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

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "kvm.h"
  19 #include "x86.h"
  20 #include "x86_emulate.h"
  21 #include "segment_descriptor.h"
  22 #include "irq.h"
  23
  24 #include <linux/kvm.h>
  25 #include <linux/module.h>
  26 #include <linux/errno.h>
  27 #include <linux/percpu.h>
  28 #include <linux/gfp.h>
  29 #include <linux/mm.h>
  30 #include <linux/miscdevice.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/reboot.h>
  33 #include <linux/debugfs.h>
  34 #include <linux/highmem.h>
  35 #include <linux/file.h>
  36 #include <linux/sysdev.h>
  37 #include <linux/cpu.h>
  38 #include <linux/sched.h>
  39 #include <linux/cpumask.h>
  40 #include <linux/smp.h>
  41 #include <linux/anon_inodes.h>
  42 #include <linux/profile.h>
  43 #include <linux/kvm_para.h>
  44 #include <linux/pagemap.h>
  45 #include <linux/mman.h>
  46
  47 #include <asm/processor.h>
  48 #include <asm/msr.h>
  49 #include <asm/io.h>
  50 #include <asm/uaccess.h>
  51 #include <asm/desc.h>
  52
  53 MODULE_AUTHOR("Qumranet");
  54 MODULE_LICENSE("GPL");
  55
  56 static DEFINE_SPINLOCK(kvm_lock);
  57 static LIST_HEAD(vm_list);
  58
  59 static cpumask_t cpus_hardware_enabled;
  60
  61 struct kvm_x86_ops *kvm_x86_ops;
  62 struct kmem_cache *kvm_vcpu_cache;
  63 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  64
  65 static __read_mostly struct preempt_ops kvm_preempt_ops;
  66
  67 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  68
  69 static struct kvm_stats_debugfs_item {
  70         const char *name;
  71         int offset;
  72         struct dentry *dentry;
  73 } debugfs_entries[] = {
  74         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  75         { "pf_guest", STAT_OFFSET(pf_guest) },
  76         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  77         { "invlpg", STAT_OFFSET(invlpg) },
  78         { "exits", STAT_OFFSET(exits) },
  79         { "io_exits", STAT_OFFSET(io_exits) },
  80         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  81         { "signal_exits", STAT_OFFSET(signal_exits) },
  82         { "irq_window", STAT_OFFSET(irq_window_exits) },
  83         { "halt_exits", STAT_OFFSET(halt_exits) },
  84         { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
  85         { "request_irq", STAT_OFFSET(request_irq_exits) },
  86         { "irq_exits", STAT_OFFSET(irq_exits) },
  87         { "light_exits", STAT_OFFSET(light_exits) },
  88         { "efer_reload", STAT_OFFSET(efer_reload) },
  89         { NULL }
  90 };
  91
  92 static struct dentry *debugfs_dir;
  93
  94 #define CR0_RESERVED_BITS                                               \
  95         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  96                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  97                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  98 #define CR4_RESERVED_BITS                                               \
  99         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
 100                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
 101                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
 102                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
 103
 104 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
 105 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
 106
 107 #ifdef CONFIG_X86_64
 108 /* LDT or TSS descriptor in the GDT. 16 bytes. */
 109 struct segment_descriptor_64 {
 110         struct segment_descriptor s;
 111         u32 base_higher;
 112         u32 pad_zero;
 113 };
 114
 115 #endif
 116
 117 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 118                            unsigned long arg);
 119
 120 unsigned long segment_base(u16 selector)
 121 {
 122         struct descriptor_table gdt;
 123         struct segment_descriptor *d;
 124         unsigned long table_base;
 125         unsigned long v;
 126
 127         if (selector == 0)
 128                 return 0;
 129
 130         asm("sgdt %0" : "=m"(gdt));
 131         table_base = gdt.base;
 132
 133         if (selector & 4) {           /* from ldt */
 134                 u16 ldt_selector;
 135
 136                 asm("sldt %0" : "=g"(ldt_selector));
 137                 table_base = segment_base(ldt_selector);
 138         }
 139         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 140         v = d->base_low | ((unsigned long)d->base_mid << 16) |
 141                 ((unsigned long)d->base_high << 24);
 142 #ifdef CONFIG_X86_64
 143         if (d->system == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 144                 v |= ((unsigned long) \
 145                       ((struct segment_descriptor_64 *)d)->base_higher) << 32;
 146 #endif
 147         return v;
 148 }
 149 EXPORT_SYMBOL_GPL(segment_base);
 150
 151 static inline int valid_vcpu(int n)
 152 {
 153         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 154 }
 155
 156 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 157 {
 158         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
 159                 return;
 160
 161         vcpu->guest_fpu_loaded = 1;
 162         fx_save(&vcpu->host_fx_image);
 163         fx_restore(&vcpu->guest_fx_image);
 164 }
 165 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
 166
 167 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 168 {
 169         if (!vcpu->guest_fpu_loaded)
 170                 return;
 171
 172         vcpu->guest_fpu_loaded = 0;
 173         fx_save(&vcpu->guest_fx_image);
 174         fx_restore(&vcpu->host_fx_image);
 175 }
 176 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
 177
 178 /*
 179  * Switches to specified vcpu, until a matching vcpu_put()
 180  */
 181 void vcpu_load(struct kvm_vcpu *vcpu)
 182 {
 183         int cpu;
 184
 185         mutex_lock(&vcpu->mutex);
 186         cpu = get_cpu();
 187         preempt_notifier_register(&vcpu->preempt_notifier);
 188         kvm_arch_vcpu_load(vcpu, cpu);
 189         put_cpu();
 190 }
 191
 192 void vcpu_put(struct kvm_vcpu *vcpu)
 193 {
 194         preempt_disable();
 195         kvm_arch_vcpu_put(vcpu);
 196         preempt_notifier_unregister(&vcpu->preempt_notifier);
 197         preempt_enable();
 198         mutex_unlock(&vcpu->mutex);
 199 }
 200
 201 static void ack_flush(void *_completed)
 202 {
 203 }
 204
 205 void kvm_flush_remote_tlbs(struct kvm *kvm)
 206 {
 207         int i, cpu;
 208         cpumask_t cpus;
 209         struct kvm_vcpu *vcpu;
 210
 211         cpus_clear(cpus);
 212         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 213                 vcpu = kvm->vcpus[i];
 214                 if (!vcpu)
 215                         continue;
 216                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 217                         continue;
 218                 cpu = vcpu->cpu;
 219                 if (cpu != -1 && cpu != raw_smp_processor_id())
 220                         cpu_set(cpu, cpus);
 221         }
 222         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 223 }
 224
 225 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 226 {
 227         struct page *page;
 228         int r;
 229
 230         mutex_init(&vcpu->mutex);
 231         vcpu->cpu = -1;
 232         vcpu->mmu.root_hpa = INVALID_PAGE;
 233         vcpu->kvm = kvm;
 234         vcpu->vcpu_id = id;
 235         if (!irqchip_in_kernel(kvm) || id == 0)
 236                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
 237         else
 238                 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
 239         init_waitqueue_head(&vcpu->wq);
 240
 241         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 242         if (!page) {
 243                 r = -ENOMEM;
 244                 goto fail;
 245         }
 246         vcpu->run = page_address(page);
 247
 248         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 249         if (!page) {
 250                 r = -ENOMEM;
 251                 goto fail_free_run;
 252         }
 253         vcpu->pio_data = page_address(page);
 254
 255         r = kvm_mmu_create(vcpu);
 256         if (r < 0)
 257                 goto fail_free_pio_data;
 258
 259         if (irqchip_in_kernel(kvm)) {
 260                 r = kvm_create_lapic(vcpu);
 261                 if (r < 0)
 262                         goto fail_mmu_destroy;
 263         }
 264
 265         return 0;
 266
 267 fail_mmu_destroy:
 268         kvm_mmu_destroy(vcpu);
 269 fail_free_pio_data:
 270         free_page((unsigned long)vcpu->pio_data);
 271 fail_free_run:
 272         free_page((unsigned long)vcpu->run);
 273 fail:
 274         return r;
 275 }
 276 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 277
 278 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 279 {
 280         kvm_free_lapic(vcpu);
 281         kvm_mmu_destroy(vcpu);
 282         free_page((unsigned long)vcpu->pio_data);
 283         free_page((unsigned long)vcpu->run);
 284 }
 285 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 286
 287 static struct kvm *kvm_create_vm(void)
 288 {
 289         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 290
 291         if (!kvm)
 292                 return ERR_PTR(-ENOMEM);
 293
 294         kvm_io_bus_init(&kvm->pio_bus);
 295         mutex_init(&kvm->lock);
 296         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 297         kvm_io_bus_init(&kvm->mmio_bus);
 298         spin_lock(&kvm_lock);
 299         list_add(&kvm->vm_list, &vm_list);
 300         spin_unlock(&kvm_lock);
 301         return kvm;
 302 }
 303
 304 /*
 305  * Free any memory in @free but not in @dont.
 306  */
 307 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 308                                   struct kvm_memory_slot *dont)
 309 {
 310         if (!dont || free->rmap != dont->rmap)
 311                 vfree(free->rmap);
 312
 313         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 314                 vfree(free->dirty_bitmap);
 315
 316         free->npages = 0;
 317         free->dirty_bitmap = NULL;
 318         free->rmap = NULL;
 319 }
 320
 321 static void kvm_free_physmem(struct kvm *kvm)
 322 {
 323         int i;
 324
 325         for (i = 0; i < kvm->nmemslots; ++i)
 326                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 327 }
 328
 329 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 330 {
 331         int i;
 332
 333         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
 334                 if (vcpu->pio.guest_pages[i]) {
 335                         kvm_release_page(vcpu->pio.guest_pages[i]);
 336                         vcpu->pio.guest_pages[i] = NULL;
 337                 }
 338 }
 339
 340 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 341 {
 342         vcpu_load(vcpu);
 343         kvm_mmu_unload(vcpu);
 344         vcpu_put(vcpu);
 345 }
 346
 347 static void kvm_free_vcpus(struct kvm *kvm)
 348 {
 349         unsigned int i;
 350
 351         /*
 352          * Unpin any mmu pages first.
 353          */
 354         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 355                 if (kvm->vcpus[i])
 356                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
 357         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 358                 if (kvm->vcpus[i]) {
 359                         kvm_x86_ops->vcpu_free(kvm->vcpus[i]);
 360                         kvm->vcpus[i] = NULL;
 361                 }
 362         }
 363
 364 }
 365
 366 static void kvm_destroy_vm(struct kvm *kvm)
 367 {
 368         spin_lock(&kvm_lock);
 369         list_del(&kvm->vm_list);
 370         spin_unlock(&kvm_lock);
 371         kvm_io_bus_destroy(&kvm->pio_bus);
 372         kvm_io_bus_destroy(&kvm->mmio_bus);
 373         kfree(kvm->vpic);
 374         kfree(kvm->vioapic);
 375         kvm_free_vcpus(kvm);
 376         kvm_free_physmem(kvm);
 377         kfree(kvm);
 378 }
 379
 380 static int kvm_vm_release(struct inode *inode, struct file *filp)
 381 {
 382         struct kvm *kvm = filp->private_data;
 383
 384         kvm_destroy_vm(kvm);
 385         return 0;
 386 }
 387
 388 static void inject_gp(struct kvm_vcpu *vcpu)
 389 {
 390         kvm_x86_ops->inject_gp(vcpu, 0);
 391 }
 392
 393 /*
 394  * Load the pae pdptrs.  Return true is they are all valid.
 395  */
 396 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 397 {
 398         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 399         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 400         int i;
 401         int ret;
 402         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
 403
 404         mutex_lock(&vcpu->kvm->lock);
 405         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 406                                   offset * sizeof(u64), sizeof(pdpte));
 407         if (ret < 0) {
 408                 ret = 0;
 409                 goto out;
 410         }
 411         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 412                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 413                         ret = 0;
 414                         goto out;
 415                 }
 416         }
 417         ret = 1;
 418
 419         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
 420 out:
 421         mutex_unlock(&vcpu->kvm->lock);
 422
 423         return ret;
 424 }
 425
 426 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 427 {
 428         if (cr0 & CR0_RESERVED_BITS) {
 429                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 430                        cr0, vcpu->cr0);
 431                 inject_gp(vcpu);
 432                 return;
 433         }
 434
 435         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 436                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 437                 inject_gp(vcpu);
 438                 return;
 439         }
 440
 441         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 442                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 443                        "and a clear PE flag\n");
 444                 inject_gp(vcpu);
 445                 return;
 446         }
 447
 448         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 449 #ifdef CONFIG_X86_64
 450                 if ((vcpu->shadow_efer & EFER_LME)) {
 451                         int cs_db, cs_l;
 452
 453                         if (!is_pae(vcpu)) {
 454                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 455                                        "in long mode while PAE is disabled\n");
 456                                 inject_gp(vcpu);
 457                                 return;
 458                         }
 459                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 460                         if (cs_l) {
 461                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 462                                        "in long mode while CS.L == 1\n");
 463                                 inject_gp(vcpu);
 464                                 return;
 465
 466                         }
 467                 } else
 468 #endif
 469                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 470                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 471                                "reserved bits\n");
 472                         inject_gp(vcpu);
 473                         return;
 474                 }
 475
 476         }
 477
 478         kvm_x86_ops->set_cr0(vcpu, cr0);
 479         vcpu->cr0 = cr0;
 480
 481         mutex_lock(&vcpu->kvm->lock);
 482         kvm_mmu_reset_context(vcpu);
 483         mutex_unlock(&vcpu->kvm->lock);
 484         return;
 485 }
 486 EXPORT_SYMBOL_GPL(set_cr0);
 487
 488 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 489 {
 490         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 491 }
 492 EXPORT_SYMBOL_GPL(lmsw);
 493
 494 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 495 {
 496         if (cr4 & CR4_RESERVED_BITS) {
 497                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 498                 inject_gp(vcpu);
 499                 return;
 500         }
 501
 502         if (is_long_mode(vcpu)) {
 503                 if (!(cr4 & X86_CR4_PAE)) {
 504                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 505                                "in long mode\n");
 506                         inject_gp(vcpu);
 507                         return;
 508                 }
 509         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 510                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 511                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 512                 inject_gp(vcpu);
 513                 return;
 514         }
 515
 516         if (cr4 & X86_CR4_VMXE) {
 517                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 518                 inject_gp(vcpu);
 519                 return;
 520         }
 521         kvm_x86_ops->set_cr4(vcpu, cr4);
 522         vcpu->cr4 = cr4;
 523         mutex_lock(&vcpu->kvm->lock);
 524         kvm_mmu_reset_context(vcpu);
 525         mutex_unlock(&vcpu->kvm->lock);
 526 }
 527 EXPORT_SYMBOL_GPL(set_cr4);
 528
 529 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 530 {
 531         if (is_long_mode(vcpu)) {
 532                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 533                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 534                         inject_gp(vcpu);
 535                         return;
 536                 }
 537         } else {
 538                 if (is_pae(vcpu)) {
 539                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 540                                 printk(KERN_DEBUG
 541                                        "set_cr3: #GP, reserved bits\n");
 542                                 inject_gp(vcpu);
 543                                 return;
 544                         }
 545                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 546                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 547                                        "reserved bits\n");
 548                                 inject_gp(vcpu);
 549                                 return;
 550                         }
 551                 }
 552                 /*
 553                  * We don't check reserved bits in nonpae mode, because
 554                  * this isn't enforced, and VMware depends on this.
 555                  */
 556         }
 557
 558         mutex_lock(&vcpu->kvm->lock);
 559         /*
 560          * Does the new cr3 value map to physical memory? (Note, we
 561          * catch an invalid cr3 even in real-mode, because it would
 562          * cause trouble later on when we turn on paging anyway.)
 563          *
 564          * A real CPU would silently accept an invalid cr3 and would
 565          * attempt to use it - with largely undefined (and often hard
 566          * to debug) behavior on the guest side.
 567          */
 568         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 569                 inject_gp(vcpu);
 570         else {
 571                 vcpu->cr3 = cr3;
 572                 vcpu->mmu.new_cr3(vcpu);
 573         }
 574         mutex_unlock(&vcpu->kvm->lock);
 575 }
 576 EXPORT_SYMBOL_GPL(set_cr3);
 577
 578 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 579 {
 580         if (cr8 & CR8_RESERVED_BITS) {
 581                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 582                 inject_gp(vcpu);
 583                 return;
 584         }
 585         if (irqchip_in_kernel(vcpu->kvm))
 586                 kvm_lapic_set_tpr(vcpu, cr8);
 587         else
 588                 vcpu->cr8 = cr8;
 589 }
 590 EXPORT_SYMBOL_GPL(set_cr8);
 591
 592 unsigned long get_cr8(struct kvm_vcpu *vcpu)
 593 {
 594         if (irqchip_in_kernel(vcpu->kvm))
 595                 return kvm_lapic_get_cr8(vcpu);
 596         else
 597                 return vcpu->cr8;
 598 }
 599 EXPORT_SYMBOL_GPL(get_cr8);
 600
 601 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 602 {
 603         if (irqchip_in_kernel(vcpu->kvm))
 604                 return vcpu->apic_base;
 605         else
 606                 return vcpu->apic_base;
 607 }
 608 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 609
 610 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 611 {
 612         /* TODO: reserve bits check */
 613         if (irqchip_in_kernel(vcpu->kvm))
 614                 kvm_lapic_set_base(vcpu, data);
 615         else
 616                 vcpu->apic_base = data;
 617 }
 618 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 619
 620 void fx_init(struct kvm_vcpu *vcpu)
 621 {
 622         unsigned after_mxcsr_mask;
 623
 624         /* Initialize guest FPU by resetting ours and saving into guest's */
 625         preempt_disable();
 626         fx_save(&vcpu->host_fx_image);
 627         fpu_init();
 628         fx_save(&vcpu->guest_fx_image);
 629         fx_restore(&vcpu->host_fx_image);
 630         preempt_enable();
 631
 632         vcpu->cr0 |= X86_CR0_ET;
 633         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
 634         vcpu->guest_fx_image.mxcsr = 0x1f80;
 635         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
 636                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
 637 }
 638 EXPORT_SYMBOL_GPL(fx_init);
 639
 640 /*
 641  * Allocate some memory and give it an address in the guest physical address
 642  * space.
 643  *
 644  * Discontiguous memory is allowed, mostly for framebuffers.
 645  */
 646 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 647                                           struct
 648                                           kvm_userspace_memory_region *mem,
 649                                           int user_alloc)
 650 {
 651         int r;
 652         gfn_t base_gfn;
 653         unsigned long npages;
 654         unsigned long i;
 655         struct kvm_memory_slot *memslot;
 656         struct kvm_memory_slot old, new;
 657
 658         r = -EINVAL;
 659         /* General sanity checks */
 660         if (mem->memory_size & (PAGE_SIZE - 1))
 661                 goto out;
 662         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 663                 goto out;
 664         if (mem->slot >= KVM_MEMORY_SLOTS)
 665                 goto out;
 666         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 667                 goto out;
 668
 669         memslot = &kvm->memslots[mem->slot];
 670         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 671         npages = mem->memory_size >> PAGE_SHIFT;
 672
 673         if (!npages)
 674                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 675
 676         mutex_lock(&kvm->lock);
 677
 678         new = old = *memslot;
 679
 680         new.base_gfn = base_gfn;
 681         new.npages = npages;
 682         new.flags = mem->flags;
 683
 684         /* Disallow changing a memory slot's size. */
 685         r = -EINVAL;
 686         if (npages && old.npages && npages != old.npages)
 687                 goto out_unlock;
 688
 689         /* Check for overlaps */
 690         r = -EEXIST;
 691         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 692                 struct kvm_memory_slot *s = &kvm->memslots[i];
 693
 694                 if (s == memslot)
 695                         continue;
 696                 if (!((base_gfn + npages <= s->base_gfn) ||
 697                       (base_gfn >= s->base_gfn + s->npages)))
 698                         goto out_unlock;
 699         }
 700
 701         /* Free page dirty bitmap if unneeded */
 702         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 703                 new.dirty_bitmap = NULL;
 704
 705         r = -ENOMEM;
 706
 707         /* Allocate if a slot is being created */
 708         if (npages && !new.rmap) {
 709                 new.rmap = vmalloc(npages * sizeof(struct page *));
 710
 711                 if (!new.rmap)
 712                         goto out_unlock;
 713
 714                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 715
 716                 if (user_alloc)
 717                         new.userspace_addr = mem->userspace_addr;
 718                 else {
 719                         down_write(&current->mm->mmap_sem);
 720                         new.userspace_addr = do_mmap(NULL, 0,
 721                                                      npages * PAGE_SIZE,
 722                                                      PROT_READ | PROT_WRITE,
 723                                                      MAP_SHARED | MAP_ANONYMOUS,
 724                                                      0);
 725                         up_write(&current->mm->mmap_sem);
 726
 727                         if (IS_ERR((void *)new.userspace_addr))
 728                                 goto out_unlock;
 729                 }
 730         }
 731
 732         /* Allocate page dirty bitmap if needed */
 733         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 734                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 735
 736                 new.dirty_bitmap = vmalloc(dirty_bytes);
 737                 if (!new.dirty_bitmap)
 738                         goto out_unlock;
 739                 memset(new.dirty_bitmap, 0, dirty_bytes);
 740         }
 741
 742         if (mem->slot >= kvm->nmemslots)
 743                 kvm->nmemslots = mem->slot + 1;
 744
 745         if (!kvm->n_requested_mmu_pages) {
 746                 unsigned int n_pages;
 747
 748                 if (npages) {
 749                         n_pages = npages * KVM_PERMILLE_MMU_PAGES / 1000;
 750                         kvm_mmu_change_mmu_pages(kvm, kvm->n_alloc_mmu_pages +
 751                                                  n_pages);
 752                 } else {
 753                         unsigned int nr_mmu_pages;
 754
 755                         n_pages = old.npages * KVM_PERMILLE_MMU_PAGES / 1000;
 756                         nr_mmu_pages = kvm->n_alloc_mmu_pages - n_pages;
 757                         nr_mmu_pages = max(nr_mmu_pages,
 758                                         (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
 759                         kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
 760                 }
 761         }
 762
 763         *memslot = new;
 764
 765         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 766         kvm_flush_remote_tlbs(kvm);
 767
 768         mutex_unlock(&kvm->lock);
 769
 770         kvm_free_physmem_slot(&old, &new);
 771         return 0;
 772
 773 out_unlock:
 774         mutex_unlock(&kvm->lock);
 775         kvm_free_physmem_slot(&new, &old);
 776 out:
 777         return r;
 778 }
 779
 780 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
 781                                           u32 kvm_nr_mmu_pages)
 782 {
 783         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
 784                 return -EINVAL;
 785
 786         mutex_lock(&kvm->lock);
 787
 788         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
 789         kvm->n_requested_mmu_pages = kvm_nr_mmu_pages;
 790
 791         mutex_unlock(&kvm->lock);
 792         return 0;
 793 }
 794
 795 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
 796 {
 797         return kvm->n_alloc_mmu_pages;
 798 }
 799
 800 /*
 801  * Get (and clear) the dirty memory log for a memory slot.
 802  */
 803 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 804                                       struct kvm_dirty_log *log)
 805 {
 806         struct kvm_memory_slot *memslot;
 807         int r, i;
 808         int n;
 809         unsigned long any = 0;
 810
 811         mutex_lock(&kvm->lock);
 812
 813         r = -EINVAL;
 814         if (log->slot >= KVM_MEMORY_SLOTS)
 815                 goto out;
 816
 817         memslot = &kvm->memslots[log->slot];
 818         r = -ENOENT;
 819         if (!memslot->dirty_bitmap)
 820                 goto out;
 821
 822         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 823
 824         for (i = 0; !any && i < n/sizeof(long); ++i)
 825                 any = memslot->dirty_bitmap[i];
 826
 827         r = -EFAULT;
 828         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 829                 goto out;
 830
 831         /* If nothing is dirty, don't bother messing with page tables. */
 832         if (any) {
 833                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 834                 kvm_flush_remote_tlbs(kvm);
 835                 memset(memslot->dirty_bitmap, 0, n);
 836         }
 837
 838         r = 0;
 839
 840 out:
 841         mutex_unlock(&kvm->lock);
 842         return r;
 843 }
 844
 845 /*
 846  * Set a new alias region.  Aliases map a portion of physical memory into
 847  * another portion.  This is useful for memory windows, for example the PC
 848  * VGA region.
 849  */
 850 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 851                                          struct kvm_memory_alias *alias)
 852 {
 853         int r, n;
 854         struct kvm_mem_alias *p;
 855
 856         r = -EINVAL;
 857         /* General sanity checks */
 858         if (alias->memory_size & (PAGE_SIZE - 1))
 859                 goto out;
 860         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 861                 goto out;
 862         if (alias->slot >= KVM_ALIAS_SLOTS)
 863                 goto out;
 864         if (alias->guest_phys_addr + alias->memory_size
 865             < alias->guest_phys_addr)
 866                 goto out;
 867         if (alias->target_phys_addr + alias->memory_size
 868             < alias->target_phys_addr)
 869                 goto out;
 870
 871         mutex_lock(&kvm->lock);
 872
 873         p = &kvm->aliases[alias->slot];
 874         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 875         p->npages = alias->memory_size >> PAGE_SHIFT;
 876         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 877
 878         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 879                 if (kvm->aliases[n - 1].npages)
 880                         break;
 881         kvm->naliases = n;
 882
 883         kvm_mmu_zap_all(kvm);
 884
 885         mutex_unlock(&kvm->lock);
 886
 887         return 0;
 888
 889 out:
 890         return r;
 891 }
 892
 893 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 894 {
 895         int r;
 896
 897         r = 0;
 898         switch (chip->chip_id) {
 899         case KVM_IRQCHIP_PIC_MASTER:
 900                 memcpy(&chip->chip.pic,
 901                         &pic_irqchip(kvm)->pics[0],
 902                         sizeof(struct kvm_pic_state));
 903                 break;
 904         case KVM_IRQCHIP_PIC_SLAVE:
 905                 memcpy(&chip->chip.pic,
 906                         &pic_irqchip(kvm)->pics[1],
 907                         sizeof(struct kvm_pic_state));
 908                 break;
 909         case KVM_IRQCHIP_IOAPIC:
 910                 memcpy(&chip->chip.ioapic,
 911                         ioapic_irqchip(kvm),
 912                         sizeof(struct kvm_ioapic_state));
 913                 break;
 914         default:
 915                 r = -EINVAL;
 916                 break;
 917         }
 918         return r;
 919 }
 920
 921 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 922 {
 923         int r;
 924
 925         r = 0;
 926         switch (chip->chip_id) {
 927         case KVM_IRQCHIP_PIC_MASTER:
 928                 memcpy(&pic_irqchip(kvm)->pics[0],
 929                         &chip->chip.pic,
 930                         sizeof(struct kvm_pic_state));
 931                 break;
 932         case KVM_IRQCHIP_PIC_SLAVE:
 933                 memcpy(&pic_irqchip(kvm)->pics[1],
 934                         &chip->chip.pic,
 935                         sizeof(struct kvm_pic_state));
 936                 break;
 937         case KVM_IRQCHIP_IOAPIC:
 938                 memcpy(ioapic_irqchip(kvm),
 939                         &chip->chip.ioapic,
 940                         sizeof(struct kvm_ioapic_state));
 941                 break;
 942         default:
 943                 r = -EINVAL;
 944                 break;
 945         }
 946         kvm_pic_update_irq(pic_irqchip(kvm));
 947         return r;
 948 }
 949
 950 int is_error_page(struct page *page)
 951 {
 952         return page == bad_page;
 953 }
 954 EXPORT_SYMBOL_GPL(is_error_page);
 955
 956 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 957 {
 958         int i;
 959         struct kvm_mem_alias *alias;
 960
 961         for (i = 0; i < kvm->naliases; ++i) {
 962                 alias = &kvm->aliases[i];
 963                 if (gfn >= alias->base_gfn
 964                     && gfn < alias->base_gfn + alias->npages)
 965                         return alias->target_gfn + gfn - alias->base_gfn;
 966         }
 967         return gfn;
 968 }
 969
 970 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 971 {
 972         int i;
 973
 974         for (i = 0; i < kvm->nmemslots; ++i) {
 975                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 976
 977                 if (gfn >= memslot->base_gfn
 978                     && gfn < memslot->base_gfn + memslot->npages)
 979                         return memslot;
 980         }
 981         return NULL;
 982 }
 983
 984 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 985 {
 986         gfn = unalias_gfn(kvm, gfn);
 987         return __gfn_to_memslot(kvm, gfn);
 988 }
 989
 990 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 991 {
 992         struct kvm_memory_slot *slot;
 993         struct page *page[1];
 994         int npages;
 995
 996         might_sleep();
 997
 998         gfn = unalias_gfn(kvm, gfn);
 999         slot = __gfn_to_memslot(kvm, gfn);
1000         if (!slot) {
1001                 get_page(bad_page);
1002                 return bad_page;
1003         }
1004
1005         down_read(&current->mm->mmap_sem);
1006         npages = get_user_pages(current, current->mm,
1007                                 slot->userspace_addr
1008                                 + (gfn - slot->base_gfn) * PAGE_SIZE, 1,
1009                                 1, 1, page, NULL);
1010         up_read(&current->mm->mmap_sem);
1011         if (npages != 1) {
1012                 get_page(bad_page);
1013                 return bad_page;
1014         }
1015
1016         return page[0];
1017 }
1018 EXPORT_SYMBOL_GPL(gfn_to_page);
1019
1020 void kvm_release_page(struct page *page)
1021 {
1022         if (!PageReserved(page))
1023                 SetPageDirty(page);
1024         put_page(page);
1025 }
1026 EXPORT_SYMBOL_GPL(kvm_release_page);
1027
1028 static int next_segment(unsigned long len, int offset)
1029 {
1030         if (len > PAGE_SIZE - offset)
1031                 return PAGE_SIZE - offset;
1032         else
1033                 return len;
1034 }
1035
1036 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1037                         int len)
1038 {
1039         void *page_virt;
1040         struct page *page;
1041
1042         page = gfn_to_page(kvm, gfn);
1043         if (is_error_page(page)) {
1044                 kvm_release_page(page);
1045                 return -EFAULT;
1046         }
1047         page_virt = kmap_atomic(page, KM_USER0);
1048
1049         memcpy(data, page_virt + offset, len);
1050
1051         kunmap_atomic(page_virt, KM_USER0);
1052         kvm_release_page(page);
1053         return 0;
1054 }
1055 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1056
1057 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1058 {
1059         gfn_t gfn = gpa >> PAGE_SHIFT;
1060         int seg;
1061         int offset = offset_in_page(gpa);
1062         int ret;
1063
1064         while ((seg = next_segment(len, offset)) != 0) {
1065                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1066                 if (ret < 0)
1067                         return ret;
1068                 offset = 0;
1069                 len -= seg;
1070                 data += seg;
1071                 ++gfn;
1072         }
1073         return 0;
1074 }
1075 EXPORT_SYMBOL_GPL(kvm_read_guest);
1076
1077 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1078                          int offset, int len)
1079 {
1080         void *page_virt;
1081         struct page *page;
1082
1083         page = gfn_to_page(kvm, gfn);
1084         if (is_error_page(page)) {
1085                 kvm_release_page(page);
1086                 return -EFAULT;
1087         }
1088         page_virt = kmap_atomic(page, KM_USER0);
1089
1090         memcpy(page_virt + offset, data, len);
1091
1092         kunmap_atomic(page_virt, KM_USER0);
1093         mark_page_dirty(kvm, gfn);
1094         kvm_release_page(page);
1095         return 0;
1096 }
1097 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1098
1099 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1100                     unsigned long len)
1101 {
1102         gfn_t gfn = gpa >> PAGE_SHIFT;
1103         int seg;
1104         int offset = offset_in_page(gpa);
1105         int ret;
1106
1107         while ((seg = next_segment(len, offset)) != 0) {
1108                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1109                 if (ret < 0)
1110                         return ret;
1111                 offset = 0;
1112                 len -= seg;
1113                 data += seg;
1114                 ++gfn;
1115         }
1116         return 0;
1117 }
1118
1119 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1120 {
1121         void *page_virt;
1122         struct page *page;
1123
1124         page = gfn_to_page(kvm, gfn);
1125         if (is_error_page(page)) {
1126                 kvm_release_page(page);
1127                 return -EFAULT;
1128         }
1129         page_virt = kmap_atomic(page, KM_USER0);
1130
1131         memset(page_virt + offset, 0, len);
1132
1133         kunmap_atomic(page_virt, KM_USER0);
1134         kvm_release_page(page);
1135         return 0;
1136 }
1137 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1138
1139 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1140 {
1141         gfn_t gfn = gpa >> PAGE_SHIFT;
1142         int seg;
1143         int offset = offset_in_page(gpa);
1144         int ret;
1145
1146         while ((seg = next_segment(len, offset)) != 0) {
1147                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1148                 if (ret < 0)
1149                         return ret;
1150                 offset = 0;
1151                 len -= seg;
1152                 ++gfn;
1153         }
1154         return 0;
1155 }
1156 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1157
1158 /* WARNING: Does not work on aliased pages. */
1159 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1160 {
1161         struct kvm_memory_slot *memslot;
1162
1163         memslot = __gfn_to_memslot(kvm, gfn);
1164         if (memslot && memslot->dirty_bitmap) {
1165                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1166
1167                 /* avoid RMW */
1168                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1169                         set_bit(rel_gfn, memslot->dirty_bitmap);
1170         }
1171 }
1172
1173 int emulator_read_std(unsigned long addr,
1174                              void *val,
1175                              unsigned int bytes,
1176                              struct kvm_vcpu *vcpu)
1177 {
1178         void *data = val;
1179
1180         while (bytes) {
1181                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1182                 unsigned offset = addr & (PAGE_SIZE-1);
1183                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1184                 int ret;
1185
1186                 if (gpa == UNMAPPED_GVA)
1187                         return X86EMUL_PROPAGATE_FAULT;
1188                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1189                 if (ret < 0)
1190                         return X86EMUL_UNHANDLEABLE;
1191
1192                 bytes -= tocopy;
1193                 data += tocopy;
1194                 addr += tocopy;
1195         }
1196
1197         return X86EMUL_CONTINUE;
1198 }
1199 EXPORT_SYMBOL_GPL(emulator_read_std);
1200
1201 static int emulator_write_std(unsigned long addr,
1202                               const void *val,
1203                               unsigned int bytes,
1204                               struct kvm_vcpu *vcpu)
1205 {
1206         pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1207         return X86EMUL_UNHANDLEABLE;
1208 }
1209
1210 /*
1211  * Only apic need an MMIO device hook, so shortcut now..
1212  */
1213 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1214                                                 gpa_t addr)
1215 {
1216         struct kvm_io_device *dev;
1217
1218         if (vcpu->apic) {
1219                 dev = &vcpu->apic->dev;
1220                 if (dev->in_range(dev, addr))
1221                         return dev;
1222         }
1223         return NULL;
1224 }
1225
1226 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1227                                                 gpa_t addr)
1228 {
1229         struct kvm_io_device *dev;
1230
1231         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1232         if (dev == NULL)
1233                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1234         return dev;
1235 }
1236
1237 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1238                                                gpa_t addr)
1239 {
1240         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1241 }
1242
1243 static int emulator_read_emulated(unsigned long addr,
1244                                   void *val,
1245                                   unsigned int bytes,
1246                                   struct kvm_vcpu *vcpu)
1247 {
1248         struct kvm_io_device *mmio_dev;
1249         gpa_t                 gpa;
1250
1251         if (vcpu->mmio_read_completed) {
1252                 memcpy(val, vcpu->mmio_data, bytes);
1253                 vcpu->mmio_read_completed = 0;
1254                 return X86EMUL_CONTINUE;
1255         } else if (emulator_read_std(addr, val, bytes, vcpu)
1256                    == X86EMUL_CONTINUE)
1257                 return X86EMUL_CONTINUE;
1258
1259         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1260         if (gpa == UNMAPPED_GVA)
1261                 return X86EMUL_PROPAGATE_FAULT;
1262
1263         /*
1264          * Is this MMIO handled locally?
1265          */
1266         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1267         if (mmio_dev) {
1268                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1269                 return X86EMUL_CONTINUE;
1270         }
1271
1272         vcpu->mmio_needed = 1;
1273         vcpu->mmio_phys_addr = gpa;
1274         vcpu->mmio_size = bytes;
1275         vcpu->mmio_is_write = 0;
1276
1277         return X86EMUL_UNHANDLEABLE;
1278 }
1279
1280 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1281                                const void *val, int bytes)
1282 {
1283         int ret;
1284
1285         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1286         if (ret < 0)
1287                 return 0;
1288         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1289         return 1;
1290 }
1291
1292 static int emulator_write_emulated_onepage(unsigned long addr,
1293                                            const void *val,
1294                                            unsigned int bytes,
1295                                            struct kvm_vcpu *vcpu)
1296 {
1297         struct kvm_io_device *mmio_dev;
1298         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1299
1300         if (gpa == UNMAPPED_GVA) {
1301                 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1302                 return X86EMUL_PROPAGATE_FAULT;
1303         }
1304
1305         if (emulator_write_phys(vcpu, gpa, val, bytes))
1306                 return X86EMUL_CONTINUE;
1307
1308         /*
1309          * Is this MMIO handled locally?
1310          */
1311         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1312         if (mmio_dev) {
1313                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1314                 return X86EMUL_CONTINUE;
1315         }
1316
1317         vcpu->mmio_needed = 1;
1318         vcpu->mmio_phys_addr = gpa;
1319         vcpu->mmio_size = bytes;
1320         vcpu->mmio_is_write = 1;
1321         memcpy(vcpu->mmio_data, val, bytes);
1322
1323         return X86EMUL_CONTINUE;
1324 }
1325
1326 int emulator_write_emulated(unsigned long addr,
1327                                    const void *val,
1328                                    unsigned int bytes,
1329                                    struct kvm_vcpu *vcpu)
1330 {
1331         /* Crossing a page boundary? */
1332         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1333                 int rc, now;
1334
1335                 now = -addr & ~PAGE_MASK;
1336                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1337                 if (rc != X86EMUL_CONTINUE)
1338                         return rc;
1339                 addr += now;
1340                 val += now;
1341                 bytes -= now;
1342         }
1343         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1344 }
1345 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1346
1347 static int emulator_cmpxchg_emulated(unsigned long addr,
1348                                      const void *old,
1349                                      const void *new,
1350                                      unsigned int bytes,
1351                                      struct kvm_vcpu *vcpu)
1352 {
1353         static int reported;
1354
1355         if (!reported) {
1356                 reported = 1;
1357                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1358         }
1359         return emulator_write_emulated(addr, new, bytes, vcpu);
1360 }
1361
1362 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1363 {
1364         return kvm_x86_ops->get_segment_base(vcpu, seg);
1365 }
1366
1367 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1368 {
1369         return X86EMUL_CONTINUE;
1370 }
1371
1372 int emulate_clts(struct kvm_vcpu *vcpu)
1373 {
1374         kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1375         return X86EMUL_CONTINUE;
1376 }
1377
1378 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1379 {
1380         struct kvm_vcpu *vcpu = ctxt->vcpu;
1381
1382         switch (dr) {
1383         case 0 ... 3:
1384                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1385                 return X86EMUL_CONTINUE;
1386         default:
1387                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1388                 return X86EMUL_UNHANDLEABLE;
1389         }
1390 }
1391
1392 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1393 {
1394         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1395         int exception;
1396
1397         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1398         if (exception) {
1399                 /* FIXME: better handling */
1400                 return X86EMUL_UNHANDLEABLE;
1401         }
1402         return X86EMUL_CONTINUE;
1403 }
1404
1405 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1406 {
1407         static int reported;
1408         u8 opcodes[4];
1409         unsigned long rip = vcpu->rip;
1410         unsigned long rip_linear;
1411
1412         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1413
1414         if (reported)
1415                 return;
1416
1417         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1418
1419         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1420                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1421         reported = 1;
1422 }
1423 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1424
1425 struct x86_emulate_ops emulate_ops = {
1426         .read_std            = emulator_read_std,
1427         .write_std           = emulator_write_std,
1428         .read_emulated       = emulator_read_emulated,
1429         .write_emulated      = emulator_write_emulated,
1430         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1431 };
1432
1433 int emulate_instruction(struct kvm_vcpu *vcpu,
1434                         struct kvm_run *run,
1435                         unsigned long cr2,
1436                         u16 error_code,
1437                         int no_decode)
1438 {
1439         int r;
1440
1441         vcpu->mmio_fault_cr2 = cr2;
1442         kvm_x86_ops->cache_regs(vcpu);
1443
1444         vcpu->mmio_is_write = 0;
1445         vcpu->pio.string = 0;
1446
1447         if (!no_decode) {
1448                 int cs_db, cs_l;
1449                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1450
1451                 vcpu->emulate_ctxt.vcpu = vcpu;
1452                 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1453                 vcpu->emulate_ctxt.cr2 = cr2;
1454                 vcpu->emulate_ctxt.mode =
1455                         (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1456                         ? X86EMUL_MODE_REAL : cs_l
1457                         ? X86EMUL_MODE_PROT64 : cs_db
1458                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1459
1460                 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1461                         vcpu->emulate_ctxt.cs_base = 0;
1462                         vcpu->emulate_ctxt.ds_base = 0;
1463                         vcpu->emulate_ctxt.es_base = 0;
1464                         vcpu->emulate_ctxt.ss_base = 0;
1465                 } else {
1466                         vcpu->emulate_ctxt.cs_base =
1467                                         get_segment_base(vcpu, VCPU_SREG_CS);
1468                         vcpu->emulate_ctxt.ds_base =
1469                                         get_segment_base(vcpu, VCPU_SREG_DS);
1470                         vcpu->emulate_ctxt.es_base =
1471                                         get_segment_base(vcpu, VCPU_SREG_ES);
1472                         vcpu->emulate_ctxt.ss_base =
1473                                         get_segment_base(vcpu, VCPU_SREG_SS);
1474                 }
1475
1476                 vcpu->emulate_ctxt.gs_base =
1477                                         get_segment_base(vcpu, VCPU_SREG_GS);
1478                 vcpu->emulate_ctxt.fs_base =
1479                                         get_segment_base(vcpu, VCPU_SREG_FS);
1480
1481                 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
1482                 if (r)  {
1483                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1484                                 return EMULATE_DONE;
1485                         return EMULATE_FAIL;
1486                 }
1487         }
1488
1489         r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1490
1491         if (vcpu->pio.string)
1492                 return EMULATE_DO_MMIO;
1493
1494         if ((r || vcpu->mmio_is_write) && run) {
1495                 run->exit_reason = KVM_EXIT_MMIO;
1496                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1497                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1498                 run->mmio.len = vcpu->mmio_size;
1499                 run->mmio.is_write = vcpu->mmio_is_write;
1500         }
1501
1502         if (r) {
1503                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1504                         return EMULATE_DONE;
1505                 if (!vcpu->mmio_needed) {
1506                         kvm_report_emulation_failure(vcpu, "mmio");
1507                         return EMULATE_FAIL;
1508                 }
1509                 return EMULATE_DO_MMIO;
1510         }
1511
1512         kvm_x86_ops->decache_regs(vcpu);
1513         kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1514
1515         if (vcpu->mmio_is_write) {
1516                 vcpu->mmio_needed = 0;
1517                 return EMULATE_DO_MMIO;
1518         }
1519
1520         return EMULATE_DONE;
1521 }
1522 EXPORT_SYMBOL_GPL(emulate_instruction);
1523
1524 /*
1525  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1526  */
1527 static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1528 {
1529         DECLARE_WAITQUEUE(wait, current);
1530
1531         add_wait_queue(&vcpu->wq, &wait);
1532
1533         /*
1534          * We will block until either an interrupt or a signal wakes us up
1535          */
1536         while (!kvm_cpu_has_interrupt(vcpu)
1537                && !signal_pending(current)
1538                && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
1539                && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
1540                 set_current_state(TASK_INTERRUPTIBLE);
1541                 vcpu_put(vcpu);
1542                 schedule();
1543                 vcpu_load(vcpu);
1544         }
1545
1546         __set_current_state(TASK_RUNNING);
1547         remove_wait_queue(&vcpu->wq, &wait);
1548 }
1549
1550 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1551 {
1552         ++vcpu->stat.halt_exits;
1553         if (irqchip_in_kernel(vcpu->kvm)) {
1554                 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1555                 kvm_vcpu_block(vcpu);
1556                 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1557                         return -EINTR;
1558                 return 1;
1559         } else {
1560                 vcpu->run->exit_reason = KVM_EXIT_HLT;
1561                 return 0;
1562         }
1563 }
1564 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1565
1566 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
1567 {
1568         unsigned long nr, a0, a1, a2, a3, ret;
1569
1570         kvm_x86_ops->cache_regs(vcpu);
1571
1572         nr = vcpu->regs[VCPU_REGS_RAX];
1573         a0 = vcpu->regs[VCPU_REGS_RBX];
1574         a1 = vcpu->regs[VCPU_REGS_RCX];
1575         a2 = vcpu->regs[VCPU_REGS_RDX];
1576         a3 = vcpu->regs[VCPU_REGS_RSI];
1577
1578         if (!is_long_mode(vcpu)) {
1579                 nr &= 0xFFFFFFFF;
1580                 a0 &= 0xFFFFFFFF;
1581                 a1 &= 0xFFFFFFFF;
1582                 a2 &= 0xFFFFFFFF;
1583                 a3 &= 0xFFFFFFFF;
1584         }
1585
1586         switch (nr) {
1587         default:
1588                 ret = -KVM_ENOSYS;
1589                 break;
1590         }
1591         vcpu->regs[VCPU_REGS_RAX] = ret;
1592         kvm_x86_ops->decache_regs(vcpu);
1593         return 0;
1594 }
1595 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1596
1597 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1598 {
1599         char instruction[3];
1600         int ret = 0;
1601
1602         mutex_lock(&vcpu->kvm->lock);
1603
1604         /*
1605          * Blow out the MMU to ensure that no other VCPU has an active mapping
1606          * to ensure that the updated hypercall appears atomically across all
1607          * VCPUs.
1608          */
1609         kvm_mmu_zap_all(vcpu->kvm);
1610
1611         kvm_x86_ops->cache_regs(vcpu);
1612         kvm_x86_ops->patch_hypercall(vcpu, instruction);
1613         if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1614             != X86EMUL_CONTINUE)
1615                 ret = -EFAULT;
1616
1617         mutex_unlock(&vcpu->kvm->lock);
1618
1619         return ret;
1620 }
1621
1622 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1623 {
1624         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1625 }
1626
1627 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1628 {
1629         struct descriptor_table dt = { limit, base };
1630
1631         kvm_x86_ops->set_gdt(vcpu, &dt);
1632 }
1633
1634 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1635 {
1636         struct descriptor_table dt = { limit, base };
1637
1638         kvm_x86_ops->set_idt(vcpu, &dt);
1639 }
1640
1641 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1642                    unsigned long *rflags)
1643 {
1644         lmsw(vcpu, msw);
1645         *rflags = kvm_x86_ops->get_rflags(vcpu);
1646 }
1647
1648 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1649 {
1650         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1651         switch (cr) {
1652         case 0:
1653                 return vcpu->cr0;
1654         case 2:
1655                 return vcpu->cr2;
1656         case 3:
1657                 return vcpu->cr3;
1658         case 4:
1659                 return vcpu->cr4;
1660         default:
1661                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1662                 return 0;
1663         }
1664 }
1665
1666 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1667                      unsigned long *rflags)
1668 {
1669         switch (cr) {
1670         case 0:
1671                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1672                 *rflags = kvm_x86_ops->get_rflags(vcpu);
1673                 break;
1674         case 2:
1675                 vcpu->cr2 = val;
1676                 break;
1677         case 3:
1678                 set_cr3(vcpu, val);
1679                 break;
1680         case 4:
1681                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1682                 break;
1683         default:
1684                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1685         }
1686 }
1687
1688 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1689 {
1690         u64 data;
1691
1692         switch (msr) {
1693         case 0xc0010010: /* SYSCFG */
1694         case 0xc0010015: /* HWCR */
1695         case MSR_IA32_PLATFORM_ID:
1696         case MSR_IA32_P5_MC_ADDR:
1697         case MSR_IA32_P5_MC_TYPE:
1698         case MSR_IA32_MC0_CTL:
1699         case MSR_IA32_MCG_STATUS:
1700         case MSR_IA32_MCG_CAP:
1701         case MSR_IA32_MC0_MISC:
1702         case MSR_IA32_MC0_MISC+4:
1703         case MSR_IA32_MC0_MISC+8:
1704         case MSR_IA32_MC0_MISC+12:
1705         case MSR_IA32_MC0_MISC+16:
1706         case MSR_IA32_UCODE_REV:
1707         case MSR_IA32_PERF_STATUS:
1708         case MSR_IA32_EBL_CR_POWERON:
1709                 /* MTRR registers */
1710         case 0xfe:
1711         case 0x200 ... 0x2ff:
1712                 data = 0;
1713                 break;
1714         case 0xcd: /* fsb frequency */
1715                 data = 3;
1716                 break;
1717         case MSR_IA32_APICBASE:
1718                 data = kvm_get_apic_base(vcpu);
1719                 break;
1720         case MSR_IA32_MISC_ENABLE:
1721                 data = vcpu->ia32_misc_enable_msr;
1722                 break;
1723 #ifdef CONFIG_X86_64
1724         case MSR_EFER:
1725                 data = vcpu->shadow_efer;
1726                 break;
1727 #endif
1728         default:
1729                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1730                 return 1;
1731         }
1732         *pdata = data;
1733         return 0;
1734 }
1735 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1736
1737 /*
1738  * Reads an msr value (of 'msr_index') into 'pdata'.
1739  * Returns 0 on success, non-0 otherwise.
1740  * Assumes vcpu_load() was already called.
1741  */
1742 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1743 {
1744         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1745 }
1746
1747 #ifdef CONFIG_X86_64
1748
1749 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1750 {
1751         if (efer & EFER_RESERVED_BITS) {
1752                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1753                        efer);
1754                 inject_gp(vcpu);
1755                 return;
1756         }
1757
1758         if (is_paging(vcpu)
1759             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1760                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1761                 inject_gp(vcpu);
1762                 return;
1763         }
1764
1765         kvm_x86_ops->set_efer(vcpu, efer);
1766
1767         efer &= ~EFER_LMA;
1768         efer |= vcpu->shadow_efer & EFER_LMA;
1769
1770         vcpu->shadow_efer = efer;
1771 }
1772
1773 #endif
1774
1775 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1776 {
1777         switch (msr) {
1778 #ifdef CONFIG_X86_64
1779         case MSR_EFER:
1780                 set_efer(vcpu, data);
1781                 break;
1782 #endif
1783         case MSR_IA32_MC0_STATUS:
1784                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1785                        __FUNCTION__, data);
1786                 break;
1787         case MSR_IA32_MCG_STATUS:
1788                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1789                         __FUNCTION__, data);
1790                 break;
1791         case MSR_IA32_UCODE_REV:
1792         case MSR_IA32_UCODE_WRITE:
1793         case 0x200 ... 0x2ff: /* MTRRs */
1794                 break;
1795         case MSR_IA32_APICBASE:
1796                 kvm_set_apic_base(vcpu, data);
1797                 break;
1798         case MSR_IA32_MISC_ENABLE:
1799                 vcpu->ia32_misc_enable_msr = data;
1800                 break;
1801         default:
1802                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1803                 return 1;
1804         }
1805         return 0;
1806 }
1807 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1808
1809 /*
1810  * Writes msr value into into the appropriate "register".
1811  * Returns 0 on success, non-0 otherwise.
1812  * Assumes vcpu_load() was already called.
1813  */
1814 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1815 {
1816         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
1817 }
1818
1819 void kvm_resched(struct kvm_vcpu *vcpu)
1820 {
1821         if (!need_resched())
1822                 return;
1823         cond_resched();
1824 }
1825 EXPORT_SYMBOL_GPL(kvm_resched);
1826
1827 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1828 {
1829         int i;
1830         u32 function;
1831         struct kvm_cpuid_entry *e, *best;
1832
1833         kvm_x86_ops->cache_regs(vcpu);
1834         function = vcpu->regs[VCPU_REGS_RAX];
1835         vcpu->regs[VCPU_REGS_RAX] = 0;
1836         vcpu->regs[VCPU_REGS_RBX] = 0;
1837         vcpu->regs[VCPU_REGS_RCX] = 0;
1838         vcpu->regs[VCPU_REGS_RDX] = 0;
1839         best = NULL;
1840         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1841                 e = &vcpu->cpuid_entries[i];
1842                 if (e->function == function) {
1843                         best = e;
1844                         break;
1845                 }
1846                 /*
1847                  * Both basic or both extended?
1848                  */
1849                 if (((e->function ^ function) & 0x80000000) == 0)
1850                         if (!best || e->function > best->function)
1851                                 best = e;
1852         }
1853         if (best) {
1854                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1855                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1856                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1857                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1858         }
1859         kvm_x86_ops->decache_regs(vcpu);
1860         kvm_x86_ops->skip_emulated_instruction(vcpu);
1861 }
1862 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1863
1864 static int pio_copy_data(struct kvm_vcpu *vcpu)
1865 {
1866         void *p = vcpu->pio_data;
1867         void *q;
1868         unsigned bytes;
1869         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1870
1871         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1872                  PAGE_KERNEL);
1873         if (!q) {
1874                 free_pio_guest_pages(vcpu);
1875                 return -ENOMEM;
1876         }
1877         q += vcpu->pio.guest_page_offset;
1878         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1879         if (vcpu->pio.in)
1880                 memcpy(q, p, bytes);
1881         else
1882                 memcpy(p, q, bytes);
1883         q -= vcpu->pio.guest_page_offset;
1884         vunmap(q);
1885         free_pio_guest_pages(vcpu);
1886         return 0;
1887 }
1888
1889 static int complete_pio(struct kvm_vcpu *vcpu)
1890 {
1891         struct kvm_pio_request *io = &vcpu->pio;
1892         long delta;
1893         int r;
1894
1895         kvm_x86_ops->cache_regs(vcpu);
1896
1897         if (!io->string) {
1898                 if (io->in)
1899                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1900                                io->size);
1901         } else {
1902                 if (io->in) {
1903                         r = pio_copy_data(vcpu);
1904                         if (r) {
1905                                 kvm_x86_ops->cache_regs(vcpu);
1906                                 return r;
1907                         }
1908                 }
1909
1910                 delta = 1;
1911                 if (io->rep) {
1912                         delta *= io->cur_count;
1913                         /*
1914                          * The size of the register should really depend on
1915                          * current address size.
1916                          */
1917                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1918                 }
1919                 if (io->down)
1920                         delta = -delta;
1921                 delta *= io->size;
1922                 if (io->in)
1923                         vcpu->regs[VCPU_REGS_RDI] += delta;
1924                 else
1925                         vcpu->regs[VCPU_REGS_RSI] += delta;
1926         }
1927
1928         kvm_x86_ops->decache_regs(vcpu);
1929
1930         io->count -= io->cur_count;
1931         io->cur_count = 0;
1932
1933         return 0;
1934 }
1935
1936 static void kernel_pio(struct kvm_io_device *pio_dev,
1937                        struct kvm_vcpu *vcpu,
1938                        void *pd)
1939 {
1940         /* TODO: String I/O for in kernel device */
1941
1942         mutex_lock(&vcpu->kvm->lock);
1943         if (vcpu->pio.in)
1944                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1945                                   vcpu->pio.size,
1946                                   pd);
1947         else
1948                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1949                                    vcpu->pio.size,
1950                                    pd);
1951         mutex_unlock(&vcpu->kvm->lock);
1952 }
1953
1954 static void pio_string_write(struct kvm_io_device *pio_dev,
1955                              struct kvm_vcpu *vcpu)
1956 {
1957         struct kvm_pio_request *io = &vcpu->pio;
1958         void *pd = vcpu->pio_data;
1959         int i;
1960
1961         mutex_lock(&vcpu->kvm->lock);
1962         for (i = 0; i < io->cur_count; i++) {
1963                 kvm_iodevice_write(pio_dev, io->port,
1964                                    io->size,
1965                                    pd);
1966                 pd += io->size;
1967         }
1968         mutex_unlock(&vcpu->kvm->lock);
1969 }
1970
1971 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1972                   int size, unsigned port)
1973 {
1974         struct kvm_io_device *pio_dev;
1975
1976         vcpu->run->exit_reason = KVM_EXIT_IO;
1977         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1978         vcpu->run->io.size = vcpu->pio.size = size;
1979         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1980         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1981         vcpu->run->io.port = vcpu->pio.port = port;
1982         vcpu->pio.in = in;
1983         vcpu->pio.string = 0;
1984         vcpu->pio.down = 0;
1985         vcpu->pio.guest_page_offset = 0;
1986         vcpu->pio.rep = 0;
1987
1988         kvm_x86_ops->cache_regs(vcpu);
1989         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1990         kvm_x86_ops->decache_regs(vcpu);
1991
1992         kvm_x86_ops->skip_emulated_instruction(vcpu);
1993
1994         pio_dev = vcpu_find_pio_dev(vcpu, port);
1995         if (pio_dev) {
1996                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1997                 complete_pio(vcpu);
1998                 return 1;
1999         }
2000         return 0;
2001 }
2002 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2003
2004 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2005                   int size, unsigned long count, int down,
2006                   gva_t address, int rep, unsigned port)
2007 {
2008         unsigned now, in_page;
2009         int i, ret = 0;
2010         int nr_pages = 1;
2011         struct page *page;
2012         struct kvm_io_device *pio_dev;
2013
2014         vcpu->run->exit_reason = KVM_EXIT_IO;
2015         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2016         vcpu->run->io.size = vcpu->pio.size = size;
2017         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2018         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
2019         vcpu->run->io.port = vcpu->pio.port = port;
2020         vcpu->pio.in = in;
2021         vcpu->pio.string = 1;
2022         vcpu->pio.down = down;
2023         vcpu->pio.guest_page_offset = offset_in_page(address);
2024         vcpu->pio.rep = rep;
2025
2026         if (!count) {
2027                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2028                 return 1;
2029         }
2030
2031         if (!down)
2032                 in_page = PAGE_SIZE - offset_in_page(address);
2033         else
2034                 in_page = offset_in_page(address) + size;
2035         now = min(count, (unsigned long)in_page / size);
2036         if (!now) {
2037                 /*
2038                  * String I/O straddles page boundary.  Pin two guest pages
2039                  * so that we satisfy atomicity constraints.  Do just one
2040                  * transaction to avoid complexity.
2041                  */
2042                 nr_pages = 2;
2043                 now = 1;
2044         }
2045         if (down) {
2046                 /*
2047                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2048                  */
2049                 pr_unimpl(vcpu, "guest string pio down\n");
2050                 inject_gp(vcpu);
2051                 return 1;
2052         }
2053         vcpu->run->io.count = now;
2054         vcpu->pio.cur_count = now;
2055
2056         if (vcpu->pio.cur_count == vcpu->pio.count)
2057                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2058
2059         for (i = 0; i < nr_pages; ++i) {
2060                 mutex_lock(&vcpu->kvm->lock);
2061                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2062                 vcpu->pio.guest_pages[i] = page;
2063                 mutex_unlock(&vcpu->kvm->lock);
2064                 if (!page) {
2065                         inject_gp(vcpu);
2066                         free_pio_guest_pages(vcpu);
2067                         return 1;
2068                 }
2069         }
2070
2071         pio_dev = vcpu_find_pio_dev(vcpu, port);
2072         if (!vcpu->pio.in) {
2073                 /* string PIO write */
2074                 ret = pio_copy_data(vcpu);
2075                 if (ret >= 0 && pio_dev) {
2076                         pio_string_write(pio_dev, vcpu);
2077                         complete_pio(vcpu);
2078                         if (vcpu->pio.count == 0)
2079                                 ret = 1;
2080                 }
2081         } else if (pio_dev)
2082                 pr_unimpl(vcpu, "no string pio read support yet, "
2083                        "port %x size %d count %ld\n",
2084                         port, size, count);
2085
2086         return ret;
2087 }
2088 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2089
2090 /*
2091  * Check if userspace requested an interrupt window, and that the
2092  * interrupt window is open.
2093  *
2094  * No need to exit to userspace if we already have an interrupt queued.
2095  */
2096 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2097                                           struct kvm_run *kvm_run)
2098 {
2099         return (!vcpu->irq_summary &&
2100                 kvm_run->request_interrupt_window &&
2101                 vcpu->interrupt_window_open &&
2102                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2103 }
2104
2105 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2106                               struct kvm_run *kvm_run)
2107 {
2108         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2109         kvm_run->cr8 = get_cr8(vcpu);
2110         kvm_run->apic_base = kvm_get_apic_base(vcpu);
2111         if (irqchip_in_kernel(vcpu->kvm))
2112                 kvm_run->ready_for_interrupt_injection = 1;
2113         else
2114                 kvm_run->ready_for_interrupt_injection =
2115                                         (vcpu->interrupt_window_open &&
2116                                          vcpu->irq_summary == 0);
2117 }
2118
2119 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2120 {
2121         int r;
2122
2123         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
2124                 pr_debug("vcpu %d received sipi with vector # %x\n",
2125                        vcpu->vcpu_id, vcpu->sipi_vector);
2126                 kvm_lapic_reset(vcpu);
2127                 r = kvm_x86_ops->vcpu_reset(vcpu);
2128                 if (r)
2129                         return r;
2130                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2131         }
2132
2133 preempted:
2134         if (vcpu->guest_debug.enabled)
2135                 kvm_x86_ops->guest_debug_pre(vcpu);
2136
2137 again:
2138         r = kvm_mmu_reload(vcpu);
2139         if (unlikely(r))
2140                 goto out;
2141
2142         kvm_inject_pending_timer_irqs(vcpu);
2143
2144         preempt_disable();
2145
2146         kvm_x86_ops->prepare_guest_switch(vcpu);
2147         kvm_load_guest_fpu(vcpu);
2148
2149         local_irq_disable();
2150
2151         if (signal_pending(current)) {
2152                 local_irq_enable();
2153                 preempt_enable();
2154                 r = -EINTR;
2155                 kvm_run->exit_reason = KVM_EXIT_INTR;
2156                 ++vcpu->stat.signal_exits;
2157                 goto out;
2158         }
2159
2160         if (irqchip_in_kernel(vcpu->kvm))
2161                 kvm_x86_ops->inject_pending_irq(vcpu);
2162         else if (!vcpu->mmio_read_completed)
2163                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2164
2165         vcpu->guest_mode = 1;
2166         kvm_guest_enter();
2167
2168         if (vcpu->requests)
2169                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2170                         kvm_x86_ops->tlb_flush(vcpu);
2171
2172         kvm_x86_ops->run(vcpu, kvm_run);
2173
2174         vcpu->guest_mode = 0;
2175         local_irq_enable();
2176
2177         ++vcpu->stat.exits;
2178
2179         /*
2180          * We must have an instruction between local_irq_enable() and
2181          * kvm_guest_exit(), so the timer interrupt isn't delayed by
2182          * the interrupt shadow.  The stat.exits increment will do nicely.
2183          * But we need to prevent reordering, hence this barrier():
2184          */
2185         barrier();
2186
2187         kvm_guest_exit();
2188
2189         preempt_enable();
2190
2191         /*
2192          * Profile KVM exit RIPs:
2193          */
2194         if (unlikely(prof_on == KVM_PROFILING)) {
2195                 kvm_x86_ops->cache_regs(vcpu);
2196                 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
2197         }
2198
2199         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2200
2201         if (r > 0) {
2202                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2203                         r = -EINTR;
2204                         kvm_run->exit_reason = KVM_EXIT_INTR;
2205                         ++vcpu->stat.request_irq_exits;
2206                         goto out;
2207                 }
2208                 if (!need_resched()) {
2209                         ++vcpu->stat.light_exits;
2210                         goto again;
2211                 }
2212         }
2213
2214 out:
2215         if (r > 0) {
2216                 kvm_resched(vcpu);
2217                 goto preempted;
2218         }
2219
2220         post_kvm_run_save(vcpu, kvm_run);
2221
2222         return r;
2223 }
2224
2225
2226 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2227 {
2228         int r;
2229         sigset_t sigsaved;
2230
2231         vcpu_load(vcpu);
2232
2233         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2234                 kvm_vcpu_block(vcpu);
2235                 vcpu_put(vcpu);
2236                 return -EAGAIN;
2237         }
2238
2239         if (vcpu->sigset_active)
2240                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2241
2242         /* re-sync apic's tpr */
2243         if (!irqchip_in_kernel(vcpu->kvm))
2244                 set_cr8(vcpu, kvm_run->cr8);
2245
2246         if (vcpu->pio.cur_count) {
2247                 r = complete_pio(vcpu);
2248                 if (r)
2249                         goto out;
2250         }
2251 #if CONFIG_HAS_IOMEM
2252         if (vcpu->mmio_needed) {
2253                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2254                 vcpu->mmio_read_completed = 1;
2255                 vcpu->mmio_needed = 0;
2256                 r = emulate_instruction(vcpu, kvm_run,
2257                                         vcpu->mmio_fault_cr2, 0, 1);
2258                 if (r == EMULATE_DO_MMIO) {
2259                         /*
2260                          * Read-modify-write.  Back to userspace.
2261                          */
2262                         r = 0;
2263                         goto out;
2264                 }
2265         }
2266 #endif
2267         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2268                 kvm_x86_ops->cache_regs(vcpu);
2269                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2270                 kvm_x86_ops->decache_regs(vcpu);
2271         }
2272
2273         r = __vcpu_run(vcpu, kvm_run);
2274
2275 out:
2276         if (vcpu->sigset_active)
2277                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2278
2279         vcpu_put(vcpu);
2280         return r;
2281 }
2282
2283 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2284                                    struct kvm_regs *regs)
2285 {
2286         vcpu_load(vcpu);
2287
2288         kvm_x86_ops->cache_regs(vcpu);
2289
2290         regs->rax = vcpu->regs[VCPU_REGS_RAX];
2291         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2292         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2293         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2294         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2295         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2296         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2297         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2298 #ifdef CONFIG_X86_64
2299         regs->r8 = vcpu->regs[VCPU_REGS_R8];
2300         regs->r9 = vcpu->regs[VCPU_REGS_R9];
2301         regs->r10 = vcpu->regs[VCPU_REGS_R10];
2302         regs->r11 = vcpu->regs[VCPU_REGS_R11];
2303         regs->r12 = vcpu->regs[VCPU_REGS_R12];
2304         regs->r13 = vcpu->regs[VCPU_REGS_R13];
2305         regs->r14 = vcpu->regs[VCPU_REGS_R14];
2306         regs->r15 = vcpu->regs[VCPU_REGS_R15];
2307 #endif
2308
2309         regs->rip = vcpu->rip;
2310         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2311
2312         /*
2313          * Don't leak debug flags in case they were set for guest debugging
2314          */
2315         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2316                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2317
2318         vcpu_put(vcpu);
2319
2320         return 0;
2321 }
2322
2323 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2324                                    struct kvm_regs *regs)
2325 {
2326         vcpu_load(vcpu);
2327
2328         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2329         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2330         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2331         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2332         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2333         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2334         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2335         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2336 #ifdef CONFIG_X86_64
2337         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2338         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2339         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2340         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2341         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2342         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2343         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2344         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2345 #endif
2346
2347         vcpu->rip = regs->rip;
2348         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2349
2350         kvm_x86_ops->decache_regs(vcpu);
2351
2352         vcpu_put(vcpu);
2353
2354         return 0;
2355 }
2356
2357 static void get_segment(struct kvm_vcpu *vcpu,
2358                         struct kvm_segment *var, int seg)
2359 {
2360         return kvm_x86_ops->get_segment(vcpu, var, seg);
2361 }
2362
2363 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2364                                     struct kvm_sregs *sregs)
2365 {
2366         struct descriptor_table dt;
2367         int pending_vec;
2368
2369         vcpu_load(vcpu);
2370
2371         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2372         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2373         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2374         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2375         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2376         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2377
2378         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2379         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2380
2381         kvm_x86_ops->get_idt(vcpu, &dt);
2382         sregs->idt.limit = dt.limit;
2383         sregs->idt.base = dt.base;
2384         kvm_x86_ops->get_gdt(vcpu, &dt);
2385         sregs->gdt.limit = dt.limit;
2386         sregs->gdt.base = dt.base;
2387
2388         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2389         sregs->cr0 = vcpu->cr0;
2390         sregs->cr2 = vcpu->cr2;
2391         sregs->cr3 = vcpu->cr3;
2392         sregs->cr4 = vcpu->cr4;
2393         sregs->cr8 = get_cr8(vcpu);
2394         sregs->efer = vcpu->shadow_efer;
2395         sregs->apic_base = kvm_get_apic_base(vcpu);
2396
2397         if (irqchip_in_kernel(vcpu->kvm)) {
2398                 memset(sregs->interrupt_bitmap, 0,
2399                        sizeof sregs->interrupt_bitmap);
2400                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2401                 if (pending_vec >= 0)
2402                         set_bit(pending_vec,
2403                                 (unsigned long *)sregs->interrupt_bitmap);
2404         } else
2405                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2406                        sizeof sregs->interrupt_bitmap);
2407
2408         vcpu_put(vcpu);
2409
2410         return 0;
2411 }
2412
2413 static void set_segment(struct kvm_vcpu *vcpu,
2414                         struct kvm_segment *var, int seg)
2415 {
2416         return kvm_x86_ops->set_segment(vcpu, var, seg);
2417 }
2418
2419 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2420                                     struct kvm_sregs *sregs)
2421 {
2422         int mmu_reset_needed = 0;
2423         int i, pending_vec, max_bits;
2424         struct descriptor_table dt;
2425
2426         vcpu_load(vcpu);
2427
2428         dt.limit = sregs->idt.limit;
2429         dt.base = sregs->idt.base;
2430         kvm_x86_ops->set_idt(vcpu, &dt);
2431         dt.limit = sregs->gdt.limit;
2432         dt.base = sregs->gdt.base;
2433         kvm_x86_ops->set_gdt(vcpu, &dt);
2434
2435         vcpu->cr2 = sregs->cr2;
2436         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2437         vcpu->cr3 = sregs->cr3;
2438
2439         set_cr8(vcpu, sregs->cr8);
2440
2441         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2442 #ifdef CONFIG_X86_64
2443         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2444 #endif
2445         kvm_set_apic_base(vcpu, sregs->apic_base);
2446
2447         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2448
2449         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2450         vcpu->cr0 = sregs->cr0;
2451         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2452
2453         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2454         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2455         if (!is_long_mode(vcpu) && is_pae(vcpu))
2456                 load_pdptrs(vcpu, vcpu->cr3);
2457
2458         if (mmu_reset_needed)
2459                 kvm_mmu_reset_context(vcpu);
2460
2461         if (!irqchip_in_kernel(vcpu->kvm)) {
2462                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2463                        sizeof vcpu->irq_pending);
2464                 vcpu->irq_summary = 0;
2465                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2466                         if (vcpu->irq_pending[i])
2467                                 __set_bit(i, &vcpu->irq_summary);
2468         } else {
2469                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2470                 pending_vec = find_first_bit(
2471                         (const unsigned long *)sregs->interrupt_bitmap,
2472                         max_bits);
2473                 /* Only pending external irq is handled here */
2474                 if (pending_vec < max_bits) {
2475                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2476                         pr_debug("Set back pending irq %d\n",
2477                                  pending_vec);
2478                 }
2479         }
2480
2481         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2482         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2483         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2484         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2485         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2486         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2487
2488         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2489         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2490
2491         vcpu_put(vcpu);
2492
2493         return 0;
2494 }
2495
2496 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2497 {
2498         struct kvm_segment cs;
2499
2500         get_segment(vcpu, &cs, VCPU_SREG_CS);
2501         *db = cs.db;
2502         *l = cs.l;
2503 }
2504 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2505
2506 /*
2507  * Translate a guest virtual address to a guest physical address.
2508  */
2509 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2510                                     struct kvm_translation *tr)
2511 {
2512         unsigned long vaddr = tr->linear_address;
2513         gpa_t gpa;
2514
2515         vcpu_load(vcpu);
2516         mutex_lock(&vcpu->kvm->lock);
2517         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2518         tr->physical_address = gpa;
2519         tr->valid = gpa != UNMAPPED_GVA;
2520         tr->writeable = 1;
2521         tr->usermode = 0;
2522         mutex_unlock(&vcpu->kvm->lock);
2523         vcpu_put(vcpu);
2524
2525         return 0;
2526 }
2527
2528 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2529                                     struct kvm_interrupt *irq)
2530 {
2531         if (irq->irq < 0 || irq->irq >= 256)
2532                 return -EINVAL;
2533         if (irqchip_in_kernel(vcpu->kvm))
2534                 return -ENXIO;
2535         vcpu_load(vcpu);
2536
2537         set_bit(irq->irq, vcpu->irq_pending);
2538         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2539
2540         vcpu_put(vcpu);
2541
2542         return 0;
2543 }
2544
2545 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2546                                       struct kvm_debug_guest *dbg)
2547 {
2548         int r;
2549
2550         vcpu_load(vcpu);
2551
2552         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2553
2554         vcpu_put(vcpu);
2555
2556         return r;
2557 }
2558
2559 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2560                                     unsigned long address,
2561                                     int *type)
2562 {
2563         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2564         unsigned long pgoff;
2565         struct page *page;
2566
2567         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2568         if (pgoff == 0)
2569                 page = virt_to_page(vcpu->run);
2570         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2571                 page = virt_to_page(vcpu->pio_data);
2572         else
2573                 return NOPAGE_SIGBUS;
2574         get_page(page);
2575         if (type != NULL)
2576                 *type = VM_FAULT_MINOR;
2577
2578         return page;
2579 }
2580
2581 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2582         .nopage = kvm_vcpu_nopage,
2583 };
2584
2585 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2586 {
2587         vma->vm_ops = &kvm_vcpu_vm_ops;
2588         return 0;
2589 }
2590
2591 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2592 {
2593         struct kvm_vcpu *vcpu = filp->private_data;
2594
2595         fput(vcpu->kvm->filp);
2596         return 0;
2597 }
2598
2599 static struct file_operations kvm_vcpu_fops = {
2600         .release        = kvm_vcpu_release,
2601         .unlocked_ioctl = kvm_vcpu_ioctl,
2602         .compat_ioctl   = kvm_vcpu_ioctl,
2603         .mmap           = kvm_vcpu_mmap,
2604 };
2605
2606 /*
2607  * Allocates an inode for the vcpu.
2608  */
2609 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2610 {
2611         int fd, r;
2612         struct inode *inode;
2613         struct file *file;
2614
2615         r = anon_inode_getfd(&fd, &inode, &file,
2616                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2617         if (r)
2618                 return r;
2619         atomic_inc(&vcpu->kvm->filp->f_count);
2620         return fd;
2621 }
2622
2623 /*
2624  * Creates some virtual cpus.  Good luck creating more than one.
2625  */
2626 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2627 {
2628         int r;
2629         struct kvm_vcpu *vcpu;
2630
2631         if (!valid_vcpu(n))
2632                 return -EINVAL;
2633
2634         vcpu = kvm_x86_ops->vcpu_create(kvm, n);
2635         if (IS_ERR(vcpu))
2636                 return PTR_ERR(vcpu);
2637
2638         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2639
2640         /* We do fxsave: this must be aligned. */
2641         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2642
2643         vcpu_load(vcpu);
2644         r = kvm_x86_ops->vcpu_reset(vcpu);
2645         if (r == 0)
2646                 r = kvm_mmu_setup(vcpu);
2647         vcpu_put(vcpu);
2648         if (r < 0)
2649                 goto free_vcpu;
2650
2651         mutex_lock(&kvm->lock);
2652         if (kvm->vcpus[n]) {
2653                 r = -EEXIST;
2654                 mutex_unlock(&kvm->lock);
2655                 goto mmu_unload;
2656         }
2657         kvm->vcpus[n] = vcpu;
2658         mutex_unlock(&kvm->lock);
2659
2660         /* Now it's all set up, let userspace reach it */
2661         r = create_vcpu_fd(vcpu);
2662         if (r < 0)
2663                 goto unlink;
2664         return r;
2665
2666 unlink:
2667         mutex_lock(&kvm->lock);
2668         kvm->vcpus[n] = NULL;
2669         mutex_unlock(&kvm->lock);
2670
2671 mmu_unload:
2672         vcpu_load(vcpu);
2673         kvm_mmu_unload(vcpu);
2674         vcpu_put(vcpu);
2675
2676 free_vcpu:
2677         kvm_x86_ops->vcpu_free(vcpu);
2678         return r;
2679 }
2680
2681 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2682 {
2683         if (sigset) {
2684                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2685                 vcpu->sigset_active = 1;
2686                 vcpu->sigset = *sigset;
2687         } else
2688                 vcpu->sigset_active = 0;
2689         return 0;
2690 }
2691
2692 /*
2693  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2694  * we have asm/x86/processor.h
2695  */
2696 struct fxsave {
2697         u16     cwd;
2698         u16     swd;
2699         u16     twd;
2700         u16     fop;
2701         u64     rip;
2702         u64     rdp;
2703         u32     mxcsr;
2704         u32     mxcsr_mask;
2705         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2706 #ifdef CONFIG_X86_64
2707         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2708 #else
2709         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2710 #endif
2711 };
2712
2713 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2714 {
2715         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2716
2717         vcpu_load(vcpu);
2718
2719         memcpy(fpu->fpr, fxsave->st_space, 128);
2720         fpu->fcw = fxsave->cwd;
2721         fpu->fsw = fxsave->swd;
2722         fpu->ftwx = fxsave->twd;
2723         fpu->last_opcode = fxsave->fop;
2724         fpu->last_ip = fxsave->rip;
2725         fpu->last_dp = fxsave->rdp;
2726         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2727
2728         vcpu_put(vcpu);
2729
2730         return 0;
2731 }
2732
2733 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2734 {
2735         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2736
2737         vcpu_load(vcpu);
2738
2739         memcpy(fxsave->st_space, fpu->fpr, 128);
2740         fxsave->cwd = fpu->fcw;
2741         fxsave->swd = fpu->fsw;
2742         fxsave->twd = fpu->ftwx;
2743         fxsave->fop = fpu->last_opcode;
2744         fxsave->rip = fpu->last_ip;
2745         fxsave->rdp = fpu->last_dp;
2746         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2747
2748         vcpu_put(vcpu);
2749
2750         return 0;
2751 }
2752
2753 static long kvm_vcpu_ioctl(struct file *filp,
2754                            unsigned int ioctl, unsigned long arg)
2755 {
2756         struct kvm_vcpu *vcpu = filp->private_data;
2757         void __user *argp = (void __user *)arg;
2758         int r;
2759
2760         switch (ioctl) {
2761         case KVM_RUN:
2762                 r = -EINVAL;
2763                 if (arg)
2764                         goto out;
2765                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2766                 break;
2767         case KVM_GET_REGS: {
2768                 struct kvm_regs kvm_regs;
2769
2770                 memset(&kvm_regs, 0, sizeof kvm_regs);
2771                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2772                 if (r)
2773                         goto out;
2774                 r = -EFAULT;
2775                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2776                         goto out;
2777                 r = 0;
2778                 break;
2779         }
2780         case KVM_SET_REGS: {
2781                 struct kvm_regs kvm_regs;
2782
2783                 r = -EFAULT;
2784                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2785                         goto out;
2786                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2787                 if (r)
2788                         goto out;
2789                 r = 0;
2790                 break;
2791         }
2792         case KVM_GET_SREGS: {
2793                 struct kvm_sregs kvm_sregs;
2794
2795                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2796                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2797                 if (r)
2798                         goto out;
2799                 r = -EFAULT;
2800                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2801                         goto out;
2802                 r = 0;
2803                 break;
2804         }
2805         case KVM_SET_SREGS: {
2806                 struct kvm_sregs kvm_sregs;
2807
2808                 r = -EFAULT;
2809                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2810                         goto out;
2811                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2812                 if (r)
2813                         goto out;
2814                 r = 0;
2815                 break;
2816         }
2817         case KVM_TRANSLATE: {
2818                 struct kvm_translation tr;
2819
2820                 r = -EFAULT;
2821                 if (copy_from_user(&tr, argp, sizeof tr))
2822                         goto out;
2823                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2824                 if (r)
2825                         goto out;
2826                 r = -EFAULT;
2827                 if (copy_to_user(argp, &tr, sizeof tr))
2828                         goto out;
2829                 r = 0;
2830                 break;
2831         }
2832         case KVM_INTERRUPT: {
2833                 struct kvm_interrupt irq;
2834
2835                 r = -EFAULT;
2836                 if (copy_from_user(&irq, argp, sizeof irq))
2837                         goto out;
2838                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2839                 if (r)
2840                         goto out;
2841                 r = 0;
2842                 break;
2843         }
2844         case KVM_DEBUG_GUEST: {
2845                 struct kvm_debug_guest dbg;
2846
2847                 r = -EFAULT;
2848                 if (copy_from_user(&dbg, argp, sizeof dbg))
2849                         goto out;
2850                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2851                 if (r)
2852                         goto out;
2853                 r = 0;
2854                 break;
2855         }
2856         case KVM_SET_SIGNAL_MASK: {
2857                 struct kvm_signal_mask __user *sigmask_arg = argp;
2858                 struct kvm_signal_mask kvm_sigmask;
2859                 sigset_t sigset, *p;
2860
2861                 p = NULL;
2862                 if (argp) {
2863                         r = -EFAULT;
2864                         if (copy_from_user(&kvm_sigmask, argp,
2865                                            sizeof kvm_sigmask))
2866                                 goto out;
2867                         r = -EINVAL;
2868                         if (kvm_sigmask.len != sizeof sigset)
2869                                 goto out;
2870                         r = -EFAULT;
2871                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2872                                            sizeof sigset))
2873                                 goto out;
2874                         p = &sigset;
2875                 }
2876                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2877                 break;
2878         }
2879         case KVM_GET_FPU: {
2880                 struct kvm_fpu fpu;
2881
2882                 memset(&fpu, 0, sizeof fpu);
2883                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2884                 if (r)
2885                         goto out;
2886                 r = -EFAULT;
2887                 if (copy_to_user(argp, &fpu, sizeof fpu))
2888                         goto out;
2889                 r = 0;
2890                 break;
2891         }
2892         case KVM_SET_FPU: {
2893                 struct kvm_fpu fpu;
2894
2895                 r = -EFAULT;
2896                 if (copy_from_user(&fpu, argp, sizeof fpu))
2897                         goto out;
2898                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2899                 if (r)
2900                         goto out;
2901                 r = 0;
2902                 break;
2903         }
2904         default:
2905                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2906         }
2907 out:
2908         return r;
2909 }
2910
2911 static long kvm_vm_ioctl(struct file *filp,
2912                            unsigned int ioctl, unsigned long arg)
2913 {
2914         struct kvm *kvm = filp->private_data;
2915         void __user *argp = (void __user *)arg;
2916         int r = -EINVAL;
2917
2918         switch (ioctl) {
2919         case KVM_CREATE_VCPU:
2920                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2921                 if (r < 0)
2922                         goto out;
2923                 break;
2924         case KVM_SET_MEMORY_REGION: {
2925                 struct kvm_memory_region kvm_mem;
2926                 struct kvm_userspace_memory_region kvm_userspace_mem;
2927
2928                 r = -EFAULT;
2929                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2930                         goto out;
2931                 kvm_userspace_mem.slot = kvm_mem.slot;
2932                 kvm_userspace_mem.flags = kvm_mem.flags;
2933                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2934                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2935                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2936                 if (r)
2937                         goto out;
2938                 break;
2939         }
2940         case KVM_SET_USER_MEMORY_REGION: {
2941                 struct kvm_userspace_memory_region kvm_userspace_mem;
2942
2943                 r = -EFAULT;
2944                 if (copy_from_user(&kvm_userspace_mem, argp,
2945                                                 sizeof kvm_userspace_mem))
2946                         goto out;
2947
2948                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2949                 if (r)
2950                         goto out;
2951                 break;
2952         }
2953         case KVM_SET_NR_MMU_PAGES:
2954                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2955                 if (r)
2956                         goto out;
2957                 break;
2958         case KVM_GET_NR_MMU_PAGES:
2959                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2960                 break;
2961         case KVM_GET_DIRTY_LOG: {
2962                 struct kvm_dirty_log log;
2963
2964                 r = -EFAULT;
2965                 if (copy_from_user(&log, argp, sizeof log))
2966                         goto out;
2967                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2968                 if (r)
2969                         goto out;
2970                 break;
2971         }
2972         case KVM_SET_MEMORY_ALIAS: {
2973                 struct kvm_memory_alias alias;
2974
2975                 r = -EFAULT;
2976                 if (copy_from_user(&alias, argp, sizeof alias))
2977                         goto out;
2978                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2979                 if (r)
2980                         goto out;
2981                 break;
2982         }
2983         case KVM_CREATE_IRQCHIP:
2984                 r = -ENOMEM;
2985                 kvm->vpic = kvm_create_pic(kvm);
2986                 if (kvm->vpic) {
2987                         r = kvm_ioapic_init(kvm);
2988                         if (r) {
2989                                 kfree(kvm->vpic);
2990                                 kvm->vpic = NULL;
2991                                 goto out;
2992                         }
2993                 } else
2994                         goto out;
2995                 break;
2996         case KVM_IRQ_LINE: {
2997                 struct kvm_irq_level irq_event;
2998
2999                 r = -EFAULT;
3000                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3001                         goto out;
3002                 if (irqchip_in_kernel(kvm)) {
3003                         mutex_lock(&kvm->lock);
3004                         if (irq_event.irq < 16)
3005                                 kvm_pic_set_irq(pic_irqchip(kvm),
3006                                         irq_event.irq,
3007                                         irq_event.level);
3008                         kvm_ioapic_set_irq(kvm->vioapic,
3009                                         irq_event.irq,
3010                                         irq_event.level);
3011                         mutex_unlock(&kvm->lock);
3012                         r = 0;
3013                 }
3014                 break;
3015         }
3016         case KVM_GET_IRQCHIP: {
3017                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3018                 struct kvm_irqchip chip;
3019
3020                 r = -EFAULT;
3021                 if (copy_from_user(&chip, argp, sizeof chip))
3022                         goto out;
3023                 r = -ENXIO;
3024                 if (!irqchip_in_kernel(kvm))
3025                         goto out;
3026                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
3027                 if (r)
3028                         goto out;
3029                 r = -EFAULT;
3030                 if (copy_to_user(argp, &chip, sizeof chip))
3031                         goto out;
3032                 r = 0;
3033                 break;
3034         }
3035         case KVM_SET_IRQCHIP: {
3036                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3037                 struct kvm_irqchip chip;
3038
3039                 r = -EFAULT;
3040                 if (copy_from_user(&chip, argp, sizeof chip))
3041                         goto out;
3042                 r = -ENXIO;
3043                 if (!irqchip_in_kernel(kvm))
3044                         goto out;
3045                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
3046                 if (r)
3047                         goto out;
3048                 r = 0;
3049                 break;
3050         }
3051         default:
3052                 ;
3053         }
3054 out:
3055         return r;
3056 }
3057
3058 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
3059                                   unsigned long address,
3060                                   int *type)
3061 {
3062         struct kvm *kvm = vma->vm_file->private_data;
3063         unsigned long pgoff;
3064         struct page *page;
3065
3066         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
3067         page = gfn_to_page(kvm, pgoff);
3068         if (is_error_page(page)) {
3069                 kvm_release_page(page);
3070                 return NOPAGE_SIGBUS;
3071         }
3072         if (type != NULL)
3073                 *type = VM_FAULT_MINOR;
3074
3075         return page;
3076 }
3077
3078 static struct vm_operations_struct kvm_vm_vm_ops = {
3079         .nopage = kvm_vm_nopage,
3080 };
3081
3082 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
3083 {
3084         vma->vm_ops = &kvm_vm_vm_ops;
3085         return 0;
3086 }
3087
3088 static struct file_operations kvm_vm_fops = {
3089         .release        = kvm_vm_release,
3090         .unlocked_ioctl = kvm_vm_ioctl,
3091         .compat_ioctl   = kvm_vm_ioctl,
3092         .mmap           = kvm_vm_mmap,
3093 };
3094
3095 static int kvm_dev_ioctl_create_vm(void)
3096 {
3097         int fd, r;
3098         struct inode *inode;
3099         struct file *file;
3100         struct kvm *kvm;
3101
3102         kvm = kvm_create_vm();
3103         if (IS_ERR(kvm))
3104                 return PTR_ERR(kvm);
3105         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
3106         if (r) {
3107                 kvm_destroy_vm(kvm);
3108                 return r;
3109         }
3110
3111         kvm->filp = file;
3112
3113         return fd;
3114 }
3115
3116 static long kvm_dev_ioctl(struct file *filp,
3117                           unsigned int ioctl, unsigned long arg)
3118 {
3119         void __user *argp = (void __user *)arg;
3120         long r = -EINVAL;
3121
3122         switch (ioctl) {
3123         case KVM_GET_API_VERSION:
3124                 r = -EINVAL;
3125                 if (arg)
3126                         goto out;
3127                 r = KVM_API_VERSION;
3128                 break;
3129         case KVM_CREATE_VM:
3130                 r = -EINVAL;
3131                 if (arg)
3132                         goto out;
3133                 r = kvm_dev_ioctl_create_vm();
3134                 break;
3135         case KVM_CHECK_EXTENSION: {
3136                 int ext = (long)argp;
3137
3138                 switch (ext) {
3139                 case KVM_CAP_IRQCHIP:
3140                 case KVM_CAP_HLT:
3141                 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3142                 case KVM_CAP_USER_MEMORY:
3143                         r = 1;
3144                         break;
3145                 default:
3146                         r = 0;
3147                         break;
3148                 }
3149                 break;
3150         }
3151         case KVM_GET_VCPU_MMAP_SIZE:
3152                 r = -EINVAL;
3153                 if (arg)
3154                         goto out;
3155                 r = 2 * PAGE_SIZE;
3156                 break;
3157         default:
3158                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3159         }
3160 out:
3161         return r;
3162 }
3163
3164 static struct file_operations kvm_chardev_ops = {
3165         .unlocked_ioctl = kvm_dev_ioctl,
3166         .compat_ioctl   = kvm_dev_ioctl,
3167 };
3168
3169 static struct miscdevice kvm_dev = {
3170         KVM_MINOR,
3171         "kvm",
3172         &kvm_chardev_ops,
3173 };
3174
3175 /*
3176  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
3177  * cached on it.
3178  */
3179 static void decache_vcpus_on_cpu(int cpu)
3180 {
3181         struct kvm *vm;
3182         struct kvm_vcpu *vcpu;
3183         int i;
3184
3185         spin_lock(&kvm_lock);
3186         list_for_each_entry(vm, &vm_list, vm_list)
3187                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3188                         vcpu = vm->vcpus[i];
3189                         if (!vcpu)
3190                                 continue;
3191                         /*
3192                          * If the vcpu is locked, then it is running on some
3193                          * other cpu and therefore it is not cached on the
3194                          * cpu in question.
3195                          *
3196                          * If it's not locked, check the last cpu it executed
3197                          * on.
3198                          */
3199                         if (mutex_trylock(&vcpu->mutex)) {
3200                                 if (vcpu->cpu == cpu) {
3201                                         kvm_x86_ops->vcpu_decache(vcpu);
3202                                         vcpu->cpu = -1;
3203                                 }
3204                                 mutex_unlock(&vcpu->mutex);
3205                         }
3206                 }
3207         spin_unlock(&kvm_lock);
3208 }
3209
3210 static void hardware_enable(void *junk)
3211 {
3212         int cpu = raw_smp_processor_id();
3213
3214         if (cpu_isset(cpu, cpus_hardware_enabled))
3215                 return;
3216         cpu_set(cpu, cpus_hardware_enabled);
3217         kvm_x86_ops->hardware_enable(NULL);
3218 }
3219
3220 static void hardware_disable(void *junk)
3221 {
3222         int cpu = raw_smp_processor_id();
3223
3224         if (!cpu_isset(cpu, cpus_hardware_enabled))
3225                 return;
3226         cpu_clear(cpu, cpus_hardware_enabled);
3227         decache_vcpus_on_cpu(cpu);
3228         kvm_x86_ops->hardware_disable(NULL);
3229 }
3230
3231 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3232                            void *v)
3233 {
3234         int cpu = (long)v;
3235
3236         switch (val) {
3237         case CPU_DYING:
3238         case CPU_DYING_FROZEN:
3239                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3240                        cpu);
3241                 hardware_disable(NULL);
3242                 break;
3243         case CPU_UP_CANCELED:
3244         case CPU_UP_CANCELED_FROZEN:
3245                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3246                        cpu);
3247                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3248                 break;
3249         case CPU_ONLINE:
3250         case CPU_ONLINE_FROZEN:
3251                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3252                        cpu);
3253                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3254                 break;
3255         }
3256         return NOTIFY_OK;
3257 }
3258
3259 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3260                       void *v)
3261 {
3262         if (val == SYS_RESTART) {
3263                 /*
3264                  * Some (well, at least mine) BIOSes hang on reboot if
3265                  * in vmx root mode.
3266                  */
3267                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3268                 on_each_cpu(hardware_disable, NULL, 0, 1);
3269         }
3270         return NOTIFY_OK;
3271 }
3272
3273 static struct notifier_block kvm_reboot_notifier = {
3274         .notifier_call = kvm_reboot,
3275         .priority = 0,
3276 };
3277
3278 void kvm_io_bus_init(struct kvm_io_bus *bus)
3279 {
3280         memset(bus, 0, sizeof(*bus));
3281 }
3282
3283 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3284 {
3285         int i;
3286
3287         for (i = 0; i < bus->dev_count; i++) {
3288                 struct kvm_io_device *pos = bus->devs[i];
3289
3290                 kvm_iodevice_destructor(pos);
3291         }
3292 }
3293
3294 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3295 {
3296         int i;
3297
3298         for (i = 0; i < bus->dev_count; i++) {
3299                 struct kvm_io_device *pos = bus->devs[i];
3300
3301                 if (pos->in_range(pos, addr))
3302                         return pos;
3303         }
3304
3305         return NULL;
3306 }
3307
3308 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3309 {
3310         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3311
3312         bus->devs[bus->dev_count++] = dev;
3313 }
3314
3315 static struct notifier_block kvm_cpu_notifier = {
3316         .notifier_call = kvm_cpu_hotplug,
3317         .priority = 20, /* must be > scheduler priority */
3318 };
3319
3320 static u64 stat_get(void *_offset)
3321 {
3322         unsigned offset = (long)_offset;
3323         u64 total = 0;
3324         struct kvm *kvm;
3325         struct kvm_vcpu *vcpu;
3326         int i;
3327
3328         spin_lock(&kvm_lock);
3329         list_for_each_entry(kvm, &vm_list, vm_list)
3330                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3331                         vcpu = kvm->vcpus[i];
3332                         if (vcpu)
3333                                 total += *(u32 *)((void *)vcpu + offset);
3334                 }
3335         spin_unlock(&kvm_lock);
3336         return total;
3337 }
3338
3339 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3340
3341 static __init void kvm_init_debug(void)
3342 {
3343         struct kvm_stats_debugfs_item *p;
3344
3345         debugfs_dir = debugfs_create_dir("kvm", NULL);
3346         for (p = debugfs_entries; p->name; ++p)
3347                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3348                                                 (void *)(long)p->offset,
3349                                                 &stat_fops);
3350 }
3351
3352 static void kvm_exit_debug(void)
3353 {
3354         struct kvm_stats_debugfs_item *p;
3355
3356         for (p = debugfs_entries; p->name; ++p)
3357                 debugfs_remove(p->dentry);
3358         debugfs_remove(debugfs_dir);
3359 }
3360
3361 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3362 {
3363         hardware_disable(NULL);
3364         return 0;
3365 }
3366
3367 static int kvm_resume(struct sys_device *dev)
3368 {
3369         hardware_enable(NULL);
3370         return 0;
3371 }
3372
3373 static struct sysdev_class kvm_sysdev_class = {
3374         .name = "kvm",
3375         .suspend = kvm_suspend,
3376         .resume = kvm_resume,
3377 };
3378
3379 static struct sys_device kvm_sysdev = {
3380         .id = 0,
3381         .cls = &kvm_sysdev_class,
3382 };
3383
3384 struct page *bad_page;
3385
3386 static inline
3387 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3388 {
3389         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3390 }
3391
3392 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3393 {
3394         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3395
3396         kvm_x86_ops->vcpu_load(vcpu, cpu);
3397 }
3398
3399 static void kvm_sched_out(struct preempt_notifier *pn,
3400                           struct task_struct *next)
3401 {
3402         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3403
3404         kvm_x86_ops->vcpu_put(vcpu);
3405 }
3406
3407 int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
3408                   struct module *module)
3409 {
3410         int r;
3411         int cpu;
3412
3413         if (kvm_x86_ops) {
3414                 printk(KERN_ERR "kvm: already loaded the other module\n");
3415                 return -EEXIST;
3416         }
3417
3418         if (!ops->cpu_has_kvm_support()) {
3419                 printk(KERN_ERR "kvm: no hardware support\n");
3420                 return -EOPNOTSUPP;
3421         }
3422         if (ops->disabled_by_bios()) {
3423                 printk(KERN_ERR "kvm: disabled by bios\n");
3424                 return -EOPNOTSUPP;
3425         }
3426
3427         kvm_x86_ops = ops;
3428
3429         r = kvm_x86_ops->hardware_setup();
3430         if (r < 0)
3431                 goto out;
3432
3433         for_each_online_cpu(cpu) {
3434                 smp_call_function_single(cpu,
3435                                 kvm_x86_ops->check_processor_compatibility,
3436                                 &r, 0, 1);
3437                 if (r < 0)
3438                         goto out_free_0;
3439         }
3440
3441         on_each_cpu(hardware_enable, NULL, 0, 1);
3442         r = register_cpu_notifier(&kvm_cpu_notifier);
3443         if (r)
3444                 goto out_free_1;
3445         register_reboot_notifier(&kvm_reboot_notifier);
3446
3447         r = sysdev_class_register(&kvm_sysdev_class);
3448         if (r)
3449                 goto out_free_2;
3450
3451         r = sysdev_register(&kvm_sysdev);
3452         if (r)
3453                 goto out_free_3;
3454
3455         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3456         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3457                                            __alignof__(struct kvm_vcpu), 0, 0);
3458         if (!kvm_vcpu_cache) {
3459                 r = -ENOMEM;
3460                 goto out_free_4;
3461         }
3462
3463         kvm_chardev_ops.owner = module;
3464
3465         r = misc_register(&kvm_dev);
3466         if (r) {
3467                 printk(KERN_ERR "kvm: misc device register failed\n");
3468                 goto out_free;
3469         }
3470
3471         kvm_preempt_ops.sched_in = kvm_sched_in;
3472         kvm_preempt_ops.sched_out = kvm_sched_out;
3473
3474         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3475
3476         return 0;
3477
3478 out_free:
3479         kmem_cache_destroy(kvm_vcpu_cache);
3480 out_free_4:
3481         sysdev_unregister(&kvm_sysdev);
3482 out_free_3:
3483         sysdev_class_unregister(&kvm_sysdev_class);
3484 out_free_2:
3485         unregister_reboot_notifier(&kvm_reboot_notifier);
3486         unregister_cpu_notifier(&kvm_cpu_notifier);
3487 out_free_1:
3488         on_each_cpu(hardware_disable, NULL, 0, 1);
3489 out_free_0:
3490         kvm_x86_ops->hardware_unsetup();
3491 out:
3492         kvm_x86_ops = NULL;
3493         return r;
3494 }
3495 EXPORT_SYMBOL_GPL(kvm_init_x86);
3496
3497 void kvm_exit_x86(void)
3498 {
3499         misc_deregister(&kvm_dev);
3500         kmem_cache_destroy(kvm_vcpu_cache);
3501         sysdev_unregister(&kvm_sysdev);
3502         sysdev_class_unregister(&kvm_sysdev_class);
3503         unregister_reboot_notifier(&kvm_reboot_notifier);
3504         unregister_cpu_notifier(&kvm_cpu_notifier);
3505         on_each_cpu(hardware_disable, NULL, 0, 1);
3506         kvm_x86_ops->hardware_unsetup();
3507         kvm_x86_ops = NULL;
3508 }
3509 EXPORT_SYMBOL_GPL(kvm_exit_x86);
3510
3511 static __init int kvm_init(void)
3512 {
3513         int r;
3514
3515         r = kvm_mmu_module_init();
3516         if (r)
3517                 goto out4;
3518
3519         kvm_init_debug();
3520
3521         kvm_arch_init();
3522
3523         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3524
3525         if (bad_page == NULL) {
3526                 r = -ENOMEM;
3527                 goto out;
3528         }
3529
3530         return 0;
3531
3532 out:
3533         kvm_exit_debug();
3534         kvm_mmu_module_exit();
3535 out4:
3536         return r;
3537 }
3538
3539 static __exit void kvm_exit(void)
3540 {
3541         kvm_exit_debug();
3542         __free_page(bad_page);
3543         kvm_mmu_module_exit();
3544 }
3545
3546 module_init(kvm_init)
3547 module_exit(kvm_exit)