drivers/kvm/x86.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * derived from drivers/kvm/kvm_main.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  *
   8  * Authors:
   9  *   Avi Kivity   <avi@qumranet.com>
  10  *   Yaniv Kamay  <yaniv@qumranet.com>
  11  *
  12  * This work is licensed under the terms of the GNU GPL, version 2.  See
  13  * the COPYING file in the top-level directory.
  14  *
  15  */
  16
  17 #include "kvm.h"
  18 #include "x86.h"
  19 #include "x86_emulate.h"
  20 #include "segment_descriptor.h"
  21 #include "irq.h"
  22
  23 #include <linux/kvm.h>
  24 #include <linux/fs.h>
  25 #include <linux/vmalloc.h>
  26 #include <linux/module.h>
  27
  28 #include <asm/uaccess.h>
  29 #include <asm/msr.h>
  30
  31 #define MAX_IO_MSRS 256
  32 #define CR0_RESERVED_BITS                                               \
  33         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  34                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  35                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  36 #define CR4_RESERVED_BITS                                               \
  37         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  38                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  39                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  40                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  41
  42 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  43 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  44
  45 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  46
  47 struct kvm_x86_ops *kvm_x86_ops;
  48
  49 struct kvm_stats_debugfs_item debugfs_entries[] = {
  50         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  51         { "pf_guest", STAT_OFFSET(pf_guest) },
  52         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  53         { "invlpg", STAT_OFFSET(invlpg) },
  54         { "exits", STAT_OFFSET(exits) },
  55         { "io_exits", STAT_OFFSET(io_exits) },
  56         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  57         { "signal_exits", STAT_OFFSET(signal_exits) },
  58         { "irq_window", STAT_OFFSET(irq_window_exits) },
  59         { "halt_exits", STAT_OFFSET(halt_exits) },
  60         { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
  61         { "request_irq", STAT_OFFSET(request_irq_exits) },
  62         { "irq_exits", STAT_OFFSET(irq_exits) },
  63         { "light_exits", STAT_OFFSET(light_exits) },
  64         { "efer_reload", STAT_OFFSET(efer_reload) },
  65         { NULL }
  66 };
  67
  68
  69 unsigned long segment_base(u16 selector)
  70 {
  71         struct descriptor_table gdt;
  72         struct segment_descriptor *d;
  73         unsigned long table_base;
  74         unsigned long v;
  75
  76         if (selector == 0)
  77                 return 0;
  78
  79         asm("sgdt %0" : "=m"(gdt));
  80         table_base = gdt.base;
  81
  82         if (selector & 4) {           /* from ldt */
  83                 u16 ldt_selector;
  84
  85                 asm("sldt %0" : "=g"(ldt_selector));
  86                 table_base = segment_base(ldt_selector);
  87         }
  88         d = (struct segment_descriptor *)(table_base + (selector & ~7));
  89         v = d->base_low | ((unsigned long)d->base_mid << 16) |
  90                 ((unsigned long)d->base_high << 24);
  91 #ifdef CONFIG_X86_64
  92         if (d->system == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
  93                 v |= ((unsigned long) \
  94                       ((struct segment_descriptor_64 *)d)->base_higher) << 32;
  95 #endif
  96         return v;
  97 }
  98 EXPORT_SYMBOL_GPL(segment_base);
  99
 100 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 101 {
 102         if (irqchip_in_kernel(vcpu->kvm))
 103                 return vcpu->apic_base;
 104         else
 105                 return vcpu->apic_base;
 106 }
 107 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 108
 109 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 110 {
 111         /* TODO: reserve bits check */
 112         if (irqchip_in_kernel(vcpu->kvm))
 113                 kvm_lapic_set_base(vcpu, data);
 114         else
 115                 vcpu->apic_base = data;
 116 }
 117 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 118
 119 static void inject_gp(struct kvm_vcpu *vcpu)
 120 {
 121         kvm_x86_ops->inject_gp(vcpu, 0);
 122 }
 123
 124 /*
 125  * Load the pae pdptrs.  Return true is they are all valid.
 126  */
 127 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 128 {
 129         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 130         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 131         int i;
 132         int ret;
 133         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
 134
 135         mutex_lock(&vcpu->kvm->lock);
 136         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 137                                   offset * sizeof(u64), sizeof(pdpte));
 138         if (ret < 0) {
 139                 ret = 0;
 140                 goto out;
 141         }
 142         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 143                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 144                         ret = 0;
 145                         goto out;
 146                 }
 147         }
 148         ret = 1;
 149
 150         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
 151 out:
 152         mutex_unlock(&vcpu->kvm->lock);
 153
 154         return ret;
 155 }
 156
 157 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 158 {
 159         if (cr0 & CR0_RESERVED_BITS) {
 160                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 161                        cr0, vcpu->cr0);
 162                 inject_gp(vcpu);
 163                 return;
 164         }
 165
 166         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 167                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 168                 inject_gp(vcpu);
 169                 return;
 170         }
 171
 172         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 173                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 174                        "and a clear PE flag\n");
 175                 inject_gp(vcpu);
 176                 return;
 177         }
 178
 179         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 180 #ifdef CONFIG_X86_64
 181                 if ((vcpu->shadow_efer & EFER_LME)) {
 182                         int cs_db, cs_l;
 183
 184                         if (!is_pae(vcpu)) {
 185                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 186                                        "in long mode while PAE is disabled\n");
 187                                 inject_gp(vcpu);
 188                                 return;
 189                         }
 190                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 191                         if (cs_l) {
 192                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 193                                        "in long mode while CS.L == 1\n");
 194                                 inject_gp(vcpu);
 195                                 return;
 196
 197                         }
 198                 } else
 199 #endif
 200                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 201                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 202                                "reserved bits\n");
 203                         inject_gp(vcpu);
 204                         return;
 205                 }
 206
 207         }
 208
 209         kvm_x86_ops->set_cr0(vcpu, cr0);
 210         vcpu->cr0 = cr0;
 211
 212         mutex_lock(&vcpu->kvm->lock);
 213         kvm_mmu_reset_context(vcpu);
 214         mutex_unlock(&vcpu->kvm->lock);
 215         return;
 216 }
 217 EXPORT_SYMBOL_GPL(set_cr0);
 218
 219 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 220 {
 221         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 222 }
 223 EXPORT_SYMBOL_GPL(lmsw);
 224
 225 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 226 {
 227         if (cr4 & CR4_RESERVED_BITS) {
 228                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 229                 inject_gp(vcpu);
 230                 return;
 231         }
 232
 233         if (is_long_mode(vcpu)) {
 234                 if (!(cr4 & X86_CR4_PAE)) {
 235                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 236                                "in long mode\n");
 237                         inject_gp(vcpu);
 238                         return;
 239                 }
 240         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 241                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 242                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 243                 inject_gp(vcpu);
 244                 return;
 245         }
 246
 247         if (cr4 & X86_CR4_VMXE) {
 248                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 249                 inject_gp(vcpu);
 250                 return;
 251         }
 252         kvm_x86_ops->set_cr4(vcpu, cr4);
 253         vcpu->cr4 = cr4;
 254         mutex_lock(&vcpu->kvm->lock);
 255         kvm_mmu_reset_context(vcpu);
 256         mutex_unlock(&vcpu->kvm->lock);
 257 }
 258 EXPORT_SYMBOL_GPL(set_cr4);
 259
 260 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 261 {
 262         if (is_long_mode(vcpu)) {
 263                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 264                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 265                         inject_gp(vcpu);
 266                         return;
 267                 }
 268         } else {
 269                 if (is_pae(vcpu)) {
 270                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 271                                 printk(KERN_DEBUG
 272                                        "set_cr3: #GP, reserved bits\n");
 273                                 inject_gp(vcpu);
 274                                 return;
 275                         }
 276                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 277                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 278                                        "reserved bits\n");
 279                                 inject_gp(vcpu);
 280                                 return;
 281                         }
 282                 }
 283                 /*
 284                  * We don't check reserved bits in nonpae mode, because
 285                  * this isn't enforced, and VMware depends on this.
 286                  */
 287         }
 288
 289         mutex_lock(&vcpu->kvm->lock);
 290         /*
 291          * Does the new cr3 value map to physical memory? (Note, we
 292          * catch an invalid cr3 even in real-mode, because it would
 293          * cause trouble later on when we turn on paging anyway.)
 294          *
 295          * A real CPU would silently accept an invalid cr3 and would
 296          * attempt to use it - with largely undefined (and often hard
 297          * to debug) behavior on the guest side.
 298          */
 299         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 300                 inject_gp(vcpu);
 301         else {
 302                 vcpu->cr3 = cr3;
 303                 vcpu->mmu.new_cr3(vcpu);
 304         }
 305         mutex_unlock(&vcpu->kvm->lock);
 306 }
 307 EXPORT_SYMBOL_GPL(set_cr3);
 308
 309 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 310 {
 311         if (cr8 & CR8_RESERVED_BITS) {
 312                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 313                 inject_gp(vcpu);
 314                 return;
 315         }
 316         if (irqchip_in_kernel(vcpu->kvm))
 317                 kvm_lapic_set_tpr(vcpu, cr8);
 318         else
 319                 vcpu->cr8 = cr8;
 320 }
 321 EXPORT_SYMBOL_GPL(set_cr8);
 322
 323 unsigned long get_cr8(struct kvm_vcpu *vcpu)
 324 {
 325         if (irqchip_in_kernel(vcpu->kvm))
 326                 return kvm_lapic_get_cr8(vcpu);
 327         else
 328                 return vcpu->cr8;
 329 }
 330 EXPORT_SYMBOL_GPL(get_cr8);
 331
 332 /*
 333  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 334  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 335  *
 336  * This list is modified at module load time to reflect the
 337  * capabilities of the host cpu.
 338  */
 339 static u32 msrs_to_save[] = {
 340         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 341         MSR_K6_STAR,
 342 #ifdef CONFIG_X86_64
 343         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 344 #endif
 345         MSR_IA32_TIME_STAMP_COUNTER,
 346 };
 347
 348 static unsigned num_msrs_to_save;
 349
 350 static u32 emulated_msrs[] = {
 351         MSR_IA32_MISC_ENABLE,
 352 };
 353
 354 #ifdef CONFIG_X86_64
 355
 356 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 357 {
 358         if (efer & EFER_RESERVED_BITS) {
 359                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 360                        efer);
 361                 inject_gp(vcpu);
 362                 return;
 363         }
 364
 365         if (is_paging(vcpu)
 366             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 367                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 368                 inject_gp(vcpu);
 369                 return;
 370         }
 371
 372         kvm_x86_ops->set_efer(vcpu, efer);
 373
 374         efer &= ~EFER_LMA;
 375         efer |= vcpu->shadow_efer & EFER_LMA;
 376
 377         vcpu->shadow_efer = efer;
 378 }
 379
 380 #endif
 381
 382 /*
 383  * Writes msr value into into the appropriate "register".
 384  * Returns 0 on success, non-0 otherwise.
 385  * Assumes vcpu_load() was already called.
 386  */
 387 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 388 {
 389         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 390 }
 391
 392 /*
 393  * Adapt set_msr() to msr_io()'s calling convention
 394  */
 395 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 396 {
 397         return kvm_set_msr(vcpu, index, *data);
 398 }
 399
 400
 401 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 402 {
 403         switch (msr) {
 404 #ifdef CONFIG_X86_64
 405         case MSR_EFER:
 406                 set_efer(vcpu, data);
 407                 break;
 408 #endif
 409         case MSR_IA32_MC0_STATUS:
 410                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 411                        __FUNCTION__, data);
 412                 break;
 413         case MSR_IA32_MCG_STATUS:
 414                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 415                         __FUNCTION__, data);
 416                 break;
 417         case MSR_IA32_UCODE_REV:
 418         case MSR_IA32_UCODE_WRITE:
 419         case 0x200 ... 0x2ff: /* MTRRs */
 420                 break;
 421         case MSR_IA32_APICBASE:
 422                 kvm_set_apic_base(vcpu, data);
 423                 break;
 424         case MSR_IA32_MISC_ENABLE:
 425                 vcpu->ia32_misc_enable_msr = data;
 426                 break;
 427         default:
 428                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
 429                 return 1;
 430         }
 431         return 0;
 432 }
 433 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 434
 435
 436 /*
 437  * Reads an msr value (of 'msr_index') into 'pdata'.
 438  * Returns 0 on success, non-0 otherwise.
 439  * Assumes vcpu_load() was already called.
 440  */
 441 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 442 {
 443         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 444 }
 445
 446 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 447 {
 448         u64 data;
 449
 450         switch (msr) {
 451         case 0xc0010010: /* SYSCFG */
 452         case 0xc0010015: /* HWCR */
 453         case MSR_IA32_PLATFORM_ID:
 454         case MSR_IA32_P5_MC_ADDR:
 455         case MSR_IA32_P5_MC_TYPE:
 456         case MSR_IA32_MC0_CTL:
 457         case MSR_IA32_MCG_STATUS:
 458         case MSR_IA32_MCG_CAP:
 459         case MSR_IA32_MC0_MISC:
 460         case MSR_IA32_MC0_MISC+4:
 461         case MSR_IA32_MC0_MISC+8:
 462         case MSR_IA32_MC0_MISC+12:
 463         case MSR_IA32_MC0_MISC+16:
 464         case MSR_IA32_UCODE_REV:
 465         case MSR_IA32_PERF_STATUS:
 466         case MSR_IA32_EBL_CR_POWERON:
 467                 /* MTRR registers */
 468         case 0xfe:
 469         case 0x200 ... 0x2ff:
 470                 data = 0;
 471                 break;
 472         case 0xcd: /* fsb frequency */
 473                 data = 3;
 474                 break;
 475         case MSR_IA32_APICBASE:
 476                 data = kvm_get_apic_base(vcpu);
 477                 break;
 478         case MSR_IA32_MISC_ENABLE:
 479                 data = vcpu->ia32_misc_enable_msr;
 480                 break;
 481 #ifdef CONFIG_X86_64
 482         case MSR_EFER:
 483                 data = vcpu->shadow_efer;
 484                 break;
 485 #endif
 486         default:
 487                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 488                 return 1;
 489         }
 490         *pdata = data;
 491         return 0;
 492 }
 493 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 494
 495 /*
 496  * Read or write a bunch of msrs. All parameters are kernel addresses.
 497  *
 498  * @return number of msrs set successfully.
 499  */
 500 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 501                     struct kvm_msr_entry *entries,
 502                     int (*do_msr)(struct kvm_vcpu *vcpu,
 503                                   unsigned index, u64 *data))
 504 {
 505         int i;
 506
 507         vcpu_load(vcpu);
 508
 509         for (i = 0; i < msrs->nmsrs; ++i)
 510                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 511                         break;
 512
 513         vcpu_put(vcpu);
 514
 515         return i;
 516 }
 517
 518 /*
 519  * Read or write a bunch of msrs. Parameters are user addresses.
 520  *
 521  * @return number of msrs set successfully.
 522  */
 523 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 524                   int (*do_msr)(struct kvm_vcpu *vcpu,
 525                                 unsigned index, u64 *data),
 526                   int writeback)
 527 {
 528         struct kvm_msrs msrs;
 529         struct kvm_msr_entry *entries;
 530         int r, n;
 531         unsigned size;
 532
 533         r = -EFAULT;
 534         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
 535                 goto out;
 536
 537         r = -E2BIG;
 538         if (msrs.nmsrs >= MAX_IO_MSRS)
 539                 goto out;
 540
 541         r = -ENOMEM;
 542         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
 543         entries = vmalloc(size);
 544         if (!entries)
 545                 goto out;
 546
 547         r = -EFAULT;
 548         if (copy_from_user(entries, user_msrs->entries, size))
 549                 goto out_free;
 550
 551         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
 552         if (r < 0)
 553                 goto out_free;
 554
 555         r = -EFAULT;
 556         if (writeback && copy_to_user(user_msrs->entries, entries, size))
 557                 goto out_free;
 558
 559         r = n;
 560
 561 out_free:
 562         vfree(entries);
 563 out:
 564         return r;
 565 }
 566
 567 long kvm_arch_dev_ioctl(struct file *filp,
 568                         unsigned int ioctl, unsigned long arg)
 569 {
 570         void __user *argp = (void __user *)arg;
 571         long r;
 572
 573         switch (ioctl) {
 574         case KVM_GET_MSR_INDEX_LIST: {
 575                 struct kvm_msr_list __user *user_msr_list = argp;
 576                 struct kvm_msr_list msr_list;
 577                 unsigned n;
 578
 579                 r = -EFAULT;
 580                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
 581                         goto out;
 582                 n = msr_list.nmsrs;
 583                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
 584                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
 585                         goto out;
 586                 r = -E2BIG;
 587                 if (n < num_msrs_to_save)
 588                         goto out;
 589                 r = -EFAULT;
 590                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
 591                                  num_msrs_to_save * sizeof(u32)))
 592                         goto out;
 593                 if (copy_to_user(user_msr_list->indices
 594                                  + num_msrs_to_save * sizeof(u32),
 595                                  &emulated_msrs,
 596                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
 597                         goto out;
 598                 r = 0;
 599                 break;
 600         }
 601         default:
 602                 r = -EINVAL;
 603         }
 604 out:
 605         return r;
 606 }
 607
 608 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 609 {
 610         kvm_x86_ops->vcpu_load(vcpu, cpu);
 611 }
 612
 613 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 614 {
 615         kvm_x86_ops->vcpu_put(vcpu);
 616 }
 617
 618 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 619 {
 620         u64 efer;
 621         int i;
 622         struct kvm_cpuid_entry *e, *entry;
 623
 624         rdmsrl(MSR_EFER, efer);
 625         entry = NULL;
 626         for (i = 0; i < vcpu->cpuid_nent; ++i) {
 627                 e = &vcpu->cpuid_entries[i];
 628                 if (e->function == 0x80000001) {
 629                         entry = e;
 630                         break;
 631                 }
 632         }
 633         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
 634                 entry->edx &= ~(1 << 20);
 635                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 636         }
 637 }
 638
 639 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 640                                     struct kvm_cpuid *cpuid,
 641                                     struct kvm_cpuid_entry __user *entries)
 642 {
 643         int r;
 644
 645         r = -E2BIG;
 646         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 647                 goto out;
 648         r = -EFAULT;
 649         if (copy_from_user(&vcpu->cpuid_entries, entries,
 650                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 651                 goto out;
 652         vcpu->cpuid_nent = cpuid->nent;
 653         cpuid_fix_nx_cap(vcpu);
 654         return 0;
 655
 656 out:
 657         return r;
 658 }
 659
 660 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
 661                                     struct kvm_lapic_state *s)
 662 {
 663         vcpu_load(vcpu);
 664         memcpy(s->regs, vcpu->apic->regs, sizeof *s);
 665         vcpu_put(vcpu);
 666
 667         return 0;
 668 }
 669
 670 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
 671                                     struct kvm_lapic_state *s)
 672 {
 673         vcpu_load(vcpu);
 674         memcpy(vcpu->apic->regs, s->regs, sizeof *s);
 675         kvm_apic_post_state_restore(vcpu);
 676         vcpu_put(vcpu);
 677
 678         return 0;
 679 }
 680
 681 long kvm_arch_vcpu_ioctl(struct file *filp,
 682                          unsigned int ioctl, unsigned long arg)
 683 {
 684         struct kvm_vcpu *vcpu = filp->private_data;
 685         void __user *argp = (void __user *)arg;
 686         int r;
 687
 688         switch (ioctl) {
 689         case KVM_GET_LAPIC: {
 690                 struct kvm_lapic_state lapic;
 691
 692                 memset(&lapic, 0, sizeof lapic);
 693                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
 694                 if (r)
 695                         goto out;
 696                 r = -EFAULT;
 697                 if (copy_to_user(argp, &lapic, sizeof lapic))
 698                         goto out;
 699                 r = 0;
 700                 break;
 701         }
 702         case KVM_SET_LAPIC: {
 703                 struct kvm_lapic_state lapic;
 704
 705                 r = -EFAULT;
 706                 if (copy_from_user(&lapic, argp, sizeof lapic))
 707                         goto out;
 708                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
 709                 if (r)
 710                         goto out;
 711                 r = 0;
 712                 break;
 713         }
 714         case KVM_SET_CPUID: {
 715                 struct kvm_cpuid __user *cpuid_arg = argp;
 716                 struct kvm_cpuid cpuid;
 717
 718                 r = -EFAULT;
 719                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
 720                         goto out;
 721                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
 722                 if (r)
 723                         goto out;
 724                 break;
 725         }
 726         case KVM_GET_MSRS:
 727                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
 728                 break;
 729         case KVM_SET_MSRS:
 730                 r = msr_io(vcpu, argp, do_set_msr, 0);
 731                 break;
 732         default:
 733                 r = -EINVAL;
 734         }
 735 out:
 736         return r;
 737 }
 738
 739 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
 740 {
 741         int ret;
 742
 743         if (addr > (unsigned int)(-3 * PAGE_SIZE))
 744                 return -1;
 745         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
 746         return ret;
 747 }
 748
 749 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
 750                                           u32 kvm_nr_mmu_pages)
 751 {
 752         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
 753                 return -EINVAL;
 754
 755         mutex_lock(&kvm->lock);
 756
 757         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
 758         kvm->n_requested_mmu_pages = kvm_nr_mmu_pages;
 759
 760         mutex_unlock(&kvm->lock);
 761         return 0;
 762 }
 763
 764 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
 765 {
 766         return kvm->n_alloc_mmu_pages;
 767 }
 768
 769 /*
 770  * Set a new alias region.  Aliases map a portion of physical memory into
 771  * another portion.  This is useful for memory windows, for example the PC
 772  * VGA region.
 773  */
 774 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 775                                          struct kvm_memory_alias *alias)
 776 {
 777         int r, n;
 778         struct kvm_mem_alias *p;
 779
 780         r = -EINVAL;
 781         /* General sanity checks */
 782         if (alias->memory_size & (PAGE_SIZE - 1))
 783                 goto out;
 784         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 785                 goto out;
 786         if (alias->slot >= KVM_ALIAS_SLOTS)
 787                 goto out;
 788         if (alias->guest_phys_addr + alias->memory_size
 789             < alias->guest_phys_addr)
 790                 goto out;
 791         if (alias->target_phys_addr + alias->memory_size
 792             < alias->target_phys_addr)
 793                 goto out;
 794
 795         mutex_lock(&kvm->lock);
 796
 797         p = &kvm->aliases[alias->slot];
 798         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 799         p->npages = alias->memory_size >> PAGE_SHIFT;
 800         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 801
 802         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 803                 if (kvm->aliases[n - 1].npages)
 804                         break;
 805         kvm->naliases = n;
 806
 807         kvm_mmu_zap_all(kvm);
 808
 809         mutex_unlock(&kvm->lock);
 810
 811         return 0;
 812
 813 out:
 814         return r;
 815 }
 816
 817 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 818 {
 819         int r;
 820
 821         r = 0;
 822         switch (chip->chip_id) {
 823         case KVM_IRQCHIP_PIC_MASTER:
 824                 memcpy(&chip->chip.pic,
 825                         &pic_irqchip(kvm)->pics[0],
 826                         sizeof(struct kvm_pic_state));
 827                 break;
 828         case KVM_IRQCHIP_PIC_SLAVE:
 829                 memcpy(&chip->chip.pic,
 830                         &pic_irqchip(kvm)->pics[1],
 831                         sizeof(struct kvm_pic_state));
 832                 break;
 833         case KVM_IRQCHIP_IOAPIC:
 834                 memcpy(&chip->chip.ioapic,
 835                         ioapic_irqchip(kvm),
 836                         sizeof(struct kvm_ioapic_state));
 837                 break;
 838         default:
 839                 r = -EINVAL;
 840                 break;
 841         }
 842         return r;
 843 }
 844
 845 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 846 {
 847         int r;
 848
 849         r = 0;
 850         switch (chip->chip_id) {
 851         case KVM_IRQCHIP_PIC_MASTER:
 852                 memcpy(&pic_irqchip(kvm)->pics[0],
 853                         &chip->chip.pic,
 854                         sizeof(struct kvm_pic_state));
 855                 break;
 856         case KVM_IRQCHIP_PIC_SLAVE:
 857                 memcpy(&pic_irqchip(kvm)->pics[1],
 858                         &chip->chip.pic,
 859                         sizeof(struct kvm_pic_state));
 860                 break;
 861         case KVM_IRQCHIP_IOAPIC:
 862                 memcpy(ioapic_irqchip(kvm),
 863                         &chip->chip.ioapic,
 864                         sizeof(struct kvm_ioapic_state));
 865                 break;
 866         default:
 867                 r = -EINVAL;
 868                 break;
 869         }
 870         kvm_pic_update_irq(pic_irqchip(kvm));
 871         return r;
 872 }
 873
 874 long kvm_arch_vm_ioctl(struct file *filp,
 875                        unsigned int ioctl, unsigned long arg)
 876 {
 877         struct kvm *kvm = filp->private_data;
 878         void __user *argp = (void __user *)arg;
 879         int r = -EINVAL;
 880
 881         switch (ioctl) {
 882         case KVM_SET_TSS_ADDR:
 883                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
 884                 if (r < 0)
 885                         goto out;
 886                 break;
 887         case KVM_SET_MEMORY_REGION: {
 888                 struct kvm_memory_region kvm_mem;
 889                 struct kvm_userspace_memory_region kvm_userspace_mem;
 890
 891                 r = -EFAULT;
 892                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
 893                         goto out;
 894                 kvm_userspace_mem.slot = kvm_mem.slot;
 895                 kvm_userspace_mem.flags = kvm_mem.flags;
 896                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
 897                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
 898                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
 899                 if (r)
 900                         goto out;
 901                 break;
 902         }
 903         case KVM_SET_NR_MMU_PAGES:
 904                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
 905                 if (r)
 906                         goto out;
 907                 break;
 908         case KVM_GET_NR_MMU_PAGES:
 909                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
 910                 break;
 911         case KVM_SET_MEMORY_ALIAS: {
 912                 struct kvm_memory_alias alias;
 913
 914                 r = -EFAULT;
 915                 if (copy_from_user(&alias, argp, sizeof alias))
 916                         goto out;
 917                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
 918                 if (r)
 919                         goto out;
 920                 break;
 921         }
 922         case KVM_CREATE_IRQCHIP:
 923                 r = -ENOMEM;
 924                 kvm->vpic = kvm_create_pic(kvm);
 925                 if (kvm->vpic) {
 926                         r = kvm_ioapic_init(kvm);
 927                         if (r) {
 928                                 kfree(kvm->vpic);
 929                                 kvm->vpic = NULL;
 930                                 goto out;
 931                         }
 932                 } else
 933                         goto out;
 934                 break;
 935         case KVM_IRQ_LINE: {
 936                 struct kvm_irq_level irq_event;
 937
 938                 r = -EFAULT;
 939                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
 940                         goto out;
 941                 if (irqchip_in_kernel(kvm)) {
 942                         mutex_lock(&kvm->lock);
 943                         if (irq_event.irq < 16)
 944                                 kvm_pic_set_irq(pic_irqchip(kvm),
 945                                         irq_event.irq,
 946                                         irq_event.level);
 947                         kvm_ioapic_set_irq(kvm->vioapic,
 948                                         irq_event.irq,
 949                                         irq_event.level);
 950                         mutex_unlock(&kvm->lock);
 951                         r = 0;
 952                 }
 953                 break;
 954         }
 955         case KVM_GET_IRQCHIP: {
 956                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
 957                 struct kvm_irqchip chip;
 958
 959                 r = -EFAULT;
 960                 if (copy_from_user(&chip, argp, sizeof chip))
 961                         goto out;
 962                 r = -ENXIO;
 963                 if (!irqchip_in_kernel(kvm))
 964                         goto out;
 965                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
 966                 if (r)
 967                         goto out;
 968                 r = -EFAULT;
 969                 if (copy_to_user(argp, &chip, sizeof chip))
 970                         goto out;
 971                 r = 0;
 972                 break;
 973         }
 974         case KVM_SET_IRQCHIP: {
 975                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
 976                 struct kvm_irqchip chip;
 977
 978                 r = -EFAULT;
 979                 if (copy_from_user(&chip, argp, sizeof chip))
 980                         goto out;
 981                 r = -ENXIO;
 982                 if (!irqchip_in_kernel(kvm))
 983                         goto out;
 984                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
 985                 if (r)
 986                         goto out;
 987                 r = 0;
 988                 break;
 989         }
 990         default:
 991                 ;
 992         }
 993 out:
 994         return r;
 995 }
 996
 997 static __init void kvm_init_msr_list(void)
 998 {
 999         u32 dummy[2];
1000         unsigned i, j;
1001
1002         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1003                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1004                         continue;
1005                 if (j < i)
1006                         msrs_to_save[j] = msrs_to_save[i];
1007                 j++;
1008         }
1009         num_msrs_to_save = j;
1010 }
1011
1012 /*
1013  * Only apic need an MMIO device hook, so shortcut now..
1014  */
1015 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1016                                                 gpa_t addr)
1017 {
1018         struct kvm_io_device *dev;
1019
1020         if (vcpu->apic) {
1021                 dev = &vcpu->apic->dev;
1022                 if (dev->in_range(dev, addr))
1023                         return dev;
1024         }
1025         return NULL;
1026 }
1027
1028
1029 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1030                                                 gpa_t addr)
1031 {
1032         struct kvm_io_device *dev;
1033
1034         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1035         if (dev == NULL)
1036                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1037         return dev;
1038 }
1039
1040 int emulator_read_std(unsigned long addr,
1041                              void *val,
1042                              unsigned int bytes,
1043                              struct kvm_vcpu *vcpu)
1044 {
1045         void *data = val;
1046
1047         while (bytes) {
1048                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1049                 unsigned offset = addr & (PAGE_SIZE-1);
1050                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1051                 int ret;
1052
1053                 if (gpa == UNMAPPED_GVA)
1054                         return X86EMUL_PROPAGATE_FAULT;
1055                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1056                 if (ret < 0)
1057                         return X86EMUL_UNHANDLEABLE;
1058
1059                 bytes -= tocopy;
1060                 data += tocopy;
1061                 addr += tocopy;
1062         }
1063
1064         return X86EMUL_CONTINUE;
1065 }
1066 EXPORT_SYMBOL_GPL(emulator_read_std);
1067
1068 static int emulator_write_std(unsigned long addr,
1069                               const void *val,
1070                               unsigned int bytes,
1071                               struct kvm_vcpu *vcpu)
1072 {
1073         pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1074         return X86EMUL_UNHANDLEABLE;
1075 }
1076
1077 static int emulator_read_emulated(unsigned long addr,
1078                                   void *val,
1079                                   unsigned int bytes,
1080                                   struct kvm_vcpu *vcpu)
1081 {
1082         struct kvm_io_device *mmio_dev;
1083         gpa_t                 gpa;
1084
1085         if (vcpu->mmio_read_completed) {
1086                 memcpy(val, vcpu->mmio_data, bytes);
1087                 vcpu->mmio_read_completed = 0;
1088                 return X86EMUL_CONTINUE;
1089         }
1090
1091         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1092
1093         /* For APIC access vmexit */
1094         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1095                 goto mmio;
1096
1097         if (emulator_read_std(addr, val, bytes, vcpu)
1098                         == X86EMUL_CONTINUE)
1099                 return X86EMUL_CONTINUE;
1100         if (gpa == UNMAPPED_GVA)
1101                 return X86EMUL_PROPAGATE_FAULT;
1102
1103 mmio:
1104         /*
1105          * Is this MMIO handled locally?
1106          */
1107         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1108         if (mmio_dev) {
1109                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1110                 return X86EMUL_CONTINUE;
1111         }
1112
1113         vcpu->mmio_needed = 1;
1114         vcpu->mmio_phys_addr = gpa;
1115         vcpu->mmio_size = bytes;
1116         vcpu->mmio_is_write = 0;
1117
1118         return X86EMUL_UNHANDLEABLE;
1119 }
1120
1121 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1122                                const void *val, int bytes)
1123 {
1124         int ret;
1125
1126         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1127         if (ret < 0)
1128                 return 0;
1129         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1130         return 1;
1131 }
1132
1133 static int emulator_write_emulated_onepage(unsigned long addr,
1134                                            const void *val,
1135                                            unsigned int bytes,
1136                                            struct kvm_vcpu *vcpu)
1137 {
1138         struct kvm_io_device *mmio_dev;
1139         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1140
1141         if (gpa == UNMAPPED_GVA) {
1142                 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1143                 return X86EMUL_PROPAGATE_FAULT;
1144         }
1145
1146         /* For APIC access vmexit */
1147         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1148                 goto mmio;
1149
1150         if (emulator_write_phys(vcpu, gpa, val, bytes))
1151                 return X86EMUL_CONTINUE;
1152
1153 mmio:
1154         /*
1155          * Is this MMIO handled locally?
1156          */
1157         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1158         if (mmio_dev) {
1159                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1160                 return X86EMUL_CONTINUE;
1161         }
1162
1163         vcpu->mmio_needed = 1;
1164         vcpu->mmio_phys_addr = gpa;
1165         vcpu->mmio_size = bytes;
1166         vcpu->mmio_is_write = 1;
1167         memcpy(vcpu->mmio_data, val, bytes);
1168
1169         return X86EMUL_CONTINUE;
1170 }
1171
1172 int emulator_write_emulated(unsigned long addr,
1173                                    const void *val,
1174                                    unsigned int bytes,
1175                                    struct kvm_vcpu *vcpu)
1176 {
1177         /* Crossing a page boundary? */
1178         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1179                 int rc, now;
1180
1181                 now = -addr & ~PAGE_MASK;
1182                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1183                 if (rc != X86EMUL_CONTINUE)
1184                         return rc;
1185                 addr += now;
1186                 val += now;
1187                 bytes -= now;
1188         }
1189         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1190 }
1191 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1192
1193 static int emulator_cmpxchg_emulated(unsigned long addr,
1194                                      const void *old,
1195                                      const void *new,
1196                                      unsigned int bytes,
1197                                      struct kvm_vcpu *vcpu)
1198 {
1199         static int reported;
1200
1201         if (!reported) {
1202                 reported = 1;
1203                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1204         }
1205         return emulator_write_emulated(addr, new, bytes, vcpu);
1206 }
1207
1208 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1209 {
1210         return kvm_x86_ops->get_segment_base(vcpu, seg);
1211 }
1212
1213 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1214 {
1215         return X86EMUL_CONTINUE;
1216 }
1217
1218 int emulate_clts(struct kvm_vcpu *vcpu)
1219 {
1220         kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1221         return X86EMUL_CONTINUE;
1222 }
1223
1224 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1225 {
1226         struct kvm_vcpu *vcpu = ctxt->vcpu;
1227
1228         switch (dr) {
1229         case 0 ... 3:
1230                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1231                 return X86EMUL_CONTINUE;
1232         default:
1233                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1234                 return X86EMUL_UNHANDLEABLE;
1235         }
1236 }
1237
1238 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1239 {
1240         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1241         int exception;
1242
1243         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1244         if (exception) {
1245                 /* FIXME: better handling */
1246                 return X86EMUL_UNHANDLEABLE;
1247         }
1248         return X86EMUL_CONTINUE;
1249 }
1250
1251 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1252 {
1253         static int reported;
1254         u8 opcodes[4];
1255         unsigned long rip = vcpu->rip;
1256         unsigned long rip_linear;
1257
1258         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1259
1260         if (reported)
1261                 return;
1262
1263         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1264
1265         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1266                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1267         reported = 1;
1268 }
1269 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1270
1271 struct x86_emulate_ops emulate_ops = {
1272         .read_std            = emulator_read_std,
1273         .write_std           = emulator_write_std,
1274         .read_emulated       = emulator_read_emulated,
1275         .write_emulated      = emulator_write_emulated,
1276         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1277 };
1278
1279 int emulate_instruction(struct kvm_vcpu *vcpu,
1280                         struct kvm_run *run,
1281                         unsigned long cr2,
1282                         u16 error_code,
1283                         int no_decode)
1284 {
1285         int r;
1286
1287         vcpu->mmio_fault_cr2 = cr2;
1288         kvm_x86_ops->cache_regs(vcpu);
1289
1290         vcpu->mmio_is_write = 0;
1291         vcpu->pio.string = 0;
1292
1293         if (!no_decode) {
1294                 int cs_db, cs_l;
1295                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1296
1297                 vcpu->emulate_ctxt.vcpu = vcpu;
1298                 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1299                 vcpu->emulate_ctxt.cr2 = cr2;
1300                 vcpu->emulate_ctxt.mode =
1301                         (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1302                         ? X86EMUL_MODE_REAL : cs_l
1303                         ? X86EMUL_MODE_PROT64 : cs_db
1304                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1305
1306                 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1307                         vcpu->emulate_ctxt.cs_base = 0;
1308                         vcpu->emulate_ctxt.ds_base = 0;
1309                         vcpu->emulate_ctxt.es_base = 0;
1310                         vcpu->emulate_ctxt.ss_base = 0;
1311                 } else {
1312                         vcpu->emulate_ctxt.cs_base =
1313                                         get_segment_base(vcpu, VCPU_SREG_CS);
1314                         vcpu->emulate_ctxt.ds_base =
1315                                         get_segment_base(vcpu, VCPU_SREG_DS);
1316                         vcpu->emulate_ctxt.es_base =
1317                                         get_segment_base(vcpu, VCPU_SREG_ES);
1318                         vcpu->emulate_ctxt.ss_base =
1319                                         get_segment_base(vcpu, VCPU_SREG_SS);
1320                 }
1321
1322                 vcpu->emulate_ctxt.gs_base =
1323                                         get_segment_base(vcpu, VCPU_SREG_GS);
1324                 vcpu->emulate_ctxt.fs_base =
1325                                         get_segment_base(vcpu, VCPU_SREG_FS);
1326
1327                 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
1328                 if (r)  {
1329                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1330                                 return EMULATE_DONE;
1331                         return EMULATE_FAIL;
1332                 }
1333         }
1334
1335         r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1336
1337         if (vcpu->pio.string)
1338                 return EMULATE_DO_MMIO;
1339
1340         if ((r || vcpu->mmio_is_write) && run) {
1341                 run->exit_reason = KVM_EXIT_MMIO;
1342                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1343                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1344                 run->mmio.len = vcpu->mmio_size;
1345                 run->mmio.is_write = vcpu->mmio_is_write;
1346         }
1347
1348         if (r) {
1349                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1350                         return EMULATE_DONE;
1351                 if (!vcpu->mmio_needed) {
1352                         kvm_report_emulation_failure(vcpu, "mmio");
1353                         return EMULATE_FAIL;
1354                 }
1355                 return EMULATE_DO_MMIO;
1356         }
1357
1358         kvm_x86_ops->decache_regs(vcpu);
1359         kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1360
1361         if (vcpu->mmio_is_write) {
1362                 vcpu->mmio_needed = 0;
1363                 return EMULATE_DO_MMIO;
1364         }
1365
1366         return EMULATE_DONE;
1367 }
1368 EXPORT_SYMBOL_GPL(emulate_instruction);
1369
1370 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
1371 {
1372         int i;
1373
1374         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
1375                 if (vcpu->pio.guest_pages[i]) {
1376                         kvm_release_page(vcpu->pio.guest_pages[i]);
1377                         vcpu->pio.guest_pages[i] = NULL;
1378                 }
1379 }
1380
1381 static int pio_copy_data(struct kvm_vcpu *vcpu)
1382 {
1383         void *p = vcpu->pio_data;
1384         void *q;
1385         unsigned bytes;
1386         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1387
1388         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1389                  PAGE_KERNEL);
1390         if (!q) {
1391                 free_pio_guest_pages(vcpu);
1392                 return -ENOMEM;
1393         }
1394         q += vcpu->pio.guest_page_offset;
1395         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1396         if (vcpu->pio.in)
1397                 memcpy(q, p, bytes);
1398         else
1399                 memcpy(p, q, bytes);
1400         q -= vcpu->pio.guest_page_offset;
1401         vunmap(q);
1402         free_pio_guest_pages(vcpu);
1403         return 0;
1404 }
1405
1406 int complete_pio(struct kvm_vcpu *vcpu)
1407 {
1408         struct kvm_pio_request *io = &vcpu->pio;
1409         long delta;
1410         int r;
1411
1412         kvm_x86_ops->cache_regs(vcpu);
1413
1414         if (!io->string) {
1415                 if (io->in)
1416                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1417                                io->size);
1418         } else {
1419                 if (io->in) {
1420                         r = pio_copy_data(vcpu);
1421                         if (r) {
1422                                 kvm_x86_ops->cache_regs(vcpu);
1423                                 return r;
1424                         }
1425                 }
1426
1427                 delta = 1;
1428                 if (io->rep) {
1429                         delta *= io->cur_count;
1430                         /*
1431                          * The size of the register should really depend on
1432                          * current address size.
1433                          */
1434                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1435                 }
1436                 if (io->down)
1437                         delta = -delta;
1438                 delta *= io->size;
1439                 if (io->in)
1440                         vcpu->regs[VCPU_REGS_RDI] += delta;
1441                 else
1442                         vcpu->regs[VCPU_REGS_RSI] += delta;
1443         }
1444
1445         kvm_x86_ops->decache_regs(vcpu);
1446
1447         io->count -= io->cur_count;
1448         io->cur_count = 0;
1449
1450         return 0;
1451 }
1452
1453 static void kernel_pio(struct kvm_io_device *pio_dev,
1454                        struct kvm_vcpu *vcpu,
1455                        void *pd)
1456 {
1457         /* TODO: String I/O for in kernel device */
1458
1459         mutex_lock(&vcpu->kvm->lock);
1460         if (vcpu->pio.in)
1461                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1462                                   vcpu->pio.size,
1463                                   pd);
1464         else
1465                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1466                                    vcpu->pio.size,
1467                                    pd);
1468         mutex_unlock(&vcpu->kvm->lock);
1469 }
1470
1471 static void pio_string_write(struct kvm_io_device *pio_dev,
1472                              struct kvm_vcpu *vcpu)
1473 {
1474         struct kvm_pio_request *io = &vcpu->pio;
1475         void *pd = vcpu->pio_data;
1476         int i;
1477
1478         mutex_lock(&vcpu->kvm->lock);
1479         for (i = 0; i < io->cur_count; i++) {
1480                 kvm_iodevice_write(pio_dev, io->port,
1481                                    io->size,
1482                                    pd);
1483                 pd += io->size;
1484         }
1485         mutex_unlock(&vcpu->kvm->lock);
1486 }
1487
1488 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1489                                                gpa_t addr)
1490 {
1491         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1492 }
1493
1494 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1495                   int size, unsigned port)
1496 {
1497         struct kvm_io_device *pio_dev;
1498
1499         vcpu->run->exit_reason = KVM_EXIT_IO;
1500         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1501         vcpu->run->io.size = vcpu->pio.size = size;
1502         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1503         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1504         vcpu->run->io.port = vcpu->pio.port = port;
1505         vcpu->pio.in = in;
1506         vcpu->pio.string = 0;
1507         vcpu->pio.down = 0;
1508         vcpu->pio.guest_page_offset = 0;
1509         vcpu->pio.rep = 0;
1510
1511         kvm_x86_ops->cache_regs(vcpu);
1512         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1513         kvm_x86_ops->decache_regs(vcpu);
1514
1515         kvm_x86_ops->skip_emulated_instruction(vcpu);
1516
1517         pio_dev = vcpu_find_pio_dev(vcpu, port);
1518         if (pio_dev) {
1519                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1520                 complete_pio(vcpu);
1521                 return 1;
1522         }
1523         return 0;
1524 }
1525 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1526
1527 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1528                   int size, unsigned long count, int down,
1529                   gva_t address, int rep, unsigned port)
1530 {
1531         unsigned now, in_page;
1532         int i, ret = 0;
1533         int nr_pages = 1;
1534         struct page *page;
1535         struct kvm_io_device *pio_dev;
1536
1537         vcpu->run->exit_reason = KVM_EXIT_IO;
1538         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1539         vcpu->run->io.size = vcpu->pio.size = size;
1540         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1541         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1542         vcpu->run->io.port = vcpu->pio.port = port;
1543         vcpu->pio.in = in;
1544         vcpu->pio.string = 1;
1545         vcpu->pio.down = down;
1546         vcpu->pio.guest_page_offset = offset_in_page(address);
1547         vcpu->pio.rep = rep;
1548
1549         if (!count) {
1550                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1551                 return 1;
1552         }
1553
1554         if (!down)
1555                 in_page = PAGE_SIZE - offset_in_page(address);
1556         else
1557                 in_page = offset_in_page(address) + size;
1558         now = min(count, (unsigned long)in_page / size);
1559         if (!now) {
1560                 /*
1561                  * String I/O straddles page boundary.  Pin two guest pages
1562                  * so that we satisfy atomicity constraints.  Do just one
1563                  * transaction to avoid complexity.
1564                  */
1565                 nr_pages = 2;
1566                 now = 1;
1567         }
1568         if (down) {
1569                 /*
1570                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1571                  */
1572                 pr_unimpl(vcpu, "guest string pio down\n");
1573                 inject_gp(vcpu);
1574                 return 1;
1575         }
1576         vcpu->run->io.count = now;
1577         vcpu->pio.cur_count = now;
1578
1579         if (vcpu->pio.cur_count == vcpu->pio.count)
1580                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1581
1582         for (i = 0; i < nr_pages; ++i) {
1583                 mutex_lock(&vcpu->kvm->lock);
1584                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1585                 vcpu->pio.guest_pages[i] = page;
1586                 mutex_unlock(&vcpu->kvm->lock);
1587                 if (!page) {
1588                         inject_gp(vcpu);
1589                         free_pio_guest_pages(vcpu);
1590                         return 1;
1591                 }
1592         }
1593
1594         pio_dev = vcpu_find_pio_dev(vcpu, port);
1595         if (!vcpu->pio.in) {
1596                 /* string PIO write */
1597                 ret = pio_copy_data(vcpu);
1598                 if (ret >= 0 && pio_dev) {
1599                         pio_string_write(pio_dev, vcpu);
1600                         complete_pio(vcpu);
1601                         if (vcpu->pio.count == 0)
1602                                 ret = 1;
1603                 }
1604         } else if (pio_dev)
1605                 pr_unimpl(vcpu, "no string pio read support yet, "
1606                        "port %x size %d count %ld\n",
1607                         port, size, count);
1608
1609         return ret;
1610 }
1611 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1612
1613 __init void kvm_arch_init(void)
1614 {
1615         kvm_init_msr_list();
1616 }
1617
1618 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1619 {
1620         ++vcpu->stat.halt_exits;
1621         if (irqchip_in_kernel(vcpu->kvm)) {
1622                 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1623                 kvm_vcpu_block(vcpu);
1624                 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1625                         return -EINTR;
1626                 return 1;
1627         } else {
1628                 vcpu->run->exit_reason = KVM_EXIT_HLT;
1629                 return 0;
1630         }
1631 }
1632 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1633
1634 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
1635 {
1636         unsigned long nr, a0, a1, a2, a3, ret;
1637
1638         kvm_x86_ops->cache_regs(vcpu);
1639
1640         nr = vcpu->regs[VCPU_REGS_RAX];
1641         a0 = vcpu->regs[VCPU_REGS_RBX];
1642         a1 = vcpu->regs[VCPU_REGS_RCX];
1643         a2 = vcpu->regs[VCPU_REGS_RDX];
1644         a3 = vcpu->regs[VCPU_REGS_RSI];
1645
1646         if (!is_long_mode(vcpu)) {
1647                 nr &= 0xFFFFFFFF;
1648                 a0 &= 0xFFFFFFFF;
1649                 a1 &= 0xFFFFFFFF;
1650                 a2 &= 0xFFFFFFFF;
1651                 a3 &= 0xFFFFFFFF;
1652         }
1653
1654         switch (nr) {
1655         default:
1656                 ret = -KVM_ENOSYS;
1657                 break;
1658         }
1659         vcpu->regs[VCPU_REGS_RAX] = ret;
1660         kvm_x86_ops->decache_regs(vcpu);
1661         return 0;
1662 }
1663 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1664
1665 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1666 {
1667         char instruction[3];
1668         int ret = 0;
1669
1670         mutex_lock(&vcpu->kvm->lock);
1671
1672         /*
1673          * Blow out the MMU to ensure that no other VCPU has an active mapping
1674          * to ensure that the updated hypercall appears atomically across all
1675          * VCPUs.
1676          */
1677         kvm_mmu_zap_all(vcpu->kvm);
1678
1679         kvm_x86_ops->cache_regs(vcpu);
1680         kvm_x86_ops->patch_hypercall(vcpu, instruction);
1681         if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1682             != X86EMUL_CONTINUE)
1683                 ret = -EFAULT;
1684
1685         mutex_unlock(&vcpu->kvm->lock);
1686
1687         return ret;
1688 }
1689
1690 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1691 {
1692         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1693 }
1694
1695 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1696 {
1697         struct descriptor_table dt = { limit, base };
1698
1699         kvm_x86_ops->set_gdt(vcpu, &dt);
1700 }
1701
1702 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1703 {
1704         struct descriptor_table dt = { limit, base };
1705
1706         kvm_x86_ops->set_idt(vcpu, &dt);
1707 }
1708
1709 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1710                    unsigned long *rflags)
1711 {
1712         lmsw(vcpu, msw);
1713         *rflags = kvm_x86_ops->get_rflags(vcpu);
1714 }
1715
1716 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1717 {
1718         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1719         switch (cr) {
1720         case 0:
1721                 return vcpu->cr0;
1722         case 2:
1723                 return vcpu->cr2;
1724         case 3:
1725                 return vcpu->cr3;
1726         case 4:
1727                 return vcpu->cr4;
1728         default:
1729                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1730                 return 0;
1731         }
1732 }
1733
1734 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1735                      unsigned long *rflags)
1736 {
1737         switch (cr) {
1738         case 0:
1739                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1740                 *rflags = kvm_x86_ops->get_rflags(vcpu);
1741                 break;
1742         case 2:
1743                 vcpu->cr2 = val;
1744                 break;
1745         case 3:
1746                 set_cr3(vcpu, val);
1747                 break;
1748         case 4:
1749                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1750                 break;
1751         default:
1752                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1753         }
1754 }
1755
1756 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1757 {
1758         int i;
1759         u32 function;
1760         struct kvm_cpuid_entry *e, *best;
1761
1762         kvm_x86_ops->cache_regs(vcpu);
1763         function = vcpu->regs[VCPU_REGS_RAX];
1764         vcpu->regs[VCPU_REGS_RAX] = 0;
1765         vcpu->regs[VCPU_REGS_RBX] = 0;
1766         vcpu->regs[VCPU_REGS_RCX] = 0;
1767         vcpu->regs[VCPU_REGS_RDX] = 0;
1768         best = NULL;
1769         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1770                 e = &vcpu->cpuid_entries[i];
1771                 if (e->function == function) {
1772                         best = e;
1773                         break;
1774                 }
1775                 /*
1776                  * Both basic or both extended?
1777                  */
1778                 if (((e->function ^ function) & 0x80000000) == 0)
1779                         if (!best || e->function > best->function)
1780                                 best = e;
1781         }
1782         if (best) {
1783                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1784                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1785                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1786                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1787         }
1788         kvm_x86_ops->decache_regs(vcpu);
1789         kvm_x86_ops->skip_emulated_instruction(vcpu);
1790 }
1791 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1792
1793 /*
1794  * Check if userspace requested an interrupt window, and that the
1795  * interrupt window is open.
1796  *
1797  * No need to exit to userspace if we already have an interrupt queued.
1798  */
1799 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1800                                           struct kvm_run *kvm_run)
1801 {
1802         return (!vcpu->irq_summary &&
1803                 kvm_run->request_interrupt_window &&
1804                 vcpu->interrupt_window_open &&
1805                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
1806 }
1807
1808 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1809                               struct kvm_run *kvm_run)
1810 {
1811         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
1812         kvm_run->cr8 = get_cr8(vcpu);
1813         kvm_run->apic_base = kvm_get_apic_base(vcpu);
1814         if (irqchip_in_kernel(vcpu->kvm))
1815                 kvm_run->ready_for_interrupt_injection = 1;
1816         else
1817                 kvm_run->ready_for_interrupt_injection =
1818                                         (vcpu->interrupt_window_open &&
1819                                          vcpu->irq_summary == 0);
1820 }
1821
1822 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1823 {
1824         int r;
1825
1826         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
1827                 pr_debug("vcpu %d received sipi with vector # %x\n",
1828                        vcpu->vcpu_id, vcpu->sipi_vector);
1829                 kvm_lapic_reset(vcpu);
1830                 r = kvm_x86_ops->vcpu_reset(vcpu);
1831                 if (r)
1832                         return r;
1833                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
1834         }
1835
1836 preempted:
1837         if (vcpu->guest_debug.enabled)
1838                 kvm_x86_ops->guest_debug_pre(vcpu);
1839
1840 again:
1841         r = kvm_mmu_reload(vcpu);
1842         if (unlikely(r))
1843                 goto out;
1844
1845         kvm_inject_pending_timer_irqs(vcpu);
1846
1847         preempt_disable();
1848
1849         kvm_x86_ops->prepare_guest_switch(vcpu);
1850         kvm_load_guest_fpu(vcpu);
1851
1852         local_irq_disable();
1853
1854         if (signal_pending(current)) {
1855                 local_irq_enable();
1856                 preempt_enable();
1857                 r = -EINTR;
1858                 kvm_run->exit_reason = KVM_EXIT_INTR;
1859                 ++vcpu->stat.signal_exits;
1860                 goto out;
1861         }
1862
1863         if (irqchip_in_kernel(vcpu->kvm))
1864                 kvm_x86_ops->inject_pending_irq(vcpu);
1865         else if (!vcpu->mmio_read_completed)
1866                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
1867
1868         vcpu->guest_mode = 1;
1869         kvm_guest_enter();
1870
1871         if (vcpu->requests)
1872                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
1873                         kvm_x86_ops->tlb_flush(vcpu);
1874
1875         kvm_x86_ops->run(vcpu, kvm_run);
1876
1877         vcpu->guest_mode = 0;
1878         local_irq_enable();
1879
1880         ++vcpu->stat.exits;
1881
1882         /*
1883          * We must have an instruction between local_irq_enable() and
1884          * kvm_guest_exit(), so the timer interrupt isn't delayed by
1885          * the interrupt shadow.  The stat.exits increment will do nicely.
1886          * But we need to prevent reordering, hence this barrier():
1887          */
1888         barrier();
1889
1890         kvm_guest_exit();
1891
1892         preempt_enable();
1893
1894         /*
1895          * Profile KVM exit RIPs:
1896          */
1897         if (unlikely(prof_on == KVM_PROFILING)) {
1898                 kvm_x86_ops->cache_regs(vcpu);
1899                 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
1900         }
1901
1902         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
1903
1904         if (r > 0) {
1905                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
1906                         r = -EINTR;
1907                         kvm_run->exit_reason = KVM_EXIT_INTR;
1908                         ++vcpu->stat.request_irq_exits;
1909                         goto out;
1910                 }
1911                 if (!need_resched()) {
1912                         ++vcpu->stat.light_exits;
1913                         goto again;
1914                 }
1915         }
1916
1917 out:
1918         if (r > 0) {
1919                 kvm_resched(vcpu);
1920                 goto preempted;
1921         }
1922
1923         post_kvm_run_save(vcpu, kvm_run);
1924
1925         return r;
1926 }
1927
1928 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1929 {
1930         int r;
1931         sigset_t sigsaved;
1932
1933         vcpu_load(vcpu);
1934
1935         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
1936                 kvm_vcpu_block(vcpu);
1937                 vcpu_put(vcpu);
1938                 return -EAGAIN;
1939         }
1940
1941         if (vcpu->sigset_active)
1942                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1943
1944         /* re-sync apic's tpr */
1945         if (!irqchip_in_kernel(vcpu->kvm))
1946                 set_cr8(vcpu, kvm_run->cr8);
1947
1948         if (vcpu->pio.cur_count) {
1949                 r = complete_pio(vcpu);
1950                 if (r)
1951                         goto out;
1952         }
1953 #if CONFIG_HAS_IOMEM
1954         if (vcpu->mmio_needed) {
1955                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1956                 vcpu->mmio_read_completed = 1;
1957                 vcpu->mmio_needed = 0;
1958                 r = emulate_instruction(vcpu, kvm_run,
1959                                         vcpu->mmio_fault_cr2, 0, 1);
1960                 if (r == EMULATE_DO_MMIO) {
1961                         /*
1962                          * Read-modify-write.  Back to userspace.
1963                          */
1964                         r = 0;
1965                         goto out;
1966                 }
1967         }
1968 #endif
1969         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1970                 kvm_x86_ops->cache_regs(vcpu);
1971                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1972                 kvm_x86_ops->decache_regs(vcpu);
1973         }
1974
1975         r = __vcpu_run(vcpu, kvm_run);
1976
1977 out:
1978         if (vcpu->sigset_active)
1979                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1980
1981         vcpu_put(vcpu);
1982         return r;
1983 }
1984
1985 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1986 {
1987         vcpu_load(vcpu);
1988
1989         kvm_x86_ops->cache_regs(vcpu);
1990
1991         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1992         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1993         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1994         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1995         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1996         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1997         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1998         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1999 #ifdef CONFIG_X86_64
2000         regs->r8 = vcpu->regs[VCPU_REGS_R8];
2001         regs->r9 = vcpu->regs[VCPU_REGS_R9];
2002         regs->r10 = vcpu->regs[VCPU_REGS_R10];
2003         regs->r11 = vcpu->regs[VCPU_REGS_R11];
2004         regs->r12 = vcpu->regs[VCPU_REGS_R12];
2005         regs->r13 = vcpu->regs[VCPU_REGS_R13];
2006         regs->r14 = vcpu->regs[VCPU_REGS_R14];
2007         regs->r15 = vcpu->regs[VCPU_REGS_R15];
2008 #endif
2009
2010         regs->rip = vcpu->rip;
2011         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2012
2013         /*
2014          * Don't leak debug flags in case they were set for guest debugging
2015          */
2016         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2017                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2018
2019         vcpu_put(vcpu);
2020
2021         return 0;
2022 }
2023
2024 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2025 {
2026         vcpu_load(vcpu);
2027
2028         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2029         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2030         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2031         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2032         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2033         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2034         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2035         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2036 #ifdef CONFIG_X86_64
2037         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2038         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2039         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2040         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2041         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2042         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2043         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2044         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2045 #endif
2046
2047         vcpu->rip = regs->rip;
2048         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2049
2050         kvm_x86_ops->decache_regs(vcpu);
2051
2052         vcpu_put(vcpu);
2053
2054         return 0;
2055 }
2056
2057 static void get_segment(struct kvm_vcpu *vcpu,
2058                         struct kvm_segment *var, int seg)
2059 {
2060         return kvm_x86_ops->get_segment(vcpu, var, seg);
2061 }
2062
2063 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2064 {
2065         struct kvm_segment cs;
2066
2067         get_segment(vcpu, &cs, VCPU_SREG_CS);
2068         *db = cs.db;
2069         *l = cs.l;
2070 }
2071 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2072
2073 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2074                                   struct kvm_sregs *sregs)
2075 {
2076         struct descriptor_table dt;
2077         int pending_vec;
2078
2079         vcpu_load(vcpu);
2080
2081         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2082         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2083         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2084         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2085         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2086         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2087
2088         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2089         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2090
2091         kvm_x86_ops->get_idt(vcpu, &dt);
2092         sregs->idt.limit = dt.limit;
2093         sregs->idt.base = dt.base;
2094         kvm_x86_ops->get_gdt(vcpu, &dt);
2095         sregs->gdt.limit = dt.limit;
2096         sregs->gdt.base = dt.base;
2097
2098         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2099         sregs->cr0 = vcpu->cr0;
2100         sregs->cr2 = vcpu->cr2;
2101         sregs->cr3 = vcpu->cr3;
2102         sregs->cr4 = vcpu->cr4;
2103         sregs->cr8 = get_cr8(vcpu);
2104         sregs->efer = vcpu->shadow_efer;
2105         sregs->apic_base = kvm_get_apic_base(vcpu);
2106
2107         if (irqchip_in_kernel(vcpu->kvm)) {
2108                 memset(sregs->interrupt_bitmap, 0,
2109                        sizeof sregs->interrupt_bitmap);
2110                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2111                 if (pending_vec >= 0)
2112                         set_bit(pending_vec,
2113                                 (unsigned long *)sregs->interrupt_bitmap);
2114         } else
2115                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2116                        sizeof sregs->interrupt_bitmap);
2117
2118         vcpu_put(vcpu);
2119
2120         return 0;
2121 }
2122
2123 static void set_segment(struct kvm_vcpu *vcpu,
2124                         struct kvm_segment *var, int seg)
2125 {
2126         return kvm_x86_ops->set_segment(vcpu, var, seg);
2127 }
2128
2129 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2130                                   struct kvm_sregs *sregs)
2131 {
2132         int mmu_reset_needed = 0;
2133         int i, pending_vec, max_bits;
2134         struct descriptor_table dt;
2135
2136         vcpu_load(vcpu);
2137
2138         dt.limit = sregs->idt.limit;
2139         dt.base = sregs->idt.base;
2140         kvm_x86_ops->set_idt(vcpu, &dt);
2141         dt.limit = sregs->gdt.limit;
2142         dt.base = sregs->gdt.base;
2143         kvm_x86_ops->set_gdt(vcpu, &dt);
2144
2145         vcpu->cr2 = sregs->cr2;
2146         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2147         vcpu->cr3 = sregs->cr3;
2148
2149         set_cr8(vcpu, sregs->cr8);
2150
2151         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2152 #ifdef CONFIG_X86_64
2153         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2154 #endif
2155         kvm_set_apic_base(vcpu, sregs->apic_base);
2156
2157         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2158
2159         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2160         vcpu->cr0 = sregs->cr0;
2161         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2162
2163         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2164         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2165         if (!is_long_mode(vcpu) && is_pae(vcpu))
2166                 load_pdptrs(vcpu, vcpu->cr3);
2167
2168         if (mmu_reset_needed)
2169                 kvm_mmu_reset_context(vcpu);
2170
2171         if (!irqchip_in_kernel(vcpu->kvm)) {
2172                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2173                        sizeof vcpu->irq_pending);
2174                 vcpu->irq_summary = 0;
2175                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2176                         if (vcpu->irq_pending[i])
2177                                 __set_bit(i, &vcpu->irq_summary);
2178         } else {
2179                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2180                 pending_vec = find_first_bit(
2181                         (const unsigned long *)sregs->interrupt_bitmap,
2182                         max_bits);
2183                 /* Only pending external irq is handled here */
2184                 if (pending_vec < max_bits) {
2185                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2186                         pr_debug("Set back pending irq %d\n",
2187                                  pending_vec);
2188                 }
2189         }
2190
2191         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2192         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2193         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2194         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2195         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2196         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2197
2198         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2199         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2200
2201         vcpu_put(vcpu);
2202
2203         return 0;
2204 }
2205
2206 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2207                                     struct kvm_debug_guest *dbg)
2208 {
2209         int r;
2210
2211         vcpu_load(vcpu);
2212
2213         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2214
2215         vcpu_put(vcpu);
2216
2217         return r;
2218 }
2219
2220 /*
2221  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2222  * we have asm/x86/processor.h
2223  */
2224 struct fxsave {
2225         u16     cwd;
2226         u16     swd;
2227         u16     twd;
2228         u16     fop;
2229         u64     rip;
2230         u64     rdp;
2231         u32     mxcsr;
2232         u32     mxcsr_mask;
2233         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2234 #ifdef CONFIG_X86_64
2235         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2236 #else
2237         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2238 #endif
2239 };
2240
2241 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2242 {
2243         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2244
2245         vcpu_load(vcpu);
2246
2247         memcpy(fpu->fpr, fxsave->st_space, 128);
2248         fpu->fcw = fxsave->cwd;
2249         fpu->fsw = fxsave->swd;
2250         fpu->ftwx = fxsave->twd;
2251         fpu->last_opcode = fxsave->fop;
2252         fpu->last_ip = fxsave->rip;
2253         fpu->last_dp = fxsave->rdp;
2254         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2255
2256         vcpu_put(vcpu);
2257
2258         return 0;
2259 }
2260
2261 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2262 {
2263         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2264
2265         vcpu_load(vcpu);
2266
2267         memcpy(fxsave->st_space, fpu->fpr, 128);
2268         fxsave->cwd = fpu->fcw;
2269         fxsave->swd = fpu->fsw;
2270         fxsave->twd = fpu->ftwx;
2271         fxsave->fop = fpu->last_opcode;
2272         fxsave->rip = fpu->last_ip;
2273         fxsave->rdp = fpu->last_dp;
2274         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2275
2276         vcpu_put(vcpu);
2277
2278         return 0;
2279 }
2280
2281 void fx_init(struct kvm_vcpu *vcpu)
2282 {
2283         unsigned after_mxcsr_mask;
2284
2285         /* Initialize guest FPU by resetting ours and saving into guest's */
2286         preempt_disable();
2287         fx_save(&vcpu->host_fx_image);
2288         fpu_init();
2289         fx_save(&vcpu->guest_fx_image);
2290         fx_restore(&vcpu->host_fx_image);
2291         preempt_enable();
2292
2293         vcpu->cr0 |= X86_CR0_ET;
2294         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
2295         vcpu->guest_fx_image.mxcsr = 0x1f80;
2296         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
2297                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
2298 }
2299 EXPORT_SYMBOL_GPL(fx_init);
2300
2301 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
2302 {
2303         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
2304                 return;
2305
2306         vcpu->guest_fpu_loaded = 1;
2307         fx_save(&vcpu->host_fx_image);
2308         fx_restore(&vcpu->guest_fx_image);
2309 }
2310 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
2311
2312 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
2313 {
2314         if (!vcpu->guest_fpu_loaded)
2315                 return;
2316
2317         vcpu->guest_fpu_loaded = 0;
2318         fx_save(&vcpu->guest_fx_image);
2319         fx_restore(&vcpu->host_fx_image);
2320 }
2321 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);