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 #include <linux/pagemap.h>
  44
  45 #include <asm/processor.h>
  46 #include <asm/msr.h>
  47 #include <asm/io.h>
  48 #include <asm/uaccess.h>
  49 #include <asm/desc.h>
  50
  51 MODULE_AUTHOR("Qumranet");
  52 MODULE_LICENSE("GPL");
  53
  54 static DEFINE_SPINLOCK(kvm_lock);
  55 static LIST_HEAD(vm_list);
  56
  57 static cpumask_t cpus_hardware_enabled;
  58
  59 struct kvm_x86_ops *kvm_x86_ops;
  60 struct kmem_cache *kvm_vcpu_cache;
  61 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  62
  63 static __read_mostly struct preempt_ops kvm_preempt_ops;
  64
  65 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  66
  67 static struct kvm_stats_debugfs_item {
  68         const char *name;
  69         int offset;
  70         struct dentry *dentry;
  71 } debugfs_entries[] = {
  72         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  73         { "pf_guest", STAT_OFFSET(pf_guest) },
  74         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  75         { "invlpg", STAT_OFFSET(invlpg) },
  76         { "exits", STAT_OFFSET(exits) },
  77         { "io_exits", STAT_OFFSET(io_exits) },
  78         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  79         { "signal_exits", STAT_OFFSET(signal_exits) },
  80         { "irq_window", STAT_OFFSET(irq_window_exits) },
  81         { "halt_exits", STAT_OFFSET(halt_exits) },
  82         { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
  83         { "request_irq", STAT_OFFSET(request_irq_exits) },
  84         { "irq_exits", STAT_OFFSET(irq_exits) },
  85         { "light_exits", STAT_OFFSET(light_exits) },
  86         { "efer_reload", STAT_OFFSET(efer_reload) },
  87         { NULL }
  88 };
  89
  90 static struct dentry *debugfs_dir;
  91
  92 #define MAX_IO_MSRS 256
  93
  94 #define CR0_RESERVED_BITS                                               \
  95         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  96                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  97                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  98 #define CR4_RESERVED_BITS                                               \
  99         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
 100                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
 101                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
 102                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
 103
 104 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
 105 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
 106
 107 #ifdef CONFIG_X86_64
 108 /* LDT or TSS descriptor in the GDT. 16 bytes. */
 109 struct segment_descriptor_64 {
 110         struct segment_descriptor s;
 111         u32 base_higher;
 112         u32 pad_zero;
 113 };
 114
 115 #endif
 116
 117 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 118                            unsigned long arg);
 119
 120 unsigned long segment_base(u16 selector)
 121 {
 122         struct descriptor_table gdt;
 123         struct segment_descriptor *d;
 124         unsigned long table_base;
 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 | ((unsigned long)d->base_mid << 16) |
 141                 ((unsigned long)d->base_high << 24);
 142 #ifdef CONFIG_X86_64
 143         if (d->system == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 144                 v |= ((unsigned long) \
 145                       ((struct segment_descriptor_64 *)d)->base_higher) << 32;
 146 #endif
 147         return v;
 148 }
 149 EXPORT_SYMBOL_GPL(segment_base);
 150
 151 static inline int valid_vcpu(int n)
 152 {
 153         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 154 }
 155
 156 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 157 {
 158         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
 159                 return;
 160
 161         vcpu->guest_fpu_loaded = 1;
 162         fx_save(&vcpu->host_fx_image);
 163         fx_restore(&vcpu->guest_fx_image);
 164 }
 165 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
 166
 167 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 168 {
 169         if (!vcpu->guest_fpu_loaded)
 170                 return;
 171
 172         vcpu->guest_fpu_loaded = 0;
 173         fx_save(&vcpu->guest_fx_image);
 174         fx_restore(&vcpu->host_fx_image);
 175 }
 176 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
 177
 178 /*
 179  * Switches to specified vcpu, until a matching vcpu_put()
 180  */
 181 static void vcpu_load(struct kvm_vcpu *vcpu)
 182 {
 183         int cpu;
 184
 185         mutex_lock(&vcpu->mutex);
 186         cpu = get_cpu();
 187         preempt_notifier_register(&vcpu->preempt_notifier);
 188         kvm_x86_ops->vcpu_load(vcpu, cpu);
 189         put_cpu();
 190 }
 191
 192 static void vcpu_put(struct kvm_vcpu *vcpu)
 193 {
 194         preempt_disable();
 195         kvm_x86_ops->vcpu_put(vcpu);
 196         preempt_notifier_unregister(&vcpu->preempt_notifier);
 197         preempt_enable();
 198         mutex_unlock(&vcpu->mutex);
 199 }
 200
 201 static void ack_flush(void *_completed)
 202 {
 203 }
 204
 205 void kvm_flush_remote_tlbs(struct kvm *kvm)
 206 {
 207         int i, cpu;
 208         cpumask_t cpus;
 209         struct kvm_vcpu *vcpu;
 210
 211         cpus_clear(cpus);
 212         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 213                 vcpu = kvm->vcpus[i];
 214                 if (!vcpu)
 215                         continue;
 216                 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
 217                         continue;
 218                 cpu = vcpu->cpu;
 219                 if (cpu != -1 && cpu != raw_smp_processor_id())
 220                         cpu_set(cpu, cpus);
 221         }
 222         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 223 }
 224
 225 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 226 {
 227         struct page *page;
 228         int r;
 229
 230         mutex_init(&vcpu->mutex);
 231         vcpu->cpu = -1;
 232         vcpu->mmu.root_hpa = INVALID_PAGE;
 233         vcpu->kvm = kvm;
 234         vcpu->vcpu_id = id;
 235         if (!irqchip_in_kernel(kvm) || id == 0)
 236                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
 237         else
 238                 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
 239         init_waitqueue_head(&vcpu->wq);
 240
 241         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 242         if (!page) {
 243                 r = -ENOMEM;
 244                 goto fail;
 245         }
 246         vcpu->run = page_address(page);
 247
 248         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 249         if (!page) {
 250                 r = -ENOMEM;
 251                 goto fail_free_run;
 252         }
 253         vcpu->pio_data = page_address(page);
 254
 255         r = kvm_mmu_create(vcpu);
 256         if (r < 0)
 257                 goto fail_free_pio_data;
 258
 259         if (irqchip_in_kernel(kvm)) {
 260                 r = kvm_create_lapic(vcpu);
 261                 if (r < 0)
 262                         goto fail_mmu_destroy;
 263         }
 264
 265         return 0;
 266
 267 fail_mmu_destroy:
 268         kvm_mmu_destroy(vcpu);
 269 fail_free_pio_data:
 270         free_page((unsigned long)vcpu->pio_data);
 271 fail_free_run:
 272         free_page((unsigned long)vcpu->run);
 273 fail:
 274         return r;
 275 }
 276 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 277
 278 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 279 {
 280         kvm_free_lapic(vcpu);
 281         kvm_mmu_destroy(vcpu);
 282         free_page((unsigned long)vcpu->pio_data);
 283         free_page((unsigned long)vcpu->run);
 284 }
 285 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 286
 287 static struct kvm *kvm_create_vm(void)
 288 {
 289         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 290
 291         if (!kvm)
 292                 return ERR_PTR(-ENOMEM);
 293
 294         kvm_io_bus_init(&kvm->pio_bus);
 295         mutex_init(&kvm->lock);
 296         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 297         kvm_io_bus_init(&kvm->mmio_bus);
 298         spin_lock(&kvm_lock);
 299         list_add(&kvm->vm_list, &vm_list);
 300         spin_unlock(&kvm_lock);
 301         return kvm;
 302 }
 303
 304 static void kvm_free_userspace_physmem(struct kvm_memory_slot *free)
 305 {
 306         int i;
 307
 308         for (i = 0; i < free->npages; ++i) {
 309                 if (free->phys_mem[i]) {
 310                         if (!PageReserved(free->phys_mem[i]))
 311                                 SetPageDirty(free->phys_mem[i]);
 312                         page_cache_release(free->phys_mem[i]);
 313                 }
 314         }
 315 }
 316
 317 static void kvm_free_kernel_physmem(struct kvm_memory_slot *free)
 318 {
 319         int i;
 320
 321         for (i = 0; i < free->npages; ++i)
 322                 if (free->phys_mem[i])
 323                         __free_page(free->phys_mem[i]);
 324 }
 325
 326 /*
 327  * Free any memory in @free but not in @dont.
 328  */
 329 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 330                                   struct kvm_memory_slot *dont)
 331 {
 332         if (!dont || free->phys_mem != dont->phys_mem)
 333                 if (free->phys_mem) {
 334                         if (free->user_alloc)
 335                                 kvm_free_userspace_physmem(free);
 336                         else
 337                                 kvm_free_kernel_physmem(free);
 338                         vfree(free->phys_mem);
 339                 }
 340         if (!dont || free->rmap != dont->rmap)
 341                 vfree(free->rmap);
 342
 343         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 344                 vfree(free->dirty_bitmap);
 345
 346         free->phys_mem = NULL;
 347         free->npages = 0;
 348         free->dirty_bitmap = NULL;
 349 }
 350
 351 static void kvm_free_physmem(struct kvm *kvm)
 352 {
 353         int i;
 354
 355         for (i = 0; i < kvm->nmemslots; ++i)
 356                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 357 }
 358
 359 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 360 {
 361         int i;
 362
 363         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
 364                 if (vcpu->pio.guest_pages[i]) {
 365                         __free_page(vcpu->pio.guest_pages[i]);
 366                         vcpu->pio.guest_pages[i] = NULL;
 367                 }
 368 }
 369
 370 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 371 {
 372         vcpu_load(vcpu);
 373         kvm_mmu_unload(vcpu);
 374         vcpu_put(vcpu);
 375 }
 376
 377 static void kvm_free_vcpus(struct kvm *kvm)
 378 {
 379         unsigned int i;
 380
 381         /*
 382          * Unpin any mmu pages first.
 383          */
 384         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 385                 if (kvm->vcpus[i])
 386                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
 387         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 388                 if (kvm->vcpus[i]) {
 389                         kvm_x86_ops->vcpu_free(kvm->vcpus[i]);
 390                         kvm->vcpus[i] = NULL;
 391                 }
 392         }
 393
 394 }
 395
 396 static void kvm_destroy_vm(struct kvm *kvm)
 397 {
 398         spin_lock(&kvm_lock);
 399         list_del(&kvm->vm_list);
 400         spin_unlock(&kvm_lock);
 401         kvm_io_bus_destroy(&kvm->pio_bus);
 402         kvm_io_bus_destroy(&kvm->mmio_bus);
 403         kfree(kvm->vpic);
 404         kfree(kvm->vioapic);
 405         kvm_free_vcpus(kvm);
 406         kvm_free_physmem(kvm);
 407         kfree(kvm);
 408 }
 409
 410 static int kvm_vm_release(struct inode *inode, struct file *filp)
 411 {
 412         struct kvm *kvm = filp->private_data;
 413
 414         kvm_destroy_vm(kvm);
 415         return 0;
 416 }
 417
 418 static void inject_gp(struct kvm_vcpu *vcpu)
 419 {
 420         kvm_x86_ops->inject_gp(vcpu, 0);
 421 }
 422
 423 /*
 424  * Load the pae pdptrs.  Return true is they are all valid.
 425  */
 426 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 427 {
 428         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 429         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 430         int i;
 431         int ret;
 432         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
 433
 434         mutex_lock(&vcpu->kvm->lock);
 435         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 436                                   offset * sizeof(u64), sizeof(pdpte));
 437         if (ret < 0) {
 438                 ret = 0;
 439                 goto out;
 440         }
 441         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 442                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 443                         ret = 0;
 444                         goto out;
 445                 }
 446         }
 447         ret = 1;
 448
 449         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
 450 out:
 451         mutex_unlock(&vcpu->kvm->lock);
 452
 453         return ret;
 454 }
 455
 456 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 457 {
 458         if (cr0 & CR0_RESERVED_BITS) {
 459                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 460                        cr0, vcpu->cr0);
 461                 inject_gp(vcpu);
 462                 return;
 463         }
 464
 465         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 466                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 467                 inject_gp(vcpu);
 468                 return;
 469         }
 470
 471         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 472                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 473                        "and a clear PE flag\n");
 474                 inject_gp(vcpu);
 475                 return;
 476         }
 477
 478         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 479 #ifdef CONFIG_X86_64
 480                 if ((vcpu->shadow_efer & EFER_LME)) {
 481                         int cs_db, cs_l;
 482
 483                         if (!is_pae(vcpu)) {
 484                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 485                                        "in long mode while PAE is disabled\n");
 486                                 inject_gp(vcpu);
 487                                 return;
 488                         }
 489                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 490                         if (cs_l) {
 491                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 492                                        "in long mode while CS.L == 1\n");
 493                                 inject_gp(vcpu);
 494                                 return;
 495
 496                         }
 497                 } else
 498 #endif
 499                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 500                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 501                                "reserved bits\n");
 502                         inject_gp(vcpu);
 503                         return;
 504                 }
 505
 506         }
 507
 508         kvm_x86_ops->set_cr0(vcpu, cr0);
 509         vcpu->cr0 = cr0;
 510
 511         mutex_lock(&vcpu->kvm->lock);
 512         kvm_mmu_reset_context(vcpu);
 513         mutex_unlock(&vcpu->kvm->lock);
 514         return;
 515 }
 516 EXPORT_SYMBOL_GPL(set_cr0);
 517
 518 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 519 {
 520         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 521 }
 522 EXPORT_SYMBOL_GPL(lmsw);
 523
 524 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 525 {
 526         if (cr4 & CR4_RESERVED_BITS) {
 527                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 528                 inject_gp(vcpu);
 529                 return;
 530         }
 531
 532         if (is_long_mode(vcpu)) {
 533                 if (!(cr4 & X86_CR4_PAE)) {
 534                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 535                                "in long mode\n");
 536                         inject_gp(vcpu);
 537                         return;
 538                 }
 539         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 540                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 541                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 542                 inject_gp(vcpu);
 543                 return;
 544         }
 545
 546         if (cr4 & X86_CR4_VMXE) {
 547                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 548                 inject_gp(vcpu);
 549                 return;
 550         }
 551         kvm_x86_ops->set_cr4(vcpu, cr4);
 552         vcpu->cr4 = cr4;
 553         mutex_lock(&vcpu->kvm->lock);
 554         kvm_mmu_reset_context(vcpu);
 555         mutex_unlock(&vcpu->kvm->lock);
 556 }
 557 EXPORT_SYMBOL_GPL(set_cr4);
 558
 559 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 560 {
 561         if (is_long_mode(vcpu)) {
 562                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 563                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 564                         inject_gp(vcpu);
 565                         return;
 566                 }
 567         } else {
 568                 if (is_pae(vcpu)) {
 569                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 570                                 printk(KERN_DEBUG
 571                                        "set_cr3: #GP, reserved bits\n");
 572                                 inject_gp(vcpu);
 573                                 return;
 574                         }
 575                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 576                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 577                                        "reserved bits\n");
 578                                 inject_gp(vcpu);
 579                                 return;
 580                         }
 581                 }
 582                 /*
 583                  * We don't check reserved bits in nonpae mode, because
 584                  * this isn't enforced, and VMware depends on this.
 585                  */
 586         }
 587
 588         mutex_lock(&vcpu->kvm->lock);
 589         /*
 590          * Does the new cr3 value map to physical memory? (Note, we
 591          * catch an invalid cr3 even in real-mode, because it would
 592          * cause trouble later on when we turn on paging anyway.)
 593          *
 594          * A real CPU would silently accept an invalid cr3 and would
 595          * attempt to use it - with largely undefined (and often hard
 596          * to debug) behavior on the guest side.
 597          */
 598         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 599                 inject_gp(vcpu);
 600         else {
 601                 vcpu->cr3 = cr3;
 602                 vcpu->mmu.new_cr3(vcpu);
 603         }
 604         mutex_unlock(&vcpu->kvm->lock);
 605 }
 606 EXPORT_SYMBOL_GPL(set_cr3);
 607
 608 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 609 {
 610         if (cr8 & CR8_RESERVED_BITS) {
 611                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 612                 inject_gp(vcpu);
 613                 return;
 614         }
 615         if (irqchip_in_kernel(vcpu->kvm))
 616                 kvm_lapic_set_tpr(vcpu, cr8);
 617         else
 618                 vcpu->cr8 = cr8;
 619 }
 620 EXPORT_SYMBOL_GPL(set_cr8);
 621
 622 unsigned long get_cr8(struct kvm_vcpu *vcpu)
 623 {
 624         if (irqchip_in_kernel(vcpu->kvm))
 625                 return kvm_lapic_get_cr8(vcpu);
 626         else
 627                 return vcpu->cr8;
 628 }
 629 EXPORT_SYMBOL_GPL(get_cr8);
 630
 631 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 632 {
 633         if (irqchip_in_kernel(vcpu->kvm))
 634                 return vcpu->apic_base;
 635         else
 636                 return vcpu->apic_base;
 637 }
 638 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 639
 640 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 641 {
 642         /* TODO: reserve bits check */
 643         if (irqchip_in_kernel(vcpu->kvm))
 644                 kvm_lapic_set_base(vcpu, data);
 645         else
 646                 vcpu->apic_base = data;
 647 }
 648 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 649
 650 void fx_init(struct kvm_vcpu *vcpu)
 651 {
 652         unsigned after_mxcsr_mask;
 653
 654         /* Initialize guest FPU by resetting ours and saving into guest's */
 655         preempt_disable();
 656         fx_save(&vcpu->host_fx_image);
 657         fpu_init();
 658         fx_save(&vcpu->guest_fx_image);
 659         fx_restore(&vcpu->host_fx_image);
 660         preempt_enable();
 661
 662         vcpu->cr0 |= X86_CR0_ET;
 663         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
 664         vcpu->guest_fx_image.mxcsr = 0x1f80;
 665         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
 666                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
 667 }
 668 EXPORT_SYMBOL_GPL(fx_init);
 669
 670 /*
 671  * Allocate some memory and give it an address in the guest physical address
 672  * space.
 673  *
 674  * Discontiguous memory is allowed, mostly for framebuffers.
 675  */
 676 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 677                                           struct
 678                                           kvm_userspace_memory_region *mem,
 679                                           int user_alloc)
 680 {
 681         int r;
 682         gfn_t base_gfn;
 683         unsigned long npages;
 684         unsigned long i;
 685         struct kvm_memory_slot *memslot;
 686         struct kvm_memory_slot old, new;
 687
 688         r = -EINVAL;
 689         /* General sanity checks */
 690         if (mem->memory_size & (PAGE_SIZE - 1))
 691                 goto out;
 692         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 693                 goto out;
 694         if (mem->slot >= KVM_MEMORY_SLOTS)
 695                 goto out;
 696         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 697                 goto out;
 698
 699         memslot = &kvm->memslots[mem->slot];
 700         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 701         npages = mem->memory_size >> PAGE_SHIFT;
 702
 703         if (!npages)
 704                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 705
 706         mutex_lock(&kvm->lock);
 707
 708         new = old = *memslot;
 709
 710         new.base_gfn = base_gfn;
 711         new.npages = npages;
 712         new.flags = mem->flags;
 713
 714         /* Disallow changing a memory slot's size. */
 715         r = -EINVAL;
 716         if (npages && old.npages && npages != old.npages)
 717                 goto out_unlock;
 718
 719         /* Check for overlaps */
 720         r = -EEXIST;
 721         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 722                 struct kvm_memory_slot *s = &kvm->memslots[i];
 723
 724                 if (s == memslot)
 725                         continue;
 726                 if (!((base_gfn + npages <= s->base_gfn) ||
 727                       (base_gfn >= s->base_gfn + s->npages)))
 728                         goto out_unlock;
 729         }
 730
 731         /* Deallocate if slot is being removed */
 732         if (!npages)
 733                 new.phys_mem = NULL;
 734
 735         /* Free page dirty bitmap if unneeded */
 736         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 737                 new.dirty_bitmap = NULL;
 738
 739         r = -ENOMEM;
 740
 741         /* Allocate if a slot is being created */
 742         if (npages && !new.phys_mem) {
 743                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 744
 745                 if (!new.phys_mem)
 746                         goto out_unlock;
 747
 748                 new.rmap = vmalloc(npages * sizeof(struct page *));
 749
 750                 if (!new.rmap)
 751                         goto out_unlock;
 752
 753                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 754                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 755                 if (user_alloc) {
 756                         unsigned long pages_num;
 757
 758                         new.user_alloc = 1;
 759                         down_read(&current->mm->mmap_sem);
 760
 761                         pages_num = get_user_pages(current, current->mm,
 762                                                    mem->userspace_addr,
 763                                                    npages, 1, 1, new.phys_mem,
 764                                                    NULL);
 765
 766                         up_read(&current->mm->mmap_sem);
 767                         if (pages_num != npages)
 768                                 goto out_unlock;
 769                 } else {
 770                         for (i = 0; i < npages; ++i) {
 771                                 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 772                                                              | __GFP_ZERO);
 773                                 if (!new.phys_mem[i])
 774                                         goto out_unlock;
 775                         }
 776                 }
 777         }
 778
 779         /* Allocate page dirty bitmap if needed */
 780         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 781                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 782
 783                 new.dirty_bitmap = vmalloc(dirty_bytes);
 784                 if (!new.dirty_bitmap)
 785                         goto out_unlock;
 786                 memset(new.dirty_bitmap, 0, dirty_bytes);
 787         }
 788
 789         if (mem->slot >= kvm->nmemslots)
 790                 kvm->nmemslots = mem->slot + 1;
 791
 792         if (!kvm->n_requested_mmu_pages) {
 793                 unsigned int n_pages;
 794
 795                 if (npages) {
 796                         n_pages = npages * KVM_PERMILLE_MMU_PAGES / 1000;
 797                         kvm_mmu_change_mmu_pages(kvm, kvm->n_alloc_mmu_pages +
 798                                                  n_pages);
 799                 } else {
 800                         unsigned int nr_mmu_pages;
 801
 802                         n_pages = old.npages * KVM_PERMILLE_MMU_PAGES / 1000;
 803                         nr_mmu_pages = kvm->n_alloc_mmu_pages - n_pages;
 804                         nr_mmu_pages = max(nr_mmu_pages,
 805                                         (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
 806                         kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
 807                 }
 808         }
 809
 810         *memslot = new;
 811
 812         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 813         kvm_flush_remote_tlbs(kvm);
 814
 815         mutex_unlock(&kvm->lock);
 816
 817         kvm_free_physmem_slot(&old, &new);
 818         return 0;
 819
 820 out_unlock:
 821         mutex_unlock(&kvm->lock);
 822         kvm_free_physmem_slot(&new, &old);
 823 out:
 824         return r;
 825 }
 826
 827 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
 828                                           u32 kvm_nr_mmu_pages)
 829 {
 830         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
 831                 return -EINVAL;
 832
 833         mutex_lock(&kvm->lock);
 834
 835         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
 836         kvm->n_requested_mmu_pages = kvm_nr_mmu_pages;
 837
 838         mutex_unlock(&kvm->lock);
 839         return 0;
 840 }
 841
 842 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
 843 {
 844         return kvm->n_alloc_mmu_pages;
 845 }
 846
 847 /*
 848  * Get (and clear) the dirty memory log for a memory slot.
 849  */
 850 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 851                                       struct kvm_dirty_log *log)
 852 {
 853         struct kvm_memory_slot *memslot;
 854         int r, i;
 855         int n;
 856         unsigned long any = 0;
 857
 858         mutex_lock(&kvm->lock);
 859
 860         r = -EINVAL;
 861         if (log->slot >= KVM_MEMORY_SLOTS)
 862                 goto out;
 863
 864         memslot = &kvm->memslots[log->slot];
 865         r = -ENOENT;
 866         if (!memslot->dirty_bitmap)
 867                 goto out;
 868
 869         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 870
 871         for (i = 0; !any && i < n/sizeof(long); ++i)
 872                 any = memslot->dirty_bitmap[i];
 873
 874         r = -EFAULT;
 875         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 876                 goto out;
 877
 878         /* If nothing is dirty, don't bother messing with page tables. */
 879         if (any) {
 880                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 881                 kvm_flush_remote_tlbs(kvm);
 882                 memset(memslot->dirty_bitmap, 0, n);
 883         }
 884
 885         r = 0;
 886
 887 out:
 888         mutex_unlock(&kvm->lock);
 889         return r;
 890 }
 891
 892 /*
 893  * Set a new alias region.  Aliases map a portion of physical memory into
 894  * another portion.  This is useful for memory windows, for example the PC
 895  * VGA region.
 896  */
 897 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 898                                          struct kvm_memory_alias *alias)
 899 {
 900         int r, n;
 901         struct kvm_mem_alias *p;
 902
 903         r = -EINVAL;
 904         /* General sanity checks */
 905         if (alias->memory_size & (PAGE_SIZE - 1))
 906                 goto out;
 907         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 908                 goto out;
 909         if (alias->slot >= KVM_ALIAS_SLOTS)
 910                 goto out;
 911         if (alias->guest_phys_addr + alias->memory_size
 912             < alias->guest_phys_addr)
 913                 goto out;
 914         if (alias->target_phys_addr + alias->memory_size
 915             < alias->target_phys_addr)
 916                 goto out;
 917
 918         mutex_lock(&kvm->lock);
 919
 920         p = &kvm->aliases[alias->slot];
 921         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 922         p->npages = alias->memory_size >> PAGE_SHIFT;
 923         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 924
 925         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 926                 if (kvm->aliases[n - 1].npages)
 927                         break;
 928         kvm->naliases = n;
 929
 930         kvm_mmu_zap_all(kvm);
 931
 932         mutex_unlock(&kvm->lock);
 933
 934         return 0;
 935
 936 out:
 937         return r;
 938 }
 939
 940 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 941 {
 942         int r;
 943
 944         r = 0;
 945         switch (chip->chip_id) {
 946         case KVM_IRQCHIP_PIC_MASTER:
 947                 memcpy(&chip->chip.pic,
 948                         &pic_irqchip(kvm)->pics[0],
 949                         sizeof(struct kvm_pic_state));
 950                 break;
 951         case KVM_IRQCHIP_PIC_SLAVE:
 952                 memcpy(&chip->chip.pic,
 953                         &pic_irqchip(kvm)->pics[1],
 954                         sizeof(struct kvm_pic_state));
 955                 break;
 956         case KVM_IRQCHIP_IOAPIC:
 957                 memcpy(&chip->chip.ioapic,
 958                         ioapic_irqchip(kvm),
 959                         sizeof(struct kvm_ioapic_state));
 960                 break;
 961         default:
 962                 r = -EINVAL;
 963                 break;
 964         }
 965         return r;
 966 }
 967
 968 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 969 {
 970         int r;
 971
 972         r = 0;
 973         switch (chip->chip_id) {
 974         case KVM_IRQCHIP_PIC_MASTER:
 975                 memcpy(&pic_irqchip(kvm)->pics[0],
 976                         &chip->chip.pic,
 977                         sizeof(struct kvm_pic_state));
 978                 break;
 979         case KVM_IRQCHIP_PIC_SLAVE:
 980                 memcpy(&pic_irqchip(kvm)->pics[1],
 981                         &chip->chip.pic,
 982                         sizeof(struct kvm_pic_state));
 983                 break;
 984         case KVM_IRQCHIP_IOAPIC:
 985                 memcpy(ioapic_irqchip(kvm),
 986                         &chip->chip.ioapic,
 987                         sizeof(struct kvm_ioapic_state));
 988                 break;
 989         default:
 990                 r = -EINVAL;
 991                 break;
 992         }
 993         kvm_pic_update_irq(pic_irqchip(kvm));
 994         return r;
 995 }
 996
 997 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 998 {
 999         int i;
1000         struct kvm_mem_alias *alias;
1001
1002         for (i = 0; i < kvm->naliases; ++i) {
1003                 alias = &kvm->aliases[i];
1004                 if (gfn >= alias->base_gfn
1005                     && gfn < alias->base_gfn + alias->npages)
1006                         return alias->target_gfn + gfn - alias->base_gfn;
1007         }
1008         return gfn;
1009 }
1010
1011 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1012 {
1013         int i;
1014
1015         for (i = 0; i < kvm->nmemslots; ++i) {
1016                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1017
1018                 if (gfn >= memslot->base_gfn
1019                     && gfn < memslot->base_gfn + memslot->npages)
1020                         return memslot;
1021         }
1022         return NULL;
1023 }
1024
1025 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1026 {
1027         gfn = unalias_gfn(kvm, gfn);
1028         return __gfn_to_memslot(kvm, gfn);
1029 }
1030
1031 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1032 {
1033         struct kvm_memory_slot *slot;
1034
1035         gfn = unalias_gfn(kvm, gfn);
1036         slot = __gfn_to_memslot(kvm, gfn);
1037         if (!slot)
1038                 return NULL;
1039         return slot->phys_mem[gfn - slot->base_gfn];
1040 }
1041 EXPORT_SYMBOL_GPL(gfn_to_page);
1042
1043 static int next_segment(unsigned long len, int offset)
1044 {
1045         if (len > PAGE_SIZE - offset)
1046                 return PAGE_SIZE - offset;
1047         else
1048                 return len;
1049 }
1050
1051 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1052                         int len)
1053 {
1054         void *page_virt;
1055         struct page *page;
1056
1057         page = gfn_to_page(kvm, gfn);
1058         if (!page)
1059                 return -EFAULT;
1060         page_virt = kmap_atomic(page, KM_USER0);
1061
1062         memcpy(data, page_virt + offset, len);
1063
1064         kunmap_atomic(page_virt, KM_USER0);
1065         return 0;
1066 }
1067 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1068
1069 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, 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_read_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 EXPORT_SYMBOL_GPL(kvm_read_guest);
1088
1089 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1090                          int offset, int len)
1091 {
1092         void *page_virt;
1093         struct page *page;
1094
1095         page = gfn_to_page(kvm, gfn);
1096         if (!page)
1097                 return -EFAULT;
1098         page_virt = kmap_atomic(page, KM_USER0);
1099
1100         memcpy(page_virt + offset, data, len);
1101
1102         kunmap_atomic(page_virt, KM_USER0);
1103         mark_page_dirty(kvm, gfn);
1104         return 0;
1105 }
1106 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1107
1108 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1109                     unsigned long len)
1110 {
1111         gfn_t gfn = gpa >> PAGE_SHIFT;
1112         int seg;
1113         int offset = offset_in_page(gpa);
1114         int ret;
1115
1116         while ((seg = next_segment(len, offset)) != 0) {
1117                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1118                 if (ret < 0)
1119                         return ret;
1120                 offset = 0;
1121                 len -= seg;
1122                 data += seg;
1123                 ++gfn;
1124         }
1125         return 0;
1126 }
1127
1128 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1129 {
1130         void *page_virt;
1131         struct page *page;
1132
1133         page = gfn_to_page(kvm, gfn);
1134         if (!page)
1135                 return -EFAULT;
1136         page_virt = kmap_atomic(page, KM_USER0);
1137
1138         memset(page_virt + offset, 0, len);
1139
1140         kunmap_atomic(page_virt, KM_USER0);
1141         return 0;
1142 }
1143 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1144
1145 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1146 {
1147         gfn_t gfn = gpa >> PAGE_SHIFT;
1148         int seg;
1149         int offset = offset_in_page(gpa);
1150         int ret;
1151
1152         while ((seg = next_segment(len, offset)) != 0) {
1153                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1154                 if (ret < 0)
1155                         return ret;
1156                 offset = 0;
1157                 len -= seg;
1158                 ++gfn;
1159         }
1160         return 0;
1161 }
1162 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1163
1164 /* WARNING: Does not work on aliased pages. */
1165 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1166 {
1167         struct kvm_memory_slot *memslot;
1168
1169         memslot = __gfn_to_memslot(kvm, gfn);
1170         if (memslot && memslot->dirty_bitmap) {
1171                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1172
1173                 /* avoid RMW */
1174                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1175                         set_bit(rel_gfn, memslot->dirty_bitmap);
1176         }
1177 }
1178
1179 int emulator_read_std(unsigned long addr,
1180                              void *val,
1181                              unsigned int bytes,
1182                              struct kvm_vcpu *vcpu)
1183 {
1184         void *data = val;
1185
1186         while (bytes) {
1187                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1188                 unsigned offset = addr & (PAGE_SIZE-1);
1189                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1190                 int ret;
1191
1192                 if (gpa == UNMAPPED_GVA)
1193                         return X86EMUL_PROPAGATE_FAULT;
1194                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1195                 if (ret < 0)
1196                         return X86EMUL_UNHANDLEABLE;
1197
1198                 bytes -= tocopy;
1199                 data += tocopy;
1200                 addr += tocopy;
1201         }
1202
1203         return X86EMUL_CONTINUE;
1204 }
1205 EXPORT_SYMBOL_GPL(emulator_read_std);
1206
1207 static int emulator_write_std(unsigned long addr,
1208                               const void *val,
1209                               unsigned int bytes,
1210                               struct kvm_vcpu *vcpu)
1211 {
1212         pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1213         return X86EMUL_UNHANDLEABLE;
1214 }
1215
1216 /*
1217  * Only apic need an MMIO device hook, so shortcut now..
1218  */
1219 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1220                                                 gpa_t addr)
1221 {
1222         struct kvm_io_device *dev;
1223
1224         if (vcpu->apic) {
1225                 dev = &vcpu->apic->dev;
1226                 if (dev->in_range(dev, addr))
1227                         return dev;
1228         }
1229         return NULL;
1230 }
1231
1232 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1233                                                 gpa_t addr)
1234 {
1235         struct kvm_io_device *dev;
1236
1237         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1238         if (dev == NULL)
1239                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1240         return dev;
1241 }
1242
1243 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1244                                                gpa_t addr)
1245 {
1246         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1247 }
1248
1249 static int emulator_read_emulated(unsigned long addr,
1250                                   void *val,
1251                                   unsigned int bytes,
1252                                   struct kvm_vcpu *vcpu)
1253 {
1254         struct kvm_io_device *mmio_dev;
1255         gpa_t                 gpa;
1256
1257         if (vcpu->mmio_read_completed) {
1258                 memcpy(val, vcpu->mmio_data, bytes);
1259                 vcpu->mmio_read_completed = 0;
1260                 return X86EMUL_CONTINUE;
1261         } else if (emulator_read_std(addr, val, bytes, vcpu)
1262                    == X86EMUL_CONTINUE)
1263                 return X86EMUL_CONTINUE;
1264
1265         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1266         if (gpa == UNMAPPED_GVA)
1267                 return X86EMUL_PROPAGATE_FAULT;
1268
1269         /*
1270          * Is this MMIO handled locally?
1271          */
1272         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1273         if (mmio_dev) {
1274                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1275                 return X86EMUL_CONTINUE;
1276         }
1277
1278         vcpu->mmio_needed = 1;
1279         vcpu->mmio_phys_addr = gpa;
1280         vcpu->mmio_size = bytes;
1281         vcpu->mmio_is_write = 0;
1282
1283         return X86EMUL_UNHANDLEABLE;
1284 }
1285
1286 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1287                                const void *val, int bytes)
1288 {
1289         int ret;
1290
1291         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1292         if (ret < 0)
1293                 return 0;
1294         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1295         return 1;
1296 }
1297
1298 static int emulator_write_emulated_onepage(unsigned long addr,
1299                                            const void *val,
1300                                            unsigned int bytes,
1301                                            struct kvm_vcpu *vcpu)
1302 {
1303         struct kvm_io_device *mmio_dev;
1304         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1305
1306         if (gpa == UNMAPPED_GVA) {
1307                 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1308                 return X86EMUL_PROPAGATE_FAULT;
1309         }
1310
1311         if (emulator_write_phys(vcpu, gpa, val, bytes))
1312                 return X86EMUL_CONTINUE;
1313
1314         /*
1315          * Is this MMIO handled locally?
1316          */
1317         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1318         if (mmio_dev) {
1319                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1320                 return X86EMUL_CONTINUE;
1321         }
1322
1323         vcpu->mmio_needed = 1;
1324         vcpu->mmio_phys_addr = gpa;
1325         vcpu->mmio_size = bytes;
1326         vcpu->mmio_is_write = 1;
1327         memcpy(vcpu->mmio_data, val, bytes);
1328
1329         return X86EMUL_CONTINUE;
1330 }
1331
1332 int emulator_write_emulated(unsigned long addr,
1333                                    const void *val,
1334                                    unsigned int bytes,
1335                                    struct kvm_vcpu *vcpu)
1336 {
1337         /* Crossing a page boundary? */
1338         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1339                 int rc, now;
1340
1341                 now = -addr & ~PAGE_MASK;
1342                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1343                 if (rc != X86EMUL_CONTINUE)
1344                         return rc;
1345                 addr += now;
1346                 val += now;
1347                 bytes -= now;
1348         }
1349         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1350 }
1351 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1352
1353 static int emulator_cmpxchg_emulated(unsigned long addr,
1354                                      const void *old,
1355                                      const void *new,
1356                                      unsigned int bytes,
1357                                      struct kvm_vcpu *vcpu)
1358 {
1359         static int reported;
1360
1361         if (!reported) {
1362                 reported = 1;
1363                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1364         }
1365         return emulator_write_emulated(addr, new, bytes, vcpu);
1366 }
1367
1368 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1369 {
1370         return kvm_x86_ops->get_segment_base(vcpu, seg);
1371 }
1372
1373 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1374 {
1375         return X86EMUL_CONTINUE;
1376 }
1377
1378 int emulate_clts(struct kvm_vcpu *vcpu)
1379 {
1380         kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1381         return X86EMUL_CONTINUE;
1382 }
1383
1384 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1385 {
1386         struct kvm_vcpu *vcpu = ctxt->vcpu;
1387
1388         switch (dr) {
1389         case 0 ... 3:
1390                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1391                 return X86EMUL_CONTINUE;
1392         default:
1393                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1394                 return X86EMUL_UNHANDLEABLE;
1395         }
1396 }
1397
1398 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1399 {
1400         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1401         int exception;
1402
1403         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1404         if (exception) {
1405                 /* FIXME: better handling */
1406                 return X86EMUL_UNHANDLEABLE;
1407         }
1408         return X86EMUL_CONTINUE;
1409 }
1410
1411 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1412 {
1413         static int reported;
1414         u8 opcodes[4];
1415         unsigned long rip = vcpu->rip;
1416         unsigned long rip_linear;
1417
1418         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1419
1420         if (reported)
1421                 return;
1422
1423         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1424
1425         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1426                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1427         reported = 1;
1428 }
1429 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1430
1431 struct x86_emulate_ops emulate_ops = {
1432         .read_std            = emulator_read_std,
1433         .write_std           = emulator_write_std,
1434         .read_emulated       = emulator_read_emulated,
1435         .write_emulated      = emulator_write_emulated,
1436         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1437 };
1438
1439 int emulate_instruction(struct kvm_vcpu *vcpu,
1440                         struct kvm_run *run,
1441                         unsigned long cr2,
1442                         u16 error_code,
1443                         int no_decode)
1444 {
1445         int r;
1446
1447         vcpu->mmio_fault_cr2 = cr2;
1448         kvm_x86_ops->cache_regs(vcpu);
1449
1450         vcpu->mmio_is_write = 0;
1451         vcpu->pio.string = 0;
1452
1453         if (!no_decode) {
1454                 int cs_db, cs_l;
1455                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1456
1457                 vcpu->emulate_ctxt.vcpu = vcpu;
1458                 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1459                 vcpu->emulate_ctxt.cr2 = cr2;
1460                 vcpu->emulate_ctxt.mode =
1461                         (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1462                         ? X86EMUL_MODE_REAL : cs_l
1463                         ? X86EMUL_MODE_PROT64 : cs_db
1464                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1465
1466                 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1467                         vcpu->emulate_ctxt.cs_base = 0;
1468                         vcpu->emulate_ctxt.ds_base = 0;
1469                         vcpu->emulate_ctxt.es_base = 0;
1470                         vcpu->emulate_ctxt.ss_base = 0;
1471                 } else {
1472                         vcpu->emulate_ctxt.cs_base =
1473                                         get_segment_base(vcpu, VCPU_SREG_CS);
1474                         vcpu->emulate_ctxt.ds_base =
1475                                         get_segment_base(vcpu, VCPU_SREG_DS);
1476                         vcpu->emulate_ctxt.es_base =
1477                                         get_segment_base(vcpu, VCPU_SREG_ES);
1478                         vcpu->emulate_ctxt.ss_base =
1479                                         get_segment_base(vcpu, VCPU_SREG_SS);
1480                 }
1481
1482                 vcpu->emulate_ctxt.gs_base =
1483                                         get_segment_base(vcpu, VCPU_SREG_GS);
1484                 vcpu->emulate_ctxt.fs_base =
1485                                         get_segment_base(vcpu, VCPU_SREG_FS);
1486
1487                 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
1488                 if (r)  {
1489                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1490                                 return EMULATE_DONE;
1491                         return EMULATE_FAIL;
1492                 }
1493         }
1494
1495         r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1496
1497         if (vcpu->pio.string)
1498                 return EMULATE_DO_MMIO;
1499
1500         if ((r || vcpu->mmio_is_write) && run) {
1501                 run->exit_reason = KVM_EXIT_MMIO;
1502                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1503                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1504                 run->mmio.len = vcpu->mmio_size;
1505                 run->mmio.is_write = vcpu->mmio_is_write;
1506         }
1507
1508         if (r) {
1509                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1510                         return EMULATE_DONE;
1511                 if (!vcpu->mmio_needed) {
1512                         kvm_report_emulation_failure(vcpu, "mmio");
1513                         return EMULATE_FAIL;
1514                 }
1515                 return EMULATE_DO_MMIO;
1516         }
1517
1518         kvm_x86_ops->decache_regs(vcpu);
1519         kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1520
1521         if (vcpu->mmio_is_write) {
1522                 vcpu->mmio_needed = 0;
1523                 return EMULATE_DO_MMIO;
1524         }
1525
1526         return EMULATE_DONE;
1527 }
1528 EXPORT_SYMBOL_GPL(emulate_instruction);
1529
1530 /*
1531  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1532  */
1533 static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1534 {
1535         DECLARE_WAITQUEUE(wait, current);
1536
1537         add_wait_queue(&vcpu->wq, &wait);
1538
1539         /*
1540          * We will block until either an interrupt or a signal wakes us up
1541          */
1542         while (!kvm_cpu_has_interrupt(vcpu)
1543                && !signal_pending(current)
1544                && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
1545                && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
1546                 set_current_state(TASK_INTERRUPTIBLE);
1547                 vcpu_put(vcpu);
1548                 schedule();
1549                 vcpu_load(vcpu);
1550         }
1551
1552         __set_current_state(TASK_RUNNING);
1553         remove_wait_queue(&vcpu->wq, &wait);
1554 }
1555
1556 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1557 {
1558         ++vcpu->stat.halt_exits;
1559         if (irqchip_in_kernel(vcpu->kvm)) {
1560                 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1561                 kvm_vcpu_block(vcpu);
1562                 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1563                         return -EINTR;
1564                 return 1;
1565         } else {
1566                 vcpu->run->exit_reason = KVM_EXIT_HLT;
1567                 return 0;
1568         }
1569 }
1570 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1571
1572 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
1573 {
1574         unsigned long nr, a0, a1, a2, a3, ret;
1575
1576         kvm_x86_ops->cache_regs(vcpu);
1577
1578         nr = vcpu->regs[VCPU_REGS_RAX];
1579         a0 = vcpu->regs[VCPU_REGS_RBX];
1580         a1 = vcpu->regs[VCPU_REGS_RCX];
1581         a2 = vcpu->regs[VCPU_REGS_RDX];
1582         a3 = vcpu->regs[VCPU_REGS_RSI];
1583
1584         if (!is_long_mode(vcpu)) {
1585                 nr &= 0xFFFFFFFF;
1586                 a0 &= 0xFFFFFFFF;
1587                 a1 &= 0xFFFFFFFF;
1588                 a2 &= 0xFFFFFFFF;
1589                 a3 &= 0xFFFFFFFF;
1590         }
1591
1592         switch (nr) {
1593         default:
1594                 ret = -KVM_ENOSYS;
1595                 break;
1596         }
1597         vcpu->regs[VCPU_REGS_RAX] = ret;
1598         kvm_x86_ops->decache_regs(vcpu);
1599         return 0;
1600 }
1601 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1602
1603 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1604 {
1605         char instruction[3];
1606         int ret = 0;
1607
1608         mutex_lock(&vcpu->kvm->lock);
1609
1610         /*
1611          * Blow out the MMU to ensure that no other VCPU has an active mapping
1612          * to ensure that the updated hypercall appears atomically across all
1613          * VCPUs.
1614          */
1615         kvm_mmu_zap_all(vcpu->kvm);
1616
1617         kvm_x86_ops->cache_regs(vcpu);
1618         kvm_x86_ops->patch_hypercall(vcpu, instruction);
1619         if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1620             != X86EMUL_CONTINUE)
1621                 ret = -EFAULT;
1622
1623         mutex_unlock(&vcpu->kvm->lock);
1624
1625         return ret;
1626 }
1627
1628 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1629 {
1630         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1631 }
1632
1633 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1634 {
1635         struct descriptor_table dt = { limit, base };
1636
1637         kvm_x86_ops->set_gdt(vcpu, &dt);
1638 }
1639
1640 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1641 {
1642         struct descriptor_table dt = { limit, base };
1643
1644         kvm_x86_ops->set_idt(vcpu, &dt);
1645 }
1646
1647 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1648                    unsigned long *rflags)
1649 {
1650         lmsw(vcpu, msw);
1651         *rflags = kvm_x86_ops->get_rflags(vcpu);
1652 }
1653
1654 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1655 {
1656         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1657         switch (cr) {
1658         case 0:
1659                 return vcpu->cr0;
1660         case 2:
1661                 return vcpu->cr2;
1662         case 3:
1663                 return vcpu->cr3;
1664         case 4:
1665                 return vcpu->cr4;
1666         default:
1667                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1668                 return 0;
1669         }
1670 }
1671
1672 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1673                      unsigned long *rflags)
1674 {
1675         switch (cr) {
1676         case 0:
1677                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1678                 *rflags = kvm_x86_ops->get_rflags(vcpu);
1679                 break;
1680         case 2:
1681                 vcpu->cr2 = val;
1682                 break;
1683         case 3:
1684                 set_cr3(vcpu, val);
1685                 break;
1686         case 4:
1687                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1688                 break;
1689         default:
1690                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1691         }
1692 }
1693
1694 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1695 {
1696         u64 data;
1697
1698         switch (msr) {
1699         case 0xc0010010: /* SYSCFG */
1700         case 0xc0010015: /* HWCR */
1701         case MSR_IA32_PLATFORM_ID:
1702         case MSR_IA32_P5_MC_ADDR:
1703         case MSR_IA32_P5_MC_TYPE:
1704         case MSR_IA32_MC0_CTL:
1705         case MSR_IA32_MCG_STATUS:
1706         case MSR_IA32_MCG_CAP:
1707         case MSR_IA32_MC0_MISC:
1708         case MSR_IA32_MC0_MISC+4:
1709         case MSR_IA32_MC0_MISC+8:
1710         case MSR_IA32_MC0_MISC+12:
1711         case MSR_IA32_MC0_MISC+16:
1712         case MSR_IA32_UCODE_REV:
1713         case MSR_IA32_PERF_STATUS:
1714         case MSR_IA32_EBL_CR_POWERON:
1715                 /* MTRR registers */
1716         case 0xfe:
1717         case 0x200 ... 0x2ff:
1718                 data = 0;
1719                 break;
1720         case 0xcd: /* fsb frequency */
1721                 data = 3;
1722                 break;
1723         case MSR_IA32_APICBASE:
1724                 data = kvm_get_apic_base(vcpu);
1725                 break;
1726         case MSR_IA32_MISC_ENABLE:
1727                 data = vcpu->ia32_misc_enable_msr;
1728                 break;
1729 #ifdef CONFIG_X86_64
1730         case MSR_EFER:
1731                 data = vcpu->shadow_efer;
1732                 break;
1733 #endif
1734         default:
1735                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1736                 return 1;
1737         }
1738         *pdata = data;
1739         return 0;
1740 }
1741 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1742
1743 /*
1744  * Reads an msr value (of 'msr_index') into 'pdata'.
1745  * Returns 0 on success, non-0 otherwise.
1746  * Assumes vcpu_load() was already called.
1747  */
1748 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1749 {
1750         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1751 }
1752
1753 #ifdef CONFIG_X86_64
1754
1755 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1756 {
1757         if (efer & EFER_RESERVED_BITS) {
1758                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1759                        efer);
1760                 inject_gp(vcpu);
1761                 return;
1762         }
1763
1764         if (is_paging(vcpu)
1765             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1766                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1767                 inject_gp(vcpu);
1768                 return;
1769         }
1770
1771         kvm_x86_ops->set_efer(vcpu, efer);
1772
1773         efer &= ~EFER_LMA;
1774         efer |= vcpu->shadow_efer & EFER_LMA;
1775
1776         vcpu->shadow_efer = efer;
1777 }
1778
1779 #endif
1780
1781 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1782 {
1783         switch (msr) {
1784 #ifdef CONFIG_X86_64
1785         case MSR_EFER:
1786                 set_efer(vcpu, data);
1787                 break;
1788 #endif
1789         case MSR_IA32_MC0_STATUS:
1790                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1791                        __FUNCTION__, data);
1792                 break;
1793         case MSR_IA32_MCG_STATUS:
1794                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1795                         __FUNCTION__, data);
1796                 break;
1797         case MSR_IA32_UCODE_REV:
1798         case MSR_IA32_UCODE_WRITE:
1799         case 0x200 ... 0x2ff: /* MTRRs */
1800                 break;
1801         case MSR_IA32_APICBASE:
1802                 kvm_set_apic_base(vcpu, data);
1803                 break;
1804         case MSR_IA32_MISC_ENABLE:
1805                 vcpu->ia32_misc_enable_msr = data;
1806                 break;
1807         default:
1808                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1809                 return 1;
1810         }
1811         return 0;
1812 }
1813 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1814
1815 /*
1816  * Writes msr value into into the appropriate "register".
1817  * Returns 0 on success, non-0 otherwise.
1818  * Assumes vcpu_load() was already called.
1819  */
1820 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1821 {
1822         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
1823 }
1824
1825 void kvm_resched(struct kvm_vcpu *vcpu)
1826 {
1827         if (!need_resched())
1828                 return;
1829         cond_resched();
1830 }
1831 EXPORT_SYMBOL_GPL(kvm_resched);
1832
1833 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1834 {
1835         int i;
1836         u32 function;
1837         struct kvm_cpuid_entry *e, *best;
1838
1839         kvm_x86_ops->cache_regs(vcpu);
1840         function = vcpu->regs[VCPU_REGS_RAX];
1841         vcpu->regs[VCPU_REGS_RAX] = 0;
1842         vcpu->regs[VCPU_REGS_RBX] = 0;
1843         vcpu->regs[VCPU_REGS_RCX] = 0;
1844         vcpu->regs[VCPU_REGS_RDX] = 0;
1845         best = NULL;
1846         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1847                 e = &vcpu->cpuid_entries[i];
1848                 if (e->function == function) {
1849                         best = e;
1850                         break;
1851                 }
1852                 /*
1853                  * Both basic or both extended?
1854                  */
1855                 if (((e->function ^ function) & 0x80000000) == 0)
1856                         if (!best || e->function > best->function)
1857                                 best = e;
1858         }
1859         if (best) {
1860                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1861                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1862                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1863                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1864         }
1865         kvm_x86_ops->decache_regs(vcpu);
1866         kvm_x86_ops->skip_emulated_instruction(vcpu);
1867 }
1868 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1869
1870 static int pio_copy_data(struct kvm_vcpu *vcpu)
1871 {
1872         void *p = vcpu->pio_data;
1873         void *q;
1874         unsigned bytes;
1875         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1876
1877         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1878                  PAGE_KERNEL);
1879         if (!q) {
1880                 free_pio_guest_pages(vcpu);
1881                 return -ENOMEM;
1882         }
1883         q += vcpu->pio.guest_page_offset;
1884         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1885         if (vcpu->pio.in)
1886                 memcpy(q, p, bytes);
1887         else
1888                 memcpy(p, q, bytes);
1889         q -= vcpu->pio.guest_page_offset;
1890         vunmap(q);
1891         free_pio_guest_pages(vcpu);
1892         return 0;
1893 }
1894
1895 static int complete_pio(struct kvm_vcpu *vcpu)
1896 {
1897         struct kvm_pio_request *io = &vcpu->pio;
1898         long delta;
1899         int r;
1900
1901         kvm_x86_ops->cache_regs(vcpu);
1902
1903         if (!io->string) {
1904                 if (io->in)
1905                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1906                                io->size);
1907         } else {
1908                 if (io->in) {
1909                         r = pio_copy_data(vcpu);
1910                         if (r) {
1911                                 kvm_x86_ops->cache_regs(vcpu);
1912                                 return r;
1913                         }
1914                 }
1915
1916                 delta = 1;
1917                 if (io->rep) {
1918                         delta *= io->cur_count;
1919                         /*
1920                          * The size of the register should really depend on
1921                          * current address size.
1922                          */
1923                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1924                 }
1925                 if (io->down)
1926                         delta = -delta;
1927                 delta *= io->size;
1928                 if (io->in)
1929                         vcpu->regs[VCPU_REGS_RDI] += delta;
1930                 else
1931                         vcpu->regs[VCPU_REGS_RSI] += delta;
1932         }
1933
1934         kvm_x86_ops->decache_regs(vcpu);
1935
1936         io->count -= io->cur_count;
1937         io->cur_count = 0;
1938
1939         return 0;
1940 }
1941
1942 static void kernel_pio(struct kvm_io_device *pio_dev,
1943                        struct kvm_vcpu *vcpu,
1944                        void *pd)
1945 {
1946         /* TODO: String I/O for in kernel device */
1947
1948         mutex_lock(&vcpu->kvm->lock);
1949         if (vcpu->pio.in)
1950                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1951                                   vcpu->pio.size,
1952                                   pd);
1953         else
1954                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1955                                    vcpu->pio.size,
1956                                    pd);
1957         mutex_unlock(&vcpu->kvm->lock);
1958 }
1959
1960 static void pio_string_write(struct kvm_io_device *pio_dev,
1961                              struct kvm_vcpu *vcpu)
1962 {
1963         struct kvm_pio_request *io = &vcpu->pio;
1964         void *pd = vcpu->pio_data;
1965         int i;
1966
1967         mutex_lock(&vcpu->kvm->lock);
1968         for (i = 0; i < io->cur_count; i++) {
1969                 kvm_iodevice_write(pio_dev, io->port,
1970                                    io->size,
1971                                    pd);
1972                 pd += io->size;
1973         }
1974         mutex_unlock(&vcpu->kvm->lock);
1975 }
1976
1977 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1978                   int size, unsigned port)
1979 {
1980         struct kvm_io_device *pio_dev;
1981
1982         vcpu->run->exit_reason = KVM_EXIT_IO;
1983         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1984         vcpu->run->io.size = vcpu->pio.size = size;
1985         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1986         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1987         vcpu->run->io.port = vcpu->pio.port = port;
1988         vcpu->pio.in = in;
1989         vcpu->pio.string = 0;
1990         vcpu->pio.down = 0;
1991         vcpu->pio.guest_page_offset = 0;
1992         vcpu->pio.rep = 0;
1993
1994         kvm_x86_ops->cache_regs(vcpu);
1995         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1996         kvm_x86_ops->decache_regs(vcpu);
1997
1998         kvm_x86_ops->skip_emulated_instruction(vcpu);
1999
2000         pio_dev = vcpu_find_pio_dev(vcpu, port);
2001         if (pio_dev) {
2002                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
2003                 complete_pio(vcpu);
2004                 return 1;
2005         }
2006         return 0;
2007 }
2008 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2009
2010 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2011                   int size, unsigned long count, int down,
2012                   gva_t address, int rep, unsigned port)
2013 {
2014         unsigned now, in_page;
2015         int i, ret = 0;
2016         int nr_pages = 1;
2017         struct page *page;
2018         struct kvm_io_device *pio_dev;
2019
2020         vcpu->run->exit_reason = KVM_EXIT_IO;
2021         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2022         vcpu->run->io.size = vcpu->pio.size = size;
2023         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2024         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
2025         vcpu->run->io.port = vcpu->pio.port = port;
2026         vcpu->pio.in = in;
2027         vcpu->pio.string = 1;
2028         vcpu->pio.down = down;
2029         vcpu->pio.guest_page_offset = offset_in_page(address);
2030         vcpu->pio.rep = rep;
2031
2032         if (!count) {
2033                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2034                 return 1;
2035         }
2036
2037         if (!down)
2038                 in_page = PAGE_SIZE - offset_in_page(address);
2039         else
2040                 in_page = offset_in_page(address) + size;
2041         now = min(count, (unsigned long)in_page / size);
2042         if (!now) {
2043                 /*
2044                  * String I/O straddles page boundary.  Pin two guest pages
2045                  * so that we satisfy atomicity constraints.  Do just one
2046                  * transaction to avoid complexity.
2047                  */
2048                 nr_pages = 2;
2049                 now = 1;
2050         }
2051         if (down) {
2052                 /*
2053                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2054                  */
2055                 pr_unimpl(vcpu, "guest string pio down\n");
2056                 inject_gp(vcpu);
2057                 return 1;
2058         }
2059         vcpu->run->io.count = now;
2060         vcpu->pio.cur_count = now;
2061
2062         if (vcpu->pio.cur_count == vcpu->pio.count)
2063                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2064
2065         for (i = 0; i < nr_pages; ++i) {
2066                 mutex_lock(&vcpu->kvm->lock);
2067                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2068                 if (page)
2069                         get_page(page);
2070                 vcpu->pio.guest_pages[i] = page;
2071                 mutex_unlock(&vcpu->kvm->lock);
2072                 if (!page) {
2073                         inject_gp(vcpu);
2074                         free_pio_guest_pages(vcpu);
2075                         return 1;
2076                 }
2077         }
2078
2079         pio_dev = vcpu_find_pio_dev(vcpu, port);
2080         if (!vcpu->pio.in) {
2081                 /* string PIO write */
2082                 ret = pio_copy_data(vcpu);
2083                 if (ret >= 0 && pio_dev) {
2084                         pio_string_write(pio_dev, vcpu);
2085                         complete_pio(vcpu);
2086                         if (vcpu->pio.count == 0)
2087                                 ret = 1;
2088                 }
2089         } else if (pio_dev)
2090                 pr_unimpl(vcpu, "no string pio read support yet, "
2091                        "port %x size %d count %ld\n",
2092                         port, size, count);
2093
2094         return ret;
2095 }
2096 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2097
2098 /*
2099  * Check if userspace requested an interrupt window, and that the
2100  * interrupt window is open.
2101  *
2102  * No need to exit to userspace if we already have an interrupt queued.
2103  */
2104 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2105                                           struct kvm_run *kvm_run)
2106 {
2107         return (!vcpu->irq_summary &&
2108                 kvm_run->request_interrupt_window &&
2109                 vcpu->interrupt_window_open &&
2110                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2111 }
2112
2113 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2114                               struct kvm_run *kvm_run)
2115 {
2116         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2117         kvm_run->cr8 = get_cr8(vcpu);
2118         kvm_run->apic_base = kvm_get_apic_base(vcpu);
2119         if (irqchip_in_kernel(vcpu->kvm))
2120                 kvm_run->ready_for_interrupt_injection = 1;
2121         else
2122                 kvm_run->ready_for_interrupt_injection =
2123                                         (vcpu->interrupt_window_open &&
2124                                          vcpu->irq_summary == 0);
2125 }
2126
2127 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2128 {
2129         int r;
2130
2131         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
2132                 pr_debug("vcpu %d received sipi with vector # %x\n",
2133                        vcpu->vcpu_id, vcpu->sipi_vector);
2134                 kvm_lapic_reset(vcpu);
2135                 kvm_x86_ops->vcpu_reset(vcpu);
2136                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2137         }
2138
2139 preempted:
2140         if (vcpu->guest_debug.enabled)
2141                 kvm_x86_ops->guest_debug_pre(vcpu);
2142
2143 again:
2144         r = kvm_mmu_reload(vcpu);
2145         if (unlikely(r))
2146                 goto out;
2147
2148         preempt_disable();
2149
2150         kvm_x86_ops->prepare_guest_switch(vcpu);
2151         kvm_load_guest_fpu(vcpu);
2152
2153         local_irq_disable();
2154
2155         if (signal_pending(current)) {
2156                 local_irq_enable();
2157                 preempt_enable();
2158                 r = -EINTR;
2159                 kvm_run->exit_reason = KVM_EXIT_INTR;
2160                 ++vcpu->stat.signal_exits;
2161                 goto out;
2162         }
2163
2164         if (irqchip_in_kernel(vcpu->kvm))
2165                 kvm_x86_ops->inject_pending_irq(vcpu);
2166         else if (!vcpu->mmio_read_completed)
2167                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2168
2169         vcpu->guest_mode = 1;
2170         kvm_guest_enter();
2171
2172         if (vcpu->requests)
2173                 if (test_and_clear_bit(KVM_TLB_FLUSH, &vcpu->requests))
2174                         kvm_x86_ops->tlb_flush(vcpu);
2175
2176         kvm_x86_ops->run(vcpu, kvm_run);
2177
2178         vcpu->guest_mode = 0;
2179         local_irq_enable();
2180
2181         ++vcpu->stat.exits;
2182
2183         /*
2184          * We must have an instruction between local_irq_enable() and
2185          * kvm_guest_exit(), so the timer interrupt isn't delayed by
2186          * the interrupt shadow.  The stat.exits increment will do nicely.
2187          * But we need to prevent reordering, hence this barrier():
2188          */
2189         barrier();
2190
2191         kvm_guest_exit();
2192
2193         preempt_enable();
2194
2195         /*
2196          * Profile KVM exit RIPs:
2197          */
2198         if (unlikely(prof_on == KVM_PROFILING)) {
2199                 kvm_x86_ops->cache_regs(vcpu);
2200                 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
2201         }
2202
2203         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2204
2205         if (r > 0) {
2206                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2207                         r = -EINTR;
2208                         kvm_run->exit_reason = KVM_EXIT_INTR;
2209                         ++vcpu->stat.request_irq_exits;
2210                         goto out;
2211                 }
2212                 if (!need_resched()) {
2213                         ++vcpu->stat.light_exits;
2214                         goto again;
2215                 }
2216         }
2217
2218 out:
2219         if (r > 0) {
2220                 kvm_resched(vcpu);
2221                 goto preempted;
2222         }
2223
2224         post_kvm_run_save(vcpu, kvm_run);
2225
2226         return r;
2227 }
2228
2229
2230 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2231 {
2232         int r;
2233         sigset_t sigsaved;
2234
2235         vcpu_load(vcpu);
2236
2237         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2238                 kvm_vcpu_block(vcpu);
2239                 vcpu_put(vcpu);
2240                 return -EAGAIN;
2241         }
2242
2243         if (vcpu->sigset_active)
2244                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2245
2246         /* re-sync apic's tpr */
2247         if (!irqchip_in_kernel(vcpu->kvm))
2248                 set_cr8(vcpu, kvm_run->cr8);
2249
2250         if (vcpu->pio.cur_count) {
2251                 r = complete_pio(vcpu);
2252                 if (r)
2253                         goto out;
2254         }
2255
2256         if (vcpu->mmio_needed) {
2257                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2258                 vcpu->mmio_read_completed = 1;
2259                 vcpu->mmio_needed = 0;
2260                 r = emulate_instruction(vcpu, kvm_run,
2261                                         vcpu->mmio_fault_cr2, 0, 1);
2262                 if (r == EMULATE_DO_MMIO) {
2263                         /*
2264                          * Read-modify-write.  Back to userspace.
2265                          */
2266                         r = 0;
2267                         goto out;
2268                 }
2269         }
2270
2271         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2272                 kvm_x86_ops->cache_regs(vcpu);
2273                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2274                 kvm_x86_ops->decache_regs(vcpu);
2275         }
2276
2277         r = __vcpu_run(vcpu, kvm_run);
2278
2279 out:
2280         if (vcpu->sigset_active)
2281                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2282
2283         vcpu_put(vcpu);
2284         return r;
2285 }
2286
2287 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2288                                    struct kvm_regs *regs)
2289 {
2290         vcpu_load(vcpu);
2291
2292         kvm_x86_ops->cache_regs(vcpu);
2293
2294         regs->rax = vcpu->regs[VCPU_REGS_RAX];
2295         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2296         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2297         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2298         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2299         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2300         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2301         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2302 #ifdef CONFIG_X86_64
2303         regs->r8 = vcpu->regs[VCPU_REGS_R8];
2304         regs->r9 = vcpu->regs[VCPU_REGS_R9];
2305         regs->r10 = vcpu->regs[VCPU_REGS_R10];
2306         regs->r11 = vcpu->regs[VCPU_REGS_R11];
2307         regs->r12 = vcpu->regs[VCPU_REGS_R12];
2308         regs->r13 = vcpu->regs[VCPU_REGS_R13];
2309         regs->r14 = vcpu->regs[VCPU_REGS_R14];
2310         regs->r15 = vcpu->regs[VCPU_REGS_R15];
2311 #endif
2312
2313         regs->rip = vcpu->rip;
2314         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2315
2316         /*
2317          * Don't leak debug flags in case they were set for guest debugging
2318          */
2319         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2320                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2321
2322         vcpu_put(vcpu);
2323
2324         return 0;
2325 }
2326
2327 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2328                                    struct kvm_regs *regs)
2329 {
2330         vcpu_load(vcpu);
2331
2332         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2333         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2334         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2335         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2336         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2337         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2338         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2339         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2340 #ifdef CONFIG_X86_64
2341         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2342         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2343         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2344         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2345         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2346         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2347         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2348         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2349 #endif
2350
2351         vcpu->rip = regs->rip;
2352         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2353
2354         kvm_x86_ops->decache_regs(vcpu);
2355
2356         vcpu_put(vcpu);
2357
2358         return 0;
2359 }
2360
2361 static void get_segment(struct kvm_vcpu *vcpu,
2362                         struct kvm_segment *var, int seg)
2363 {
2364         return kvm_x86_ops->get_segment(vcpu, var, seg);
2365 }
2366
2367 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2368                                     struct kvm_sregs *sregs)
2369 {
2370         struct descriptor_table dt;
2371         int pending_vec;
2372
2373         vcpu_load(vcpu);
2374
2375         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2376         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2377         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2378         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2379         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2380         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2381
2382         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2383         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2384
2385         kvm_x86_ops->get_idt(vcpu, &dt);
2386         sregs->idt.limit = dt.limit;
2387         sregs->idt.base = dt.base;
2388         kvm_x86_ops->get_gdt(vcpu, &dt);
2389         sregs->gdt.limit = dt.limit;
2390         sregs->gdt.base = dt.base;
2391
2392         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2393         sregs->cr0 = vcpu->cr0;
2394         sregs->cr2 = vcpu->cr2;
2395         sregs->cr3 = vcpu->cr3;
2396         sregs->cr4 = vcpu->cr4;
2397         sregs->cr8 = get_cr8(vcpu);
2398         sregs->efer = vcpu->shadow_efer;
2399         sregs->apic_base = kvm_get_apic_base(vcpu);
2400
2401         if (irqchip_in_kernel(vcpu->kvm)) {
2402                 memset(sregs->interrupt_bitmap, 0,
2403                        sizeof sregs->interrupt_bitmap);
2404                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2405                 if (pending_vec >= 0)
2406                         set_bit(pending_vec,
2407                                 (unsigned long *)sregs->interrupt_bitmap);
2408         } else
2409                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2410                        sizeof sregs->interrupt_bitmap);
2411
2412         vcpu_put(vcpu);
2413
2414         return 0;
2415 }
2416
2417 static void set_segment(struct kvm_vcpu *vcpu,
2418                         struct kvm_segment *var, int seg)
2419 {
2420         return kvm_x86_ops->set_segment(vcpu, var, seg);
2421 }
2422
2423 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2424                                     struct kvm_sregs *sregs)
2425 {
2426         int mmu_reset_needed = 0;
2427         int i, pending_vec, max_bits;
2428         struct descriptor_table dt;
2429
2430         vcpu_load(vcpu);
2431
2432         dt.limit = sregs->idt.limit;
2433         dt.base = sregs->idt.base;
2434         kvm_x86_ops->set_idt(vcpu, &dt);
2435         dt.limit = sregs->gdt.limit;
2436         dt.base = sregs->gdt.base;
2437         kvm_x86_ops->set_gdt(vcpu, &dt);
2438
2439         vcpu->cr2 = sregs->cr2;
2440         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2441         vcpu->cr3 = sregs->cr3;
2442
2443         set_cr8(vcpu, sregs->cr8);
2444
2445         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2446 #ifdef CONFIG_X86_64
2447         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2448 #endif
2449         kvm_set_apic_base(vcpu, sregs->apic_base);
2450
2451         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2452
2453         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2454         vcpu->cr0 = sregs->cr0;
2455         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2456
2457         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2458         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2459         if (!is_long_mode(vcpu) && is_pae(vcpu))
2460                 load_pdptrs(vcpu, vcpu->cr3);
2461
2462         if (mmu_reset_needed)
2463                 kvm_mmu_reset_context(vcpu);
2464
2465         if (!irqchip_in_kernel(vcpu->kvm)) {
2466                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2467                        sizeof vcpu->irq_pending);
2468                 vcpu->irq_summary = 0;
2469                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2470                         if (vcpu->irq_pending[i])
2471                                 __set_bit(i, &vcpu->irq_summary);
2472         } else {
2473                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2474                 pending_vec = find_first_bit(
2475                         (const unsigned long *)sregs->interrupt_bitmap,
2476                         max_bits);
2477                 /* Only pending external irq is handled here */
2478                 if (pending_vec < max_bits) {
2479                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2480                         pr_debug("Set back pending irq %d\n",
2481                                  pending_vec);
2482                 }
2483         }
2484
2485         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2486         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2487         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2488         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2489         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2490         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2491
2492         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2493         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2494
2495         vcpu_put(vcpu);
2496
2497         return 0;
2498 }
2499
2500 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2501 {
2502         struct kvm_segment cs;
2503
2504         get_segment(vcpu, &cs, VCPU_SREG_CS);
2505         *db = cs.db;
2506         *l = cs.l;
2507 }
2508 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2509
2510 /*
2511  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2512  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2513  *
2514  * This list is modified at module load time to reflect the
2515  * capabilities of the host cpu.
2516  */
2517 static u32 msrs_to_save[] = {
2518         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2519         MSR_K6_STAR,
2520 #ifdef CONFIG_X86_64
2521         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2522 #endif
2523         MSR_IA32_TIME_STAMP_COUNTER,
2524 };
2525
2526 static unsigned num_msrs_to_save;
2527
2528 static u32 emulated_msrs[] = {
2529         MSR_IA32_MISC_ENABLE,
2530 };
2531
2532 static __init void kvm_init_msr_list(void)
2533 {
2534         u32 dummy[2];
2535         unsigned i, j;
2536
2537         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2538                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2539                         continue;
2540                 if (j < i)
2541                         msrs_to_save[j] = msrs_to_save[i];
2542                 j++;
2543         }
2544         num_msrs_to_save = j;
2545 }
2546
2547 /*
2548  * Adapt set_msr() to msr_io()'s calling convention
2549  */
2550 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2551 {
2552         return kvm_set_msr(vcpu, index, *data);
2553 }
2554
2555 /*
2556  * Read or write a bunch of msrs. All parameters are kernel addresses.
2557  *
2558  * @return number of msrs set successfully.
2559  */
2560 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2561                     struct kvm_msr_entry *entries,
2562                     int (*do_msr)(struct kvm_vcpu *vcpu,
2563                                   unsigned index, u64 *data))
2564 {
2565         int i;
2566
2567         vcpu_load(vcpu);
2568
2569         for (i = 0; i < msrs->nmsrs; ++i)
2570                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2571                         break;
2572
2573         vcpu_put(vcpu);
2574
2575         return i;
2576 }
2577
2578 /*
2579  * Read or write a bunch of msrs. Parameters are user addresses.
2580  *
2581  * @return number of msrs set successfully.
2582  */
2583 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2584                   int (*do_msr)(struct kvm_vcpu *vcpu,
2585                                 unsigned index, u64 *data),
2586                   int writeback)
2587 {
2588         struct kvm_msrs msrs;
2589         struct kvm_msr_entry *entries;
2590         int r, n;
2591         unsigned size;
2592
2593         r = -EFAULT;
2594         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2595                 goto out;
2596
2597         r = -E2BIG;
2598         if (msrs.nmsrs >= MAX_IO_MSRS)
2599                 goto out;
2600
2601         r = -ENOMEM;
2602         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2603         entries = vmalloc(size);
2604         if (!entries)
2605                 goto out;
2606
2607         r = -EFAULT;
2608         if (copy_from_user(entries, user_msrs->entries, size))
2609                 goto out_free;
2610
2611         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2612         if (r < 0)
2613                 goto out_free;
2614
2615         r = -EFAULT;
2616         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2617                 goto out_free;
2618
2619         r = n;
2620
2621 out_free:
2622         vfree(entries);
2623 out:
2624         return r;
2625 }
2626
2627 /*
2628  * Translate a guest virtual address to a guest physical address.
2629  */
2630 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2631                                     struct kvm_translation *tr)
2632 {
2633         unsigned long vaddr = tr->linear_address;
2634         gpa_t gpa;
2635
2636         vcpu_load(vcpu);
2637         mutex_lock(&vcpu->kvm->lock);
2638         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2639         tr->physical_address = gpa;
2640         tr->valid = gpa != UNMAPPED_GVA;
2641         tr->writeable = 1;
2642         tr->usermode = 0;
2643         mutex_unlock(&vcpu->kvm->lock);
2644         vcpu_put(vcpu);
2645
2646         return 0;
2647 }
2648
2649 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2650                                     struct kvm_interrupt *irq)
2651 {
2652         if (irq->irq < 0 || irq->irq >= 256)
2653                 return -EINVAL;
2654         if (irqchip_in_kernel(vcpu->kvm))
2655                 return -ENXIO;
2656         vcpu_load(vcpu);
2657
2658         set_bit(irq->irq, vcpu->irq_pending);
2659         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2660
2661         vcpu_put(vcpu);
2662
2663         return 0;
2664 }
2665
2666 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2667                                       struct kvm_debug_guest *dbg)
2668 {
2669         int r;
2670
2671         vcpu_load(vcpu);
2672
2673         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2674
2675         vcpu_put(vcpu);
2676
2677         return r;
2678 }
2679
2680 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2681                                     unsigned long address,
2682                                     int *type)
2683 {
2684         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2685         unsigned long pgoff;
2686         struct page *page;
2687
2688         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2689         if (pgoff == 0)
2690                 page = virt_to_page(vcpu->run);
2691         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2692                 page = virt_to_page(vcpu->pio_data);
2693         else
2694                 return NOPAGE_SIGBUS;
2695         get_page(page);
2696         if (type != NULL)
2697                 *type = VM_FAULT_MINOR;
2698
2699         return page;
2700 }
2701
2702 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2703         .nopage = kvm_vcpu_nopage,
2704 };
2705
2706 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2707 {
2708         vma->vm_ops = &kvm_vcpu_vm_ops;
2709         return 0;
2710 }
2711
2712 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2713 {
2714         struct kvm_vcpu *vcpu = filp->private_data;
2715
2716         fput(vcpu->kvm->filp);
2717         return 0;
2718 }
2719
2720 static struct file_operations kvm_vcpu_fops = {
2721         .release        = kvm_vcpu_release,
2722         .unlocked_ioctl = kvm_vcpu_ioctl,
2723         .compat_ioctl   = kvm_vcpu_ioctl,
2724         .mmap           = kvm_vcpu_mmap,
2725 };
2726
2727 /*
2728  * Allocates an inode for the vcpu.
2729  */
2730 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2731 {
2732         int fd, r;
2733         struct inode *inode;
2734         struct file *file;
2735
2736         r = anon_inode_getfd(&fd, &inode, &file,
2737                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2738         if (r)
2739                 return r;
2740         atomic_inc(&vcpu->kvm->filp->f_count);
2741         return fd;
2742 }
2743
2744 /*
2745  * Creates some virtual cpus.  Good luck creating more than one.
2746  */
2747 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2748 {
2749         int r;
2750         struct kvm_vcpu *vcpu;
2751
2752         if (!valid_vcpu(n))
2753                 return -EINVAL;
2754
2755         vcpu = kvm_x86_ops->vcpu_create(kvm, n);
2756         if (IS_ERR(vcpu))
2757                 return PTR_ERR(vcpu);
2758
2759         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2760
2761         /* We do fxsave: this must be aligned. */
2762         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2763
2764         vcpu_load(vcpu);
2765         r = kvm_mmu_setup(vcpu);
2766         vcpu_put(vcpu);
2767         if (r < 0)
2768                 goto free_vcpu;
2769
2770         mutex_lock(&kvm->lock);
2771         if (kvm->vcpus[n]) {
2772                 r = -EEXIST;
2773                 mutex_unlock(&kvm->lock);
2774                 goto mmu_unload;
2775         }
2776         kvm->vcpus[n] = vcpu;
2777         mutex_unlock(&kvm->lock);
2778
2779         /* Now it's all set up, let userspace reach it */
2780         r = create_vcpu_fd(vcpu);
2781         if (r < 0)
2782                 goto unlink;
2783         return r;
2784
2785 unlink:
2786         mutex_lock(&kvm->lock);
2787         kvm->vcpus[n] = NULL;
2788         mutex_unlock(&kvm->lock);
2789
2790 mmu_unload:
2791         vcpu_load(vcpu);
2792         kvm_mmu_unload(vcpu);
2793         vcpu_put(vcpu);
2794
2795 free_vcpu:
2796         kvm_x86_ops->vcpu_free(vcpu);
2797         return r;
2798 }
2799
2800 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2801 {
2802         u64 efer;
2803         int i;
2804         struct kvm_cpuid_entry *e, *entry;
2805
2806         rdmsrl(MSR_EFER, efer);
2807         entry = NULL;
2808         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2809                 e = &vcpu->cpuid_entries[i];
2810                 if (e->function == 0x80000001) {
2811                         entry = e;
2812                         break;
2813                 }
2814         }
2815         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2816                 entry->edx &= ~(1 << 20);
2817                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2818         }
2819 }
2820
2821 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2822                                     struct kvm_cpuid *cpuid,
2823                                     struct kvm_cpuid_entry __user *entries)
2824 {
2825         int r;
2826
2827         r = -E2BIG;
2828         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2829                 goto out;
2830         r = -EFAULT;
2831         if (copy_from_user(&vcpu->cpuid_entries, entries,
2832                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2833                 goto out;
2834         vcpu->cpuid_nent = cpuid->nent;
2835         cpuid_fix_nx_cap(vcpu);
2836         return 0;
2837
2838 out:
2839         return r;
2840 }
2841
2842 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2843 {
2844         if (sigset) {
2845                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2846                 vcpu->sigset_active = 1;
2847                 vcpu->sigset = *sigset;
2848         } else
2849                 vcpu->sigset_active = 0;
2850         return 0;
2851 }
2852
2853 /*
2854  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2855  * we have asm/x86/processor.h
2856  */
2857 struct fxsave {
2858         u16     cwd;
2859         u16     swd;
2860         u16     twd;
2861         u16     fop;
2862         u64     rip;
2863         u64     rdp;
2864         u32     mxcsr;
2865         u32     mxcsr_mask;
2866         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2867 #ifdef CONFIG_X86_64
2868         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2869 #else
2870         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2871 #endif
2872 };
2873
2874 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2875 {
2876         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2877
2878         vcpu_load(vcpu);
2879
2880         memcpy(fpu->fpr, fxsave->st_space, 128);
2881         fpu->fcw = fxsave->cwd;
2882         fpu->fsw = fxsave->swd;
2883         fpu->ftwx = fxsave->twd;
2884         fpu->last_opcode = fxsave->fop;
2885         fpu->last_ip = fxsave->rip;
2886         fpu->last_dp = fxsave->rdp;
2887         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2888
2889         vcpu_put(vcpu);
2890
2891         return 0;
2892 }
2893
2894 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2895 {
2896         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2897
2898         vcpu_load(vcpu);
2899
2900         memcpy(fxsave->st_space, fpu->fpr, 128);
2901         fxsave->cwd = fpu->fcw;
2902         fxsave->swd = fpu->fsw;
2903         fxsave->twd = fpu->ftwx;
2904         fxsave->fop = fpu->last_opcode;
2905         fxsave->rip = fpu->last_ip;
2906         fxsave->rdp = fpu->last_dp;
2907         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2908
2909         vcpu_put(vcpu);
2910
2911         return 0;
2912 }
2913
2914 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2915                                     struct kvm_lapic_state *s)
2916 {
2917         vcpu_load(vcpu);
2918         memcpy(s->regs, vcpu->apic->regs, sizeof *s);
2919         vcpu_put(vcpu);
2920
2921         return 0;
2922 }
2923
2924 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2925                                     struct kvm_lapic_state *s)
2926 {
2927         vcpu_load(vcpu);
2928         memcpy(vcpu->apic->regs, s->regs, sizeof *s);
2929         kvm_apic_post_state_restore(vcpu);
2930         vcpu_put(vcpu);
2931
2932         return 0;
2933 }
2934
2935 static long kvm_vcpu_ioctl(struct file *filp,
2936                            unsigned int ioctl, unsigned long arg)
2937 {
2938         struct kvm_vcpu *vcpu = filp->private_data;
2939         void __user *argp = (void __user *)arg;
2940         int r = -EINVAL;
2941
2942         switch (ioctl) {
2943         case KVM_RUN:
2944                 r = -EINVAL;
2945                 if (arg)
2946                         goto out;
2947                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2948                 break;
2949         case KVM_GET_REGS: {
2950                 struct kvm_regs kvm_regs;
2951
2952                 memset(&kvm_regs, 0, sizeof kvm_regs);
2953                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2954                 if (r)
2955                         goto out;
2956                 r = -EFAULT;
2957                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2958                         goto out;
2959                 r = 0;
2960                 break;
2961         }
2962         case KVM_SET_REGS: {
2963                 struct kvm_regs kvm_regs;
2964
2965                 r = -EFAULT;
2966                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2967                         goto out;
2968                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2969                 if (r)
2970                         goto out;
2971                 r = 0;
2972                 break;
2973         }
2974         case KVM_GET_SREGS: {
2975                 struct kvm_sregs kvm_sregs;
2976
2977                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2978                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2979                 if (r)
2980                         goto out;
2981                 r = -EFAULT;
2982                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2983                         goto out;
2984                 r = 0;
2985                 break;
2986         }
2987         case KVM_SET_SREGS: {
2988                 struct kvm_sregs kvm_sregs;
2989
2990                 r = -EFAULT;
2991                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2992                         goto out;
2993                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2994                 if (r)
2995                         goto out;
2996                 r = 0;
2997                 break;
2998         }
2999         case KVM_TRANSLATE: {
3000                 struct kvm_translation tr;
3001
3002                 r = -EFAULT;
3003                 if (copy_from_user(&tr, argp, sizeof tr))
3004                         goto out;
3005                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
3006                 if (r)
3007                         goto out;
3008                 r = -EFAULT;
3009                 if (copy_to_user(argp, &tr, sizeof tr))
3010                         goto out;
3011                 r = 0;
3012                 break;
3013         }
3014         case KVM_INTERRUPT: {
3015                 struct kvm_interrupt irq;
3016
3017                 r = -EFAULT;
3018                 if (copy_from_user(&irq, argp, sizeof irq))
3019                         goto out;
3020                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3021                 if (r)
3022                         goto out;
3023                 r = 0;
3024                 break;
3025         }
3026         case KVM_DEBUG_GUEST: {
3027                 struct kvm_debug_guest dbg;
3028
3029                 r = -EFAULT;
3030                 if (copy_from_user(&dbg, argp, sizeof dbg))
3031                         goto out;
3032                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
3033                 if (r)
3034                         goto out;
3035                 r = 0;
3036                 break;
3037         }
3038         case KVM_GET_MSRS:
3039                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3040                 break;
3041         case KVM_SET_MSRS:
3042                 r = msr_io(vcpu, argp, do_set_msr, 0);
3043                 break;
3044         case KVM_SET_CPUID: {
3045                 struct kvm_cpuid __user *cpuid_arg = argp;
3046                 struct kvm_cpuid cpuid;
3047
3048                 r = -EFAULT;
3049                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3050                         goto out;
3051                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3052                 if (r)
3053                         goto out;
3054                 break;
3055         }
3056         case KVM_SET_SIGNAL_MASK: {
3057                 struct kvm_signal_mask __user *sigmask_arg = argp;
3058                 struct kvm_signal_mask kvm_sigmask;
3059                 sigset_t sigset, *p;
3060
3061                 p = NULL;
3062                 if (argp) {
3063                         r = -EFAULT;
3064                         if (copy_from_user(&kvm_sigmask, argp,
3065                                            sizeof kvm_sigmask))
3066                                 goto out;
3067                         r = -EINVAL;
3068                         if (kvm_sigmask.len != sizeof sigset)
3069                                 goto out;
3070                         r = -EFAULT;
3071                         if (copy_from_user(&sigset, sigmask_arg->sigset,
3072                                            sizeof sigset))
3073                                 goto out;
3074                         p = &sigset;
3075                 }
3076                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3077                 break;
3078         }
3079         case KVM_GET_FPU: {
3080                 struct kvm_fpu fpu;
3081
3082                 memset(&fpu, 0, sizeof fpu);
3083                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
3084                 if (r)
3085                         goto out;
3086                 r = -EFAULT;
3087                 if (copy_to_user(argp, &fpu, sizeof fpu))
3088                         goto out;
3089                 r = 0;
3090                 break;
3091         }
3092         case KVM_SET_FPU: {
3093                 struct kvm_fpu fpu;
3094
3095                 r = -EFAULT;
3096                 if (copy_from_user(&fpu, argp, sizeof fpu))
3097                         goto out;
3098                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
3099                 if (r)
3100                         goto out;
3101                 r = 0;
3102                 break;
3103         }
3104         case KVM_GET_LAPIC: {
3105                 struct kvm_lapic_state lapic;
3106
3107                 memset(&lapic, 0, sizeof lapic);
3108                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
3109                 if (r)
3110                         goto out;
3111                 r = -EFAULT;
3112                 if (copy_to_user(argp, &lapic, sizeof lapic))
3113                         goto out;
3114                 r = 0;
3115                 break;
3116         }
3117         case KVM_SET_LAPIC: {
3118                 struct kvm_lapic_state lapic;
3119
3120                 r = -EFAULT;
3121                 if (copy_from_user(&lapic, argp, sizeof lapic))
3122                         goto out;
3123                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
3124                 if (r)
3125                         goto out;
3126                 r = 0;
3127                 break;
3128         }
3129         default:
3130                 ;
3131         }
3132 out:
3133         return r;
3134 }
3135
3136 static long kvm_vm_ioctl(struct file *filp,
3137                            unsigned int ioctl, unsigned long arg)
3138 {
3139         struct kvm *kvm = filp->private_data;
3140         void __user *argp = (void __user *)arg;
3141         int r = -EINVAL;
3142
3143         switch (ioctl) {
3144         case KVM_CREATE_VCPU:
3145                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3146                 if (r < 0)
3147                         goto out;
3148                 break;
3149         case KVM_SET_MEMORY_REGION: {
3150                 struct kvm_memory_region kvm_mem;
3151                 struct kvm_userspace_memory_region kvm_userspace_mem;
3152
3153                 r = -EFAULT;
3154                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
3155                         goto out;
3156                 kvm_userspace_mem.slot = kvm_mem.slot;
3157                 kvm_userspace_mem.flags = kvm_mem.flags;
3158                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
3159                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
3160                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
3161                 if (r)
3162                         goto out;
3163                 break;
3164         }
3165         case KVM_SET_USER_MEMORY_REGION: {
3166                 struct kvm_userspace_memory_region kvm_userspace_mem;
3167
3168                 r = -EFAULT;
3169                 if (copy_from_user(&kvm_userspace_mem, argp,
3170                                                 sizeof kvm_userspace_mem))
3171                         goto out;
3172
3173                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
3174                 if (r)
3175                         goto out;
3176                 break;
3177         }
3178         case KVM_SET_NR_MMU_PAGES:
3179                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3180                 if (r)
3181                         goto out;
3182                 break;
3183         case KVM_GET_NR_MMU_PAGES:
3184                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3185                 break;
3186         case KVM_GET_DIRTY_LOG: {
3187                 struct kvm_dirty_log log;
3188
3189                 r = -EFAULT;
3190                 if (copy_from_user(&log, argp, sizeof log))
3191                         goto out;
3192                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3193                 if (r)
3194                         goto out;
3195                 break;
3196         }
3197         case KVM_SET_MEMORY_ALIAS: {
3198                 struct kvm_memory_alias alias;
3199
3200                 r = -EFAULT;
3201                 if (copy_from_user(&alias, argp, sizeof alias))
3202                         goto out;
3203                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
3204                 if (r)
3205                         goto out;
3206                 break;
3207         }
3208         case KVM_CREATE_IRQCHIP:
3209                 r = -ENOMEM;
3210                 kvm->vpic = kvm_create_pic(kvm);
3211                 if (kvm->vpic) {
3212                         r = kvm_ioapic_init(kvm);
3213                         if (r) {
3214                                 kfree(kvm->vpic);
3215                                 kvm->vpic = NULL;
3216                                 goto out;
3217                         }
3218                 } else
3219                         goto out;
3220                 break;
3221         case KVM_IRQ_LINE: {
3222                 struct kvm_irq_level irq_event;
3223
3224                 r = -EFAULT;
3225                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3226                         goto out;
3227                 if (irqchip_in_kernel(kvm)) {
3228                         mutex_lock(&kvm->lock);
3229                         if (irq_event.irq < 16)
3230                                 kvm_pic_set_irq(pic_irqchip(kvm),
3231                                         irq_event.irq,
3232                                         irq_event.level);
3233                         kvm_ioapic_set_irq(kvm->vioapic,
3234                                         irq_event.irq,
3235                                         irq_event.level);
3236                         mutex_unlock(&kvm->lock);
3237                         r = 0;
3238                 }
3239                 break;
3240         }
3241         case KVM_GET_IRQCHIP: {
3242                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3243                 struct kvm_irqchip chip;
3244
3245                 r = -EFAULT;
3246                 if (copy_from_user(&chip, argp, sizeof chip))
3247                         goto out;
3248                 r = -ENXIO;
3249                 if (!irqchip_in_kernel(kvm))
3250                         goto out;
3251                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
3252                 if (r)
3253                         goto out;
3254                 r = -EFAULT;
3255                 if (copy_to_user(argp, &chip, sizeof chip))
3256                         goto out;
3257                 r = 0;
3258                 break;
3259         }
3260         case KVM_SET_IRQCHIP: {
3261                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3262                 struct kvm_irqchip chip;
3263
3264                 r = -EFAULT;
3265                 if (copy_from_user(&chip, argp, sizeof chip))
3266                         goto out;
3267                 r = -ENXIO;
3268                 if (!irqchip_in_kernel(kvm))
3269                         goto out;
3270                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
3271                 if (r)
3272                         goto out;
3273                 r = 0;
3274                 break;
3275         }
3276         default:
3277                 ;
3278         }
3279 out:
3280         return r;
3281 }
3282
3283 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
3284                                   unsigned long address,
3285                                   int *type)
3286 {
3287         struct kvm *kvm = vma->vm_file->private_data;
3288         unsigned long pgoff;
3289         struct page *page;
3290
3291         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
3292         page = gfn_to_page(kvm, pgoff);
3293         if (!page)
3294                 return NOPAGE_SIGBUS;
3295         get_page(page);
3296         if (type != NULL)
3297                 *type = VM_FAULT_MINOR;
3298
3299         return page;
3300 }
3301
3302 static struct vm_operations_struct kvm_vm_vm_ops = {
3303         .nopage = kvm_vm_nopage,
3304 };
3305
3306 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
3307 {
3308         vma->vm_ops = &kvm_vm_vm_ops;
3309         return 0;
3310 }
3311
3312 static struct file_operations kvm_vm_fops = {
3313         .release        = kvm_vm_release,
3314         .unlocked_ioctl = kvm_vm_ioctl,
3315         .compat_ioctl   = kvm_vm_ioctl,
3316         .mmap           = kvm_vm_mmap,
3317 };
3318
3319 static int kvm_dev_ioctl_create_vm(void)
3320 {
3321         int fd, r;
3322         struct inode *inode;
3323         struct file *file;
3324         struct kvm *kvm;
3325
3326         kvm = kvm_create_vm();
3327         if (IS_ERR(kvm))
3328                 return PTR_ERR(kvm);
3329         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
3330         if (r) {
3331                 kvm_destroy_vm(kvm);
3332                 return r;
3333         }
3334
3335         kvm->filp = file;
3336
3337         return fd;
3338 }
3339
3340 static long kvm_dev_ioctl(struct file *filp,
3341                           unsigned int ioctl, unsigned long arg)
3342 {
3343         void __user *argp = (void __user *)arg;
3344         long r = -EINVAL;
3345
3346         switch (ioctl) {
3347         case KVM_GET_API_VERSION:
3348                 r = -EINVAL;
3349                 if (arg)
3350                         goto out;
3351                 r = KVM_API_VERSION;
3352                 break;
3353         case KVM_CREATE_VM:
3354                 r = -EINVAL;
3355                 if (arg)
3356                         goto out;
3357                 r = kvm_dev_ioctl_create_vm();
3358                 break;
3359         case KVM_GET_MSR_INDEX_LIST: {
3360                 struct kvm_msr_list __user *user_msr_list = argp;
3361                 struct kvm_msr_list msr_list;
3362                 unsigned n;
3363
3364                 r = -EFAULT;
3365                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
3366                         goto out;
3367                 n = msr_list.nmsrs;
3368                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
3369                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
3370                         goto out;
3371                 r = -E2BIG;
3372                 if (n < num_msrs_to_save)
3373                         goto out;
3374                 r = -EFAULT;
3375                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3376                                  num_msrs_to_save * sizeof(u32)))
3377                         goto out;
3378                 if (copy_to_user(user_msr_list->indices
3379                                  + num_msrs_to_save * sizeof(u32),
3380                                  &emulated_msrs,
3381                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
3382                         goto out;
3383                 r = 0;
3384                 break;
3385         }
3386         case KVM_CHECK_EXTENSION: {
3387                 int ext = (long)argp;
3388
3389                 switch (ext) {
3390                 case KVM_CAP_IRQCHIP:
3391                 case KVM_CAP_HLT:
3392                 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3393                 case KVM_CAP_USER_MEMORY:
3394                         r = 1;
3395                         break;
3396                 default:
3397                         r = 0;
3398                         break;
3399                 }
3400                 break;
3401         }
3402         case KVM_GET_VCPU_MMAP_SIZE:
3403                 r = -EINVAL;
3404                 if (arg)
3405                         goto out;
3406                 r = 2 * PAGE_SIZE;
3407                 break;
3408         default:
3409                 ;
3410         }
3411 out:
3412         return r;
3413 }
3414
3415 static struct file_operations kvm_chardev_ops = {
3416         .unlocked_ioctl = kvm_dev_ioctl,
3417         .compat_ioctl   = kvm_dev_ioctl,
3418 };
3419
3420 static struct miscdevice kvm_dev = {
3421         KVM_MINOR,
3422         "kvm",
3423         &kvm_chardev_ops,
3424 };
3425
3426 /*
3427  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
3428  * cached on it.
3429  */
3430 static void decache_vcpus_on_cpu(int cpu)
3431 {
3432         struct kvm *vm;
3433         struct kvm_vcpu *vcpu;
3434         int i;
3435
3436         spin_lock(&kvm_lock);
3437         list_for_each_entry(vm, &vm_list, vm_list)
3438                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3439                         vcpu = vm->vcpus[i];
3440                         if (!vcpu)
3441                                 continue;
3442                         /*
3443                          * If the vcpu is locked, then it is running on some
3444                          * other cpu and therefore it is not cached on the
3445                          * cpu in question.
3446                          *
3447                          * If it's not locked, check the last cpu it executed
3448                          * on.
3449                          */
3450                         if (mutex_trylock(&vcpu->mutex)) {
3451                                 if (vcpu->cpu == cpu) {
3452                                         kvm_x86_ops->vcpu_decache(vcpu);
3453                                         vcpu->cpu = -1;
3454                                 }
3455                                 mutex_unlock(&vcpu->mutex);
3456                         }
3457                 }
3458         spin_unlock(&kvm_lock);
3459 }
3460
3461 static void hardware_enable(void *junk)
3462 {
3463         int cpu = raw_smp_processor_id();
3464
3465         if (cpu_isset(cpu, cpus_hardware_enabled))
3466                 return;
3467         cpu_set(cpu, cpus_hardware_enabled);
3468         kvm_x86_ops->hardware_enable(NULL);
3469 }
3470
3471 static void hardware_disable(void *junk)
3472 {
3473         int cpu = raw_smp_processor_id();
3474
3475         if (!cpu_isset(cpu, cpus_hardware_enabled))
3476                 return;
3477         cpu_clear(cpu, cpus_hardware_enabled);
3478         decache_vcpus_on_cpu(cpu);
3479         kvm_x86_ops->hardware_disable(NULL);
3480 }
3481
3482 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3483                            void *v)
3484 {
3485         int cpu = (long)v;
3486
3487         switch (val) {
3488         case CPU_DYING:
3489         case CPU_DYING_FROZEN:
3490                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3491                        cpu);
3492                 hardware_disable(NULL);
3493                 break;
3494         case CPU_UP_CANCELED:
3495         case CPU_UP_CANCELED_FROZEN:
3496                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3497                        cpu);
3498                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3499                 break;
3500         case CPU_ONLINE:
3501         case CPU_ONLINE_FROZEN:
3502                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3503                        cpu);
3504                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3505                 break;
3506         }
3507         return NOTIFY_OK;
3508 }
3509
3510 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3511                       void *v)
3512 {
3513         if (val == SYS_RESTART) {
3514                 /*
3515                  * Some (well, at least mine) BIOSes hang on reboot if
3516                  * in vmx root mode.
3517                  */
3518                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3519                 on_each_cpu(hardware_disable, NULL, 0, 1);
3520         }
3521         return NOTIFY_OK;
3522 }
3523
3524 static struct notifier_block kvm_reboot_notifier = {
3525         .notifier_call = kvm_reboot,
3526         .priority = 0,
3527 };
3528
3529 void kvm_io_bus_init(struct kvm_io_bus *bus)
3530 {
3531         memset(bus, 0, sizeof(*bus));
3532 }
3533
3534 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3535 {
3536         int i;
3537
3538         for (i = 0; i < bus->dev_count; i++) {
3539                 struct kvm_io_device *pos = bus->devs[i];
3540
3541                 kvm_iodevice_destructor(pos);
3542         }
3543 }
3544
3545 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3546 {
3547         int i;
3548
3549         for (i = 0; i < bus->dev_count; i++) {
3550                 struct kvm_io_device *pos = bus->devs[i];
3551
3552                 if (pos->in_range(pos, addr))
3553                         return pos;
3554         }
3555
3556         return NULL;
3557 }
3558
3559 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3560 {
3561         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3562
3563         bus->devs[bus->dev_count++] = dev;
3564 }
3565
3566 static struct notifier_block kvm_cpu_notifier = {
3567         .notifier_call = kvm_cpu_hotplug,
3568         .priority = 20, /* must be > scheduler priority */
3569 };
3570
3571 static u64 stat_get(void *_offset)
3572 {
3573         unsigned offset = (long)_offset;
3574         u64 total = 0;
3575         struct kvm *kvm;
3576         struct kvm_vcpu *vcpu;
3577         int i;
3578
3579         spin_lock(&kvm_lock);
3580         list_for_each_entry(kvm, &vm_list, vm_list)
3581                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3582                         vcpu = kvm->vcpus[i];
3583                         if (vcpu)
3584                                 total += *(u32 *)((void *)vcpu + offset);
3585                 }
3586         spin_unlock(&kvm_lock);
3587         return total;
3588 }
3589
3590 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3591
3592 static __init void kvm_init_debug(void)
3593 {
3594         struct kvm_stats_debugfs_item *p;
3595
3596         debugfs_dir = debugfs_create_dir("kvm", NULL);
3597         for (p = debugfs_entries; p->name; ++p)
3598                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3599                                                 (void *)(long)p->offset,
3600                                                 &stat_fops);
3601 }
3602
3603 static void kvm_exit_debug(void)
3604 {
3605         struct kvm_stats_debugfs_item *p;
3606
3607         for (p = debugfs_entries; p->name; ++p)
3608                 debugfs_remove(p->dentry);
3609         debugfs_remove(debugfs_dir);
3610 }
3611
3612 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3613 {
3614         hardware_disable(NULL);
3615         return 0;
3616 }
3617
3618 static int kvm_resume(struct sys_device *dev)
3619 {
3620         hardware_enable(NULL);
3621         return 0;
3622 }
3623
3624 static struct sysdev_class kvm_sysdev_class = {
3625         .name = "kvm",
3626         .suspend = kvm_suspend,
3627         .resume = kvm_resume,
3628 };
3629
3630 static struct sys_device kvm_sysdev = {
3631         .id = 0,
3632         .cls = &kvm_sysdev_class,
3633 };
3634
3635 hpa_t bad_page_address;
3636
3637 static inline
3638 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3639 {
3640         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3641 }
3642
3643 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3644 {
3645         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3646
3647         kvm_x86_ops->vcpu_load(vcpu, cpu);
3648 }
3649
3650 static void kvm_sched_out(struct preempt_notifier *pn,
3651                           struct task_struct *next)
3652 {
3653         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3654
3655         kvm_x86_ops->vcpu_put(vcpu);
3656 }
3657
3658 int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
3659                   struct module *module)
3660 {
3661         int r;
3662         int cpu;
3663
3664         if (kvm_x86_ops) {
3665                 printk(KERN_ERR "kvm: already loaded the other module\n");
3666                 return -EEXIST;
3667         }
3668
3669         if (!ops->cpu_has_kvm_support()) {
3670                 printk(KERN_ERR "kvm: no hardware support\n");
3671                 return -EOPNOTSUPP;
3672         }
3673         if (ops->disabled_by_bios()) {
3674                 printk(KERN_ERR "kvm: disabled by bios\n");
3675                 return -EOPNOTSUPP;
3676         }
3677
3678         kvm_x86_ops = ops;
3679
3680         r = kvm_x86_ops->hardware_setup();
3681         if (r < 0)
3682                 goto out;
3683
3684         for_each_online_cpu(cpu) {
3685                 smp_call_function_single(cpu,
3686                                 kvm_x86_ops->check_processor_compatibility,
3687                                 &r, 0, 1);
3688                 if (r < 0)
3689                         goto out_free_0;
3690         }
3691
3692         on_each_cpu(hardware_enable, NULL, 0, 1);
3693         r = register_cpu_notifier(&kvm_cpu_notifier);
3694         if (r)
3695                 goto out_free_1;
3696         register_reboot_notifier(&kvm_reboot_notifier);
3697
3698         r = sysdev_class_register(&kvm_sysdev_class);
3699         if (r)
3700                 goto out_free_2;
3701
3702         r = sysdev_register(&kvm_sysdev);
3703         if (r)
3704                 goto out_free_3;
3705
3706         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3707         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3708                                            __alignof__(struct kvm_vcpu), 0, 0);
3709         if (!kvm_vcpu_cache) {
3710                 r = -ENOMEM;
3711                 goto out_free_4;
3712         }
3713
3714         kvm_chardev_ops.owner = module;
3715
3716         r = misc_register(&kvm_dev);
3717         if (r) {
3718                 printk(KERN_ERR "kvm: misc device register failed\n");
3719                 goto out_free;
3720         }
3721
3722         kvm_preempt_ops.sched_in = kvm_sched_in;
3723         kvm_preempt_ops.sched_out = kvm_sched_out;
3724
3725         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3726
3727         return 0;
3728
3729 out_free:
3730         kmem_cache_destroy(kvm_vcpu_cache);
3731 out_free_4:
3732         sysdev_unregister(&kvm_sysdev);
3733 out_free_3:
3734         sysdev_class_unregister(&kvm_sysdev_class);
3735 out_free_2:
3736         unregister_reboot_notifier(&kvm_reboot_notifier);
3737         unregister_cpu_notifier(&kvm_cpu_notifier);
3738 out_free_1:
3739         on_each_cpu(hardware_disable, NULL, 0, 1);
3740 out_free_0:
3741         kvm_x86_ops->hardware_unsetup();
3742 out:
3743         kvm_x86_ops = NULL;
3744         return r;
3745 }
3746 EXPORT_SYMBOL_GPL(kvm_init_x86);
3747
3748 void kvm_exit_x86(void)
3749 {
3750         misc_deregister(&kvm_dev);
3751         kmem_cache_destroy(kvm_vcpu_cache);
3752         sysdev_unregister(&kvm_sysdev);
3753         sysdev_class_unregister(&kvm_sysdev_class);
3754         unregister_reboot_notifier(&kvm_reboot_notifier);
3755         unregister_cpu_notifier(&kvm_cpu_notifier);
3756         on_each_cpu(hardware_disable, NULL, 0, 1);
3757         kvm_x86_ops->hardware_unsetup();
3758         kvm_x86_ops = NULL;
3759 }
3760 EXPORT_SYMBOL_GPL(kvm_exit_x86);
3761
3762 static __init int kvm_init(void)
3763 {
3764         static struct page *bad_page;
3765         int r;
3766
3767         r = kvm_mmu_module_init();
3768         if (r)
3769                 goto out4;
3770
3771         kvm_init_debug();
3772
3773         kvm_init_msr_list();
3774
3775         bad_page = alloc_page(GFP_KERNEL);
3776
3777         if (bad_page == NULL) {
3778                 r = -ENOMEM;
3779                 goto out;
3780         }
3781
3782         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3783         memset(__va(bad_page_address), 0, PAGE_SIZE);
3784
3785         return 0;
3786
3787 out:
3788         kvm_exit_debug();
3789         kvm_mmu_module_exit();
3790 out4:
3791         return r;
3792 }
3793
3794 static __exit void kvm_exit(void)
3795 {
3796         kvm_exit_debug();
3797         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3798         kvm_mmu_module_exit();
3799 }
3800
3801 module_init(kvm_init)
3802 module_exit(kvm_exit)