BIG;
2556         if (msrs.nmsrs >= MAX_IO_MSRS)
2557                 goto out;
2558
2559         r = -ENOMEM;
2560         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2561         entries = vmalloc(size);
2562         if (!entries)
2563                 goto out;
2564
2565         r = -EFAULT;
2566         if (copy_from_user(entries, user_msrs->entries, size))
2567                 goto out_free;
2568
2569         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2570         if (r < 0)
2571                 goto out_free;
2572
2573         r = -EFAULT;
2574         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2575                 goto out_free;
2576
2577         r = n;
2578
2579 out_free:
2580         vfree(entries);
2581 out:
2582         return r;
2583 }
2584
2585 /*
2586  * Translate a guest virtual address to a guest physical address.
2587  */
2588 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2589                                     struct kvm_translation *tr)
2590 {
2591         unsigned long vaddr = tr->linear_address;
2592         gpa_t gpa;
2593
2594         vcpu_load(vcpu);
2595         mutex_lock(&vcpu->kvm->lock);
2596         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2597         tr->physical_address = gpa;
2598         tr->valid = gpa != UNMAPPED_GVA;
2599         tr->writeable = 1;
2600         tr->usermode = 0;
2601         mutex_unlock(&vcpu->kvm->lock);
2602         vcpu_put(vcpu);
2603
2604         return 0;
2605 }
2606
2607 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2608                                     struct kvm_interrupt *irq)
2609 {
2610         if (irq->irq < 0 || irq->irq >= 256)
2611                 return -EINVAL;
2612         if (irqchip_in_kernel(vcpu->kvm))
2613                 return -ENXIO;
2614         vcpu_load(vcpu);
2615
2616         set_bit(irq->irq, vcpu->irq_pending);
2617         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2618
2619         vcpu_put(vcpu);
2620
2621         return 0;
2622 }
2623
2624 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2625                                       struct kvm_debug_guest *dbg)
2626 {
2627         int r;
2628
2629         vcpu_load(vcpu);
2630
2631         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2632
2633         vcpu_put(vcpu);
2634
2635         return r;
2636 }
2637
2638 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2639                                     unsigned long address,
2640                                     int *type)
2641 {
2642         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2643         unsigned long pgoff;
2644         struct page *page;
2645
2646         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2647         if (pgoff == 0)
2648                 page = virt_to_page(vcpu->run);
2649         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2650                 page = virt_to_page(vcpu->pio_data);
2651         else
2652                 return NOPAGE_SIGBUS;
2653         get_page(page);
2654         if (type != NULL)
2655                 *type = VM_FAULT_MINOR;
2656
2657         return page;
2658 }
2659
2660 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2661         .nopage = kvm_vcpu_nopage,
2662 };
2663
2664 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2665 {
2666         vma->vm_ops = &kvm_vcpu_vm_ops;
2667         return 0;
2668 }
2669
2670 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2671 {
2672         struct kvm_vcpu *vcpu = filp->private_data;
2673
2674         fput(vcpu->kvm->filp);
2675         return 0;
2676 }
2677
2678 static struct file_operations kvm_vcpu_fops = {
2679         .release        = kvm_vcpu_release,
2680         .unlocked_ioctl = kvm_vcpu_ioctl,
2681         .compat_ioctl   = kvm_vcpu_ioctl,
2682         .mmap           = kvm_vcpu_mmap,
2683 };
2684
2685 /*
2686  * Allocates an inode for the vcpu.
2687  */
2688 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2689 {
2690         int fd, r;
2691         struct inode *inode;
2692         struct file *file;
2693
2694         r = anon_inode_getfd(&fd, &inode, &file,
2695                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2696         if (r)
2697                 return r;
2698         atomic_inc(&vcpu->kvm->filp->f_count);
2699         return fd;
2700 }
2701
2702 /*
2703  * Creates some virtual cpus.  Good luck creating more than one.
2704  */
2705 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2706 {
2707         int r;
2708         struct kvm_vcpu *vcpu;
2709
2710         if (!valid_vcpu(n))
2711                 return -EINVAL;
2712
2713         vcpu = kvm_x86_ops->vcpu_create(kvm, n);
2714         if (IS_ERR(vcpu))
2715                 return PTR_ERR(vcpu);
2716
2717         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2718
2719         /* We do fxsave: this must be aligned. */
2720         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2721
2722         vcpu_load(vcpu);
2723         r = kvm_mmu_setup(vcpu);
2724         vcpu_put(vcpu);
2725         if (r < 0)
2726                 goto free_vcpu;
2727
2728         mutex_lock(&kvm->lock);
2729         if (kvm->vcpus[n]) {
2730                 r = -EEXIST;
2731                 mutex_unlock(&kvm->lock);
2732                 goto mmu_unload;
2733         }
2734         kvm->vcpus[n] = vcpu;
2735         mutex_unlock(&kvm->lock);
2736
2737         /* Now it's all set up, let userspace reach it */
2738         r = create_vcpu_fd(vcpu);
2739         if (r < 0)
2740                 goto unlink;
2741         return r;
2742
2743 unlink:
2744         mutex_lock(&kvm->lock);
2745         kvm->vcpus[n] = NULL;
2746         mutex_unlock(&kvm->lock);
2747
2748 mmu_unload:
2749         vcpu_load(vcpu);
2750         kvm_mmu_unload(vcpu);
2751         vcpu_put(vcpu);
2752
2753 free_vcpu:
2754         kvm_x86_ops->vcpu_free(vcpu);
2755         return r;
2756 }
2757
2758 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2759 {
2760         u64 efer;
2761         int i;
2762         struct kvm_cpuid_entry *e, *entry;
2763
2764         rdmsrl(MSR_EFER, efer);
2765         entry = NULL;
2766         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2767                 e = &vcpu->cpuid_entries[i];
2768                 if (e->function == 0x80000001) {
2769                         entry = e;
2770                         break;
2771                 }
2772         }
2773         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2774                 entry->edx &= ~(1 << 20);
2775                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2776         }
2777 }
2778
2779 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2780                                     struct kvm_cpuid *cpuid,
2781                                     struct kvm_cpuid_entry __user *entries)
2782 {
2783         int r;
2784
2785         r = -E2BIG;
2786         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2787                 goto out;
2788         r = -EFAULT;
2789         if (copy_from_user(&vcpu->cpuid_entries, entries,
2790                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2791                 goto out;
2792         vcpu->cpuid_nent = cpuid->nent;
2793         cpuid_fix_nx_cap(vcpu);
2794         return 0;
2795
2796 out:
2797         return r;
2798 }
2799
2800 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2801 {
2802         if (sigset) {
2803                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2804                 vcpu->sigset_active = 1;
2805                 vcpu->sigset = *sigset;
2806         } else
2807                 vcpu->sigset_active = 0;
2808         return 0;
2809 }
2810
2811 /*
2812  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2813  * we have asm/x86/processor.h
2814  */
2815 struct fxsave {
2816         u16     cwd;
2817         u16     swd;
2818         u16     twd;
2819         u16     fop;
2820         u64     rip;
2821         u64     rdp;
2822         u32     mxcsr;
2823         u32     mxcsr_mask;
2824         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2825 #ifdef CONFIG_X86_64
2826         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2827 #else
2828         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2829 #endif
2830 };
2831
2832 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2833 {
2834         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2835
2836         vcpu_load(vcpu);
2837
2838         memcpy(fpu->fpr, fxsave->st_space, 128);
2839         fpu->fcw = fxsave->cwd;
2840         fpu->fsw = fxsave->swd;
2841         fpu->ftwx = fxsave->twd;
2842         fpu->last_opcode = fxsave->fop;
2843         fpu->last_ip = fxsave->rip;
2844         fpu->last_dp = fxsave->rdp;
2845         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2846
2847         vcpu_put(vcpu);
2848
2849         return 0;
2850 }
2851
2852 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2853 {
2854         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2855
2856         vcpu_load(vcpu);
2857
2858         memcpy(fxsave->st_space, fpu->fpr, 128);
2859         fxsave->cwd = fpu->fcw;
2860         fxsave->swd = fpu->fsw;
2861         fxsave->twd = fpu->ftwx;
2862         fxsave->fop = fpu->last_opcode;
2863         fxsave->rip = fpu->last_ip;
2864         fxsave->rdp = fpu->last_dp;
2865         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2866
2867         vcpu_put(vcpu);
2868
2869         return 0;
2870 }
2871
2872 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2873                                     struct kvm_lapic_state *s)
2874 {
2875         vcpu_load(vcpu);
2876         memcpy(s->regs, vcpu->apic->regs, sizeof *s);
2877         vcpu_put(vcpu);
2878
2879         return 0;
2880 }
2881
2882 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2883                                     struct kvm_lapic_state *s)
2884 {
2885         vcpu_load(vcpu);
2886         memcpy(vcpu->apic->regs, s->regs, sizeof *s);
2887         kvm_apic_post_state_restore(vcpu);
2888         vcpu_put(vcpu);
2889
2890         return 0;
2891 }
2892
2893 static long kvm_vcpu_ioctl(struct file *filp,
2894                            unsigned int ioctl, unsigned long arg)
2895 {
2896         struct kvm_vcpu *vcpu = filp->private_data;
2897         void __user *argp = (void __user *)arg;
2898         int r = -EINVAL;
2899
2900         switch (ioctl) {
2901         case KVM_RUN:
2902                 r = -EINVAL;
2903                 if (arg)
2904                         goto out;
2905                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2906                 break;
2907         case KVM_GET_REGS: {
2908                 struct kvm_regs kvm_regs;
2909
2910                 memset(&kvm_regs, 0, sizeof kvm_regs);
2911                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2912                 if (r)
2913                         goto out;
2914                 r = -EFAULT;
2915                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2916                         goto out;
2917                 r = 0;
2918                 break;
2919         }
2920         case KVM_SET_REGS: {
2921                 struct kvm_regs kvm_regs;
2922
2923                 r = -EFAULT;
2924                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2925                         goto out;
2926                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2927                 if (r)
2928                         goto out;
2929                 r = 0;
2930                 break;
2931         }
2932         case KVM_GET_SREGS: {
2933                 struct kvm_sregs kvm_sregs;
2934
2935                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2936                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2937                 if (r)
2938                         goto out;
2939                 r = -EFAULT;
2940                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2941                         goto out;
2942                 r = 0;
2943                 break;
2944         }
2945         case KVM_SET_SREGS: {
2946                 struct kvm_sregs kvm_sregs;
2947
2948                 r = -EFAULT;
2949                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2950                         goto out;
2951                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2952                 if (r)
2953                         goto out;
2954                 r = 0;
2955                 break;
2956         }
2957         case KVM_TRANSLATE: {
2958                 struct kvm_translation tr;
2959
2960                 r = -EFAULT;
2961                 if (copy_from_user(&tr, argp, sizeof tr))
2962                         goto out;
2963                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2964                 if (r)
2965                         goto out;
2966                 r = -EFAULT;
2967                 if (copy_to_user(argp, &tr, sizeof tr))
2968                         goto out;
2969                 r = 0;
2970                 break;
2971         }
2972         case KVM_INTERRUPT: {
2973                 struct kvm_interrupt irq;
2974
2975                 r = -EFAULT;
2976                 if (copy_from_user(&irq, argp, sizeof irq))
2977                         goto out;
2978                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2979                 if (r)
2980                         goto out;
2981                 r = 0;
2982                 break;
2983         }
2984         case KVM_DEBUG_GUEST: {
2985                 struct kvm_debug_guest dbg;
2986
2987                 r = -EFAULT;
2988                 if (copy_from_user(&dbg, argp, sizeof dbg))
2989                         goto out;
2990                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2991                 if (r)
2992                         goto out;
2993                 r = 0;
2994                 break;
2995         }
2996         case KVM_GET_MSRS:
2997                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2998                 break;
2999         case KVM_SET_MSRS:
3000                 r = msr_io(vcpu, argp, do_set_msr, 0);
3001                 break;
3002         case KVM_SET_CPUID: {
3003                 struct kvm_cpuid __user *cpuid_arg = argp;
3004                 struct kvm_cpuid cpuid;
3005
3006                 r = -EFAULT;
3007                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3008                         goto out;
3009                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3010                 if (r)
3011                         goto out;
3012                 break;
3013         }
3014         case KVM_SET_SIGNAL_MASK: {
3015                 struct kvm_signal_mask __user *sigmask_arg = argp;
3016                 struct kvm_signal_mask kvm_sigmask;
3017                 sigset_t sigset, *p;
3018
3019                 p = NULL;
3020                 if (argp) {
3021                         r = -EFAULT;
3022                         if (copy_from_user(&kvm_sigmask, argp,
3023                                            sizeof kvm_sigmask))
3024                                 goto out;
3025                         r = -EINVAL;
3026                         if (kvm_sigmask.len != sizeof sigset)
3027                                 goto out;
3028                         r = -EFAULT;
3029                         if (copy_from_user(&sigset, sigmask_arg->sigset,
3030                                            sizeof sigset))
3031                                 goto out;
3032                         p = &sigset;
3033                 }
3034                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3035                 break;
3036         }
3037         case KVM_GET_FPU: {
3038                 struct kvm_fpu fpu;
3039
3040                 memset(&fpu, 0, sizeof fpu);
3041                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
3042                 if (r)
3043                         goto out;
3044                 r = -EFAULT;
3045                 if (copy_to_user(argp, &fpu, sizeof fpu))
3046                         goto out;
3047                 r = 0;
3048                 break;
3049         }
3050         case KVM_SET_FPU: {
3051                 struct kvm_fpu fpu;
3052
3053                 r = -EFAULT;
3054                 if (copy_from_user(&fpu, argp, sizeof fpu))
3055                         goto out;
3056                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
3057                 if (r)
3058                         goto out;
3059                 r = 0;
3060                 break;
3061         }
3062         case KVM_GET_LAPIC: {
3063                 struct kvm_lapic_state lapic;
3064
3065                 memset(&lapic, 0, sizeof lapic);
3066                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
3067                 if (r)
3068                         goto out;
3069                 r = -EFAULT;
3070                 if (copy_to_user(argp, &lapic, sizeof lapic))
3071                         goto out;
3072                 r = 0;
3073                 break;
3074         }
3075         case KVM_SET_LAPIC: {
3076                 struct kvm_lapic_state lapic;
3077
3078                 r = -EFAULT;
3079                 if (copy_from_user(&lapic, argp, sizeof lapic))
3080                         goto out;
3081                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
3082                 if (r)
3083                         goto out;
3084                 r = 0;
3085                 break;
3086         }
3087         default:
3088                 ;
3089         }
3090 out:
3091         return r;
3092 }
3093
3094 static long kvm_vm_ioctl(struct file *filp,
3095                            unsigned int ioctl, unsigned long arg)
3096 {
3097         struct kvm *kvm = filp->private_data;
3098         void __user *argp = (void __user *)arg;
3099         int r = -EINVAL;
3100
3101         switch (ioctl) {
3102         case KVM_CREATE_VCPU:
3103                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3104                 if (r < 0)
3105                         goto out;
3106                 break;
3107         case KVM_SET_MEMORY_REGION: {
3108                 struct kvm_memory_region kvm_mem;
3109
3110                 r = -EFAULT;
3111                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
3112                         goto out;
3113                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
3114                 if (r)
3115                         goto out;
3116                 break;
3117         }
3118         case KVM_SET_NR_MMU_PAGES:
3119                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3120                 if (r)
3121                         goto out;
3122                 break;
3123         case KVM_GET_NR_MMU_PAGES:
3124                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3125                 break;
3126         case KVM_GET_DIRTY_LOG: {
3127                 struct kvm_dirty_log log;
3128
3129                 r = -EFAULT;
3130                 if (copy_from_user(&log, argp, sizeof log))
3131                         goto out;
3132                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3133                 if (r)
3134                         goto out;
3135                 break;
3136         }
3137         case KVM_SET_MEMORY_ALIAS: {
3138                 struct kvm_memory_alias alias;
3139
3140                 r = -EFAULT;
3141                 if (copy_from_user(&alias, argp, sizeof alias))
3142                         goto out;
3143                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
3144                 if (r)
3145                         goto out;
3146                 break;
3147         }
3148         case KVM_CREATE_IRQCHIP:
3149                 r = -ENOMEM;
3150                 kvm->vpic = kvm_create_pic(kvm);
3151                 if (kvm->vpic) {
3152                         r = kvm_ioapic_init(kvm);
3153                         if (r) {
3154                                 kfree(kvm->vpic);
3155                                 kvm->vpic = NULL;
3156                                 goto out;
3157                         }
3158                 }
3159                 else
3160                         goto out;
3161                 break;
3162         case KVM_IRQ_LINE: {
3163                 struct kvm_irq_level irq_event;
3164
3165                 r = -EFAULT;
3166                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3167                         goto out;
3168                 if (irqchip_in_kernel(kvm)) {
3169                         mutex_lock(&kvm->lock);
3170                         if (irq_event.irq < 16)
3171                                 kvm_pic_set_irq(pic_irqchip(kvm),
3172                                         irq_event.irq,
3173                                         irq_event.level);
3174                         kvm_ioapic_set_irq(kvm->vioapic,
3175                                         irq_event.irq,
3176                                         irq_event.level);
3177                         mutex_unlock(&kvm->lock);
3178                         r = 0;
3179                 }
3180                 break;
3181         }
3182         case KVM_GET_IRQCHIP: {
3183                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3184                 struct kvm_irqchip chip;
3185
3186                 r = -EFAULT;
3187                 if (copy_from_user(&chip, argp, sizeof chip))
3188                         goto out;
3189                 r = -ENXIO;
3190                 if (!irqchip_in_kernel(kvm))
3191                         goto out;
3192                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
3193                 if (r)
3194                         goto out;
3195                 r = -EFAULT;
3196                 if (copy_to_user(argp, &chip, sizeof chip))
3197                         goto out;
3198                 r = 0;
3199                 break;
3200         }
3201         case KVM_SET_IRQCHIP: {
3202                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3203                 struct kvm_irqchip chip;
3204
3205                 r = -EFAULT;
3206                 if (copy_from_user(&chip, argp, sizeof chip))
3207                         goto out;
3208                 r = -ENXIO;
3209                 if (!irqchip_in_kernel(kvm))
3210                         goto out;
3211                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
3212                 if (r)
3213                         goto out;
3214                 r = 0;
3215                 break;
3216         }
3217         default:
3218                 ;
3219         }
3220 out:
3221         return r;
3222 }
3223
3224 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
3225                                   unsigned long address,
3226                                   int *type)
3227 {
3228         struct kvm *kvm = vma->vm_file->private_data;
3229         unsigned long pgoff;
3230         struct page *page;
3231
3232         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
3233         page = gfn_to_page(kvm, pgoff);
3234         if (!page)
3235                 return NOPAGE_SIGBUS;
3236         get_page(page);
3237         if (type != NULL)
3238                 *type = VM_FAULT_MINOR;
3239
3240         return page;
3241 }
3242
3243 static struct vm_operations_struct kvm_vm_vm_ops = {
3244         .nopage = kvm_vm_nopage,
3245 };
3246
3247 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
3248 {
3249         vma->vm_ops = &kvm_vm_vm_ops;
3250         return 0;
3251 }
3252
3253 static struct file_operations kvm_vm_fops = {
3254         .release        = kvm_vm_release,
3255         .unlocked_ioctl = kvm_vm_ioctl,
3256         .compat_ioctl   = kvm_vm_ioctl,
3257         .mmap           = kvm_vm_mmap,
3258 };
3259
3260 static int kvm_dev_ioctl_create_vm(void)
3261 {
3262         int fd, r;
3263         struct inode *inode;
3264         struct file *file;
3265         struct kvm *kvm;
3266
3267         kvm = kvm_create_vm();
3268         if (IS_ERR(kvm))
3269                 return PTR_ERR(kvm);
3270         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
3271         if (r) {
3272                 kvm_destroy_vm(kvm);
3273                 return r;
3274         }
3275
3276         kvm->filp = file;
3277
3278         return fd;
3279 }
3280
3281 static long kvm_dev_ioctl(struct file *filp,
3282                           unsigned int ioctl, unsigned long arg)
3283 {
3284         void __user *argp = (void __user *)arg;
3285         long r = -EINVAL;
3286
3287         switch (ioctl) {
3288         case KVM_GET_API_VERSION:
3289                 r = -EINVAL;
3290                 if (arg)
3291                         goto out;
3292                 r = KVM_API_VERSION;
3293                 break;
3294         case KVM_CREATE_VM:
3295                 r = -EINVAL;
3296                 if (arg)
3297                         goto out;
3298                 r = kvm_dev_ioctl_create_vm();
3299                 break;
3300         case KVM_GET_MSR_INDEX_LIST: {
3301                 struct kvm_msr_list __user *user_msr_list = argp;
3302                 struct kvm_msr_list msr_list;
3303                 unsigned n;
3304
3305                 r = -EFAULT;
3306                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
3307                         goto out;
3308                 n = msr_list.nmsrs;
3309                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
3310                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
3311                         goto out;
3312                 r = -E2BIG;
3313                 if (n < num_msrs_to_save)
3314                         goto out;
3315                 r = -EFAULT;
3316                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3317                                  num_msrs_to_save * sizeof(u32)))
3318                         goto out;
3319                 if (copy_to_user(user_msr_list->indices
3320                                  + num_msrs_to_save * sizeof(u32),
3321                                  &emulated_msrs,
3322                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
3323                         goto out;
3324                 r = 0;
3325                 break;
3326         }
3327         case KVM_CHECK_EXTENSION: {
3328                 int ext = (long)argp;
3329
3330                 switch (ext) {
3331                 case KVM_CAP_IRQCHIP:
3332                 case KVM_CAP_HLT:
3333                 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3334                         r = 1;
3335                         break;
3336                 default:
3337                         r = 0;
3338                         break;
3339                 }
3340                 break;
3341         }
3342         case KVM_GET_VCPU_MMAP_SIZE:
3343                 r = -EINVAL;
3344                 if (arg)
3345                         goto out;
3346                 r = 2 * PAGE_SIZE;
3347                 break;
3348         default:
3349                 ;
3350         }
3351 out:
3352         return r;
3353 }
3354
3355 static struct file_operations kvm_chardev_ops = {
3356         .unlocked_ioctl = kvm_dev_ioctl,
3357         .compat_ioctl   = kvm_dev_ioctl,
3358 };
3359
3360 static struct miscdevice kvm_dev = {
3361         KVM_MINOR,
3362         "kvm",
3363         &kvm_chardev_ops,
3364 };
3365
3366 /*
3367  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
3368  * cached on it.
3369  */
3370 static void decache_vcpus_on_cpu(int cpu)
3371 {
3372         struct kvm *vm;
3373         struct kvm_vcpu *vcpu;
3374         int i;
3375
3376         spin_lock(&kvm_lock);
3377         list_for_each_entry(vm, &vm_list, vm_list)
3378                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3379                         vcpu = vm->vcpus[i];
3380                         if (!vcpu)
3381                                 continue;
3382                         /*
3383                          * If the vcpu is locked, then it is running on some
3384                          * other cpu and therefore it is not cached on the
3385                          * cpu in question.
3386                          *
3387                          * If it's not locked, check the last cpu it executed
3388                          * on.
3389                          */
3390                         if (mutex_trylock(&vcpu->mutex)) {
3391                                 if (vcpu->cpu == cpu) {
3392                                         kvm_x86_ops->vcpu_decache(vcpu);
3393                                         vcpu->cpu = -1;
3394                                 }
3395                                 mutex_unlock(&vcpu->mutex);
3396                         }
3397                 }
3398         spin_unlock(&kvm_lock);
3399 }
3400
3401 static void hardware_enable(void *junk)
3402 {
3403         int cpu = raw_smp_processor_id();
3404
3405         if (cpu_isset(cpu, cpus_hardware_enabled))
3406                 return;
3407         cpu_set(cpu, cpus_hardware_enabled);
3408         kvm_x86_ops->hardware_enable(NULL);
3409 }
3410
3411 static void hardware_disable(void *junk)
3412 {
3413         int cpu = raw_smp_processor_id();
3414
3415         if (!cpu_isset(cpu, cpus_hardware_enabled))
3416                 return;
3417         cpu_clear(cpu, cpus_hardware_enabled);
3418         decache_vcpus_on_cpu(cpu);
3419         kvm_x86_ops->hardware_disable(NULL);
3420 }
3421
3422 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3423                            void *v)
3424 {
3425         int cpu = (long)v;
3426
3427         switch (val) {
3428         case CPU_DYING:
3429         case CPU_DYING_FROZEN:
3430                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3431                        cpu);
3432                 hardware_disable(NULL);
3433                 break;
3434         case CPU_UP_CANCELED:
3435         case CPU_UP_CANCELED_FROZEN:
3436                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3437                        cpu);
3438                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3439                 break;
3440         case CPU_ONLINE:
3441         case CPU_ONLINE_FROZEN:
3442                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3443                        cpu);
3444                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3445                 break;
3446         }
3447         return NOTIFY_OK;
3448 }
3449
3450 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3451                        void *v)
3452 {
3453         if (val == SYS_RESTART) {
3454                 /*
3455                  * Some (well, at least mine) BIOSes hang on reboot if
3456                  * in vmx root mode.
3457                  */
3458                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3459                 on_each_cpu(hardware_disable, NULL, 0, 1);
3460         }
3461         return NOTIFY_OK;
3462 }
3463
3464 static struct notifier_block kvm_reboot_notifier = {
3465         .notifier_call = kvm_reboot,
3466         .priority = 0,
3467 };
3468
3469 void kvm_io_bus_init(struct kvm_io_bus *bus)
3470 {
3471         memset(bus, 0, sizeof(*bus));
3472 }
3473
3474 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3475 {
3476         int i;
3477
3478         for (i = 0; i < bus->dev_count; i++) {
3479                 struct kvm_io_device *pos = bus->devs[i];
3480
3481                 kvm_iodevice_destructor(pos);
3482         }
3483 }
3484
3485 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3486 {
3487         int i;
3488
3489         for (i = 0; i < bus->dev_count; i++) {
3490                 struct kvm_io_device *pos = bus->devs[i];
3491
3492                 if (pos->in_range(pos, addr))
3493                         return pos;
3494         }
3495
3496         return NULL;
3497 }
3498
3499 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3500 {
3501         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3502
3503         bus->devs[bus->dev_count++] = dev;
3504 }
3505
3506 static struct notifier_block kvm_cpu_notifier = {
3507         .notifier_call = kvm_cpu_hotplug,
3508         .priority = 20, /* must be > scheduler priority */
3509 };
3510
3511 static u64 stat_get(void *_offset)
3512 {
3513         unsigned offset = (long)_offset;
3514         u64 total = 0;
3515         struct kvm *kvm;
3516         struct kvm_vcpu *vcpu;
3517         int i;
3518
3519         spin_lock(&kvm_lock);
3520         list_for_each_entry(kvm, &vm_list, vm_list)
3521                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3522                         vcpu = kvm->vcpus[i];
3523                         if (vcpu)
3524                                 total += *(u32 *)((void *)vcpu + offset);
3525                 }
3526         spin_unlock(&kvm_lock);
3527         return total;
3528 }
3529
3530 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3531
3532 static __init void kvm_init_debug(void)
3533 {
3534         struct kvm_stats_debugfs_item *p;
3535
3536         debugfs_dir = debugfs_create_dir("kvm", NULL);
3537         for (p = debugfs_entries; p->name; ++p)
3538                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3539                                                 (void *)(long)p->offset,
3540                                                 &stat_fops);
3541 }
3542
3543 static void kvm_exit_debug(void)
3544 {
3545         struct kvm_stats_debugfs_item *p;
3546
3547         for (p = debugfs_entries; p->name; ++p)
3548                 debugfs_remove(p->dentry);
3549         debugfs_remove(debugfs_dir);
3550 }
3551
3552 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3553 {
3554         hardware_disable(NULL);
3555         return 0;
3556 }
3557
3558 static int kvm_resume(struct sys_device *dev)
3559 {
3560         hardware_enable(NULL);
3561         return 0;
3562 }
3563
3564 static struct sysdev_class kvm_sysdev_class = {
3565         .name = "kvm",
3566         .suspend = kvm_suspend,
3567         .resume = kvm_resume,
3568 };
3569
3570 static struct sys_device kvm_sysdev = {
3571         .id = 0,
3572         .cls = &kvm_sysdev_class,
3573 };
3574
3575 hpa_t bad_page_address;
3576
3577 static inline
3578 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3579 {
3580         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3581 }
3582
3583 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3584 {
3585         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3586
3587         kvm_x86_ops->vcpu_load(vcpu, cpu);
3588 }
3589
3590 static void kvm_sched_out(struct preempt_notifier *pn,
3591                           struct task_struct *next)
3592 {
3593         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3594
3595         kvm_x86_ops->vcpu_put(vcpu);
3596 }
3597
3598 int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
3599                   struct module *module)
3600 {
3601         int r;
3602         int cpu;
3603
3604         if (kvm_x86_ops) {
3605                 printk(KERN_ERR "kvm: already loaded the other module\n");
3606                 return -EEXIST;
3607         }
3608
3609         if (!ops->cpu_has_kvm_support()) {
3610                 printk(KERN_ERR "kvm: no hardware support\n");
3611                 return -EOPNOTSUPP;
3612         }
3613         if (ops->disabled_by_bios()) {
3614                 printk(KERN_ERR "kvm: disabled by bios\n");
3615                 return -EOPNOTSUPP;
3616         }
3617
3618         kvm_x86_ops = ops;
3619
3620         r = kvm_x86_ops->hardware_setup();
3621         if (r < 0)
3622                 goto out;
3623
3624         for_each_online_cpu(cpu) {
3625                 smp_call_function_single(cpu,
3626                                 kvm_x86_ops->check_processor_compatibility,
3627                                 &r, 0, 1);
3628                 if (r < 0)
3629                         goto out_free_0;
3630         }
3631
3632         on_each_cpu(hardware_enable, NULL, 0, 1);
3633         r = register_cpu_notifier(&kvm_cpu_notifier);
3634         if (r)
3635                 goto out_free_1;
3636         register_reboot_notifier(&kvm_reboot_notifier);
3637
3638         r = sysdev_class_register(&kvm_sysdev_class);
3639         if (r)
3640                 goto out_free_2;
3641
3642         r = sysdev_register(&kvm_sysdev);
3643         if (r)
3644                 goto out_free_3;
3645
3646         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3647         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3648                                            __alignof__(struct kvm_vcpu), 0, 0);
3649         if (!kvm_vcpu_cache) {
3650                 r = -ENOMEM;
3651                 goto out_free_4;
3652         }
3653
3654         kvm_chardev_ops.owner = module;
3655
3656         r = misc_register(&kvm_dev);
3657         if (r) {
3658                 printk (KERN_ERR "kvm: misc device register failed\n");
3659                 goto out_free;
3660         }
3661
3662         kvm_preempt_ops.sched_in = kvm_sched_in;
3663         kvm_preempt_ops.sched_out = kvm_sched_out;
3664
3665         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3666
3667         return 0;
3668
3669 out_free:
3670         kmem_cache_destroy(kvm_vcpu_cache);
3671 out_free_4:
3672         sysdev_unregister(&kvm_sysdev);
3673 out_free_3:
3674         sysdev_class_unregister(&kvm_sysdev_class);
3675 out_free_2:
3676         unregister_reboot_notifier(&kvm_reboot_notifier);
3677         unregister_cpu_notifier(&kvm_cpu_notifier);
3678 out_free_1:
3679         on_each_cpu(hardware_disable, NULL, 0, 1);
3680 out_free_0:
3681         kvm_x86_ops->hardware_unsetup();
3682 out:
3683         kvm_x86_ops = NULL;
3684         return r;
3685 }
3686
3687 void kvm_exit_x86(void)
3688 {
3689         misc_deregister(&kvm_dev);
3690         kmem_cache_destroy(kvm_vcpu_cache);
3691         sysdev_unregister(&kvm_sysdev);
3692         sysdev_class_unregister(&kvm_sysdev_class);
3693         unregister_reboot_notifier(&kvm_reboot_notifier);
3694         unregister_cpu_notifier(&kvm_cpu_notifier);
3695         on_each_cpu(hardware_disable, NULL, 0, 1);
3696         kvm_x86_ops->hardware_unsetup();
3697         kvm_x86_ops = NULL;
3698 }
3699
3700 static __init int kvm_init(void)
3701 {
3702         static struct page *bad_page;
3703         int r;
3704
3705         r = kvm_mmu_module_init();
3706         if (r)
3707                 goto out4;
3708
3709         kvm_init_debug();
3710
3711         kvm_init_msr_list();
3712
3713         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3714                 r = -ENOMEM;
3715                 goto out;
3716         }
3717
3718         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3719         memset(__va(bad_page_address), 0, PAGE_SIZE);
3720
3721         return 0;
3722
3723 out:
3724         kvm_exit_debug();
3725         kvm_mmu_module_exit();
3726 out4:
3727         return r;
3728 }
3729
3730 static __exit void kvm_exit(void)
3731 {
3732         kvm_exit_debug();
3733         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3734         kvm_mmu_module_exit();
3735 }
3736
3737 module_init(kvm_init)
3738 module_exit(kvm_exit)
3739
3740 EXPORT_SYMBOL_GPL(kvm_init_x86);
3741 EXPORT_SYMBOL_GPL(kvm_exit_x86);