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.h"
  20 #include "x86_emulate.h"
  21 #include "segment_descriptor.h"
  22 #include "irq.h"
  23
  24 #include <linux/kvm.h>
  25 #include <linux/module.h>
  26 #include <linux/errno.h>
  27 #include <linux/percpu.h>
  28 #include <linux/gfp.h>
  29 #include <linux/mm.h>
  30 #include <linux/miscdevice.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/reboot.h>
  33 #include <linux/debugfs.h>
  34 #include <linux/highmem.h>
  35 #include <linux/file.h>
  36 #include <linux/sysdev.h>
  37 #include <linux/cpu.h>
  38 #include <linux/sched.h>
  39 #include <linux/cpumask.h>
  40 #include <linux/smp.h>
  41 #include <linux/anon_inodes.h>
  42 #include <linux/profile.h>
  43 #include <linux/kvm_para.h>
  44 #include <linux/pagemap.h>
  45 #include <linux/mman.h>
  46
  47 #include <asm/processor.h>
  48 #include <asm/msr.h>
  49 #include <asm/io.h>
  50 #include <asm/uaccess.h>
  51 #include <asm/desc.h>
  52
  53 MODULE_AUTHOR("Qumranet");
  54 MODULE_LICENSE("GPL");
  55
  56 static DEFINE_SPINLOCK(kvm_lock);
  57 static LIST_HEAD(vm_list);
  58
  59 static cpumask_t cpus_hardware_enabled;
  60
  61 struct kvm_x86_ops *kvm_x86_ops;
  62 struct kmem_cache *kvm_vcpu_cache;
  63 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  64
  65 static __read_mostly struct preempt_ops kvm_preempt_ops;
  66
  67 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  68
  69 static struct kvm_stats_debugfs_item {
  70         const char *name;
  71         int offset;
  72         struct dentry *dentry;
  73 } debugfs_entries[] = {
  74         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  75         { "pf_guest", STAT_OFFSET(pf_guest) },
  76         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  77         { "invlpg", STAT_OFFSET(invlpg) },
  78         { "exits", STAT_OFFSET(exits) },
  79         { "io_exits", STAT_OFFSET(io_exits) },
  80         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  81         { "signal_exits", STAT_OFFSET(signal_exits) },
  82         { "irq_window", STAT_OFFSET(irq_window_exits) },
  83         { "halt_exits", STAT_OFFSET(halt_exits) },
  84         { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
  85         { "request_irq", STAT_OFFSET(request_irq_exits) },
  86         { "irq_exits", STAT_OFFSET(irq_exits) },
  87         { "light_exits", STAT_OFFSET(light_exits) },
  88         { "efer_reload", STAT_OFFSET(efer_reload) },
  89         { NULL }
  90 };
  91
  92 static struct dentry *debugfs_dir;
  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 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_arch_vcpu_load(vcpu, cpu);
 189         put_cpu();
 190 }
 191
 192 void vcpu_put(struct kvm_vcpu *vcpu)
 193 {
 194         preempt_disable();
 195         kvm_arch_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_REQ_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 /*
 305  * Free any memory in @free but not in @dont.
 306  */
 307 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 308                                   struct kvm_memory_slot *dont)
 309 {
 310         if (!dont || free->rmap != dont->rmap)
 311                 vfree(free->rmap);
 312
 313         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 314                 vfree(free->dirty_bitmap);
 315
 316         free->npages = 0;
 317         free->dirty_bitmap = NULL;
 318         free->rmap = NULL;
 319 }
 320
 321 static void kvm_free_physmem(struct kvm *kvm)
 322 {
 323         int i;
 324
 325         for (i = 0; i < kvm->nmemslots; ++i)
 326                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 327 }
 328
 329 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 330 {
 331         int i;
 332
 333         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
 334                 if (vcpu->pio.guest_pages[i]) {
 335                         kvm_release_page(vcpu->pio.guest_pages[i]);
 336                         vcpu->pio.guest_pages[i] = NULL;
 337                 }
 338 }
 339
 340 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 341 {
 342         vcpu_load(vcpu);
 343         kvm_mmu_unload(vcpu);
 344         vcpu_put(vcpu);
 345 }
 346
 347 static void kvm_free_vcpus(struct kvm *kvm)
 348 {
 349         unsigned int i;
 350
 351         /*
 352          * Unpin any mmu pages first.
 353          */
 354         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 355                 if (kvm->vcpus[i])
 356                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
 357         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 358                 if (kvm->vcpus[i]) {
 359                         kvm_x86_ops->vcpu_free(kvm->vcpus[i]);
 360                         kvm->vcpus[i] = NULL;
 361                 }
 362         }
 363
 364 }
 365
 366 static void kvm_destroy_vm(struct kvm *kvm)
 367 {
 368         spin_lock(&kvm_lock);
 369         list_del(&kvm->vm_list);
 370         spin_unlock(&kvm_lock);
 371         kvm_io_bus_destroy(&kvm->pio_bus);
 372         kvm_io_bus_destroy(&kvm->mmio_bus);
 373         kfree(kvm->vpic);
 374         kfree(kvm->vioapic);
 375         kvm_free_vcpus(kvm);
 376         kvm_free_physmem(kvm);
 377         kfree(kvm);
 378 }
 379
 380 static int kvm_vm_release(struct inode *inode, struct file *filp)
 381 {
 382         struct kvm *kvm = filp->private_data;
 383
 384         kvm_destroy_vm(kvm);
 385         return 0;
 386 }
 387
 388 static void inject_gp(struct kvm_vcpu *vcpu)
 389 {
 390         kvm_x86_ops->inject_gp(vcpu, 0);
 391 }
 392
 393 /*
 394  * Load the pae pdptrs.  Return true is they are all valid.
 395  */
 396 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 397 {
 398         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 399         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 400         int i;
 401         int ret;
 402         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
 403
 404         mutex_lock(&vcpu->kvm->lock);
 405         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 406                                   offset * sizeof(u64), sizeof(pdpte));
 407         if (ret < 0) {
 408                 ret = 0;
 409                 goto out;
 410         }
 411         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 412                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 413                         ret = 0;
 414                         goto out;
 415                 }
 416         }
 417         ret = 1;
 418
 419         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
 420 out:
 421         mutex_unlock(&vcpu->kvm->lock);
 422
 423         return ret;
 424 }
 425
 426 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 427 {
 428         if (cr0 & CR0_RESERVED_BITS) {
 429                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 430                        cr0, vcpu->cr0);
 431                 inject_gp(vcpu);
 432                 return;
 433         }
 434
 435         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 436                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 437                 inject_gp(vcpu);
 438                 return;
 439         }
 440
 441         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 442                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 443                        "and a clear PE flag\n");
 444                 inject_gp(vcpu);
 445                 return;
 446         }
 447
 448         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 449 #ifdef CONFIG_X86_64
 450                 if ((vcpu->shadow_efer & EFER_LME)) {
 451                         int cs_db, cs_l;
 452
 453                         if (!is_pae(vcpu)) {
 454                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 455                                        "in long mode while PAE is disabled\n");
 456                                 inject_gp(vcpu);
 457                                 return;
 458                         }
 459                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 460                         if (cs_l) {
 461                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 462                                        "in long mode while CS.L == 1\n");
 463                                 inject_gp(vcpu);
 464                                 return;
 465
 466                         }
 467                 } else
 468 #endif
 469                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 470                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 471                                "reserved bits\n");
 472                         inject_gp(vcpu);
 473                         return;
 474                 }
 475
 476         }
 477
 478         kvm_x86_ops->set_cr0(vcpu, cr0);
 479         vcpu->cr0 = cr0;
 480
 481         mutex_lock(&vcpu->kvm->lock);
 482         kvm_mmu_reset_context(vcpu);
 483         mutex_unlock(&vcpu->kvm->lock);
 484         return;
 485 }
 486 EXPORT_SYMBOL_GPL(set_cr0);
 487
 488 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 489 {
 490         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 491 }
 492 EXPORT_SYMBOL_GPL(lmsw);
 493
 494 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 495 {
 496         if (cr4 & CR4_RESERVED_BITS) {
 497                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 498                 inject_gp(vcpu);
 499                 return;
 500         }
 501
 502         if (is_long_mode(vcpu)) {
 503                 if (!(cr4 & X86_CR4_PAE)) {
 504                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 505                                "in long mode\n");
 506                         inject_gp(vcpu);
 507                         return;
 508                 }
 509         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 510                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 511                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 512                 inject_gp(vcpu);
 513                 return;
 514         }
 515
 516         if (cr4 & X86_CR4_VMXE) {
 517                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 518                 inject_gp(vcpu);
 519                 return;
 520         }
 521         kvm_x86_ops->set_cr4(vcpu, cr4);
 522         vcpu->cr4 = cr4;
 523         mutex_lock(&vcpu->kvm->lock);
 524         kvm_mmu_reset_context(vcpu);
 525         mutex_unlock(&vcpu->kvm->lock);
 526 }
 527 EXPORT_SYMBOL_GPL(set_cr4);
 528
 529 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 530 {
 531         if (is_long_mode(vcpu)) {
 532                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 533                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 534                         inject_gp(vcpu);
 535                         return;
 536                 }
 537         } else {
 538                 if (is_pae(vcpu)) {
 539                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 540                                 printk(KERN_DEBUG
 541                                        "set_cr3: #GP, reserved bits\n");
 542                                 inject_gp(vcpu);
 543                                 return;
 544                         }
 545                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 546                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 547                                        "reserved bits\n");
 548                                 inject_gp(vcpu);
 549                                 return;
 550                         }
 551                 }
 552                 /*
 553                  * We don't check reserved bits in nonpae mode, because
 554                  * this isn't enforced, and VMware depends on this.
 555                  */
 556         }
 557
 558         mutex_lock(&vcpu->kvm->lock);
 559         /*
 560          * Does the new cr3 value map to physical memory? (Note, we
 561          * catch an invalid cr3 even in real-mode, because it would
 562          * cause trouble later on when we turn on paging anyway.)
 563          *
 564          * A real CPU would silently accept an invalid cr3 and would
 565          * attempt to use it - with largely undefined (and often hard
 566          * to debug) behavior on the guest side.
 567          */
 568         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 569                 inject_gp(vcpu);
 570         else {
 571                 vcpu->cr3 = cr3;
 572                 vcpu->mmu.new_cr3(vcpu);
 573         }
 574         mutex_unlock(&vcpu->kvm->lock);
 575 }
 576 EXPORT_SYMBOL_GPL(set_cr3);
 577
 578 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 579 {
 580         if (cr8 & CR8_RESERVED_BITS) {
 581                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 582                 inject_gp(vcpu);
 583                 return;
 584         }
 585         if (irqchip_in_kernel(vcpu->kvm))
 586                 kvm_lapic_set_tpr(vcpu, cr8);
 587         else
 588                 vcpu->cr8 = cr8;
 589 }
 590 EXPORT_SYMBOL_GPL(set_cr8);
 591
 592 unsigned long get_cr8(struct kvm_vcpu *vcpu)
 593 {
 594         if (irqchip_in_kernel(vcpu->kvm))
 595                 return kvm_lapic_get_cr8(vcpu);
 596         else
 597                 return vcpu->cr8;
 598 }
 599 EXPORT_SYMBOL_GPL(get_cr8);
 600
 601 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 602 {
 603         if (irqchip_in_kernel(vcpu->kvm))
 604                 return vcpu->apic_base;
 605         else
 606                 return vcpu->apic_base;
 607 }
 608 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 609
 610 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 611 {
 612         /* TODO: reserve bits check */
 613         if (irqchip_in_kernel(vcpu->kvm))
 614                 kvm_lapic_set_base(vcpu, data);
 615         else
 616                 vcpu->apic_base = data;
 617 }
 618 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 619
 620 void fx_init(struct kvm_vcpu *vcpu)
 621 {
 622         unsigned after_mxcsr_mask;
 623
 624         /* Initialize guest FPU by resetting ours and saving into guest's */
 625         preempt_disable();
 626         fx_save(&vcpu->host_fx_image);
 627         fpu_init();
 628         fx_save(&vcpu->guest_fx_image);
 629         fx_restore(&vcpu->host_fx_image);
 630         preempt_enable();
 631
 632         vcpu->cr0 |= X86_CR0_ET;
 633         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
 634         vcpu->guest_fx_image.mxcsr = 0x1f80;
 635         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
 636                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
 637 }
 638 EXPORT_SYMBOL_GPL(fx_init);
 639
 640 /*
 641  * Allocate some memory and give it an address in the guest physical address
 642  * space.
 643  *
 644  * Discontiguous memory is allowed, mostly for framebuffers.
 645  */
 646 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 647                                           struct
 648                                           kvm_userspace_memory_region *mem,
 649                                           int user_alloc)
 650 {
 651         int r;
 652         gfn_t base_gfn;
 653         unsigned long npages;
 654         unsigned long i;
 655         struct kvm_memory_slot *memslot;
 656         struct kvm_memory_slot old, new;
 657
 658         r = -EINVAL;
 659         /* General sanity checks */
 660         if (mem->memory_size & (PAGE_SIZE - 1))
 661                 goto out;
 662         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 663                 goto out;
 664         if (mem->slot >= KVM_MEMORY_SLOTS)
 665                 goto out;
 666         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 667                 goto out;
 668
 669         memslot = &kvm->memslots[mem->slot];
 670         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 671         npages = mem->memory_size >> PAGE_SHIFT;
 672
 673         if (!npages)
 674                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 675
 676         mutex_lock(&kvm->lock);
 677
 678         new = old = *memslot;
 679
 680         new.base_gfn = base_gfn;
 681         new.npages = npages;
 682         new.flags = mem->flags;
 683
 684         /* Disallow changing a memory slot's size. */
 685         r = -EINVAL;
 686         if (npages && old.npages && npages != old.npages)
 687                 goto out_unlock;
 688
 689         /* Check for overlaps */
 690         r = -EEXIST;
 691         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 692                 struct kvm_memory_slot *s = &kvm->memslots[i];
 693
 694                 if (s == memslot)
 695                         continue;
 696                 if (!((base_gfn + npages <= s->base_gfn) ||
 697                       (base_gfn >= s->base_gfn + s->npages)))
 698                         goto out_unlock;
 699         }
 700
 701         /* Free page dirty bitmap if unneeded */
 702         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 703                 new.dirty_bitmap = NULL;
 704
 705         r = -ENOMEM;
 706
 707         /* Allocate if a slot is being created */
 708         if (npages && !new.rmap) {
 709                 new.rmap = vmalloc(npages * sizeof(struct page *));
 710
 711                 if (!new.rmap)
 712                         goto out_unlock;
 713
 714                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 715
 716                 if (user_alloc)
 717                         new.userspace_addr = mem->userspace_addr;
 718                 else {
 719                         down_write(&current->mm->mmap_sem);
 720                         new.userspace_addr = do_mmap(NULL, 0,
 721                                                      npages * PAGE_SIZE,
 722                                                      PROT_READ | PROT_WRITE,
 723                                                      MAP_SHARED | MAP_ANONYMOUS,
 724                                                      0);
 725                         up_write(&current->mm->mmap_sem);
 726
 727                         if (IS_ERR((void *)new.userspace_addr))
 728                                 goto out_unlock;
 729                 }
 730         }
 731
 732         /* Allocate page dirty bitmap if needed */
 733         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 734                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 735
 736                 new.dirty_bitmap = vmalloc(dirty_bytes);
 737                 if (!new.dirty_bitmap)
 738                         goto out_unlock;
 739                 memset(new.dirty_bitmap, 0, dirty_bytes);
 740         }
 741
 742         if (mem->slot >= kvm->nmemslots)
 743                 kvm->nmemslots = mem->slot + 1;
 744
 745         if (!kvm->n_requested_mmu_pages) {
 746                 unsigned int n_pages;
 747
 748                 if (npages) {
 749                         n_pages = npages * KVM_PERMILLE_MMU_PAGES / 1000;
 750                         kvm_mmu_change_mmu_pages(kvm, kvm->n_alloc_mmu_pages +
 751                                                  n_pages);
 752                 } else {
 753                         unsigned int nr_mmu_pages;
 754
 755                         n_pages = old.npages * KVM_PERMILLE_MMU_PAGES / 1000;
 756                         nr_mmu_pages = kvm->n_alloc_mmu_pages - n_pages;
 757                         nr_mmu_pages = max(nr_mmu_pages,
 758                                         (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
 759                         kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
 760                 }
 761         }
 762
 763         *memslot = new;
 764
 765         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 766         kvm_flush_remote_tlbs(kvm);
 767
 768         mutex_unlock(&kvm->lock);
 769
 770         kvm_free_physmem_slot(&old, &new);
 771         return 0;
 772
 773 out_unlock:
 774         mutex_unlock(&kvm->lock);
 775         kvm_free_physmem_slot(&new, &old);
 776 out:
 777         return r;
 778 }
 779
 780 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
 781                                           u32 kvm_nr_mmu_pages)
 782 {
 783         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
 784                 return -EINVAL;
 785
 786         mutex_lock(&kvm->lock);
 787
 788         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
 789         kvm->n_requested_mmu_pages = kvm_nr_mmu_pages;
 790
 791         mutex_unlock(&kvm->lock);
 792         return 0;
 793 }
 794
 795 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
 796 {
 797         return kvm->n_alloc_mmu_pages;
 798 }
 799
 800 /*
 801  * Get (and clear) the dirty memory log for a memory slot.
 802  */
 803 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 804                                       struct kvm_dirty_log *log)
 805 {
 806         struct kvm_memory_slot *memslot;
 807         int r, i;
 808         int n;
 809         unsigned long any = 0;
 810
 811         mutex_lock(&kvm->lock);
 812
 813         r = -EINVAL;
 814         if (log->slot >= KVM_MEMORY_SLOTS)
 815                 goto out;
 816
 817         memslot = &kvm->memslots[log->slot];
 818         r = -ENOENT;
 819         if (!memslot->dirty_bitmap)
 820                 goto out;
 821
 822         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 823
 824         for (i = 0; !any && i < n/sizeof(long); ++i)
 825                 any = memslot->dirty_bitmap[i];
 826
 827         r = -EFAULT;
 828         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 829                 goto out;
 830
 831         /* If nothing is dirty, don't bother messing with page tables. */
 832         if (any) {
 833                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 834                 kvm_flush_remote_tlbs(kvm);
 835                 memset(memslot->dirty_bitmap, 0, n);
 836         }
 837
 838         r = 0;
 839
 840 out:
 841         mutex_unlock(&kvm->lock);
 842         return r;
 843 }
 844
 845 /*
 846  * Set a new alias region.  Aliases map a portion of physical memory into
 847  * another portion.  This is useful for memory windows, for example the PC
 848  * VGA region.
 849  */
 850 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 851                                          struct kvm_memory_alias *alias)
 852 {
 853         int r, n;
 854         struct kvm_mem_alias *p;
 855
 856         r = -EINVAL;
 857         /* General sanity checks */
 858         if (alias->memory_size & (PAGE_SIZE - 1))
 859                 goto out;
 860         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 861                 goto out;
 862         if (alias->slot >= KVM_ALIAS_SLOTS)
 863                 goto out;
 864         if (alias->guest_phys_addr + alias->memory_size
 865             < alias->guest_phys_addr)
 866                 goto out;
 867         if (alias->target_phys_addr + alias->memory_size
 868             < alias->target_phys_addr)
 869                 goto out;
 870
 871         mutex_lock(&kvm->lock);
 872
 873         p = &kvm->aliases[alias->slot];
 874         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 875         p->npages = alias->memory_size >> PAGE_SHIFT;
 876         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 877
 878         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 879                 if (kvm->aliases[n - 1].npages)
 880                         break;
 881         kvm->naliases = n;
 882
 883         kvm_mmu_zap_all(kvm);
 884
 885         mutex_unlock(&kvm->lock);
 886
 887         return 0;
 888
 889 out:
 890         return r;
 891 }
 892
 893 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 894 {
 895         int r;
 896
 897         r = 0;
 898         switch (chip->chip_id) {
 899         case KVM_IRQCHIP_PIC_MASTER:
 900                 memcpy(&chip->chip.pic,
 901                         &pic_irqchip(kvm)->pics[0],
 902                         sizeof(struct kvm_pic_state));
 903                 break;
 904         case KVM_IRQCHIP_PIC_SLAVE:
 905                 memcpy(&chip->chip.pic,
 906                         &pic_irqchip(kvm)->pics[1],
 907                         sizeof(struct kvm_pic_state));
 908                 break;
 909         case KVM_IRQCHIP_IOAPIC:
 910                 memcpy(&chip->chip.ioapic,
 911                         ioapic_irqchip(kvm),
 912                         sizeof(struct kvm_ioapic_state));
 913                 break;
 914         default:
 915                 r = -EINVAL;
 916                 break;
 917         }
 918         return r;
 919 }
 920
 921 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 922 {
 923         int r;
 924
 925         r = 0;
 926         switch (chip->chip_id) {
 927         case KVM_IRQCHIP_PIC_MASTER:
 928                 memcpy(&pic_irqchip(kvm)->pics[0],
 929                         &chip->chip.pic,
 930                         sizeof(struct kvm_pic_state));
 931                 break;
 932         case KVM_IRQCHIP_PIC_SLAVE:
 933                 memcpy(&pic_irqchip(kvm)->pics[1],
 934                         &chip->chip.pic,
 935                         sizeof(struct kvm_pic_state));
 936                 break;
 937         case KVM_IRQCHIP_IOAPIC:
 938                 memcpy(ioapic_irqchip(kvm),
 939                         &chip->chip.ioapic,
 940                         sizeof(struct kvm_ioapic_state));
 941                 break;
 942         default:
 943                 r = -EINVAL;
 944                 break;
 945         }
 946         kvm_pic_update_irq(pic_irqchip(kvm));
 947         return r;
 948 }
 949
 950 int is_error_page(struct page *page)
 951 {
 952         return page == bad_page;
 953 }
 954 EXPORT_SYMBOL_GPL(is_error_page);
 955
 956 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 957 {
 958         int i;
 959         struct kvm_mem_alias *alias;
 960
 961         for (i = 0; i < kvm->naliases; ++i) {
 962                 alias = &kvm->aliases[i];
 963                 if (gfn >= alias->base_gfn
 964                     && gfn < alias->base_gfn + alias->npages)
 965                         return alias->target_gfn + gfn - alias->base_gfn;
 966         }
 967         return gfn;
 968 }
 969
 970 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 971 {
 972         int i;
 973
 974         for (i = 0; i < kvm->nmemslots; ++i) {
 975                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 976
 977                 if (gfn >= memslot->base_gfn
 978                     && gfn < memslot->base_gfn + memslot->npages)
 979                         return memslot;
 980         }
 981         return NULL;
 982 }
 983
 984 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 985 {
 986         gfn = unalias_gfn(kvm, gfn);
 987         return __gfn_to_memslot(kvm, gfn);
 988 }
 989
 990 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 991 {
 992         struct kvm_memory_slot *slot;
 993         struct page *page[1];
 994         int npages;
 995
 996         gfn = unalias_gfn(kvm, gfn);
 997         slot = __gfn_to_memslot(kvm, gfn);
 998         if (!slot) {
 999                 get_page(bad_page);
1000                 return bad_page;
1001         }
1002
1003         down_read(&current->mm->mmap_sem);
1004         npages = get_user_pages(current, current->mm,
1005                                 slot->userspace_addr
1006                                 + (gfn - slot->base_gfn) * PAGE_SIZE, 1,
1007                                 1, 1, page, NULL);
1008         up_read(&current->mm->mmap_sem);
1009         if (npages != 1) {
1010                 get_page(bad_page);
1011                 return bad_page;
1012         }
1013
1014         return page[0];
1015 }
1016 EXPORT_SYMBOL_GPL(gfn_to_page);
1017
1018 void kvm_release_page(struct page *page)
1019 {
1020         if (!PageReserved(page))
1021                 SetPageDirty(page);
1022         put_page(page);
1023 }
1024 EXPORT_SYMBOL_GPL(kvm_release_page);
1025
1026 static int next_segment(unsigned long len, int offset)
1027 {
1028         if (len > PAGE_SIZE - offset)
1029                 return PAGE_SIZE - offset;
1030         else
1031                 return len;
1032 }
1033
1034 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1035                         int len)
1036 {
1037         void *page_virt;
1038         struct page *page;
1039
1040         page = gfn_to_page(kvm, gfn);
1041         if (is_error_page(page)) {
1042                 kvm_release_page(page);
1043                 return -EFAULT;
1044         }
1045         page_virt = kmap_atomic(page, KM_USER0);
1046
1047         memcpy(data, page_virt + offset, len);
1048
1049         kunmap_atomic(page_virt, KM_USER0);
1050         kvm_release_page(page);
1051         return 0;
1052 }
1053 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1054
1055 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1056 {
1057         gfn_t gfn = gpa >> PAGE_SHIFT;
1058         int seg;
1059         int offset = offset_in_page(gpa);
1060         int ret;
1061
1062         while ((seg = next_segment(len, offset)) != 0) {
1063                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1064                 if (ret < 0)
1065                         return ret;
1066                 offset = 0;
1067                 len -= seg;
1068                 data += seg;
1069                 ++gfn;
1070         }
1071         return 0;
1072 }
1073 EXPORT_SYMBOL_GPL(kvm_read_guest);
1074
1075 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1076                          int offset, int len)
1077 {
1078         void *page_virt;
1079         struct page *page;
1080
1081         page = gfn_to_page(kvm, gfn);
1082         if (is_error_page(page)) {
1083                 kvm_release_page(page);
1084                 return -EFAULT;
1085         }
1086         page_virt = kmap_atomic(page, KM_USER0);
1087
1088         memcpy(page_virt + offset, data, len);
1089
1090         kunmap_atomic(page_virt, KM_USER0);
1091         mark_page_dirty(kvm, gfn);
1092         kvm_release_page(page);
1093         return 0;
1094 }
1095 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1096
1097 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1098                     unsigned long len)
1099 {
1100         gfn_t gfn = gpa >> PAGE_SHIFT;
1101         int seg;
1102         int offset = offset_in_page(gpa);
1103         int ret;
1104
1105         while ((seg = next_segment(len, offset)) != 0) {
1106                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1107                 if (ret < 0)
1108                         return ret;
1109                 offset = 0;
1110                 len -= seg;
1111                 data += seg;
1112                 ++gfn;
1113         }
1114         return 0;
1115 }
1116
1117 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1118 {
1119         void *page_virt;
1120         struct page *page;
1121
1122         page = gfn_to_page(kvm, gfn);
1123         if (is_error_page(page)) {
1124                 kvm_release_page(page);
1125                 return -EFAULT;
1126         }
1127         page_virt = kmap_atomic(page, KM_USER0);
1128
1129         memset(page_virt + offset, 0, len);
1130
1131         kunmap_atomic(page_virt, KM_USER0);
1132         kvm_release_page(page);
1133         return 0;
1134 }
1135 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1136
1137 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1138 {
1139         gfn_t gfn = gpa >> PAGE_SHIFT;
1140         int seg;
1141         int offset = offset_in_page(gpa);
1142         int ret;
1143
1144         while ((seg = next_segment(len, offset)) != 0) {
1145                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1146                 if (ret < 0)
1147                         return ret;
1148                 offset = 0;
1149                 len -= seg;
1150                 ++gfn;
1151         }
1152         return 0;
1153 }
1154 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1155
1156 /* WARNING: Does not work on aliased pages. */
1157 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1158 {
1159         struct kvm_memory_slot *memslot;
1160
1161         memslot = __gfn_to_memslot(kvm, gfn);
1162         if (memslot && memslot->dirty_bitmap) {
1163                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1164
1165                 /* avoid RMW */
1166                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1167                         set_bit(rel_gfn, memslot->dirty_bitmap);
1168         }
1169 }
1170
1171 int emulator_read_std(unsigned long addr,
1172                              void *val,
1173                              unsigned int bytes,
1174                              struct kvm_vcpu *vcpu)
1175 {
1176         void *data = val;
1177
1178         while (bytes) {
1179                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1180                 unsigned offset = addr & (PAGE_SIZE-1);
1181                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1182                 int ret;
1183
1184                 if (gpa == UNMAPPED_GVA)
1185                         return X86EMUL_PROPAGATE_FAULT;
1186                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1187                 if (ret < 0)
1188                         return X86EMUL_UNHANDLEABLE;
1189
1190                 bytes -= tocopy;
1191                 data += tocopy;
1192                 addr += tocopy;
1193         }
1194
1195         return X86EMUL_CONTINUE;
1196 }
1197 EXPORT_SYMBOL_GPL(emulator_read_std);
1198
1199 static int emulator_write_std(unsigned long addr,
1200                               const void *val,
1201                               unsigned int bytes,
1202                               struct kvm_vcpu *vcpu)
1203 {
1204         pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1205         return X86EMUL_UNHANDLEABLE;
1206 }
1207
1208 /*
1209  * Only apic need an MMIO device hook, so shortcut now..
1210  */
1211 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1212                                                 gpa_t addr)
1213 {
1214         struct kvm_io_device *dev;
1215
1216         if (vcpu->apic) {
1217                 dev = &vcpu->apic->dev;
1218                 if (dev->in_range(dev, addr))
1219                         return dev;
1220         }
1221         return NULL;
1222 }
1223
1224 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1225                                                 gpa_t addr)
1226 {
1227         struct kvm_io_device *dev;
1228
1229         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1230         if (dev == NULL)
1231                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1232         return dev;
1233 }
1234
1235 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1236                                                gpa_t addr)
1237 {
1238         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1239 }
1240
1241 static int emulator_read_emulated(unsigned long addr,
1242                                   void *val,
1243                                   unsigned int bytes,
1244                                   struct kvm_vcpu *vcpu)
1245 {
1246         struct kvm_io_device *mmio_dev;
1247         gpa_t                 gpa;
1248
1249         if (vcpu->mmio_read_completed) {
1250                 memcpy(val, vcpu->mmio_data, bytes);
1251                 vcpu->mmio_read_completed = 0;
1252                 return X86EMUL_CONTINUE;
1253         } else if (emulator_read_std(addr, val, bytes, vcpu)
1254                    == X86EMUL_CONTINUE)
1255                 return X86EMUL_CONTINUE;
1256
1257         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1258         if (gpa == UNMAPPED_GVA)
1259                 return X86EMUL_PROPAGATE_FAULT;
1260
1261         /*
1262          * Is this MMIO handled locally?
1263          */
1264         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1265         if (mmio_dev) {
1266                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1267                 return X86EMUL_CONTINUE;
1268         }
1269
1270         vcpu->mmio_needed = 1;
1271         vcpu->mmio_phys_addr = gpa;
1272         vcpu->mmio_size = bytes;
1273         vcpu->mmio_is_write = 0;
1274
1275         return X86EMUL_UNHANDLEABLE;
1276 }
1277
1278 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1279                                const void *val, int bytes)
1280 {
1281         int ret;
1282
1283         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1284         if (ret < 0)
1285                 return 0;
1286         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1287         return 1;
1288 }
1289
1290 static int emulator_write_emulated_onepage(unsigned long addr,
1291                                            const void *val,
1292                                            unsigned int bytes,
1293                                            struct kvm_vcpu *vcpu)
1294 {
1295         struct kvm_io_device *mmio_dev;
1296         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1297
1298         if (gpa == UNMAPPED_GVA) {
1299                 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1300                 return X86EMUL_PROPAGATE_FAULT;
1301         }
1302
1303         if (emulator_write_phys(vcpu, gpa, val, bytes))
1304                 return X86EMUL_CONTINUE;
1305
1306         /*
1307          * Is this MMIO handled locally?
1308          */
1309         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1310         if (mmio_dev) {
1311                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1312                 return X86EMUL_CONTINUE;
1313         }
1314
1315         vcpu->mmio_needed = 1;
1316         vcpu->mmio_phys_addr = gpa;
1317         vcpu->mmio_size = bytes;
1318         vcpu->mmio_is_write = 1;
1319         memcpy(vcpu->mmio_data, val, bytes);
1320
1321         return X86EMUL_CONTINUE;
1322 }
1323
1324 int emulator_write_emulated(unsigned long addr,
1325                                    const void *val,
1326                                    unsigned int bytes,
1327                                    struct kvm_vcpu *vcpu)
1328 {
1329         /* Crossing a page boundary? */
1330         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1331                 int rc, now;
1332
1333                 now = -addr & ~PAGE_MASK;
1334                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1335                 if (rc != X86EMUL_CONTINUE)
1336                         return rc;
1337                 addr += now;
1338                 val += now;
1339                 bytes -= now;
1340         }
1341         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1342 }
1343 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1344
1345 static int emulator_cmpxchg_emulated(unsigned long addr,
1346                                      const void *old,
1347                                      const void *new,
1348                                      unsigned int bytes,
1349                                      struct kvm_vcpu *vcpu)
1350 {
1351         static int reported;
1352
1353         if (!reported) {
1354                 reported = 1;
1355                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1356         }
1357         return emulator_write_emulated(addr, new, bytes, vcpu);
1358 }
1359
1360 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1361 {
1362         return kvm_x86_ops->get_segment_base(vcpu, seg);
1363 }
1364
1365 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1366 {
1367         return X86EMUL_CONTINUE;
1368 }
1369
1370 int emulate_clts(struct kvm_vcpu *vcpu)
1371 {
1372         kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1373         return X86EMUL_CONTINUE;
1374 }
1375
1376 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1377 {
1378         struct kvm_vcpu *vcpu = ctxt->vcpu;
1379
1380         switch (dr) {
1381         case 0 ... 3:
1382                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1383                 return X86EMUL_CONTINUE;
1384         default:
1385                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1386                 return X86EMUL_UNHANDLEABLE;
1387         }
1388 }
1389
1390 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1391 {
1392         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1393         int exception;
1394
1395         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1396         if (exception) {
1397                 /* FIXME: better handling */
1398                 return X86EMUL_UNHANDLEABLE;
1399         }
1400         return X86EMUL_CONTINUE;
1401 }
1402
1403 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1404 {
1405         static int reported;
1406         u8 opcodes[4];
1407         unsigned long rip = vcpu->rip;
1408         unsigned long rip_linear;
1409
1410         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1411
1412         if (reported)
1413                 return;
1414
1415         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1416
1417         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1418                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1419         reported = 1;
1420 }
1421 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1422
1423 struct x86_emulate_ops emulate_ops = {
1424         .read_std            = emulator_read_std,
1425         .write_std           = emulator_write_std,
1426         .read_emulated       = emulator_read_emulated,
1427         .write_emulated      = emulator_write_emulated,
1428         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1429 };
1430
1431 int emulate_instruction(struct kvm_vcpu *vcpu,
1432                         struct kvm_run *run,
1433                         unsigned long cr2,
1434                         u16 error_code,
1435                         int no_decode)
1436 {
1437         int r;
1438
1439         vcpu->mmio_fault_cr2 = cr2;
1440         kvm_x86_ops->cache_regs(vcpu);
1441
1442         vcpu->mmio_is_write = 0;
1443         vcpu->pio.string = 0;
1444
1445         if (!no_decode) {
1446                 int cs_db, cs_l;
1447                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1448
1449                 vcpu->emulate_ctxt.vcpu = vcpu;
1450                 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1451                 vcpu->emulate_ctxt.cr2 = cr2;
1452                 vcpu->emulate_ctxt.mode =
1453                         (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1454                         ? X86EMUL_MODE_REAL : cs_l
1455                         ? X86EMUL_MODE_PROT64 : cs_db
1456                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1457
1458                 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1459                         vcpu->emulate_ctxt.cs_base = 0;
1460                         vcpu->emulate_ctxt.ds_base = 0;
1461                         vcpu->emulate_ctxt.es_base = 0;
1462                         vcpu->emulate_ctxt.ss_base = 0;
1463                 } else {
1464                         vcpu->emulate_ctxt.cs_base =
1465                                         get_segment_base(vcpu, VCPU_SREG_CS);
1466                         vcpu->emulate_ctxt.ds_base =
1467                                         get_segment_base(vcpu, VCPU_SREG_DS);
1468                         vcpu->emulate_ctxt.es_base =
1469                                         get_segment_base(vcpu, VCPU_SREG_ES);
1470                         vcpu->emulate_ctxt.ss_base =
1471                                         get_segment_base(vcpu, VCPU_SREG_SS);
1472                 }
1473
1474                 vcpu->emulate_ctxt.gs_base =
1475                                         get_segment_base(vcpu, VCPU_SREG_GS);
1476                 vcpu->emulate_ctxt.fs_base =
1477                                         get_segment_base(vcpu, VCPU_SREG_FS);
1478
1479                 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
1480                 if (r)  {
1481                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1482                                 return EMULATE_DONE;
1483                         return EMULATE_FAIL;
1484                 }
1485         }
1486
1487         r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1488
1489         if (vcpu->pio.string)
1490                 return EMULATE_DO_MMIO;
1491
1492         if ((r || vcpu->mmio_is_write) && run) {
1493                 run->exit_reason = KVM_EXIT_MMIO;
1494                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1495                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1496                 run->mmio.len = vcpu->mmio_size;
1497                 run->mmio.is_write = vcpu->mmio_is_write;
1498         }
1499
1500         if (r) {
1501                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1502                         return EMULATE_DONE;
1503                 if (!vcpu->mmio_needed) {
1504                         kvm_report_emulation_failure(vcpu, "mmio");
1505                         return EMULATE_FAIL;
1506                 }
1507                 return EMULATE_DO_MMIO;
1508         }
1509
1510         kvm_x86_ops->decache_regs(vcpu);
1511         kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1512
1513         if (vcpu->mmio_is_write) {
1514                 vcpu->mmio_needed = 0;
1515                 return EMULATE_DO_MMIO;
1516         }
1517
1518         return EMULATE_DONE;
1519 }
1520 EXPORT_SYMBOL_GPL(emulate_instruction);
1521
1522 /*
1523  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1524  */
1525 static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1526 {
1527         DECLARE_WAITQUEUE(wait, current);
1528
1529         add_wait_queue(&vcpu->wq, &wait);
1530
1531         /*
1532          * We will block until either an interrupt or a signal wakes us up
1533          */
1534         while (!kvm_cpu_has_interrupt(vcpu)
1535                && !signal_pending(current)
1536                && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
1537                && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
1538                 set_current_state(TASK_INTERRUPTIBLE);
1539                 vcpu_put(vcpu);
1540                 schedule();
1541                 vcpu_load(vcpu);
1542         }
1543
1544         __set_current_state(TASK_RUNNING);
1545         remove_wait_queue(&vcpu->wq, &wait);
1546 }
1547
1548 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1549 {
1550         ++vcpu->stat.halt_exits;
1551         if (irqchip_in_kernel(vcpu->kvm)) {
1552                 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1553                 kvm_vcpu_block(vcpu);
1554                 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1555                         return -EINTR;
1556                 return 1;
1557         } else {
1558                 vcpu->run->exit_reason = KVM_EXIT_HLT;
1559                 return 0;
1560         }
1561 }
1562 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1563
1564 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
1565 {
1566         unsigned long nr, a0, a1, a2, a3, ret;
1567
1568         kvm_x86_ops->cache_regs(vcpu);
1569
1570         nr = vcpu->regs[VCPU_REGS_RAX];
1571         a0 = vcpu->regs[VCPU_REGS_RBX];
1572         a1 = vcpu->regs[VCPU_REGS_RCX];
1573         a2 = vcpu->regs[VCPU_REGS_RDX];
1574         a3 = vcpu->regs[VCPU_REGS_RSI];
1575
1576         if (!is_long_mode(vcpu)) {
1577                 nr &= 0xFFFFFFFF;
1578                 a0 &= 0xFFFFFFFF;
1579                 a1 &= 0xFFFFFFFF;
1580                 a2 &= 0xFFFFFFFF;
1581                 a3 &= 0xFFFFFFFF;
1582         }
1583
1584         switch (nr) {
1585         default:
1586                 ret = -KVM_ENOSYS;
1587                 break;
1588         }
1589         vcpu->regs[VCPU_REGS_RAX] = ret;
1590         kvm_x86_ops->decache_regs(vcpu);
1591         return 0;
1592 }
1593 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1594
1595 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1596 {
1597         char instruction[3];
1598         int ret = 0;
1599
1600         mutex_lock(&vcpu->kvm->lock);
1601
1602         /*
1603          * Blow out the MMU to ensure that no other VCPU has an active mapping
1604          * to ensure that the updated hypercall appears atomically across all
1605          * VCPUs.
1606          */
1607         kvm_mmu_zap_all(vcpu->kvm);
1608
1609         kvm_x86_ops->cache_regs(vcpu);
1610         kvm_x86_ops->patch_hypercall(vcpu, instruction);
1611         if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1612             != X86EMUL_CONTINUE)
1613                 ret = -EFAULT;
1614
1615         mutex_unlock(&vcpu->kvm->lock);
1616
1617         return ret;
1618 }
1619
1620 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1621 {
1622         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1623 }
1624
1625 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1626 {
1627         struct descriptor_table dt = { limit, base };
1628
1629         kvm_x86_ops->set_gdt(vcpu, &dt);
1630 }
1631
1632 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1633 {
1634         struct descriptor_table dt = { limit, base };
1635
1636         kvm_x86_ops->set_idt(vcpu, &dt);
1637 }
1638
1639 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1640                    unsigned long *rflags)
1641 {
1642         lmsw(vcpu, msw);
1643         *rflags = kvm_x86_ops->get_rflags(vcpu);
1644 }
1645
1646 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1647 {
1648         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1649         switch (cr) {
1650         case 0:
1651                 return vcpu->cr0;
1652         case 2:
1653                 return vcpu->cr2;
1654         case 3:
1655                 return vcpu->cr3;
1656         case 4:
1657                 return vcpu->cr4;
1658         default:
1659                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1660                 return 0;
1661         }
1662 }
1663
1664 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1665                      unsigned long *rflags)
1666 {
1667         switch (cr) {
1668         case 0:
1669                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1670                 *rflags = kvm_x86_ops->get_rflags(vcpu);
1671                 break;
1672         case 2:
1673                 vcpu->cr2 = val;
1674                 break;
1675         case 3:
1676                 set_cr3(vcpu, val);
1677                 break;
1678         case 4:
1679                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1680                 break;
1681         default:
1682                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1683         }
1684 }
1685
1686 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1687 {
1688         u64 data;
1689
1690         switch (msr) {
1691         case 0xc0010010: /* SYSCFG */
1692         case 0xc0010015: /* HWCR */
1693         case MSR_IA32_PLATFORM_ID:
1694         case MSR_IA32_P5_MC_ADDR:
1695         case MSR_IA32_P5_MC_TYPE:
1696         case MSR_IA32_MC0_CTL:
1697         case MSR_IA32_MCG_STATUS:
1698         case MSR_IA32_MCG_CAP:
1699         case MSR_IA32_MC0_MISC:
1700         case MSR_IA32_MC0_MISC+4:
1701         case MSR_IA32_MC0_MISC+8:
1702         case MSR_IA32_MC0_MISC+12:
1703         case MSR_IA32_MC0_MISC+16:
1704         case MSR_IA32_UCODE_REV:
1705         case MSR_IA32_PERF_STATUS:
1706         case MSR_IA32_EBL_CR_POWERON:
1707                 /* MTRR registers */
1708         case 0xfe:
1709         case 0x200 ... 0x2ff:
1710                 data = 0;
1711                 break;
1712         case 0xcd: /* fsb frequency */
1713                 data = 3;
1714                 break;
1715         case MSR_IA32_APICBASE:
1716                 data = kvm_get_apic_base(vcpu);
1717                 break;
1718         case MSR_IA32_MISC_ENABLE:
1719                 data = vcpu->ia32_misc_enable_msr;
1720                 break;
1721 #ifdef CONFIG_X86_64
1722         case MSR_EFER:
1723                 data = vcpu->shadow_efer;
1724                 break;
1725 #endif
1726         default:
1727                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1728                 return 1;
1729         }
1730         *pdata = data;
1731         return 0;
1732 }
1733 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1734
1735 /*
1736  * Reads an msr value (of 'msr_index') into 'pdata'.
1737  * Returns 0 on success, non-0 otherwise.
1738  * Assumes vcpu_load() was already called.
1739  */
1740 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1741 {
1742         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1743 }
1744
1745 #ifdef CONFIG_X86_64
1746
1747 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1748 {
1749         if (efer & EFER_RESERVED_BITS) {
1750                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1751                        efer);
1752                 inject_gp(vcpu);
1753                 return;
1754         }
1755
1756         if (is_paging(vcpu)
1757             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1758                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1759                 inject_gp(vcpu);
1760                 return;
1761         }
1762
1763         kvm_x86_ops->set_efer(vcpu, efer);
1764
1765         efer &= ~EFER_LMA;
1766         efer |= vcpu->shadow_efer & EFER_LMA;
1767
1768         vcpu->shadow_efer = efer;
1769 }
1770
1771 #endif
1772
1773 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1774 {
1775         switch (msr) {
1776 #ifdef CONFIG_X86_64
1777         case MSR_EFER:
1778                 set_efer(vcpu, data);
1779                 break;
1780 #endif
1781         case MSR_IA32_MC0_STATUS:
1782                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1783                        __FUNCTION__, data);
1784                 break;
1785         case MSR_IA32_MCG_STATUS:
1786                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1787                         __FUNCTION__, data);
1788                 break;
1789         case MSR_IA32_UCODE_REV:
1790         case MSR_IA32_UCODE_WRITE:
1791         case 0x200 ... 0x2ff: /* MTRRs */
1792                 break;
1793         case MSR_IA32_APICBASE:
1794                 kvm_set_apic_base(vcpu, data);
1795                 break;
1796         case MSR_IA32_MISC_ENABLE:
1797                 vcpu->ia32_misc_enable_msr = data;
1798                 break;
1799         default:
1800                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1801                 return 1;
1802         }
1803         return 0;
1804 }
1805 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1806
1807 /*
1808  * Writes msr value into into the appropriate "register".
1809  * Returns 0 on success, non-0 otherwise.
1810  * Assumes vcpu_load() was already called.
1811  */
1812 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1813 {
1814         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
1815 }
1816
1817 void kvm_resched(struct kvm_vcpu *vcpu)
1818 {
1819         if (!need_resched())
1820                 return;
1821         cond_resched();
1822 }
1823 EXPORT_SYMBOL_GPL(kvm_resched);
1824
1825 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1826 {
1827         int i;
1828         u32 function;
1829         struct kvm_cpuid_entry *e, *best;
1830
1831         kvm_x86_ops->cache_regs(vcpu);
1832         function = vcpu->regs[VCPU_REGS_RAX];
1833         vcpu->regs[VCPU_REGS_RAX] = 0;
1834         vcpu->regs[VCPU_REGS_RBX] = 0;
1835         vcpu->regs[VCPU_REGS_RCX] = 0;
1836         vcpu->regs[VCPU_REGS_RDX] = 0;
1837         best = NULL;
1838         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1839                 e = &vcpu->cpuid_entries[i];
1840                 if (e->function == function) {
1841                         best = e;
1842                         break;
1843                 }
1844                 /*
1845                  * Both basic or both extended?
1846                  */
1847                 if (((e->function ^ function) & 0x80000000) == 0)
1848                         if (!best || e->function > best->function)
1849                                 best = e;
1850         }
1851         if (best) {
1852                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1853                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1854                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1855                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1856         }
1857         kvm_x86_ops->decache_regs(vcpu);
1858         kvm_x86_ops->skip_emulated_instruction(vcpu);
1859 }
1860 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1861
1862 static int pio_copy_data(struct kvm_vcpu *vcpu)
1863 {
1864         void *p = vcpu->pio_data;
1865         void *q;
1866         unsigned bytes;
1867         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1868
1869         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1870                  PAGE_KERNEL);
1871         if (!q) {
1872                 free_pio_guest_pages(vcpu);
1873                 return -ENOMEM;
1874         }
1875         q += vcpu->pio.guest_page_offset;
1876         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1877         if (vcpu->pio.in)
1878                 memcpy(q, p, bytes);
1879         else
1880                 memcpy(p, q, bytes);
1881         q -= vcpu->pio.guest_page_offset;
1882         vunmap(q);
1883         free_pio_guest_pages(vcpu);
1884         return 0;
1885 }
1886
1887 static int complete_pio(struct kvm_vcpu *vcpu)
1888 {
1889         struct kvm_pio_request *io = &vcpu->pio;
1890         long delta;
1891         int r;
1892
1893         kvm_x86_ops->cache_regs(vcpu);
1894
1895         if (!io->string) {
1896                 if (io->in)
1897                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1898                                io->size);
1899         } else {
1900                 if (io->in) {
1901                         r = pio_copy_data(vcpu);
1902                         if (r) {
1903                                 kvm_x86_ops->cache_regs(vcpu);
1904                                 return r;
1905                         }
1906                 }
1907
1908                 delta = 1;
1909                 if (io->rep) {
1910                         delta *= io->cur_count;
1911                         /*
1912                          * The size of the register should really depend on
1913                          * current address size.
1914                          */
1915                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1916                 }
1917                 if (io->down)
1918                         delta = -delta;
1919                 delta *= io->size;
1920                 if (io->in)
1921                         vcpu->regs[VCPU_REGS_RDI] += delta;
1922                 else
1923                         vcpu->regs[VCPU_REGS_RSI] += delta;
1924         }
1925
1926         kvm_x86_ops->decache_regs(vcpu);
1927
1928         io->count -= io->cur_count;
1929         io->cur_count = 0;
1930
1931         return 0;
1932 }
1933
1934 static void kernel_pio(struct kvm_io_device *pio_dev,
1935                        struct kvm_vcpu *vcpu,
1936                        void *pd)
1937 {
1938         /* TODO: String I/O for in kernel device */
1939
1940         mutex_lock(&vcpu->kvm->lock);
1941         if (vcpu->pio.in)
1942                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1943                                   vcpu->pio.size,
1944                                   pd);
1945         else
1946                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1947                                    vcpu->pio.size,
1948                                    pd);
1949         mutex_unlock(&vcpu->kvm->lock);
1950 }
1951
1952 static void pio_string_write(struct kvm_io_device *pio_dev,
1953                              struct kvm_vcpu *vcpu)
1954 {
1955         struct kvm_pio_request *io = &vcpu->pio;
1956         void *pd = vcpu->pio_data;
1957         int i;
1958
1959         mutex_lock(&vcpu->kvm->lock);
1960         for (i = 0; i < io->cur_count; i++) {
1961                 kvm_iodevice_write(pio_dev, io->port,
1962                                    io->size,
1963                                    pd);
1964                 pd += io->size;
1965         }
1966         mutex_unlock(&vcpu->kvm->lock);
1967 }
1968
1969 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1970                   int size, unsigned port)
1971 {
1972         struct kvm_io_device *pio_dev;
1973
1974         vcpu->run->exit_reason = KVM_EXIT_IO;
1975         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1976         vcpu->run->io.size = vcpu->pio.size = size;
1977         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1978         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1979         vcpu->run->io.port = vcpu->pio.port = port;
1980         vcpu->pio.in = in;
1981         vcpu->pio.string = 0;
1982         vcpu->pio.down = 0;
1983         vcpu->pio.guest_page_offset = 0;
1984         vcpu->pio.rep = 0;
1985
1986         kvm_x86_ops->cache_regs(vcpu);
1987         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1988         kvm_x86_ops->decache_regs(vcpu);
1989
1990         kvm_x86_ops->skip_emulated_instruction(vcpu);
1991
1992         pio_dev = vcpu_find_pio_dev(vcpu, port);
1993         if (pio_dev) {
1994                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1995                 complete_pio(vcpu);
1996                 return 1;
1997         }
1998         return 0;
1999 }
2000 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2001
2002 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2003                   int size, unsigned long count, int down,
2004                   gva_t address, int rep, unsigned port)
2005 {
2006         unsigned now, in_page;
2007         int i, ret = 0;
2008         int nr_pages = 1;
2009         struct page *page;
2010         struct kvm_io_device *pio_dev;
2011
2012         vcpu->run->exit_reason = KVM_EXIT_IO;
2013         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2014         vcpu->run->io.size = vcpu->pio.size = size;
2015         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2016         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
2017         vcpu->run->io.port = vcpu->pio.port = port;
2018         vcpu->pio.in = in;
2019         vcpu->pio.string = 1;
2020         vcpu->pio.down = down;
2021         vcpu->pio.guest_page_offset = offset_in_page(address);
2022         vcpu->pio.rep = rep;
2023
2024         if (!count) {
2025                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2026                 return 1;
2027         }
2028
2029         if (!down)
2030                 in_page = PAGE_SIZE - offset_in_page(address);
2031         else
2032                 in_page = offset_in_page(address) + size;
2033         now = min(count, (unsigned long)in_page / size);
2034         if (!now) {
2035                 /*
2036                  * String I/O straddles page boundary.  Pin two guest pages
2037                  * so that we satisfy atomicity constraints.  Do just one
2038                  * transaction to avoid complexity.
2039                  */
2040                 nr_pages = 2;
2041                 now = 1;
2042         }
2043         if (down) {
2044                 /*
2045                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2046                  */
2047                 pr_unimpl(vcpu, "guest string pio down\n");
2048                 inject_gp(vcpu);
2049                 return 1;
2050         }
2051         vcpu->run->io.count = now;
2052         vcpu->pio.cur_count = now;
2053
2054         if (vcpu->pio.cur_count == vcpu->pio.count)
2055                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2056
2057         for (i = 0; i < nr_pages; ++i) {
2058                 mutex_lock(&vcpu->kvm->lock);
2059                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2060                 vcpu->pio.guest_pages[i] = page;
2061                 mutex_unlock(&vcpu->kvm->lock);
2062                 if (!page) {
2063                         inject_gp(vcpu);
2064                         free_pio_guest_pages(vcpu);
2065                         return 1;
2066                 }
2067         }
2068
2069         pio_dev = vcpu_find_pio_dev(vcpu, port);
2070         if (!vcpu->pio.in) {
2071                 /* string PIO write */
2072                 ret = pio_copy_data(vcpu);
2073                 if (ret >= 0 && pio_dev) {
2074                         pio_string_write(pio_dev, vcpu);
2075                         complete_pio(vcpu);
2076                         if (vcpu->pio.count == 0)
2077                                 ret = 1;
2078                 }
2079         } else if (pio_dev)
2080                 pr_unimpl(vcpu, "no string pio read support yet, "
2081                        "port %x size %d count %ld\n",
2082                         port, size, count);
2083
2084         return ret;
2085 }
2086 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2087
2088 /*
2089  * Check if userspace requested an interrupt window, and that the
2090  * interrupt window is open.
2091  *
2092  * No need to exit to userspace if we already have an interrupt queued.
2093  */
2094 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2095                                           struct kvm_run *kvm_run)
2096 {
2097         return (!vcpu->irq_summary &&
2098                 kvm_run->request_interrupt_window &&
2099                 vcpu->interrupt_window_open &&
2100                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2101 }
2102
2103 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2104                               struct kvm_run *kvm_run)
2105 {
2106         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2107         kvm_run->cr8 = get_cr8(vcpu);
2108         kvm_run->apic_base = kvm_get_apic_base(vcpu);
2109         if (irqchip_in_kernel(vcpu->kvm))
2110                 kvm_run->ready_for_interrupt_injection = 1;
2111         else
2112                 kvm_run->ready_for_interrupt_injection =
2113                                         (vcpu->interrupt_window_open &&
2114                                          vcpu->irq_summary == 0);
2115 }
2116
2117 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2118 {
2119         int r;
2120
2121         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
2122                 pr_debug("vcpu %d received sipi with vector # %x\n",
2123                        vcpu->vcpu_id, vcpu->sipi_vector);
2124                 kvm_lapic_reset(vcpu);
2125                 r = kvm_x86_ops->vcpu_reset(vcpu);
2126                 if (r)
2127                         return r;
2128                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2129         }
2130
2131 preempted:
2132         if (vcpu->guest_debug.enabled)
2133                 kvm_x86_ops->guest_debug_pre(vcpu);
2134
2135 again:
2136         r = kvm_mmu_reload(vcpu);
2137         if (unlikely(r))
2138                 goto out;
2139
2140         kvm_inject_pending_timer_irqs(vcpu);
2141
2142         preempt_disable();
2143
2144         kvm_x86_ops->prepare_guest_switch(vcpu);
2145         kvm_load_guest_fpu(vcpu);
2146
2147         local_irq_disable();
2148
2149         if (signal_pending(current)) {
2150                 local_irq_enable();
2151                 preempt_enable();
2152                 r = -EINTR;
2153                 kvm_run->exit_reason = KVM_EXIT_INTR;
2154                 ++vcpu->stat.signal_exits;
2155                 goto out;
2156         }
2157
2158         if (irqchip_in_kernel(vcpu->kvm))
2159                 kvm_x86_ops->inject_pending_irq(vcpu);
2160         else if (!vcpu->mmio_read_completed)
2161                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2162
2163         vcpu->guest_mode = 1;
2164         kvm_guest_enter();
2165
2166         if (vcpu->requests)
2167                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2168                         kvm_x86_ops->tlb_flush(vcpu);
2169
2170         kvm_x86_ops->run(vcpu, kvm_run);
2171
2172         vcpu->guest_mode = 0;
2173         local_irq_enable();
2174
2175         ++vcpu->stat.exits;
2176
2177         /*
2178          * We must have an instruction between local_irq_enable() and
2179          * kvm_guest_exit(), so the timer interrupt isn't delayed by
2180          * the interrupt shadow.  The stat.exits increment will do nicely.
2181          * But we need to prevent reordering, hence this barrier():
2182          */
2183         barrier();
2184
2185         kvm_guest_exit();
2186
2187         preempt_enable();
2188
2189         /*
2190          * Profile KVM exit RIPs:
2191          */
2192         if (unlikely(prof_on == KVM_PROFILING)) {
2193                 kvm_x86_ops->cache_regs(vcpu);
2194                 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
2195         }
2196
2197         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2198
2199         if (r > 0) {
2200                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2201                         r = -EINTR;
2202                         kvm_run->exit_reason = KVM_EXIT_INTR;
2203                         ++vcpu->stat.request_irq_exits;
2204                         goto out;
2205                 }
2206                 if (!need_resched()) {
2207                         ++vcpu->stat.light_exits;
2208                         goto again;
2209                 }
2210         }
2211
2212 out:
2213         if (r > 0) {
2214                 kvm_resched(vcpu);
2215                 goto preempted;
2216         }
2217
2218         post_kvm_run_save(vcpu, kvm_run);
2219
2220         return r;
2221 }
2222
2223
2224 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2225 {
2226         int r;
2227         sigset_t sigsaved;
2228
2229         vcpu_load(vcpu);
2230
2231         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2232                 kvm_vcpu_block(vcpu);
2233                 vcpu_put(vcpu);
2234                 return -EAGAIN;
2235         }
2236
2237         if (vcpu->sigset_active)
2238                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2239
2240         /* re-sync apic's tpr */
2241         if (!irqchip_in_kernel(vcpu->kvm))
2242                 set_cr8(vcpu, kvm_run->cr8);
2243
2244         if (vcpu->pio.cur_count) {
2245                 r = complete_pio(vcpu);
2246                 if (r)
2247                         goto out;
2248         }
2249 #if CONFIG_HAS_IOMEM
2250         if (vcpu->mmio_needed) {
2251                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2252                 vcpu->mmio_read_completed = 1;
2253                 vcpu->mmio_needed = 0;
2254                 r = emulate_instruction(vcpu, kvm_run,
2255                                         vcpu->mmio_fault_cr2, 0, 1);
2256                 if (r == EMULATE_DO_MMIO) {
2257                         /*
2258                          * Read-modify-write.  Back to userspace.
2259                          */
2260                         r = 0;
2261                         goto out;
2262                 }
2263         }
2264 #endif
2265         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2266                 kvm_x86_ops->cache_regs(vcpu);
2267                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2268                 kvm_x86_ops->decache_regs(vcpu);
2269         }
2270
2271         r = __vcpu_run(vcpu, kvm_run);
2272
2273 out:
2274         if (vcpu->sigset_active)
2275                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2276
2277         vcpu_put(vcpu);
2278         return r;
2279 }
2280
2281 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2282                                    struct kvm_regs *regs)
2283 {
2284         vcpu_load(vcpu);
2285
2286         kvm_x86_ops->cache_regs(vcpu);
2287
2288         regs->rax = vcpu->regs[VCPU_REGS_RAX];
2289         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2290         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2291         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2292         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2293         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2294         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2295         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2296 #ifdef CONFIG_X86_64
2297         regs->r8 = vcpu->regs[VCPU_REGS_R8];
2298         regs->r9 = vcpu->regs[VCPU_REGS_R9];
2299         regs->r10 = vcpu->regs[VCPU_REGS_R10];
2300         regs->r11 = vcpu->regs[VCPU_REGS_R11];
2301         regs->r12 = vcpu->regs[VCPU_REGS_R12];
2302         regs->r13 = vcpu->regs[VCPU_REGS_R13];
2303         regs->r14 = vcpu->regs[VCPU_REGS_R14];
2304         regs->r15 = vcpu->regs[VCPU_REGS_R15];
2305 #endif
2306
2307         regs->rip = vcpu->rip;
2308         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2309
2310         /*
2311          * Don't leak debug flags in case they were set for guest debugging
2312          */
2313         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2314                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2315
2316         vcpu_put(vcpu);
2317
2318         return 0;
2319 }
2320
2321 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2322                                    struct kvm_regs *regs)
2323 {
2324         vcpu_load(vcpu);
2325
2326         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2327         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2328         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2329         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2330         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2331         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2332         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2333         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2334 #ifdef CONFIG_X86_64
2335         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2336         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2337         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2338         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2339         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2340         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2341         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2342         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2343 #endif
2344
2345         vcpu->rip = regs->rip;
2346         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2347
2348         kvm_x86_ops->decache_regs(vcpu);
2349
2350         vcpu_put(vcpu);
2351
2352         return 0;
2353 }
2354
2355 static void get_segment(struct kvm_vcpu *vcpu,
2356                         struct kvm_segment *var, int seg)
2357 {
2358         return kvm_x86_ops->get_segment(vcpu, var, seg);
2359 }
2360
2361 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2362                                     struct kvm_sregs *sregs)
2363 {
2364         struct descriptor_table dt;
2365         int pending_vec;
2366
2367         vcpu_load(vcpu);
2368
2369         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2370         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2371         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2372         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2373         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2374         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2375
2376         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2377         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2378
2379         kvm_x86_ops->get_idt(vcpu, &dt);
2380         sregs->idt.limit = dt.limit;
2381         sregs->idt.base = dt.base;
2382         kvm_x86_ops->get_gdt(vcpu, &dt);
2383         sregs->gdt.limit = dt.limit;
2384         sregs->gdt.base = dt.base;
2385
2386         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2387         sregs->cr0 = vcpu->cr0;
2388         sregs->cr2 = vcpu->cr2;
2389         sregs->cr3 = vcpu->cr3;
2390         sregs->cr4 = vcpu->cr4;
2391         sregs->cr8 = get_cr8(vcpu);
2392         sregs->efer = vcpu->shadow_efer;
2393         sregs->apic_base = kvm_get_apic_base(vcpu);
2394
2395         if (irqchip_in_kernel(vcpu->kvm)) {
2396                 memset(sregs->interrupt_bitmap, 0,
2397                        sizeof sregs->interrupt_bitmap);
2398                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2399                 if (pending_vec >= 0)
2400                         set_bit(pending_vec,
2401                                 (unsigned long *)sregs->interrupt_bitmap);
2402         } else
2403                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2404                        sizeof sregs->interrupt_bitmap);
2405
2406         vcpu_put(vcpu);
2407
2408         return 0;
2409 }
2410
2411 static void set_segment(struct kvm_vcpu *vcpu,
2412                         struct kvm_segment *var, int seg)
2413 {
2414         return kvm_x86_ops->set_segment(vcpu, var, seg);
2415 }
2416
2417 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2418                                     struct kvm_sregs *sregs)
2419 {
2420         int mmu_reset_needed = 0;
2421         int i, pending_vec, max_bits;
2422         struct descriptor_table dt;
2423
2424         vcpu_load(vcpu);
2425
2426         dt.limit = sregs->idt.limit;
2427         dt.base = sregs->idt.base;
2428         kvm_x86_ops->set_idt(vcpu, &dt);
2429         dt.limit = sregs->gdt.limit;
2430         dt.base = sregs->gdt.base;
2431         kvm_x86_ops->set_gdt(vcpu, &dt);
2432
2433         vcpu->cr2 = sregs->cr2;
2434         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2435         vcpu->cr3 = sregs->cr3;
2436
2437         set_cr8(vcpu, sregs->cr8);
2438
2439         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2440 #ifdef CONFIG_X86_64
2441         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2442 #endif
2443         kvm_set_apic_base(vcpu, sregs->apic_base);
2444
2445         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2446
2447         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2448         vcpu->cr0 = sregs->cr0;
2449         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2450
2451         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2452         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2453         if (!is_long_mode(vcpu) && is_pae(vcpu))
2454                 load_pdptrs(vcpu, vcpu->cr3);
2455
2456         if (mmu_reset_needed)
2457                 kvm_mmu_reset_context(vcpu);
2458
2459         if (!irqchip_in_kernel(vcpu->kvm)) {
2460                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2461                        sizeof vcpu->irq_pending);
2462                 vcpu->irq_summary = 0;
2463                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2464                         if (vcpu->irq_pending[i])
2465                                 __set_bit(i, &vcpu->irq_summary);
2466         } else {
2467                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2468                 pending_vec = find_first_bit(
2469                         (const unsigned long *)sregs->interrupt_bitmap,
2470                         max_bits);
2471                 /* Only pending external irq is handled here */
2472                 if (pending_vec < max_bits) {
2473                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2474                         pr_debug("Set back pending irq %d\n",
2475                                  pending_vec);
2476                 }
2477         }
2478
2479         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2480         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2481         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2482         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2483         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2484         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2485
2486         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2487         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2488
2489         vcpu_put(vcpu);
2490
2491         return 0;
2492 }
2493
2494 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2495 {
2496         struct kvm_segment cs;
2497
2498         get_segment(vcpu, &cs, VCPU_SREG_CS);
2499         *db = cs.db;
2500         *l = cs.l;
2501 }
2502 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2503
2504 /*
2505  * Translate a guest virtual address to a guest physical address.
2506  */
2507 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2508                                     struct kvm_translation *tr)
2509 {
2510         unsigned long vaddr = tr->linear_address;
2511         gpa_t gpa;
2512
2513         vcpu_load(vcpu);
2514         mutex_lock(&vcpu->kvm->lock);
2515         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2516         tr->physical_address = gpa;
2517         tr->valid = gpa != UNMAPPED_GVA;
2518         tr->writeable = 1;
2519         tr->usermode = 0;
2520         mutex_unlock(&vcpu->kvm->lock);
2521         vcpu_put(vcpu);
2522
2523         return 0;
2524 }
2525
2526 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2527                                     struct kvm_interrupt *irq)
2528 {
2529         if (irq->irq < 0 || irq->irq >= 256)
2530                 return -EINVAL;
2531         if (irqchip_in_kernel(vcpu->kvm))
2532                 return -ENXIO;
2533         vcpu_load(vcpu);
2534
2535         set_bit(irq->irq, vcpu->irq_pending);
2536         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2537
2538         vcpu_put(vcpu);
2539
2540         return 0;
2541 }
2542
2543 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2544                                       struct kvm_debug_guest *dbg)
2545 {
2546         int r;
2547
2548         vcpu_load(vcpu);
2549
2550         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2551
2552         vcpu_put(vcpu);
2553
2554         return r;
2555 }
2556
2557 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2558                                     unsigned long address,
2559                                     int *type)
2560 {
2561         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2562         unsigned long pgoff;
2563         struct page *page;
2564
2565         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2566         if (pgoff == 0)
2567                 page = virt_to_page(vcpu->run);
2568         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2569                 page = virt_to_page(vcpu->pio_data);
2570         else
2571                 return NOPAGE_SIGBUS;
2572         get_page(page);
2573         if (type != NULL)
2574                 *type = VM_FAULT_MINOR;
2575
2576         return page;
2577 }
2578
2579 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2580         .nopage = kvm_vcpu_nopage,
2581 };
2582
2583 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2584 {
2585         vma->vm_ops = &kvm_vcpu_vm_ops;
2586         return 0;
2587 }
2588
2589 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2590 {
2591         struct kvm_vcpu *vcpu = filp->private_data;
2592
2593         fput(vcpu->kvm->filp);
2594         return 0;
2595 }
2596
2597 static struct file_operations kvm_vcpu_fops = {
2598         .release        = kvm_vcpu_release,
2599         .unlocked_ioctl = kvm_vcpu_ioctl,
2600         .compat_ioctl   = kvm_vcpu_ioctl,
2601         .mmap           = kvm_vcpu_mmap,
2602 };
2603
2604 /*
2605  * Allocates an inode for the vcpu.
2606  */
2607 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2608 {
2609         int fd, r;
2610         struct inode *inode;
2611         struct file *file;
2612
2613         r = anon_inode_getfd(&fd, &inode, &file,
2614                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2615         if (r)
2616                 return r;
2617         atomic_inc(&vcpu->kvm->filp->f_count);
2618         return fd;
2619 }
2620
2621 /*
2622  * Creates some virtual cpus.  Good luck creating more than one.
2623  */
2624 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2625 {
2626         int r;
2627         struct kvm_vcpu *vcpu;
2628
2629         if (!valid_vcpu(n))
2630                 return -EINVAL;
2631
2632         vcpu = kvm_x86_ops->vcpu_create(kvm, n);
2633         if (IS_ERR(vcpu))
2634                 return PTR_ERR(vcpu);
2635
2636         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2637
2638         /* We do fxsave: this must be aligned. */
2639         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2640
2641         vcpu_load(vcpu);
2642         r = kvm_x86_ops->vcpu_reset(vcpu);
2643         if (r == 0)
2644                 r = kvm_mmu_setup(vcpu);
2645         vcpu_put(vcpu);
2646         if (r < 0)
2647                 goto free_vcpu;
2648
2649         mutex_lock(&kvm->lock);
2650         if (kvm->vcpus[n]) {
2651                 r = -EEXIST;
2652                 mutex_unlock(&kvm->lock);
2653                 goto mmu_unload;
2654         }
2655         kvm->vcpus[n] = vcpu;
2656         mutex_unlock(&kvm->lock);
2657
2658         /* Now it's all set up, let userspace reach it */
2659         r = create_vcpu_fd(vcpu);
2660         if (r < 0)
2661                 goto unlink;
2662         return r;
2663
2664 unlink:
2665         mutex_lock(&kvm->lock);
2666         kvm->vcpus[n] = NULL;
2667         mutex_unlock(&kvm->lock);
2668
2669 mmu_unload:
2670         vcpu_load(vcpu);
2671         kvm_mmu_unload(vcpu);
2672         vcpu_put(vcpu);
2673
2674 free_vcpu:
2675         kvm_x86_ops->vcpu_free(vcpu);
2676         return r;
2677 }
2678
2679 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2680 {
2681         if (sigset) {
2682                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2683                 vcpu->sigset_active = 1;
2684                 vcpu->sigset = *sigset;
2685         } else
2686                 vcpu->sigset_active = 0;
2687         return 0;
2688 }
2689
2690 /*
2691  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2692  * we have asm/x86/processor.h
2693  */
2694 struct fxsave {
2695         u16     cwd;
2696         u16     swd;
2697         u16     twd;
2698         u16     fop;
2699         u64     rip;
2700         u64     rdp;
2701         u32     mxcsr;
2702         u32     mxcsr_mask;
2703         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2704 #ifdef CONFIG_X86_64
2705         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2706 #else
2707         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2708 #endif
2709 };
2710
2711 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2712 {
2713         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2714
2715         vcpu_load(vcpu);
2716
2717         memcpy(fpu->fpr, fxsave->st_space, 128);
2718         fpu->fcw = fxsave->cwd;
2719         fpu->fsw = fxsave->swd;
2720         fpu->ftwx = fxsave->twd;
2721         fpu->last_opcode = fxsave->fop;
2722         fpu->last_ip = fxsave->rip;
2723         fpu->last_dp = fxsave->rdp;
2724         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2725
2726         vcpu_put(vcpu);
2727
2728         return 0;
2729 }
2730
2731 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2732 {
2733         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2734
2735         vcpu_load(vcpu);
2736
2737         memcpy(fxsave->st_space, fpu->fpr, 128);
2738         fxsave->cwd = fpu->fcw;
2739         fxsave->swd = fpu->fsw;
2740         fxsave->twd = fpu->ftwx;
2741         fxsave->fop = fpu->last_opcode;
2742         fxsave->rip = fpu->last_ip;
2743         fxsave->rdp = fpu->last_dp;
2744         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2745
2746         vcpu_put(vcpu);
2747
2748         return 0;
2749 }
2750
2751 static long kvm_vcpu_ioctl(struct file *filp,
2752                            unsigned int ioctl, unsigned long arg)
2753 {
2754         struct kvm_vcpu *vcpu = filp->private_data;
2755         void __user *argp = (void __user *)arg;
2756         int r;
2757
2758         switch (ioctl) {
2759         case KVM_RUN:
2760                 r = -EINVAL;
2761                 if (arg)
2762                         goto out;
2763                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2764                 break;
2765         case KVM_GET_REGS: {
2766                 struct kvm_regs kvm_regs;
2767
2768                 memset(&kvm_regs, 0, sizeof kvm_regs);
2769                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2770                 if (r)
2771                         goto out;
2772                 r = -EFAULT;
2773                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2774                         goto out;
2775                 r = 0;
2776                 break;
2777         }
2778         case KVM_SET_REGS: {
2779                 struct kvm_regs kvm_regs;
2780
2781                 r = -EFAULT;
2782                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2783                         goto out;
2784                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2785                 if (r)
2786                         goto out;
2787                 r = 0;
2788                 break;
2789         }
2790         case KVM_GET_SREGS: {
2791                 struct kvm_sregs kvm_sregs;
2792
2793                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2794                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2795                 if (r)
2796                         goto out;
2797                 r = -EFAULT;
2798                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2799                         goto out;
2800                 r = 0;
2801                 break;
2802         }
2803         case KVM_SET_SREGS: {
2804                 struct kvm_sregs kvm_sregs;
2805
2806                 r = -EFAULT;
2807                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2808                         goto out;
2809                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2810                 if (r)
2811                         goto out;
2812                 r = 0;
2813                 break;
2814         }
2815         case KVM_TRANSLATE: {
2816                 struct kvm_translation tr;
2817
2818                 r = -EFAULT;
2819                 if (copy_from_user(&tr, argp, sizeof tr))
2820                         goto out;
2821                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2822                 if (r)
2823                         goto out;
2824                 r = -EFAULT;
2825                 if (copy_to_user(argp, &tr, sizeof tr))
2826                         goto out;
2827                 r = 0;
2828                 break;
2829         }
2830         case KVM_INTERRUPT: {
2831                 struct kvm_interrupt irq;
2832
2833                 r = -EFAULT;
2834                 if (copy_from_user(&irq, argp, sizeof irq))
2835                         goto out;
2836                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2837                 if (r)
2838                         goto out;
2839                 r = 0;
2840                 break;
2841         }
2842         case KVM_DEBUG_GUEST: {
2843                 struct kvm_debug_guest dbg;
2844
2845                 r = -EFAULT;
2846                 if (copy_from_user(&dbg, argp, sizeof dbg))
2847                         goto out;
2848                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2849                 if (r)
2850                         goto out;
2851                 r = 0;
2852                 break;
2853         }
2854         case KVM_SET_SIGNAL_MASK: {
2855                 struct kvm_signal_mask __user *sigmask_arg = argp;
2856                 struct kvm_signal_mask kvm_sigmask;
2857                 sigset_t sigset, *p;
2858
2859                 p = NULL;
2860                 if (argp) {
2861                         r = -EFAULT;
2862                         if (copy_from_user(&kvm_sigmask, argp,
2863                                            sizeof kvm_sigmask))
2864                                 goto out;
2865                         r = -EINVAL;
2866                         if (kvm_sigmask.len != sizeof sigset)
2867                                 goto out;
2868                         r = -EFAULT;
2869                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2870                                            sizeof sigset))
2871                                 goto out;
2872                         p = &sigset;
2873                 }
2874                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2875                 break;
2876         }
2877         case KVM_GET_FPU: {
2878                 struct kvm_fpu fpu;
2879
2880                 memset(&fpu, 0, sizeof fpu);
2881                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2882                 if (r)
2883                         goto out;
2884                 r = -EFAULT;
2885                 if (copy_to_user(argp, &fpu, sizeof fpu))
2886                         goto out;
2887                 r = 0;
2888                 break;
2889         }
2890         case KVM_SET_FPU: {
2891                 struct kvm_fpu fpu;
2892
2893                 r = -EFAULT;
2894                 if (copy_from_user(&fpu, argp, sizeof fpu))
2895                         goto out;
2896                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2897                 if (r)
2898                         goto out;
2899                 r = 0;
2900                 break;
2901         }
2902         default:
2903                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2904         }
2905 out:
2906         return r;
2907 }
2908
2909 static long kvm_vm_ioctl(struct file *filp,
2910                            unsigned int ioctl, unsigned long arg)
2911 {
2912         struct kvm *kvm = filp->private_data;
2913         void __user *argp = (void __user *)arg;
2914         int r = -EINVAL;
2915
2916         switch (ioctl) {
2917         case KVM_CREATE_VCPU:
2918                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2919                 if (r < 0)
2920                         goto out;
2921                 break;
2922         case KVM_SET_MEMORY_REGION: {
2923                 struct kvm_memory_region kvm_mem;
2924                 struct kvm_userspace_memory_region kvm_userspace_mem;
2925
2926                 r = -EFAULT;
2927                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2928                         goto out;
2929                 kvm_userspace_mem.slot = kvm_mem.slot;
2930                 kvm_userspace_mem.flags = kvm_mem.flags;
2931                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2932                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2933                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2934                 if (r)
2935                         goto out;
2936                 break;
2937         }
2938         case KVM_SET_USER_MEMORY_REGION: {
2939                 struct kvm_userspace_memory_region kvm_userspace_mem;
2940
2941                 r = -EFAULT;
2942                 if (copy_from_user(&kvm_userspace_mem, argp,
2943                                                 sizeof kvm_userspace_mem))
2944                         goto out;
2945
2946                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2947                 if (r)
2948                         goto out;
2949                 break;
2950         }
2951         case KVM_SET_NR_MMU_PAGES:
2952                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2953                 if (r)
2954                         goto out;
2955                 break;
2956         case KVM_GET_NR_MMU_PAGES:
2957                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2958                 break;
2959         case KVM_GET_DIRTY_LOG: {
2960                 struct kvm_dirty_log log;
2961
2962                 r = -EFAULT;
2963                 if (copy_from_user(&log, argp, sizeof log))
2964                         goto out;
2965                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2966                 if (r)
2967                         goto out;
2968                 break;
2969         }
2970         case KVM_SET_MEMORY_ALIAS: {
2971                 struct kvm_memory_alias alias;
2972
2973                 r = -EFAULT;
2974                 if (copy_from_user(&alias, argp, sizeof alias))
2975                         goto out;
2976                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2977                 if (r)
2978                         goto out;
2979                 break;
2980         }
2981         case KVM_CREATE_IRQCHIP:
2982                 r = -ENOMEM;
2983                 kvm->vpic = kvm_create_pic(kvm);
2984                 if (kvm->vpic) {
2985                         r = kvm_ioapic_init(kvm);
2986                         if (r) {
2987                                 kfree(kvm->vpic);
2988                                 kvm->vpic = NULL;
2989                                 goto out;
2990                         }
2991                 } else
2992                         goto out;
2993                 break;
2994         case KVM_IRQ_LINE: {
2995                 struct kvm_irq_level irq_event;
2996
2997                 r = -EFAULT;
2998                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2999                         goto out;
3000                 if (irqchip_in_kernel(kvm)) {
3001                         mutex_lock(&kvm->lock);
3002                         if (irq_event.irq < 16)
3003                                 kvm_pic_set_irq(pic_irqchip(kvm),
3004                                         irq_event.irq,
3005                                         irq_event.level);
3006                         kvm_ioapic_set_irq(kvm->vioapic,
3007                                         irq_event.irq,
3008                                         irq_event.level);
3009                         mutex_unlock(&kvm->lock);
3010                         r = 0;
3011                 }
3012                 break;
3013         }
3014         case KVM_GET_IRQCHIP: {
3015                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3016                 struct kvm_irqchip chip;
3017
3018                 r = -EFAULT;
3019                 if (copy_from_user(&chip, argp, sizeof chip))
3020                         goto out;
3021                 r = -ENXIO;
3022                 if (!irqchip_in_kernel(kvm))
3023                         goto out;
3024                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
3025                 if (r)
3026                         goto out;
3027                 r = -EFAULT;
3028                 if (copy_to_user(argp, &chip, sizeof chip))
3029                         goto out;
3030                 r = 0;
3031                 break;
3032         }
3033         case KVM_SET_IRQCHIP: {
3034                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3035                 struct kvm_irqchip chip;
3036
3037                 r = -EFAULT;
3038                 if (copy_from_user(&chip, argp, sizeof chip))
3039                         goto out;
3040                 r = -ENXIO;
3041                 if (!irqchip_in_kernel(kvm))
3042                         goto out;
3043                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
3044                 if (r)
3045                         goto out;
3046                 r = 0;
3047                 break;
3048         }
3049         default:
3050                 ;
3051         }
3052 out:
3053         return r;
3054 }
3055
3056 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
3057                                   unsigned long address,
3058                                   int *type)
3059 {
3060         struct kvm *kvm = vma->vm_file->private_data;
3061         unsigned long pgoff;
3062         struct page *page;
3063
3064         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
3065         page = gfn_to_page(kvm, pgoff);
3066         if (is_error_page(page)) {
3067                 kvm_release_page(page);
3068                 return NOPAGE_SIGBUS;
3069         }
3070         if (type != NULL)
3071                 *type = VM_FAULT_MINOR;
3072
3073         return page;
3074 }
3075
3076 static struct vm_operations_struct kvm_vm_vm_ops = {
3077         .nopage = kvm_vm_nopage,
3078 };
3079
3080 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
3081 {
3082         vma->vm_ops = &kvm_vm_vm_ops;
3083         return 0;
3084 }
3085
3086 static struct file_operations kvm_vm_fops = {
3087         .release        = kvm_vm_release,
3088         .unlocked_ioctl = kvm_vm_ioctl,
3089         .compat_ioctl   = kvm_vm_ioctl,
3090         .mmap           = kvm_vm_mmap,
3091 };
3092
3093 static int kvm_dev_ioctl_create_vm(void)
3094 {
3095         int fd, r;
3096         struct inode *inode;
3097         struct file *file;
3098         struct kvm *kvm;
3099
3100         kvm = kvm_create_vm();
3101         if (IS_ERR(kvm))
3102                 return PTR_ERR(kvm);
3103         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
3104         if (r) {
3105                 kvm_destroy_vm(kvm);
3106                 return r;
3107         }
3108
3109         kvm->filp = file;
3110
3111         return fd;
3112 }
3113
3114 static long kvm_dev_ioctl(struct file *filp,
3115                           unsigned int ioctl, unsigned long arg)
3116 {
3117         void __user *argp = (void __user *)arg;
3118         long r = -EINVAL;
3119
3120         switch (ioctl) {
3121         case KVM_GET_API_VERSION:
3122                 r = -EINVAL;
3123                 if (arg)
3124                         goto out;
3125                 r = KVM_API_VERSION;
3126                 break;
3127         case KVM_CREATE_VM:
3128                 r = -EINVAL;
3129                 if (arg)
3130                         goto out;
3131                 r = kvm_dev_ioctl_create_vm();
3132                 break;
3133         case KVM_CHECK_EXTENSION: {
3134                 int ext = (long)argp;
3135
3136                 switch (ext) {
3137                 case KVM_CAP_IRQCHIP:
3138                 case KVM_CAP_HLT:
3139                 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3140                 case KVM_CAP_USER_MEMORY:
3141                         r = 1;
3142                         break;
3143                 default:
3144                         r = 0;
3145                         break;
3146                 }
3147                 break;
3148         }
3149         case KVM_GET_VCPU_MMAP_SIZE:
3150                 r = -EINVAL;
3151                 if (arg)
3152                         goto out;
3153                 r = 2 * PAGE_SIZE;
3154                 break;
3155         default:
3156                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3157         }
3158 out:
3159         return r;
3160 }
3161
3162 static struct file_operations kvm_chardev_ops = {
3163         .unlocked_ioctl = kvm_dev_ioctl,
3164         .compat_ioctl   = kvm_dev_ioctl,
3165 };
3166
3167 static struct miscdevice kvm_dev = {
3168         KVM_MINOR,
3169         "kvm",
3170         &kvm_chardev_ops,
3171 };
3172
3173 /*
3174  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
3175  * cached on it.
3176  */
3177 static void decache_vcpus_on_cpu(int cpu)
3178 {
3179         struct kvm *vm;
3180         struct kvm_vcpu *vcpu;
3181         int i;
3182
3183         spin_lock(&kvm_lock);
3184         list_for_each_entry(vm, &vm_list, vm_list)
3185                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3186                         vcpu = vm->vcpus[i];
3187                         if (!vcpu)
3188                                 continue;
3189                         /*
3190                          * If the vcpu is locked, then it is running on some
3191                          * other cpu and therefore it is not cached on the
3192                          * cpu in question.
3193                          *
3194                          * If it's not locked, check the last cpu it executed
3195                          * on.
3196                          */
3197                         if (mutex_trylock(&vcpu->mutex)) {
3198                                 if (vcpu->cpu == cpu) {
3199                                         kvm_x86_ops->vcpu_decache(vcpu);
3200                                         vcpu->cpu = -1;
3201                                 }
3202                                 mutex_unlock(&vcpu->mutex);
3203                         }
3204                 }
3205         spin_unlock(&kvm_lock);
3206 }
3207
3208 static void hardware_enable(void *junk)
3209 {
3210         int cpu = raw_smp_processor_id();
3211
3212         if (cpu_isset(cpu, cpus_hardware_enabled))
3213                 return;
3214         cpu_set(cpu, cpus_hardware_enabled);
3215         kvm_x86_ops->hardware_enable(NULL);
3216 }
3217
3218 static void hardware_disable(void *junk)
3219 {
3220         int cpu = raw_smp_processor_id();
3221
3222         if (!cpu_isset(cpu, cpus_hardware_enabled))
3223                 return;
3224         cpu_clear(cpu, cpus_hardware_enabled);
3225         decache_vcpus_on_cpu(cpu);
3226         kvm_x86_ops->hardware_disable(NULL);
3227 }
3228
3229 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3230                            void *v)
3231 {
3232         int cpu = (long)v;
3233
3234         switch (val) {
3235         case CPU_DYING:
3236         case CPU_DYING_FROZEN:
3237                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3238                        cpu);
3239                 hardware_disable(NULL);
3240                 break;
3241         case CPU_UP_CANCELED:
3242         case CPU_UP_CANCELED_FROZEN:
3243                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3244                        cpu);
3245                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3246                 break;
3247         case CPU_ONLINE:
3248         case CPU_ONLINE_FROZEN:
3249                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3250                        cpu);
3251                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3252                 break;
3253         }
3254         return NOTIFY_OK;
3255 }
3256
3257 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3258                       void *v)
3259 {
3260         if (val == SYS_RESTART) {
3261                 /*
3262                  * Some (well, at least mine) BIOSes hang on reboot if
3263                  * in vmx root mode.
3264                  */
3265                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3266                 on_each_cpu(hardware_disable, NULL, 0, 1);
3267         }
3268         return NOTIFY_OK;
3269 }
3270
3271 static struct notifier_block kvm_reboot_notifier = {
3272         .notifier_call = kvm_reboot,
3273         .priority = 0,
3274 };
3275
3276 void kvm_io_bus_init(struct kvm_io_bus *bus)
3277 {
3278         memset(bus, 0, sizeof(*bus));
3279 }
3280
3281 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3282 {
3283         int i;
3284
3285         for (i = 0; i < bus->dev_count; i++) {
3286                 struct kvm_io_device *pos = bus->devs[i];
3287
3288                 kvm_iodevice_destructor(pos);
3289         }
3290 }
3291
3292 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3293 {
3294         int i;
3295
3296         for (i = 0; i < bus->dev_count; i++) {
3297                 struct kvm_io_device *pos = bus->devs[i];
3298
3299                 if (pos->in_range(pos, addr))
3300                         return pos;
3301         }
3302
3303         return NULL;
3304 }
3305
3306 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3307 {
3308         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3309
3310         bus->devs[bus->dev_count++] = dev;
3311 }
3312
3313 static struct notifier_block kvm_cpu_notifier = {
3314         .notifier_call = kvm_cpu_hotplug,
3315         .priority = 20, /* must be > scheduler priority */
3316 };
3317
3318 static u64 stat_get(void *_offset)
3319 {
3320         unsigned offset = (long)_offset;
3321         u64 total = 0;
3322         struct kvm *kvm;
3323         struct kvm_vcpu *vcpu;
3324         int i;
3325
3326         spin_lock(&kvm_lock);
3327         list_for_each_entry(kvm, &vm_list, vm_list)
3328                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3329                         vcpu = kvm->vcpus[i];
3330                         if (vcpu)
3331                                 total += *(u32 *)((void *)vcpu + offset);
3332                 }
3333         spin_unlock(&kvm_lock);
3334         return total;
3335 }
3336
3337 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3338
3339 static __init void kvm_init_debug(void)
3340 {
3341         struct kvm_stats_debugfs_item *p;
3342
3343         debugfs_dir = debugfs_create_dir("kvm", NULL);
3344         for (p = debugfs_entries; p->name; ++p)
3345                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3346                                                 (void *)(long)p->offset,
3347                                                 &stat_fops);
3348 }
3349
3350 static void kvm_exit_debug(void)
3351 {
3352         struct kvm_stats_debugfs_item *p;
3353
3354         for (p = debugfs_entries; p->name; ++p)
3355                 debugfs_remove(p->dentry);
3356         debugfs_remove(debugfs_dir);
3357 }
3358
3359 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3360 {
3361         hardware_disable(NULL);
3362         return 0;
3363 }
3364
3365 static int kvm_resume(struct sys_device *dev)
3366 {
3367         hardware_enable(NULL);
3368         return 0;
3369 }
3370
3371 static struct sysdev_class kvm_sysdev_class = {
3372         .name = "kvm",
3373         .suspend = kvm_suspend,
3374         .resume = kvm_resume,
3375 };
3376
3377 static struct sys_device kvm_sysdev = {
3378         .id = 0,
3379         .cls = &kvm_sysdev_class,
3380 };
3381
3382 struct page *bad_page;
3383
3384 static inline
3385 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3386 {
3387         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3388 }
3389
3390 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3391 {
3392         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3393
3394         kvm_x86_ops->vcpu_load(vcpu, cpu);
3395 }
3396
3397 static void kvm_sched_out(struct preempt_notifier *pn,
3398                           struct task_struct *next)
3399 {
3400         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3401
3402         kvm_x86_ops->vcpu_put(vcpu);
3403 }
3404
3405 int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
3406                   struct module *module)
3407 {
3408         int r;
3409         int cpu;
3410
3411         if (kvm_x86_ops) {
3412                 printk(KERN_ERR "kvm: already loaded the other module\n");
3413                 return -EEXIST;
3414         }
3415
3416         if (!ops->cpu_has_kvm_support()) {
3417                 printk(KERN_ERR "kvm: no hardware support\n");
3418                 return -EOPNOTSUPP;
3419         }
3420         if (ops->disabled_by_bios()) {
3421                 printk(KERN_ERR "kvm: disabled by bios\n");
3422                 return -EOPNOTSUPP;
3423         }
3424
3425         kvm_x86_ops = ops;
3426
3427         r = kvm_x86_ops->hardware_setup();
3428         if (r < 0)
3429                 goto out;
3430
3431         for_each_online_cpu(cpu) {
3432                 smp_call_function_single(cpu,
3433                                 kvm_x86_ops->check_processor_compatibility,
3434                                 &r, 0, 1);
3435                 if (r < 0)
3436                         goto out_free_0;
3437         }
3438
3439         on_each_cpu(hardware_enable, NULL, 0, 1);
3440         r = register_cpu_notifier(&kvm_cpu_notifier);
3441         if (r)
3442                 goto out_free_1;
3443         register_reboot_notifier(&kvm_reboot_notifier);
3444
3445         r = sysdev_class_register(&kvm_sysdev_class);
3446         if (r)
3447                 goto out_free_2;
3448
3449         r = sysdev_register(&kvm_sysdev);
3450         if (r)
3451                 goto out_free_3;
3452
3453         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3454         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3455                                            __alignof__(struct kvm_vcpu), 0, 0);
3456         if (!kvm_vcpu_cache) {
3457                 r = -ENOMEM;
3458                 goto out_free_4;
3459         }
3460
3461         kvm_chardev_ops.owner = module;
3462
3463         r = misc_register(&kvm_dev);
3464         if (r) {
3465                 printk(KERN_ERR "kvm: misc device register failed\n");
3466                 goto out_free;
3467         }
3468
3469         kvm_preempt_ops.sched_in = kvm_sched_in;
3470         kvm_preempt_ops.sched_out = kvm_sched_out;
3471
3472         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3473
3474         return 0;
3475
3476 out_free:
3477         kmem_cache_destroy(kvm_vcpu_cache);
3478 out_free_4:
3479         sysdev_unregister(&kvm_sysdev);
3480 out_free_3:
3481         sysdev_class_unregister(&kvm_sysdev_class);
3482 out_free_2:
3483         unregister_reboot_notifier(&kvm_reboot_notifier);
3484         unregister_cpu_notifier(&kvm_cpu_notifier);
3485 out_free_1:
3486         on_each_cpu(hardware_disable, NULL, 0, 1);
3487 out_free_0:
3488         kvm_x86_ops->hardware_unsetup();
3489 out:
3490         kvm_x86_ops = NULL;
3491         return r;
3492 }
3493 EXPORT_SYMBOL_GPL(kvm_init_x86);
3494
3495 void kvm_exit_x86(void)
3496 {
3497         misc_deregister(&kvm_dev);
3498         kmem_cache_destroy(kvm_vcpu_cache);
3499         sysdev_unregister(&kvm_sysdev);
3500         sysdev_class_unregister(&kvm_sysdev_class);
3501         unregister_reboot_notifier(&kvm_reboot_notifier);
3502         unregister_cpu_notifier(&kvm_cpu_notifier);
3503         on_each_cpu(hardware_disable, NULL, 0, 1);
3504         kvm_x86_ops->hardware_unsetup();
3505         kvm_x86_ops = NULL;
3506 }
3507 EXPORT_SYMBOL_GPL(kvm_exit_x86);
3508
3509 static __init int kvm_init(void)
3510 {
3511         int r;
3512
3513         r = kvm_mmu_module_init();
3514         if (r)
3515                 goto out4;
3516
3517         kvm_init_debug();
3518
3519         kvm_arch_init();
3520
3521         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3522
3523         if (bad_page == NULL) {
3524                 r = -ENOMEM;
3525                 goto out;
3526         }
3527
3528         return 0;
3529
3530 out:
3531         kvm_exit_debug();
3532         kvm_mmu_module_exit();
3533 out4:
3534         return r;
3535 }
3536
3537 static __exit void kvm_exit(void)
3538 {
3539         kvm_exit_debug();
3540         __free_page(bad_page);
3541         kvm_mmu_module_exit();
3542 }
3543
3544 module_init(kvm_init)
3545 module_exit(kvm_exit)