SAFE public projects git trees. - safe/jmp/linux-2.6/blob - drivers/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "kvm.h"
  19 #include "x86.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                 new.user_alloc = user_alloc;
 717                 if (user_alloc)
 718                         new.userspace_addr = mem->userspace_addr;
 719                 else {
 720                         down_write(&current->mm->mmap_sem);
 721                         new.userspace_addr = do_mmap(NULL, 0,
 722                                                      npages * PAGE_SIZE,
 723                                                      PROT_READ | PROT_WRITE,
 724                                                      MAP_SHARED | MAP_ANONYMOUS,
 725                                                      0);
 726                         up_write(&current->mm->mmap_sem);
 727
 728                         if (IS_ERR((void *)new.userspace_addr))
 729                                 goto out_unlock;
 730                 }
 731         } else {
 732                 if (!old.user_alloc && old.rmap) {
 733                         int ret;
 734
 735                         down_write(&current->mm->mmap_sem);
 736                         ret = do_munmap(current->mm, old.userspace_addr,
 737                                         old.npages * PAGE_SIZE);
 738                         up_write(&current->mm->mmap_sem);
 739                         if (ret < 0)
 740                                 printk(KERN_WARNING
 741                                        "kvm_vm_ioctl_set_memory_region: "
 742                                        "failed to munmap memory\n");
 743                 }
 744         }
 745
 746         /* Allocate page dirty bitmap if needed */
 747         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 748                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 749
 750                 new.dirty_bitmap = vmalloc(dirty_bytes);
 751                 if (!new.dirty_bitmap)
 752                         goto out_unlock;
 753                 memset(new.dirty_bitmap, 0, dirty_bytes);
 754         }
 755
 756         if (mem->slot >= kvm->nmemslots)
 757                 kvm->nmemslots = mem->slot + 1;
 758
 759         if (!kvm->n_requested_mmu_pages) {
 760                 unsigned int n_pages;
 761
 762                 if (npages) {
 763                         n_pages = npages * KVM_PERMILLE_MMU_PAGES / 1000;
 764                         kvm_mmu_change_mmu_pages(kvm, kvm->n_alloc_mmu_pages +
 765                                                  n_pages);
 766                 } else {
 767                         unsigned int nr_mmu_pages;
 768
 769                         n_pages = old.npages * KVM_PERMILLE_MMU_PAGES / 1000;
 770                         nr_mmu_pages = kvm->n_alloc_mmu_pages - n_pages;
 771                         nr_mmu_pages = max(nr_mmu_pages,
 772                                         (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
 773                         kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
 774                 }
 775         }
 776
 777         *memslot = new;
 778
 779         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 780         kvm_flush_remote_tlbs(kvm);
 781
 782         mutex_unlock(&kvm->lock);
 783
 784         kvm_free_physmem_slot(&old, &new);
 785         return 0;
 786
 787 out_unlock:
 788         mutex_unlock(&kvm->lock);
 789         kvm_free_physmem_slot(&new, &old);
 790 out:
 791         return r;
 792 }
 793
 794 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
 795                                           u32 kvm_nr_mmu_pages)
 796 {
 797         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
 798                 return -EINVAL;
 799
 800         mutex_lock(&kvm->lock);
 801
 802         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
 803         kvm->n_requested_mmu_pages = kvm_nr_mmu_pages;
 804
 805         mutex_unlock(&kvm->lock);
 806         return 0;
 807 }
 808
 809 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
 810 {
 811         return kvm->n_alloc_mmu_pages;
 812 }
 813
 814 /*
 815  * Get (and clear) the dirty memory log for a memory slot.
 816  */
 817 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 818                                       struct kvm_dirty_log *log)
 819 {
 820         struct kvm_memory_slot *memslot;
 821         int r, i;
 822         int n;
 823         unsigned long any = 0;
 824
 825         mutex_lock(&kvm->lock);
 826
 827         r = -EINVAL;
 828         if (log->slot >= KVM_MEMORY_SLOTS)
 829                 goto out;
 830
 831         memslot = &kvm->memslots[log->slot];
 832         r = -ENOENT;
 833         if (!memslot->dirty_bitmap)
 834                 goto out;
 835
 836         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 837
 838         for (i = 0; !any && i < n/sizeof(long); ++i)
 839                 any = memslot->dirty_bitmap[i];
 840
 841         r = -EFAULT;
 842         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 843                 goto out;
 844
 845         /* If nothing is dirty, don't bother messing with page tables. */
 846         if (any) {
 847                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 848                 kvm_flush_remote_tlbs(kvm);
 849                 memset(memslot->dirty_bitmap, 0, n);
 850         }
 851
 852         r = 0;
 853
 854 out:
 855         mutex_unlock(&kvm->lock);
 856         return r;
 857 }
 858
 859 /*
 860  * Set a new alias region.  Aliases map a portion of physical memory into
 861  * another portion.  This is useful for memory windows, for example the PC
 862  * VGA region.
 863  */
 864 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 865                                          struct kvm_memory_alias *alias)
 866 {
 867         int r, n;
 868         struct kvm_mem_alias *p;
 869
 870         r = -EINVAL;
 871         /* General sanity checks */
 872         if (alias->memory_size & (PAGE_SIZE - 1))
 873                 goto out;
 874         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 875                 goto out;
 876         if (alias->slot >= KVM_ALIAS_SLOTS)
 877                 goto out;
 878         if (alias->guest_phys_addr + alias->memory_size
 879             < alias->guest_phys_addr)
 880                 goto out;
 881         if (alias->target_phys_addr + alias->memory_size
 882             < alias->target_phys_addr)
 883                 goto out;
 884
 885         mutex_lock(&kvm->lock);
 886
 887         p = &kvm->aliases[alias->slot];
 888         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 889         p->npages = alias->memory_size >> PAGE_SHIFT;
 890         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 891
 892         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 893                 if (kvm->aliases[n - 1].npages)
 894                         break;
 895         kvm->naliases = n;
 896
 897         kvm_mmu_zap_all(kvm);
 898
 899         mutex_unlock(&kvm->lock);
 900
 901         return 0;
 902
 903 out:
 904         return r;
 905 }
 906
 907 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 908 {
 909         int r;
 910
 911         r = 0;
 912         switch (chip->chip_id) {
 913         case KVM_IRQCHIP_PIC_MASTER:
 914                 memcpy(&chip->chip.pic,
 915                         &pic_irqchip(kvm)->pics[0],
 916                         sizeof(struct kvm_pic_state));
 917                 break;
 918         case KVM_IRQCHIP_PIC_SLAVE:
 919                 memcpy(&chip->chip.pic,
 920                         &pic_irqchip(kvm)->pics[1],
 921                         sizeof(struct kvm_pic_state));
 922                 break;
 923         case KVM_IRQCHIP_IOAPIC:
 924                 memcpy(&chip->chip.ioapic,
 925                         ioapic_irqchip(kvm),
 926                         sizeof(struct kvm_ioapic_state));
 927                 break;
 928         default:
 929                 r = -EINVAL;
 930                 break;
 931         }
 932         return r;
 933 }
 934
 935 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 936 {
 937         int r;
 938
 939         r = 0;
 940         switch (chip->chip_id) {
 941         case KVM_IRQCHIP_PIC_MASTER:
 942                 memcpy(&pic_irqchip(kvm)->pics[0],
 943                         &chip->chip.pic,
 944                         sizeof(struct kvm_pic_state));
 945                 break;
 946         case KVM_IRQCHIP_PIC_SLAVE:
 947                 memcpy(&pic_irqchip(kvm)->pics[1],
 948                         &chip->chip.pic,
 949                         sizeof(struct kvm_pic_state));
 950                 break;
 951         case KVM_IRQCHIP_IOAPIC:
 952                 memcpy(ioapic_irqchip(kvm),
 953                         &chip->chip.ioapic,
 954                         sizeof(struct kvm_ioapic_state));
 955                 break;
 956         default:
 957                 r = -EINVAL;
 958                 break;
 959         }
 960         kvm_pic_update_irq(pic_irqchip(kvm));
 961         return r;
 962 }
 963
 964 int is_error_page(struct page *page)
 965 {
 966         return page == bad_page;
 967 }
 968 EXPORT_SYMBOL_GPL(is_error_page);
 969
 970 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 971 {
 972         int i;
 973         struct kvm_mem_alias *alias;
 974
 975         for (i = 0; i < kvm->naliases; ++i) {
 976                 alias = &kvm->aliases[i];
 977                 if (gfn >= alias->base_gfn
 978                     && gfn < alias->base_gfn + alias->npages)
 979                         return alias->target_gfn + gfn - alias->base_gfn;
 980         }
 981         return gfn;
 982 }
 983
 984 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 985 {
 986         int i;
 987
 988         for (i = 0; i < kvm->nmemslots; ++i) {
 989                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 990
 991                 if (gfn >= memslot->base_gfn
 992                     && gfn < memslot->base_gfn + memslot->npages)
 993                         return memslot;
 994         }
 995         return NULL;
 996 }
 997
 998 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 999 {
1000         gfn = unalias_gfn(kvm, gfn);
1001         return __gfn_to_memslot(kvm, gfn);
1002 }
1003
1004 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1005 {
1006         struct kvm_memory_slot *slot;
1007         struct page *page[1];
1008         int npages;
1009
1010         might_sleep();
1011
1012         gfn = unalias_gfn(kvm, gfn);
1013         slot = __gfn_to_memslot(kvm, gfn);
1014         if (!slot) {
1015                 get_page(bad_page);
1016                 return bad_page;
1017         }
1018
1019         down_read(&current->mm->mmap_sem);
1020         npages = get_user_pages(current, current->mm,
1021                                 slot->userspace_addr
1022                                 + (gfn - slot->base_gfn) * PAGE_SIZE, 1,
1023                                 1, 1, page, NULL);
1024         up_read(&current->mm->mmap_sem);
1025         if (npages != 1) {
1026                 get_page(bad_page);
1027                 return bad_page;
1028         }
1029
1030         return page[0];
1031 }
1032 EXPORT_SYMBOL_GPL(gfn_to_page);
1033
1034 void kvm_release_page(struct page *page)
1035 {
1036         if (!PageReserved(page))
1037                 SetPageDirty(page);
1038         put_page(page);
1039 }
1040 EXPORT_SYMBOL_GPL(kvm_release_page);
1041
1042 static int next_segment(unsigned long len, int offset)
1043 {
1044         if (len > PAGE_SIZE - offset)
1045                 return PAGE_SIZE - offset;
1046         else
1047                 return len;
1048 }
1049
1050 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1051                         int len)
1052 {
1053         void *page_virt;
1054         struct page *page;
1055
1056         page = gfn_to_page(kvm, gfn);
1057         if (is_error_page(page)) {
1058                 kvm_release_page(page);
1059                 return -EFAULT;
1060         }
1061         page_virt = kmap_atomic(page, KM_USER0);
1062
1063         memcpy(data, page_virt + offset, len);
1064
1065         kunmap_atomic(page_virt, KM_USER0);
1066         kvm_release_page(page);
1067         return 0;
1068 }
1069 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1070
1071 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1072 {
1073         gfn_t gfn = gpa >> PAGE_SHIFT;
1074         int seg;
1075         int offset = offset_in_page(gpa);
1076         int ret;
1077
1078         while ((seg = next_segment(len, offset)) != 0) {
1079                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1080                 if (ret < 0)
1081                         return ret;
1082                 offset = 0;
1083                 len -= seg;
1084                 data += seg;
1085                 ++gfn;
1086         }
1087         return 0;
1088 }
1089 EXPORT_SYMBOL_GPL(kvm_read_guest);
1090
1091 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1092                          int offset, int len)
1093 {
1094         void *page_virt;
1095         struct page *page;
1096
1097         page = gfn_to_page(kvm, gfn);
1098         if (is_error_page(page)) {
1099                 kvm_release_page(page);
1100                 return -EFAULT;
1101         }
1102         page_virt = kmap_atomic(page, KM_USER0);
1103
1104         memcpy(page_virt + offset, data, len);
1105
1106         kunmap_atomic(page_virt, KM_USER0);
1107         mark_page_dirty(kvm, gfn);
1108         kvm_release_page(page);
1109         return 0;
1110 }
1111 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1112
1113 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1114                     unsigned long len)
1115 {
1116         gfn_t gfn = gpa >> PAGE_SHIFT;
1117         int seg;
1118         int offset = offset_in_page(gpa);
1119         int ret;
1120
1121         while ((seg = next_segment(len, offset)) != 0) {
1122                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1123                 if (ret < 0)
1124                         return ret;
1125                 offset = 0;
1126                 len -= seg;
1127                 data += seg;
1128                 ++gfn;
1129         }
1130         return 0;
1131 }
1132
1133 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1134 {
1135         void *page_virt;
1136         struct page *page;
1137
1138         page = gfn_to_page(kvm, gfn);
1139         if (is_error_page(page)) {
1140                 kvm_release_page(page);
1141                 return -EFAULT;
1142         }
1143         page_virt = kmap_atomic(page, KM_USER0);
1144
1145         memset(page_virt + offset, 0, len);
1146
1147         kunmap_atomic(page_virt, KM_USER0);
1148         kvm_release_page(page);
1149         return 0;
1150 }
1151 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1152
1153 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1154 {
1155         gfn_t gfn = gpa >> PAGE_SHIFT;
1156         int seg;
1157         int offset = offset_in_page(gpa);
1158         int ret;
1159
1160         while ((seg = next_segment(len, offset)) != 0) {
1161                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1162                 if (ret < 0)
1163                         return ret;
1164                 offset = 0;
1165                 len -= seg;
1166                 ++gfn;
1167         }
1168         return 0;
1169 }
1170 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1171
1172 /* WARNING: Does not work on aliased pages. */
1173 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1174 {
1175         struct kvm_memory_slot *memslot;
1176
1177         memslot = __gfn_to_memslot(kvm, gfn);
1178         if (memslot && memslot->dirty_bitmap) {
1179                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1180
1181                 /* avoid RMW */
1182                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1183                         set_bit(rel_gfn, memslot->dirty_bitmap);
1184         }
1185 }
1186
1187 int emulator_read_std(unsigned long addr,
1188                              void *val,
1189                              unsigned int bytes,
1190                              struct kvm_vcpu *vcpu)
1191 {
1192         void *data = val;
1193
1194         while (bytes) {
1195                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1196                 unsigned offset = addr & (PAGE_SIZE-1);
1197                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1198                 int ret;
1199
1200                 if (gpa == UNMAPPED_GVA)
1201                         return X86EMUL_PROPAGATE_FAULT;
1202                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1203                 if (ret < 0)
1204                         return X86EMUL_UNHANDLEABLE;
1205
1206                 bytes -= tocopy;
1207                 data += tocopy;
1208                 addr += tocopy;
1209         }
1210
1211         return X86EMUL_CONTINUE;
1212 }
1213 EXPORT_SYMBOL_GPL(emulator_read_std);
1214
1215 static int emulator_write_std(unsigned long addr,
1216                               const void *val,
1217                               unsigned int bytes,
1218                               struct kvm_vcpu *vcpu)
1219 {
1220         pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1221         return X86EMUL_UNHANDLEABLE;
1222 }
1223
1224 /*
1225  * Only apic need an MMIO device hook, so shortcut now..
1226  */
1227 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1228                                                 gpa_t addr)
1229 {
1230         struct kvm_io_device *dev;
1231
1232         if (vcpu->apic) {
1233                 dev = &vcpu->apic->dev;
1234                 if (dev->in_range(dev, addr))
1235                         return dev;
1236         }
1237         return NULL;
1238 }
1239
1240 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1241                                                 gpa_t addr)
1242 {
1243         struct kvm_io_device *dev;
1244
1245         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1246         if (dev == NULL)
1247                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1248         return dev;
1249 }
1250
1251 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1252                                                gpa_t addr)
1253 {
1254         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1255 }
1256
1257 static int emulator_read_emulated(unsigned long addr,
1258                                   void *val,
1259                                   unsigned int bytes,
1260                                   struct kvm_vcpu *vcpu)
1261 {
1262         struct kvm_io_device *mmio_dev;
1263         gpa_t                 gpa;
1264
1265         if (vcpu->mmio_read_completed) {
1266                 memcpy(val, vcpu->mmio_data, bytes);
1267                 vcpu->mmio_read_completed = 0;
1268                 return X86EMUL_CONTINUE;
1269         } else if (emulator_read_std(addr, val, bytes, vcpu)
1270                    == X86EMUL_CONTINUE)
1271                 return X86EMUL_CONTINUE;
1272
1273         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1274         if (gpa == UNMAPPED_GVA)
1275                 return X86EMUL_PROPAGATE_FAULT;
1276
1277         /*
1278          * Is this MMIO handled locally?
1279          */
1280         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1281         if (mmio_dev) {
1282                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1283                 return X86EMUL_CONTINUE;
1284         }
1285
1286         vcpu->mmio_needed = 1;
1287         vcpu->mmio_phys_addr = gpa;
1288         vcpu->mmio_size = bytes;
1289         vcpu->mmio_is_write = 0;
1290
1291         return X86EMUL_UNHANDLEABLE;
1292 }
1293
1294 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1295                                const void *val, int bytes)
1296 {
1297         int ret;
1298
1299         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1300         if (ret < 0)
1301                 return 0;
1302         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1303         return 1;
1304 }
1305
1306 static int emulator_write_emulated_onepage(unsigned long addr,
1307                                            const void *val,
1308                                            unsigned int bytes,
1309                                            struct kvm_vcpu *vcpu)
1310 {
1311         struct kvm_io_device *mmio_dev;
1312         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1313
1314         if (gpa == UNMAPPED_GVA) {
1315                 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1316                 return X86EMUL_PROPAGATE_FAULT;
1317         }
1318
1319         if (emulator_write_phys(vcpu, gpa, val, bytes))
1320                 return X86EMUL_CONTINUE;
1321
1322         /*
1323          * Is this MMIO handled locally?
1324          */
1325         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1326         if (mmio_dev) {
1327                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1328                 return X86EMUL_CONTINUE;
1329         }
1330
1331         vcpu->mmio_needed = 1;
1332         vcpu->mmio_phys_addr = gpa;
1333         vcpu->mmio_size = bytes;
1334         vcpu->mmio_is_write = 1;
1335         memcpy(vcpu->mmio_data, val, bytes);
1336
1337         return X86EMUL_CONTINUE;
1338 }
1339
1340 int emulator_write_emulated(unsigned long addr,
1341                                    const void *val,
1342                                    unsigned int bytes,
1343                                    struct kvm_vcpu *vcpu)
1344 {
1345         /* Crossing a page boundary? */
1346         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1347                 int rc, now;
1348
1349                 now = -addr & ~PAGE_MASK;
1350                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1351                 if (rc != X86EMUL_CONTINUE)
1352                         return rc;
1353                 addr += now;
1354                 val += now;
1355                 bytes -= now;
1356         }
1357         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1358 }
1359 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1360
1361 static int emulator_cmpxchg_emulated(unsigned long addr,
1362                                      const void *old,
1363                                      const void *new,
1364                                      unsigned int bytes,
1365                                      struct kvm_vcpu *vcpu)
1366 {
1367         static int reported;
1368
1369         if (!reported) {
1370                 reported = 1;
1371                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1372         }
1373         return emulator_write_emulated(addr, new, bytes, vcpu);
1374 }
1375
1376 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1377 {
1378         return kvm_x86_ops->get_segment_base(vcpu, seg);
1379 }
1380
1381 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1382 {
1383         return X86EMUL_CONTINUE;
1384 }
1385
1386 int emulate_clts(struct kvm_vcpu *vcpu)
1387 {
1388         kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1389         return X86EMUL_CONTINUE;
1390 }
1391
1392 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1393 {
1394         struct kvm_vcpu *vcpu = ctxt->vcpu;
1395
1396         switch (dr) {
1397         case 0 ... 3:
1398                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1399                 return X86EMUL_CONTINUE;
1400         default:
1401                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1402                 return X86EMUL_UNHANDLEABLE;
1403         }
1404 }
1405
1406 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1407 {
1408         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1409         int exception;
1410
1411         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1412         if (exception) {
1413                 /* FIXME: better handling */
1414                 return X86EMUL_UNHANDLEABLE;
1415         }
1416         return X86EMUL_CONTINUE;
1417 }
1418
1419 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1420 {
1421         static int reported;
1422         u8 opcodes[4];
1423         unsigned long rip = vcpu->rip;
1424         unsigned long rip_linear;
1425
1426         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1427
1428         if (reported)
1429                 return;
1430
1431         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1432
1433         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1434                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1435         reported = 1;
1436 }
1437 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1438
1439 struct x86_emulate_ops emulate_ops = {
1440         .read_std            = emulator_read_std,
1441         .write_std           = emulator_write_std,
1442         .read_emulated       = emulator_read_emulated,
1443         .write_emulated      = emulator_write_emulated,
1444         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1445 };
1446
1447 int emulate_instruction(struct kvm_vcpu *vcpu,
1448                         struct kvm_run *run,
1449                         unsigned long cr2,
1450                         u16 error_code,
1451                         int no_decode)
1452 {
1453         int r;
1454
1455         vcpu->mmio_fault_cr2 = cr2;
1456         kvm_x86_ops->cache_regs(vcpu);
1457
1458         vcpu->mmio_is_write = 0;
1459         vcpu->pio.string = 0;
1460
1461         if (!no_decode) {
1462                 int cs_db, cs_l;
1463                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1464
1465                 vcpu->emulate_ctxt.vcpu = vcpu;
1466                 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1467                 vcpu->emulate_ctxt.cr2 = cr2;
1468                 vcpu->emulate_ctxt.mode =
1469                         (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1470                         ? X86EMUL_MODE_REAL : cs_l
1471                         ? X86EMUL_MODE_PROT64 : cs_db
1472                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1473
1474                 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1475                         vcpu->emulate_ctxt.cs_base = 0;
1476                         vcpu->emulate_ctxt.ds_base = 0;
1477                         vcpu->emulate_ctxt.es_base = 0;
1478                         vcpu->emulate_ctxt.ss_base = 0;
1479                 } else {
1480                         vcpu->emulate_ctxt.cs_base =
1481                                         get_segment_base(vcpu, VCPU_SREG_CS);
1482                         vcpu->emulate_ctxt.ds_base =
1483                                         get_segment_base(vcpu, VCPU_SREG_DS);
1484                         vcpu->emulate_ctxt.es_base =
1485                                         get_segment_base(vcpu, VCPU_SREG_ES);
1486                         vcpu->emulate_ctxt.ss_base =
1487                                         get_segment_base(vcpu, VCPU_SREG_SS);
1488                 }
1489
1490                 vcpu->emulate_ctxt.gs_base =
1491                                         get_segment_base(vcpu, VCPU_SREG_GS);
1492                 vcpu->emulate_ctxt.fs_base =
1493                                         get_segment_base(vcpu, VCPU_SREG_FS);
1494
1495                 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
1496                 if (r)  {
1497                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1498                                 return EMULATE_DONE;
1499                         return EMULATE_FAIL;
1500                 }
1501         }
1502
1503         r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1504
1505         if (vcpu->pio.string)
1506                 return EMULATE_DO_MMIO;
1507
1508         if ((r || vcpu->mmio_is_write) && run) {
1509                 run->exit_reason = KVM_EXIT_MMIO;
1510                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1511                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1512                 run->mmio.len = vcpu->mmio_size;
1513                 run->mmio.is_write = vcpu->mmio_is_write;
1514         }
1515
1516         if (r) {
1517                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1518                         return EMULATE_DONE;
1519                 if (!vcpu->mmio_needed) {
1520                         kvm_report_emulation_failure(vcpu, "mmio");
1521                         return EMULATE_FAIL;
1522                 }
1523                 return EMULATE_DO_MMIO;
1524         }
1525
1526         kvm_x86_ops->decache_regs(vcpu);
1527         kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1528
1529         if (vcpu->mmio_is_write) {
1530                 vcpu->mmio_needed = 0;
1531                 return EMULATE_DO_MMIO;
1532         }
1533
1534         return EMULATE_DONE;
1535 }
1536 EXPORT_SYMBOL_GPL(emulate_instruction);
1537
1538 /*
1539  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1540  */
1541 static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1542 {
1543         DECLARE_WAITQUEUE(wait, current);
1544
1545         add_wait_queue(&vcpu->wq, &wait);
1546
1547         /*
1548          * We will block until either an interrupt or a signal wakes us up
1549          */
1550         while (!kvm_cpu_has_interrupt(vcpu)
1551                && !signal_pending(current)
1552                && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
1553                && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
1554                 set_current_state(TASK_INTERRUPTIBLE);
1555                 vcpu_put(vcpu);
1556                 schedule();
1557                 vcpu_load(vcpu);
1558         }
1559
1560         __set_current_state(TASK_RUNNING);
1561         remove_wait_queue(&vcpu->wq, &wait);
1562 }
1563
1564 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1565 {
1566         ++vcpu->stat.halt_exits;
1567         if (irqchip_in_kernel(vcpu->kvm)) {
1568                 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1569                 kvm_vcpu_block(vcpu);
1570                 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1571                         return -EINTR;
1572                 return 1;
1573         } else {
1574                 vcpu->run->exit_reason = KVM_EXIT_HLT;
1575                 return 0;
1576         }
1577 }
1578 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1579
1580 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
1581 {
1582         unsigned long nr, a0, a1, a2, a3, ret;
1583
1584         kvm_x86_ops->cache_regs(vcpu);
1585
1586         nr = vcpu->regs[VCPU_REGS_RAX];
1587         a0 = vcpu->regs[VCPU_REGS_RBX];
1588         a1 = vcpu->regs[VCPU_REGS_RCX];
1589         a2 = vcpu->regs[VCPU_REGS_RDX];
1590         a3 = vcpu->regs[VCPU_REGS_RSI];
1591
1592         if (!is_long_mode(vcpu)) {
1593                 nr &= 0xFFFFFFFF;
1594                 a0 &= 0xFFFFFFFF;
1595                 a1 &= 0xFFFFFFFF;
1596                 a2 &= 0xFFFFFFFF;
1597                 a3 &= 0xFFFFFFFF;
1598         }
1599
1600         switch (nr) {
1601         default:
1602                 ret = -KVM_ENOSYS;
1603                 break;
1604         }
1605         vcpu->regs[VCPU_REGS_RAX] = ret;
1606         kvm_x86_ops->decache_regs(vcpu);
1607         return 0;
1608 }
1609 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1610
1611 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1612 {
1613         char instruction[3];
1614         int ret = 0;
1615
1616         mutex_lock(&vcpu->kvm->lock);
1617
1618         /*
1619          * Blow out the MMU to ensure that no other VCPU has an active mapping
1620          * to ensure that the updated hypercall appears atomically across all
1621          * VCPUs.
1622          */
1623         kvm_mmu_zap_all(vcpu->kvm);
1624
1625         kvm_x86_ops->cache_regs(vcpu);
1626         kvm_x86_ops->patch_hypercall(vcpu, instruction);
1627         if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1628             != X86EMUL_CONTINUE)
1629                 ret = -EFAULT;
1630
1631         mutex_unlock(&vcpu->kvm->lock);
1632
1633         return ret;
1634 }
1635
1636 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1637 {
1638         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1639 }
1640
1641 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1642 {
1643         struct descriptor_table dt = { limit, base };
1644
1645         kvm_x86_ops->set_gdt(vcpu, &dt);
1646 }
1647
1648 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1649 {
1650         struct descriptor_table dt = { limit, base };
1651
1652         kvm_x86_ops->set_idt(vcpu, &dt);
1653 }
1654
1655 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1656                    unsigned long *rflags)
1657 {
1658         lmsw(vcpu, msw);
1659         *rflags = kvm_x86_ops->get_rflags(vcpu);
1660 }
1661
1662 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1663 {
1664         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1665         switch (cr) {
1666         case 0:
1667                 return vcpu->cr0;
1668         case 2:
1669                 return vcpu->cr2;
1670         case 3:
1671                 return vcpu->cr3;
1672         case 4:
1673                 return vcpu->cr4;
1674         default:
1675                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1676                 return 0;
1677         }
1678 }
1679
1680 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1681                      unsigned long *rflags)
1682 {
1683         switch (cr) {
1684         case 0:
1685                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1686                 *rflags = kvm_x86_ops->get_rflags(vcpu);
1687                 break;
1688         case 2:
1689                 vcpu->cr2 = val;
1690                 break;
1691         case 3:
1692                 set_cr3(vcpu, val);
1693                 break;
1694         case 4:
1695                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1696                 break;
1697         default:
1698                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1699         }
1700 }
1701
1702 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1703 {
1704         u64 data;
1705
1706         switch (msr) {
1707         case 0xc0010010: /* SYSCFG */
1708         case 0xc0010015: /* HWCR */
1709         case MSR_IA32_PLATFORM_ID:
1710         case MSR_IA32_P5_MC_ADDR:
1711         case MSR_IA32_P5_MC_TYPE:
1712         case MSR_IA32_MC0_CTL:
1713         case MSR_IA32_MCG_STATUS:
1714         case MSR_IA32_MCG_CAP:
1715         case MSR_IA32_MC0_MISC:
1716         case MSR_IA32_MC0_MISC+4:
1717         case MSR_IA32_MC0_MISC+8:
1718         case MSR_IA32_MC0_MISC+12:
1719         case MSR_IA32_MC0_MISC+16:
1720         case MSR_IA32_UCODE_REV:
1721         case MSR_IA32_PERF_STATUS:
1722         case MSR_IA32_EBL_CR_POWERON:
1723                 /* MTRR registers */
1724         case 0xfe:
1725         case 0x200 ... 0x2ff:
1726                 data = 0;
1727                 break;
1728         case 0xcd: /* fsb frequency */
1729                 data = 3;
1730                 break;
1731         case MSR_IA32_APICBASE:
1732                 data = kvm_get_apic_base(vcpu);
1733                 break;
1734         case MSR_IA32_MISC_ENABLE:
1735                 data = vcpu->ia32_misc_enable_msr;
1736                 break;
1737 #ifdef CONFIG_X86_64
1738         case MSR_EFER:
1739                 data = vcpu->shadow_efer;
1740                 break;
1741 #endif
1742         default:
1743                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1744                 return 1;
1745         }
1746         *pdata = data;
1747         return 0;
1748 }
1749 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1750
1751 /*
1752  * Reads an msr value (of 'msr_index') into 'pdata'.
1753  * Returns 0 on success, non-0 otherwise.
1754  * Assumes vcpu_load() was already called.
1755  */
1756 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1757 {
1758         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1759 }
1760
1761 #ifdef CONFIG_X86_64
1762
1763 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1764 {
1765         if (efer & EFER_RESERVED_BITS) {
1766                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1767                        efer);
1768                 inject_gp(vcpu);
1769                 return;
1770         }
1771
1772         if (is_paging(vcpu)
1773             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1774                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1775                 inject_gp(vcpu);
1776                 return;
1777         }
1778
1779         kvm_x86_ops->set_efer(vcpu, efer);
1780
1781         efer &= ~EFER_LMA;
1782         efer |= vcpu->shadow_efer & EFER_LMA;
1783
1784         vcpu->shadow_efer = efer;
1785 }
1786
1787 #endif
1788
1789 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1790 {
1791         switch (msr) {
1792 #ifdef CONFIG_X86_64
1793         case MSR_EFER:
1794                 set_efer(vcpu, data);
1795                 break;
1796 #endif
1797         case MSR_IA32_MC0_STATUS:
1798                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1799                        __FUNCTION__, data);
1800                 break;
1801         case MSR_IA32_MCG_STATUS:
1802                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1803                         __FUNCTION__, data);
1804                 break;
1805         case MSR_IA32_UCODE_REV:
1806         case MSR_IA32_UCODE_WRITE:
1807         case 0x200 ... 0x2ff: /* MTRRs */
1808                 break;
1809         case MSR_IA32_APICBASE:
1810                 kvm_set_apic_base(vcpu, data);
1811                 break;
1812         case MSR_IA32_MISC_ENABLE:
1813                 vcpu->ia32_misc_enable_msr = data;
1814                 break;
1815         default:
1816                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1817                 return 1;
1818         }
1819         return 0;
1820 }
1821 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1822
1823 /*
1824  * Writes msr value into into the appropriate "register".
1825  * Returns 0 on success, non-0 otherwise.
1826  * Assumes vcpu_load() was already called.
1827  */
1828 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1829 {
1830         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
1831 }
1832
1833 void kvm_resched(struct kvm_vcpu *vcpu)
1834 {
1835         if (!need_resched())
1836                 return;
1837         cond_resched();
1838 }
1839 EXPORT_SYMBOL_GPL(kvm_resched);
1840
1841 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1842 {
1843         int i;
1844         u32 function;
1845         struct kvm_cpuid_entry *e, *best;
1846
1847         kvm_x86_ops->cache_regs(vcpu);
1848         function = vcpu->regs[VCPU_REGS_RAX];
1849         vcpu->regs[VCPU_REGS_RAX] = 0;
1850         vcpu->regs[VCPU_REGS_RBX] = 0;
1851         vcpu->regs[VCPU_REGS_RCX] = 0;
1852         vcpu->regs[VCPU_REGS_RDX] = 0;
1853         best = NULL;
1854         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1855                 e = &vcpu->cpuid_entries[i];
1856                 if (e->function == function) {
1857                         best = e;
1858                         break;
1859                 }
1860                 /*
1861                  * Both basic or both extended?
1862                  */
1863                 if (((e->function ^ function) & 0x80000000) == 0)
1864                         if (!best || e->function > best->function)
1865                                 best = e;
1866         }
1867         if (best) {
1868                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1869                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1870                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1871                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1872         }
1873         kvm_x86_ops->decache_regs(vcpu);
1874         kvm_x86_ops->skip_emulated_instruction(vcpu);
1875 }
1876 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1877
1878 static int pio_copy_data(struct kvm_vcpu *vcpu)
1879 {
1880         void *p = vcpu->pio_data;
1881         void *q;
1882         unsigned bytes;
1883         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1884
1885         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1886                  PAGE_KERNEL);
1887         if (!q) {
1888                 free_pio_guest_pages(vcpu);
1889                 return -ENOMEM;
1890         }
1891         q += vcpu->pio.guest_page_offset;
1892         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1893         if (vcpu->pio.in)
1894                 memcpy(q, p, bytes);
1895         else
1896                 memcpy(p, q, bytes);
1897         q -= vcpu->pio.guest_page_offset;
1898         vunmap(q);
1899         free_pio_guest_pages(vcpu);
1900         return 0;
1901 }
1902
1903 static int complete_pio(struct kvm_vcpu *vcpu)
1904 {
1905         struct kvm_pio_request *io = &vcpu->pio;
1906         long delta;
1907         int r;
1908
1909         kvm_x86_ops->cache_regs(vcpu);
1910
1911         if (!io->string) {
1912                 if (io->in)
1913                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1914                                io->size);
1915         } else {
1916                 if (io->in) {
1917                         r = pio_copy_data(vcpu);
1918                         if (r) {
1919                                 kvm_x86_ops->cache_regs(vcpu);
1920                                 return r;
1921                         }
1922                 }
1923
1924                 delta = 1;
1925                 if (io->rep) {
1926                         delta *= io->cur_count;
1927                         /*
1928                          * The size of the register should really depend on
1929                          * current address size.
1930                          */
1931                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1932                 }
1933                 if (io->down)
1934                         delta = -delta;
1935                 delta *= io->size;
1936                 if (io->in)
1937                         vcpu->regs[VCPU_REGS_RDI] += delta;
1938                 else
1939                         vcpu->regs[VCPU_REGS_RSI] += delta;
1940         }
1941
1942         kvm_x86_ops->decache_regs(vcpu);
1943
1944         io->count -= io->cur_count;
1945         io->cur_count = 0;
1946
1947         return 0;
1948 }
1949
1950 static void kernel_pio(struct kvm_io_device *pio_dev,
1951                        struct kvm_vcpu *vcpu,
1952                        void *pd)
1953 {
1954         /* TODO: String I/O for in kernel device */
1955
1956         mutex_lock(&vcpu->kvm->lock);
1957         if (vcpu->pio.in)
1958                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1959                                   vcpu->pio.size,
1960                                   pd);
1961         else
1962                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1963                                    vcpu->pio.size,
1964                                    pd);
1965         mutex_unlock(&vcpu->kvm->lock);
1966 }
1967
1968 static void pio_string_write(struct kvm_io_device *pio_dev,
1969                              struct kvm_vcpu *vcpu)
1970 {
1971         struct kvm_pio_request *io = &vcpu->pio;
1972         void *pd = vcpu->pio_data;
1973         int i;
1974
1975         mutex_lock(&vcpu->kvm->lock);
1976         for (i = 0; i < io->cur_count; i++) {
1977                 kvm_iodevice_write(pio_dev, io->port,
1978                                    io->size,
1979                                    pd);
1980                 pd += io->size;
1981         }
1982         mutex_unlock(&vcpu->kvm->lock);
1983 }
1984
1985 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1986                   int size, unsigned port)
1987 {
1988         struct kvm_io_device *pio_dev;
1989
1990         vcpu->run->exit_reason = KVM_EXIT_IO;
1991         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1992         vcpu->run->io.size = vcpu->pio.size = size;
1993         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1994         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1995         vcpu->run->io.port = vcpu->pio.port = port;
1996         vcpu->pio.in = in;
1997         vcpu->pio.string = 0;
1998         vcpu->pio.down = 0;
1999         vcpu->pio.guest_page_offset = 0;
2000         vcpu->pio.rep = 0;
2001
2002         kvm_x86_ops->cache_regs(vcpu);
2003         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
2004         kvm_x86_ops->decache_regs(vcpu);
2005
2006         kvm_x86_ops->skip_emulated_instruction(vcpu);
2007
2008         pio_dev = vcpu_find_pio_dev(vcpu, port);
2009         if (pio_dev) {
2010                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
2011                 complete_pio(vcpu);
2012                 return 1;
2013         }
2014         return 0;
2015 }
2016 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2017
2018 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2019                   int size, unsigned long count, int down,
2020                   gva_t address, int rep, unsigned port)
2021 {
2022         unsigned now, in_page;
2023         int i, ret = 0;
2024         int nr_pages = 1;
2025         struct page *page;
2026         struct kvm_io_device *pio_dev;
2027
2028         vcpu->run->exit_reason = KVM_EXIT_IO;
2029         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2030         vcpu->run->io.size = vcpu->pio.size = size;
2031         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2032         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
2033         vcpu->run->io.port = vcpu->pio.port = port;
2034         vcpu->pio.in = in;
2035         vcpu->pio.string = 1;
2036         vcpu->pio.down = down;
2037         vcpu->pio.guest_page_offset = offset_in_page(address);
2038         vcpu->pio.rep = rep;
2039
2040         if (!count) {
2041                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2042                 return 1;
2043         }
2044
2045         if (!down)
2046                 in_page = PAGE_SIZE - offset_in_page(address);
2047         else
2048                 in_page = offset_in_page(address) + size;
2049         now = min(count, (unsigned long)in_page / size);
2050         if (!now) {
2051                 /*
2052                  * String I/O straddles page boundary.  Pin two guest pages
2053                  * so that we satisfy atomicity constraints.  Do just one
2054                  * transaction to avoid complexity.
2055                  */
2056                 nr_pages = 2;
2057                 now = 1;
2058         }
2059         if (down) {
2060                 /*
2061                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2062                  */
2063                 pr_unimpl(vcpu, "guest string pio down\n");
2064                 inject_gp(vcpu);
2065                 return 1;
2066         }
2067         vcpu->run->io.count = now;
2068         vcpu->pio.cur_count = now;
2069
2070         if (vcpu->pio.cur_count == vcpu->pio.count)
2071                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2072
2073         for (i = 0; i < nr_pages; ++i) {
2074                 mutex_lock(&vcpu->kvm->lock);
2075                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2076                 vcpu->pio.guest_pages[i] = page;
2077                 mutex_unlock(&vcpu->kvm->lock);
2078                 if (!page) {
2079                         inject_gp(vcpu);
2080                         free_pio_guest_pages(vcpu);
2081                         return 1;
2082                 }
2083         }
2084
2085         pio_dev = vcpu_find_pio_dev(vcpu, port);
2086         if (!vcpu->pio.in) {
2087                 /* string PIO write */
2088                 ret = pio_copy_data(vcpu);
2089                 if (ret >= 0 && pio_dev) {
2090                         pio_string_write(pio_dev, vcpu);
2091                         complete_pio(vcpu);
2092                         if (vcpu->pio.count == 0)
2093                                 ret = 1;
2094                 }
2095         } else if (pio_dev)
2096                 pr_unimpl(vcpu, "no string pio read support yet, "
2097                        "port %x size %d count %ld\n",
2098                         port, size, count);
2099
2100         return ret;
2101 }
2102 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2103
2104 /*
2105  * Check if userspace requested an interrupt window, and that the
2106  * interrupt window is open.
2107  *
2108  * No need to exit to userspace if we already have an interrupt queued.
2109  */
2110 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2111                                           struct kvm_run *kvm_run)
2112 {
2113         return (!vcpu->irq_summary &&
2114                 kvm_run->request_interrupt_window &&
2115                 vcpu->interrupt_window_open &&
2116                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2117 }
2118
2119 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2120                               struct kvm_run *kvm_run)
2121 {
2122         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2123         kvm_run->cr8 = get_cr8(vcpu);
2124         kvm_run->apic_base = kvm_get_apic_base(vcpu);
2125         if (irqchip_in_kernel(vcpu->kvm))
2126                 kvm_run->ready_for_interrupt_injection = 1;
2127         else
2128                 kvm_run->ready_for_interrupt_injection =
2129                                         (vcpu->interrupt_window_open &&
2130                                          vcpu->irq_summary == 0);
2131 }
2132
2133 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2134 {
2135         int r;
2136
2137         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
2138                 pr_debug("vcpu %d received sipi with vector # %x\n",
2139                        vcpu->vcpu_id, vcpu->sipi_vector);
2140                 kvm_lapic_reset(vcpu);
2141                 r = kvm_x86_ops->vcpu_reset(vcpu);
2142                 if (r)
2143                         return r;
2144                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2145         }
2146
2147 preempted:
2148         if (vcpu->guest_debug.enabled)
2149                 kvm_x86_ops->guest_debug_pre(vcpu);
2150
2151 again:
2152         r = kvm_mmu_reload(vcpu);
2153         if (unlikely(r))
2154                 goto out;
2155
2156         kvm_inject_pending_timer_irqs(vcpu);
2157
2158         preempt_disable();
2159
2160         kvm_x86_ops->prepare_guest_switch(vcpu);
2161         kvm_load_guest_fpu(vcpu);
2162
2163         local_irq_disable();
2164
2165         if (signal_pending(current)) {
2166                 local_irq_enable();
2167                 preempt_enable();
2168                 r = -EINTR;
2169                 kvm_run->exit_reason = KVM_EXIT_INTR;
2170                 ++vcpu->stat.signal_exits;
2171                 goto out;
2172         }
2173
2174         if (irqchip_in_kernel(vcpu->kvm))
2175                 kvm_x86_ops->inject_pending_irq(vcpu);
2176         else if (!vcpu->mmio_read_completed)
2177                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2178
2179         vcpu->guest_mode = 1;
2180         kvm_guest_enter();
2181
2182         if (vcpu->requests)
2183                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2184                         kvm_x86_ops->tlb_flush(vcpu);
2185
2186         kvm_x86_ops->run(vcpu, kvm_run);
2187
2188         vcpu->guest_mode = 0;
2189         local_irq_enable();
2190
2191         ++vcpu->stat.exits;
2192
2193         /*
2194          * We must have an instruction between local_irq_enable() and
2195          * kvm_guest_exit(), so the timer interrupt isn't delayed by
2196          * the interrupt shadow.  The stat.exits increment will do nicely.
2197          * But we need to prevent reordering, hence this barrier():
2198          */
2199         barrier();
2200
2201         kvm_guest_exit();
2202
2203         preempt_enable();
2204
2205         /*
2206          * Profile KVM exit RIPs:
2207          */
2208         if (unlikely(prof_on == KVM_PROFILING)) {
2209                 kvm_x86_ops->cache_regs(vcpu);
2210                 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
2211         }
2212
2213         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2214
2215         if (r > 0) {
2216                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2217                         r = -EINTR;
2218                         kvm_run->exit_reason = KVM_EXIT_INTR;
2219                         ++vcpu->stat.request_irq_exits;
2220                         goto out;
2221                 }
2222                 if (!need_resched()) {
2223                         ++vcpu->stat.light_exits;
2224                         goto again;
2225                 }
2226         }
2227
2228 out:
2229         if (r > 0) {
2230                 kvm_resched(vcpu);
2231                 goto preempted;
2232         }
2233
2234         post_kvm_run_save(vcpu, kvm_run);
2235
2236         return r;
2237 }
2238
2239
2240 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2241 {
2242         int r;
2243         sigset_t sigsaved;
2244
2245         vcpu_load(vcpu);
2246
2247         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2248                 kvm_vcpu_block(vcpu);
2249                 vcpu_put(vcpu);
2250                 return -EAGAIN;
2251         }
2252
2253         if (vcpu->sigset_active)
2254                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2255
2256         /* re-sync apic's tpr */
2257         if (!irqchip_in_kernel(vcpu->kvm))
2258                 set_cr8(vcpu, kvm_run->cr8);
2259
2260         if (vcpu->pio.cur_count) {
2261                 r = complete_pio(vcpu);
2262                 if (r)
2263                         goto out;
2264         }
2265 #if CONFIG_HAS_IOMEM
2266         if (vcpu->mmio_needed) {
2267                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2268                 vcpu->mmio_read_completed = 1;
2269                 vcpu->mmio_needed = 0;
2270                 r = emulate_instruction(vcpu, kvm_run,
2271                                         vcpu->mmio_fault_cr2, 0, 1);
2272                 if (r == EMULATE_DO_MMIO) {
2273                         /*
2274                          * Read-modify-write.  Back to userspace.
2275                          */
2276                         r = 0;
2277                         goto out;
2278                 }
2279         }
2280 #endif
2281         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2282                 kvm_x86_ops->cache_regs(vcpu);
2283                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2284                 kvm_x86_ops->decache_regs(vcpu);
2285         }
2286
2287         r = __vcpu_run(vcpu, kvm_run);
2288
2289 out:
2290         if (vcpu->sigset_active)
2291                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2292
2293         vcpu_put(vcpu);
2294         return r;
2295 }
2296
2297 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2298                                    struct kvm_regs *regs)
2299 {
2300         vcpu_load(vcpu);
2301
2302         kvm_x86_ops->cache_regs(vcpu);
2303
2304         regs->rax = vcpu->regs[VCPU_REGS_RAX];
2305         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2306         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2307         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2308         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2309         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2310         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2311         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2312 #ifdef CONFIG_X86_64
2313         regs->r8 = vcpu->regs[VCPU_REGS_R8];
2314         regs->r9 = vcpu->regs[VCPU_REGS_R9];
2315         regs->r10 = vcpu->regs[VCPU_REGS_R10];
2316         regs->r11 = vcpu->regs[VCPU_REGS_R11];
2317         regs->r12 = vcpu->regs[VCPU_REGS_R12];
2318         regs->r13 = vcpu->regs[VCPU_REGS_R13];
2319         regs->r14 = vcpu->regs[VCPU_REGS_R14];
2320         regs->r15 = vcpu->regs[VCPU_REGS_R15];
2321 #endif
2322
2323         regs->rip = vcpu->rip;
2324         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2325
2326         /*
2327          * Don't leak debug flags in case they were set for guest debugging
2328          */
2329         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2330                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2331
2332         vcpu_put(vcpu);
2333
2334         return 0;
2335 }
2336
2337 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2338                                    struct kvm_regs *regs)
2339 {
2340         vcpu_load(vcpu);
2341
2342         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2343         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2344         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2345         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2346         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2347         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2348         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2349         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2350 #ifdef CONFIG_X86_64
2351         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2352         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2353         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2354         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2355         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2356         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2357         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2358         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2359 #endif
2360
2361         vcpu->rip = regs->rip;
2362         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2363
2364         kvm_x86_ops->decache_regs(vcpu);
2365
2366         vcpu_put(vcpu);
2367
2368         return 0;
2369 }
2370
2371 static void get_segment(struct kvm_vcpu *vcpu,
2372                         struct kvm_segment *var, int seg)
2373 {
2374         return kvm_x86_ops->get_segment(vcpu, var, seg);
2375 }
2376
2377 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2378                                     struct kvm_sregs *sregs)
2379 {
2380         struct descriptor_table dt;
2381         int pending_vec;
2382
2383         vcpu_load(vcpu);
2384
2385         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2386         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2387         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2388         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2389         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2390         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2391
2392         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2393         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2394
2395         kvm_x86_ops->get_idt(vcpu, &dt);
2396         sregs->idt.limit = dt.limit;
2397         sregs->idt.base = dt.base;
2398         kvm_x86_ops->get_gdt(vcpu, &dt);
2399         sregs->gdt.limit = dt.limit;
2400         sregs->gdt.base = dt.base;
2401
2402         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2403         sregs->cr0 = vcpu->cr0;
2404         sregs->cr2 = vcpu->cr2;
2405         sregs->cr3 = vcpu->cr3;
2406         sregs->cr4 = vcpu->cr4;
2407         sregs->cr8 = get_cr8(vcpu);
2408         sregs->efer = vcpu->shadow_efer;
2409         sregs->apic_base = kvm_get_apic_base(vcpu);
2410
2411         if (irqchip_in_kernel(vcpu->kvm)) {
2412                 memset(sregs->interrupt_bitmap, 0,
2413                        sizeof sregs->interrupt_bitmap);
2414                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2415                 if (pending_vec >= 0)
2416                         set_bit(pending_vec,
2417                                 (unsigned long *)sregs->interrupt_bitmap);
2418         } else
2419                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2420                        sizeof sregs->interrupt_bitmap);
2421
2422         vcpu_put(vcpu);
2423
2424         return 0;
2425 }
2426
2427 static void set_segment(struct kvm_vcpu *vcpu,
2428                         struct kvm_segment *var, int seg)
2429 {
2430         return kvm_x86_ops->set_segment(vcpu, var, seg);
2431 }
2432
2433 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2434                                     struct kvm_sregs *sregs)
2435 {
2436         int mmu_reset_needed = 0;
2437         int i, pending_vec, max_bits;
2438         struct descriptor_table dt;
2439
2440         vcpu_load(vcpu);
2441
2442         dt.limit = sregs->idt.limit;
2443         dt.base = sregs->idt.base;
2444         kvm_x86_ops->set_idt(vcpu, &dt);
2445         dt.limit = sregs->gdt.limit;
2446         dt.base = sregs->gdt.base;
2447         kvm_x86_ops->set_gdt(vcpu, &dt);
2448
2449         vcpu->cr2 = sregs->cr2;
2450         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2451         vcpu->cr3 = sregs->cr3;
2452
2453         set_cr8(vcpu, sregs->cr8);
2454
2455         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2456 #ifdef CONFIG_X86_64
2457         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2458 #endif
2459         kvm_set_apic_base(vcpu, sregs->apic_base);
2460
2461         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2462
2463         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2464         vcpu->cr0 = sregs->cr0;
2465         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2466
2467         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2468         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2469         if (!is_long_mode(vcpu) && is_pae(vcpu))
2470                 load_pdptrs(vcpu, vcpu->cr3);
2471
2472         if (mmu_reset_needed)
2473                 kvm_mmu_reset_context(vcpu);
2474
2475         if (!irqchip_in_kernel(vcpu->kvm)) {
2476                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2477                        sizeof vcpu->irq_pending);
2478                 vcpu->irq_summary = 0;
2479                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2480                         if (vcpu->irq_pending[i])
2481                                 __set_bit(i, &vcpu->irq_summary);
2482         } else {
2483                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2484                 pending_vec = find_first_bit(
2485                         (const unsigned long *)sregs->interrupt_bitmap,
2486                         max_bits);
2487                 /* Only pending external irq is handled here */
2488                 if (pending_vec < max_bits) {
2489                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2490                         pr_debug("Set back pending irq %d\n",
2491                                  pending_vec);
2492                 }
2493         }
2494
2495         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2496         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2497         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2498         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2499         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2500         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2501
2502         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2503         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2504
2505         vcpu_put(vcpu);
2506
2507         return 0;
2508 }
2509
2510 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2511 {
2512         struct kvm_segment cs;
2513
2514         get_segment(vcpu, &cs, VCPU_SREG_CS);
2515         *db = cs.db;
2516         *l = cs.l;
2517 }
2518 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2519
2520 /*
2521  * Translate a guest virtual address to a guest physical address.
2522  */
2523 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2524                                     struct kvm_translation *tr)
2525 {
2526         unsigned long vaddr = tr->linear_address;
2527         gpa_t gpa;
2528
2529         vcpu_load(vcpu);
2530         mutex_lock(&vcpu->kvm->lock);
2531         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2532         tr->physical_address = gpa;
2533         tr->valid = gpa != UNMAPPED_GVA;
2534         tr->writeable = 1;
2535         tr->usermode = 0;
2536         mutex_unlock(&vcpu->kvm->lock);
2537         vcpu_put(vcpu);
2538
2539         return 0;
2540 }
2541
2542 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2543                                     struct kvm_interrupt *irq)
2544 {
2545         if (irq->irq < 0 || irq->irq >= 256)
2546                 return -EINVAL;
2547         if (irqchip_in_kernel(vcpu->kvm))
2548                 return -ENXIO;
2549         vcpu_load(vcpu);
2550
2551         set_bit(irq->irq, vcpu->irq_pending);
2552         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2553
2554         vcpu_put(vcpu);
2555
2556         return 0;
2557 }
2558
2559 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2560                                       struct kvm_debug_guest *dbg)
2561 {
2562         int r;
2563
2564         vcpu_load(vcpu);
2565
2566         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2567
2568         vcpu_put(vcpu);
2569
2570         return r;
2571 }
2572
2573 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2574                                     unsigned long address,
2575                                     int *type)
2576 {
2577         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2578         unsigned long pgoff;
2579         struct page *page;
2580
2581         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2582         if (pgoff == 0)
2583                 page = virt_to_page(vcpu->run);
2584         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2585                 page = virt_to_page(vcpu->pio_data);
2586         else
2587                 return NOPAGE_SIGBUS;
2588         get_page(page);
2589         if (type != NULL)
2590                 *type = VM_FAULT_MINOR;
2591
2592         return page;
2593 }
2594
2595 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2596         .nopage = kvm_vcpu_nopage,
2597 };
2598
2599 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2600 {
2601         vma->vm_ops = &kvm_vcpu_vm_ops;
2602         return 0;
2603 }
2604
2605 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2606 {
2607         struct kvm_vcpu *vcpu = filp->private_data;
2608
2609         fput(vcpu->kvm->filp);
2610         return 0;
2611 }
2612
2613 static struct file_operations kvm_vcpu_fops = {
2614         .release        = kvm_vcpu_release,
2615         .unlocked_ioctl = kvm_vcpu_ioctl,
2616         .compat_ioctl   = kvm_vcpu_ioctl,
2617         .mmap           = kvm_vcpu_mmap,
2618 };
2619
2620 /*
2621  * Allocates an inode for the vcpu.
2622  */
2623 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2624 {
2625         int fd, r;
2626         struct inode *inode;
2627         struct file *file;
2628
2629         r = anon_inode_getfd(&fd, &inode, &file,
2630                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2631         if (r)
2632                 return r;
2633         atomic_inc(&vcpu->kvm->filp->f_count);
2634         return fd;
2635 }
2636
2637 /*
2638  * Creates some virtual cpus.  Good luck creating more than one.
2639  */
2640 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2641 {
2642         int r;
2643         struct kvm_vcpu *vcpu;
2644
2645         if (!valid_vcpu(n))
2646                 return -EINVAL;
2647
2648         vcpu = kvm_x86_ops->vcpu_create(kvm, n);
2649         if (IS_ERR(vcpu))
2650                 return PTR_ERR(vcpu);
2651
2652         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2653
2654         /* We do fxsave: this must be aligned. */
2655         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2656
2657         vcpu_load(vcpu);
2658         r = kvm_x86_ops->vcpu_reset(vcpu);
2659         if (r == 0)
2660                 r = kvm_mmu_setup(vcpu);
2661         vcpu_put(vcpu);
2662         if (r < 0)
2663                 goto free_vcpu;
2664
2665         mutex_lock(&kvm->lock);
2666         if (kvm->vcpus[n]) {
2667                 r = -EEXIST;
2668                 mutex_unlock(&kvm->lock);
2669                 goto mmu_unload;
2670         }
2671         kvm->vcpus[n] = vcpu;
2672         mutex_unlock(&kvm->lock);
2673
2674         /* Now it's all set up, let userspace reach it */
2675         r = create_vcpu_fd(vcpu);
2676         if (r < 0)
2677                 goto unlink;
2678         return r;
2679
2680 unlink:
2681         mutex_lock(&kvm->lock);
2682         kvm->vcpus[n] = NULL;
2683         mutex_unlock(&kvm->lock);
2684
2685 mmu_unload:
2686         vcpu_load(vcpu);
2687         kvm_mmu_unload(vcpu);
2688         vcpu_put(vcpu);
2689
2690 free_vcpu:
2691         kvm_x86_ops->vcpu_free(vcpu);
2692         return r;
2693 }
2694
2695 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2696 {
2697         if (sigset) {
2698                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2699                 vcpu->sigset_active = 1;
2700                 vcpu->sigset = *sigset;
2701         } else
2702                 vcpu->sigset_active = 0;
2703         return 0;
2704 }
2705
2706 /*
2707  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2708  * we have asm/x86/processor.h
2709  */
2710 struct fxsave {
2711         u16     cwd;
2712         u16     swd;
2713         u16     twd;
2714         u16     fop;
2715         u64     rip;
2716         u64     rdp;
2717         u32     mxcsr;
2718         u32     mxcsr_mask;
2719         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2720 #ifdef CONFIG_X86_64
2721         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2722 #else
2723         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2724 #endif
2725 };
2726
2727 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2728 {
2729         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2730
2731         vcpu_load(vcpu);
2732
2733         memcpy(fpu->fpr, fxsave->st_space, 128);
2734         fpu->fcw = fxsave->cwd;
2735         fpu->fsw = fxsave->swd;
2736         fpu->ftwx = fxsave->twd;
2737         fpu->last_opcode = fxsave->fop;
2738         fpu->last_ip = fxsave->rip;
2739         fpu->last_dp = fxsave->rdp;
2740         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2741
2742         vcpu_put(vcpu);
2743
2744         return 0;
2745 }
2746
2747 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2748 {
2749         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2750
2751         vcpu_load(vcpu);
2752
2753         memcpy(fxsave->st_space, fpu->fpr, 128);
2754         fxsave->cwd = fpu->fcw;
2755         fxsave->swd = fpu->fsw;
2756         fxsave->twd = fpu->ftwx;
2757         fxsave->fop = fpu->last_opcode;
2758         fxsave->rip = fpu->last_ip;
2759         fxsave->rdp = fpu->last_dp;
2760         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2761
2762         vcpu_put(vcpu);
2763
2764         return 0;
2765 }
2766
2767 static long kvm_vcpu_ioctl(struct file *filp,
2768                            unsigned int ioctl, unsigned long arg)
2769 {
2770         struct kvm_vcpu *vcpu = filp->private_data;
2771         void __user *argp = (void __user *)arg;
2772         int r;
2773
2774         switch (ioctl) {
2775         case KVM_RUN:
2776                 r = -EINVAL;
2777                 if (arg)
2778                         goto out;
2779                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2780                 break;
2781         case KVM_GET_REGS: {
2782                 struct kvm_regs kvm_regs;
2783
2784                 memset(&kvm_regs, 0, sizeof kvm_regs);
2785                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2786                 if (r)
2787                         goto out;
2788                 r = -EFAULT;
2789                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2790                         goto out;
2791                 r = 0;
2792                 break;
2793         }
2794         case KVM_SET_REGS: {
2795                 struct kvm_regs kvm_regs;
2796
2797                 r = -EFAULT;
2798                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2799                         goto out;
2800                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2801                 if (r)
2802                         goto out;
2803                 r = 0;
2804                 break;
2805         }
2806         case KVM_GET_SREGS: {
2807                 struct kvm_sregs kvm_sregs;
2808
2809                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2810                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2811                 if (r)
2812                         goto out;
2813                 r = -EFAULT;
2814                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2815                         goto out;
2816                 r = 0;
2817                 break;
2818         }
2819         case KVM_SET_SREGS: {
2820                 struct kvm_sregs kvm_sregs;
2821
2822                 r = -EFAULT;
2823                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2824                         goto out;
2825                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2826                 if (r)
2827                         goto out;
2828                 r = 0;
2829                 break;
2830         }
2831         case KVM_TRANSLATE: {
2832                 struct kvm_translation tr;
2833
2834                 r = -EFAULT;
2835                 if (copy_from_user(&tr, argp, sizeof tr))
2836                         goto out;
2837                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2838                 if (r)
2839                         goto out;
2840                 r = -EFAULT;
2841                 if (copy_to_user(argp, &tr, sizeof tr))
2842                         goto out;
2843                 r = 0;
2844                 break;
2845         }
2846         case KVM_INTERRUPT: {
2847                 struct kvm_interrupt irq;
2848
2849                 r = -EFAULT;
2850                 if (copy_from_user(&irq, argp, sizeof irq))
2851                         goto out;
2852                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2853                 if (r)
2854                         goto out;
2855                 r = 0;
2856                 break;
2857         }
2858         case KVM_DEBUG_GUEST: {
2859                 struct kvm_debug_guest dbg;
2860
2861                 r = -EFAULT;
2862                 if (copy_from_user(&dbg, argp, sizeof dbg))
2863                         goto out;
2864                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2865                 if (r)
2866                         goto out;
2867                 r = 0;
2868                 break;
2869         }
2870         case KVM_SET_SIGNAL_MASK: {
2871                 struct kvm_signal_mask __user *sigmask_arg = argp;
2872                 struct kvm_signal_mask kvm_sigmask;
2873                 sigset_t sigset, *p;
2874
2875                 p = NULL;
2876                 if (argp) {
2877                         r = -EFAULT;
2878                         if (copy_from_user(&kvm_sigmask, argp,
2879                                            sizeof kvm_sigmask))
2880                                 goto out;
2881                         r = -EINVAL;
2882                         if (kvm_sigmask.len != sizeof sigset)
2883                                 goto out;
2884                         r = -EFAULT;
2885                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2886                                            sizeof sigset))
2887                                 goto out;
2888                         p = &sigset;
2889                 }
2890                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2891                 break;
2892         }
2893         case KVM_GET_FPU: {
2894                 struct kvm_fpu fpu;
2895
2896                 memset(&fpu, 0, sizeof fpu);
2897                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2898                 if (r)
2899                         goto out;
2900                 r = -EFAULT;
2901                 if (copy_to_user(argp, &fpu, sizeof fpu))
2902                         goto out;
2903                 r = 0;
2904                 break;
2905         }
2906         case KVM_SET_FPU: {
2907                 struct kvm_fpu fpu;
2908
2909                 r = -EFAULT;
2910                 if (copy_from_user(&fpu, argp, sizeof fpu))
2911                         goto out;
2912                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2913                 if (r)
2914                         goto out;
2915                 r = 0;
2916                 break;
2917         }
2918         default:
2919                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2920         }
2921 out:
2922         return r;
2923 }
2924
2925 static long kvm_vm_ioctl(struct file *filp,
2926                            unsigned int ioctl, unsigned long arg)
2927 {
2928         struct kvm *kvm = filp->private_data;
2929         void __user *argp = (void __user *)arg;
2930         int r = -EINVAL;
2931
2932         switch (ioctl) {
2933         case KVM_CREATE_VCPU:
2934                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2935                 if (r < 0)
2936                         goto out;
2937                 break;
2938         case KVM_SET_MEMORY_REGION: {
2939                 struct kvm_memory_region kvm_mem;
2940                 struct kvm_userspace_memory_region kvm_userspace_mem;
2941
2942                 r = -EFAULT;
2943                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2944                         goto out;
2945                 kvm_userspace_mem.slot = kvm_mem.slot;
2946                 kvm_userspace_mem.flags = kvm_mem.flags;
2947                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2948                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2949                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2950                 if (r)
2951                         goto out;
2952                 break;
2953         }
2954         case KVM_SET_USER_MEMORY_REGION: {
2955                 struct kvm_userspace_memory_region kvm_userspace_mem;
2956
2957                 r = -EFAULT;
2958                 if (copy_from_user(&kvm_userspace_mem, argp,
2959                                                 sizeof kvm_userspace_mem))
2960                         goto out;
2961
2962                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2963                 if (r)
2964                         goto out;
2965                 break;
2966         }
2967         case KVM_SET_NR_MMU_PAGES:
2968                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2969                 if (r)
2970                         goto out;
2971                 break;
2972         case KVM_GET_NR_MMU_PAGES:
2973                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2974                 break;
2975         case KVM_GET_DIRTY_LOG: {
2976                 struct kvm_dirty_log log;
2977
2978                 r = -EFAULT;
2979                 if (copy_from_user(&log, argp, sizeof log))
2980                         goto out;
2981                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2982                 if (r)
2983                         goto out;
2984                 break;
2985         }
2986         case KVM_SET_MEMORY_ALIAS: {
2987                 struct kvm_memory_alias alias;
2988
2989                 r = -EFAULT;
2990                 if (copy_from_user(&alias, argp, sizeof alias))
2991                         goto out;
2992                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2993                 if (r)
2994                         goto out;
2995                 break;
2996         }
2997         case KVM_CREATE_IRQCHIP:
2998                 r = -ENOMEM;
2999                 kvm->vpic = kvm_create_pic(kvm);
3000                 if (kvm->vpic) {
3001                         r = kvm_ioapic_init(kvm);
3002                         if (r) {
3003                                 kfree(kvm->vpic);
3004                                 kvm->vpic = NULL;
3005                                 goto out;
3006                         }
3007                 } else
3008                         goto out;
3009                 break;
3010         case KVM_IRQ_LINE: {
3011                 struct kvm_irq_level irq_event;
3012
3013                 r = -EFAULT;
3014                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3015                         goto out;
3016                 if (irqchip_in_kernel(kvm)) {
3017                         mutex_lock(&kvm->lock);
3018                         if (irq_event.irq < 16)
3019                                 kvm_pic_set_irq(pic_irqchip(kvm),
3020                                         irq_event.irq,
3021                                         irq_event.level);
3022                         kvm_ioapic_set_irq(kvm->vioapic,
3023                                         irq_event.irq,
3024                                         irq_event.level);
3025                         mutex_unlock(&kvm->lock);
3026                         r = 0;
3027                 }
3028                 break;
3029         }
3030         case KVM_GET_IRQCHIP: {
3031                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3032                 struct kvm_irqchip chip;
3033
3034                 r = -EFAULT;
3035                 if (copy_from_user(&chip, argp, sizeof chip))
3036                         goto out;
3037                 r = -ENXIO;
3038                 if (!irqchip_in_kernel(kvm))
3039                         goto out;
3040                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
3041                 if (r)
3042                         goto out;
3043                 r = -EFAULT;
3044                 if (copy_to_user(argp, &chip, sizeof chip))
3045                         goto out;
3046                 r = 0;
3047                 break;
3048         }
3049         case KVM_SET_IRQCHIP: {
3050                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3051                 struct kvm_irqchip chip;
3052
3053                 r = -EFAULT;
3054                 if (copy_from_user(&chip, argp, sizeof chip))
3055                         goto out;
3056                 r = -ENXIO;
3057                 if (!irqchip_in_kernel(kvm))
3058                         goto out;
3059                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
3060                 if (r)
3061                         goto out;
3062                 r = 0;
3063                 break;
3064         }
3065         default:
3066                 ;
3067         }
3068 out:
3069         return r;
3070 }
3071
3072 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
3073                                   unsigned long address,
3074                                   int *type)
3075 {
3076         struct kvm *kvm = vma->vm_file->private_data;
3077         unsigned long pgoff;
3078         struct page *page;
3079
3080         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
3081         page = gfn_to_page(kvm, pgoff);
3082         if (is_error_page(page)) {
3083                 kvm_release_page(page);
3084                 return NOPAGE_SIGBUS;
3085         }
3086         if (type != NULL)
3087                 *type = VM_FAULT_MINOR;
3088
3089         return page;
3090 }
3091
3092 static struct vm_operations_struct kvm_vm_vm_ops = {
3093         .nopage = kvm_vm_nopage,
3094 };
3095
3096 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
3097 {
3098         vma->vm_ops = &kvm_vm_vm_ops;
3099         return 0;
3100 }
3101
3102 static struct file_operations kvm_vm_fops = {
3103         .release        = kvm_vm_release,
3104         .unlocked_ioctl = kvm_vm_ioctl,
3105         .compat_ioctl   = kvm_vm_ioctl,
3106         .mmap           = kvm_vm_mmap,
3107 };
3108
3109 static int kvm_dev_ioctl_create_vm(void)
3110 {
3111         int fd, r;
3112         struct inode *inode;
3113         struct file *file;
3114         struct kvm *kvm;
3115
3116         kvm = kvm_create_vm();
3117         if (IS_ERR(kvm))
3118                 return PTR_ERR(kvm);
3119         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
3120         if (r) {
3121                 kvm_destroy_vm(kvm);
3122                 return r;
3123         }
3124
3125         kvm->filp = file;
3126
3127         return fd;
3128 }
3129
3130 static long kvm_dev_ioctl(struct file *filp,
3131                           unsigned int ioctl, unsigned long arg)
3132 {
3133         void __user *argp = (void __user *)arg;
3134         long r = -EINVAL;
3135
3136         switch (ioctl) {
3137         case KVM_GET_API_VERSION:
3138                 r = -EINVAL;
3139                 if (arg)
3140                         goto out;
3141                 r = KVM_API_VERSION;
3142                 break;
3143         case KVM_CREATE_VM:
3144                 r = -EINVAL;
3145                 if (arg)
3146                         goto out;
3147                 r = kvm_dev_ioctl_create_vm();
3148                 break;
3149         case KVM_CHECK_EXTENSION: {
3150                 int ext = (long)argp;
3151
3152                 switch (ext) {
3153                 case KVM_CAP_IRQCHIP:
3154                 case KVM_CAP_HLT:
3155                 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3156                 case KVM_CAP_USER_MEMORY:
3157                         r = 1;
3158                         break;
3159                 default:
3160                         r = 0;
3161                         break;
3162                 }
3163                 break;
3164         }
3165         case KVM_GET_VCPU_MMAP_SIZE:
3166                 r = -EINVAL;
3167                 if (arg)
3168                         goto out;
3169                 r = 2 * PAGE_SIZE;
3170                 break;
3171         default:
3172                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3173         }
3174 out:
3175         return r;
3176 }
3177
3178 static struct file_operations kvm_chardev_ops = {
3179         .unlocked_ioctl = kvm_dev_ioctl,
3180         .compat_ioctl   = kvm_dev_ioctl,
3181 };
3182
3183 static struct miscdevice kvm_dev = {
3184         KVM_MINOR,
3185         "kvm",
3186         &kvm_chardev_ops,
3187 };
3188
3189 /*
3190  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
3191  * cached on it.
3192  */
3193 static void decache_vcpus_on_cpu(int cpu)
3194 {
3195         struct kvm *vm;
3196         struct kvm_vcpu *vcpu;
3197         int i;
3198
3199         spin_lock(&kvm_lock);
3200         list_for_each_entry(vm, &vm_list, vm_list)
3201                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3202                         vcpu = vm->vcpus[i];
3203                         if (!vcpu)
3204                                 continue;
3205                         /*
3206                          * If the vcpu is locked, then it is running on some
3207                          * other cpu and therefore it is not cached on the
3208                          * cpu in question.
3209                          *
3210                          * If it's not locked, check the last cpu it executed
3211                          * on.
3212                          */
3213                         if (mutex_trylock(&vcpu->mutex)) {
3214                                 if (vcpu->cpu == cpu) {
3215                                         kvm_x86_ops->vcpu_decache(vcpu);
3216                                         vcpu->cpu = -1;
3217                                 }
3218                                 mutex_unlock(&vcpu->mutex);
3219                         }
3220                 }
3221         spin_unlock(&kvm_lock);
3222 }
3223
3224 static void hardware_enable(void *junk)
3225 {
3226         int cpu = raw_smp_processor_id();
3227
3228         if (cpu_isset(cpu, cpus_hardware_enabled))
3229                 return;
3230         cpu_set(cpu, cpus_hardware_enabled);
3231         kvm_x86_ops->hardware_enable(NULL);
3232 }
3233
3234 static void hardware_disable(void *junk)
3235 {
3236         int cpu = raw_smp_processor_id();
3237
3238         if (!cpu_isset(cpu, cpus_hardware_enabled))
3239                 return;
3240         cpu_clear(cpu, cpus_hardware_enabled);
3241         decache_vcpus_on_cpu(cpu);
3242         kvm_x86_ops->hardware_disable(NULL);
3243 }
3244
3245 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3246                            void *v)
3247 {
3248         int cpu = (long)v;
3249
3250         switch (val) {
3251         case CPU_DYING:
3252         case CPU_DYING_FROZEN:
3253                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3254                        cpu);
3255                 hardware_disable(NULL);
3256                 break;
3257         case CPU_UP_CANCELED:
3258         case CPU_UP_CANCELED_FROZEN:
3259                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3260                        cpu);
3261                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3262                 break;
3263         case CPU_ONLINE:
3264         case CPU_ONLINE_FROZEN:
3265                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3266                        cpu);
3267                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3268                 break;
3269         }
3270         return NOTIFY_OK;
3271 }
3272
3273 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3274                       void *v)
3275 {
3276         if (val == SYS_RESTART) {
3277                 /*
3278                  * Some (well, at least mine) BIOSes hang on reboot if
3279                  * in vmx root mode.
3280                  */
3281                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3282                 on_each_cpu(hardware_disable, NULL, 0, 1);
3283         }
3284         return NOTIFY_OK;
3285 }
3286
3287 static struct notifier_block kvm_reboot_notifier = {
3288         .notifier_call = kvm_reboot,
3289         .priority = 0,
3290 };
3291
3292 void kvm_io_bus_init(struct kvm_io_bus *bus)
3293 {
3294         memset(bus, 0, sizeof(*bus));
3295 }
3296
3297 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3298 {
3299         int i;
3300
3301         for (i = 0; i < bus->dev_count; i++) {
3302                 struct kvm_io_device *pos = bus->devs[i];
3303
3304                 kvm_iodevice_destructor(pos);
3305         }
3306 }
3307
3308 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3309 {
3310         int i;
3311
3312         for (i = 0; i < bus->dev_count; i++) {
3313                 struct kvm_io_device *pos = bus->devs[i];
3314
3315                 if (pos->in_range(pos, addr))
3316                         return pos;
3317         }
3318
3319         return NULL;
3320 }
3321
3322 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3323 {
3324         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3325
3326         bus->devs[bus->dev_count++] = dev;
3327 }
3328
3329 static struct notifier_block kvm_cpu_notifier = {
3330         .notifier_call = kvm_cpu_hotplug,
3331         .priority = 20, /* must be > scheduler priority */
3332 };
3333
3334 static u64 stat_get(void *_offset)
3335 {
3336         unsigned offset = (long)_offset;
3337         u64 total = 0;
3338         struct kvm *kvm;
3339         struct kvm_vcpu *vcpu;
3340         int i;
3341
3342         spin_lock(&kvm_lock);
3343         list_for_each_entry(kvm, &vm_list, vm_list)
3344                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3345                         vcpu = kvm->vcpus[i];
3346                         if (vcpu)
3347                                 total += *(u32 *)((void *)vcpu + offset);
3348                 }
3349         spin_unlock(&kvm_lock);
3350         return total;
3351 }
3352
3353 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3354
3355 static __init void kvm_init_debug(void)
3356 {
3357         struct kvm_stats_debugfs_item *p;
3358
3359         debugfs_dir = debugfs_create_dir("kvm", NULL);
3360         for (p = debugfs_entries; p->name; ++p)
3361                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3362                                                 (void *)(long)p->offset,
3363                                                 &stat_fops);
3364 }
3365
3366 static void kvm_exit_debug(void)
3367 {
3368         struct kvm_stats_debugfs_item *p;
3369
3370         for (p = debugfs_entries; p->name; ++p)
3371                 debugfs_remove(p->dentry);
3372         debugfs_remove(debugfs_dir);
3373 }
3374
3375 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3376 {
3377         hardware_disable(NULL);
3378         return 0;
3379 }
3380
3381 static int kvm_resume(struct sys_device *dev)
3382 {
3383         hardware_enable(NULL);
3384         return 0;
3385 }
3386
3387 static struct sysdev_class kvm_sysdev_class = {
3388         .name = "kvm",
3389         .suspend = kvm_suspend,
3390         .resume = kvm_resume,
3391 };
3392
3393 static struct sys_device kvm_sysdev = {
3394         .id = 0,
3395         .cls = &kvm_sysdev_class,
3396 };
3397
3398 struct page *bad_page;
3399
3400 static inline
3401 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3402 {
3403         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3404 }
3405
3406 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3407 {
3408         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3409
3410         kvm_x86_ops->vcpu_load(vcpu, cpu);
3411 }
3412
3413 static void kvm_sched_out(struct preempt_notifier *pn,
3414                           struct task_struct *next)
3415 {
3416         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3417
3418         kvm_x86_ops->vcpu_put(vcpu);
3419 }
3420
3421 int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
3422                   struct module *module)
3423 {
3424         int r;
3425         int cpu;
3426
3427         if (kvm_x86_ops) {
3428                 printk(KERN_ERR "kvm: already loaded the other module\n");
3429                 return -EEXIST;
3430         }
3431
3432         if (!ops->cpu_has_kvm_support()) {
3433                 printk(KERN_ERR "kvm: no hardware support\n");
3434                 return -EOPNOTSUPP;
3435         }
3436         if (ops->disabled_by_bios()) {
3437                 printk(KERN_ERR "kvm: disabled by bios\n");
3438                 return -EOPNOTSUPP;
3439         }
3440
3441         kvm_x86_ops = ops;
3442
3443         r = kvm_x86_ops->hardware_setup();
3444         if (r < 0)
3445                 goto out;
3446
3447         for_each_online_cpu(cpu) {
3448                 smp_call_function_single(cpu,
3449                                 kvm_x86_ops->check_processor_compatibility,
3450                                 &r, 0, 1);
3451                 if (r < 0)
3452                         goto out_free_0;
3453         }
3454
3455         on_each_cpu(hardware_enable, NULL, 0, 1);
3456         r = register_cpu_notifier(&kvm_cpu_notifier);
3457         if (r)
3458                 goto out_free_1;
3459         register_reboot_notifier(&kvm_reboot_notifier);
3460
3461         r = sysdev_class_register(&kvm_sysdev_class);
3462         if (r)
3463                 goto out_free_2;
3464
3465         r = sysdev_register(&kvm_sysdev);
3466         if (r)
3467                 goto out_free_3;
3468
3469         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3470         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3471                                            __alignof__(struct kvm_vcpu), 0, 0);
3472         if (!kvm_vcpu_cache) {
3473                 r = -ENOMEM;
3474                 goto out_free_4;
3475         }
3476
3477         kvm_chardev_ops.owner = module;
3478
3479         r = misc_register(&kvm_dev);
3480         if (r) {
3481                 printk(KERN_ERR "kvm: misc device register failed\n");
3482                 goto out_free;
3483         }
3484
3485         kvm_preempt_ops.sched_in = kvm_sched_in;
3486         kvm_preempt_ops.sched_out = kvm_sched_out;
3487
3488         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3489
3490         return 0;
3491
3492 out_free:
3493         kmem_cache_destroy(kvm_vcpu_cache);
3494 out_free_4:
3495         sysdev_unregister(&kvm_sysdev);
3496 out_free_3:
3497         sysdev_class_unregister(&kvm_sysdev_class);
3498 out_free_2:
3499         unregister_reboot_notifier(&kvm_reboot_notifier);
3500         unregister_cpu_notifier(&kvm_cpu_notifier);
3501 out_free_1:
3502         on_each_cpu(hardware_disable, NULL, 0, 1);
3503 out_free_0:
3504         kvm_x86_ops->hardware_unsetup();
3505 out:
3506         kvm_x86_ops = NULL;
3507         return r;
3508 }
3509 EXPORT_SYMBOL_GPL(kvm_init_x86);
3510
3511 void kvm_exit_x86(void)
3512 {
3513         misc_deregister(&kvm_dev);
3514         kmem_cache_destroy(kvm_vcpu_cache);
3515         sysdev_unregister(&kvm_sysdev);
3516         sysdev_class_unregister(&kvm_sysdev_class);
3517         unregister_reboot_notifier(&kvm_reboot_notifier);
3518         unregister_cpu_notifier(&kvm_cpu_notifier);
3519         on_each_cpu(hardware_disable, NULL, 0, 1);
3520         kvm_x86_ops->hardware_unsetup();
3521         kvm_x86_ops = NULL;
3522 }
3523 EXPORT_SYMBOL_GPL(kvm_exit_x86);
3524
3525 static __init int kvm_init(void)
3526 {
3527         int r;
3528
3529         r = kvm_mmu_module_init();
3530         if (r)
3531                 goto out4;
3532
3533         kvm_init_debug();
3534
3535         kvm_arch_init();
3536
3537         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3538
3539         if (bad_page == NULL) {
3540                 r = -ENOMEM;
3541                 goto out;
3542         }
3543
3544         return 0;
3545
3546 out:
3547         kvm_exit_debug();
3548         kvm_mmu_module_exit();
3549 out4:
3550         return r;
3551 }
3552
3553 static __exit void kvm_exit(void)
3554 {
3555         kvm_exit_debug();
3556         __free_page(bad_page);
3557         kvm_mmu_module_exit();
3558 }
3559
3560 module_init(kvm_init)
3561 module_exit(kvm_exit)