err.no Git - linux-2.6/blob - 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 #include <linux/mman.h>
  28
  29 #include <asm/uaccess.h>
  30 #include <asm/msr.h>
  31
  32 #define MAX_IO_MSRS 256
  33 #define CR0_RESERVED_BITS                                               \
  34         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  35                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  36                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  37 #define CR4_RESERVED_BITS                                               \
  38         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  39                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  40                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  41                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  42
  43 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  44 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  45
  46 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  47 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  48
  49 struct kvm_x86_ops *kvm_x86_ops;
  50
  51 struct kvm_stats_debugfs_item debugfs_entries[] = {
  52         { "pf_fixed", VCPU_STAT(pf_fixed) },
  53         { "pf_guest", VCPU_STAT(pf_guest) },
  54         { "tlb_flush", VCPU_STAT(tlb_flush) },
  55         { "invlpg", VCPU_STAT(invlpg) },
  56         { "exits", VCPU_STAT(exits) },
  57         { "io_exits", VCPU_STAT(io_exits) },
  58         { "mmio_exits", VCPU_STAT(mmio_exits) },
  59         { "signal_exits", VCPU_STAT(signal_exits) },
  60         { "irq_window", VCPU_STAT(irq_window_exits) },
  61         { "halt_exits", VCPU_STAT(halt_exits) },
  62         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  63         { "request_irq", VCPU_STAT(request_irq_exits) },
  64         { "irq_exits", VCPU_STAT(irq_exits) },
  65         { "host_state_reload", VCPU_STAT(host_state_reload) },
  66         { "efer_reload", VCPU_STAT(efer_reload) },
  67         { "fpu_reload", VCPU_STAT(fpu_reload) },
  68         { "insn_emulation", VCPU_STAT(insn_emulation) },
  69         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
  70         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
  71         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
  72         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
  73         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
  74         { "mmu_flooded", VM_STAT(mmu_flooded) },
  75         { "mmu_recycled", VM_STAT(mmu_recycled) },
  76         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
  77         { NULL }
  78 };
  79
  80
  81 unsigned long segment_base(u16 selector)
  82 {
  83         struct descriptor_table gdt;
  84         struct segment_descriptor *d;
  85         unsigned long table_base;
  86         unsigned long v;
  87
  88         if (selector == 0)
  89                 return 0;
  90
  91         asm("sgdt %0" : "=m"(gdt));
  92         table_base = gdt.base;
  93
  94         if (selector & 4) {           /* from ldt */
  95                 u16 ldt_selector;
  96
  97                 asm("sldt %0" : "=g"(ldt_selector));
  98                 table_base = segment_base(ldt_selector);
  99         }
 100         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 101         v = d->base_low | ((unsigned long)d->base_mid << 16) |
 102                 ((unsigned long)d->base_high << 24);
 103 #ifdef CONFIG_X86_64
 104         if (d->system == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 105                 v |= ((unsigned long) \
 106                       ((struct segment_descriptor_64 *)d)->base_higher) << 32;
 107 #endif
 108         return v;
 109 }
 110 EXPORT_SYMBOL_GPL(segment_base);
 111
 112 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 113 {
 114         if (irqchip_in_kernel(vcpu->kvm))
 115                 return vcpu->apic_base;
 116         else
 117                 return vcpu->apic_base;
 118 }
 119 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 120
 121 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 122 {
 123         /* TODO: reserve bits check */
 124         if (irqchip_in_kernel(vcpu->kvm))
 125                 kvm_lapic_set_base(vcpu, data);
 126         else
 127                 vcpu->apic_base = data;
 128 }
 129 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 130
 131 static void inject_gp(struct kvm_vcpu *vcpu)
 132 {
 133         kvm_x86_ops->inject_gp(vcpu, 0);
 134 }
 135
 136 /*
 137  * Load the pae pdptrs.  Return true is they are all valid.
 138  */
 139 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 140 {
 141         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 142         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 143         int i;
 144         int ret;
 145         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
 146
 147         mutex_lock(&vcpu->kvm->lock);
 148         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 149                                   offset * sizeof(u64), sizeof(pdpte));
 150         if (ret < 0) {
 151                 ret = 0;
 152                 goto out;
 153         }
 154         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 155                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 156                         ret = 0;
 157                         goto out;
 158                 }
 159         }
 160         ret = 1;
 161
 162         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
 163 out:
 164         mutex_unlock(&vcpu->kvm->lock);
 165
 166         return ret;
 167 }
 168
 169 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 170 {
 171         if (cr0 & CR0_RESERVED_BITS) {
 172                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 173                        cr0, vcpu->cr0);
 174                 inject_gp(vcpu);
 175                 return;
 176         }
 177
 178         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 179                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 180                 inject_gp(vcpu);
 181                 return;
 182         }
 183
 184         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 185                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 186                        "and a clear PE flag\n");
 187                 inject_gp(vcpu);
 188                 return;
 189         }
 190
 191         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 192 #ifdef CONFIG_X86_64
 193                 if ((vcpu->shadow_efer & EFER_LME)) {
 194                         int cs_db, cs_l;
 195
 196                         if (!is_pae(vcpu)) {
 197                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 198                                        "in long mode while PAE is disabled\n");
 199                                 inject_gp(vcpu);
 200                                 return;
 201                         }
 202                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 203                         if (cs_l) {
 204                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 205                                        "in long mode while CS.L == 1\n");
 206                                 inject_gp(vcpu);
 207                                 return;
 208
 209                         }
 210                 } else
 211 #endif
 212                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 213                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 214                                "reserved bits\n");
 215                         inject_gp(vcpu);
 216                         return;
 217                 }
 218
 219         }
 220
 221         kvm_x86_ops->set_cr0(vcpu, cr0);
 222         vcpu->cr0 = cr0;
 223
 224         mutex_lock(&vcpu->kvm->lock);
 225         kvm_mmu_reset_context(vcpu);
 226         mutex_unlock(&vcpu->kvm->lock);
 227         return;
 228 }
 229 EXPORT_SYMBOL_GPL(set_cr0);
 230
 231 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 232 {
 233         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 234 }
 235 EXPORT_SYMBOL_GPL(lmsw);
 236
 237 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 238 {
 239         if (cr4 & CR4_RESERVED_BITS) {
 240                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 241                 inject_gp(vcpu);
 242                 return;
 243         }
 244
 245         if (is_long_mode(vcpu)) {
 246                 if (!(cr4 & X86_CR4_PAE)) {
 247                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 248                                "in long mode\n");
 249                         inject_gp(vcpu);
 250                         return;
 251                 }
 252         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 253                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 254                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 255                 inject_gp(vcpu);
 256                 return;
 257         }
 258
 259         if (cr4 & X86_CR4_VMXE) {
 260                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 261                 inject_gp(vcpu);
 262                 return;
 263         }
 264         kvm_x86_ops->set_cr4(vcpu, cr4);
 265         vcpu->cr4 = cr4;
 266         mutex_lock(&vcpu->kvm->lock);
 267         kvm_mmu_reset_context(vcpu);
 268         mutex_unlock(&vcpu->kvm->lock);
 269 }
 270 EXPORT_SYMBOL_GPL(set_cr4);
 271
 272 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 273 {
 274         if (is_long_mode(vcpu)) {
 275                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 276                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 277                         inject_gp(vcpu);
 278                         return;
 279                 }
 280         } else {
 281                 if (is_pae(vcpu)) {
 282                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 283                                 printk(KERN_DEBUG
 284                                        "set_cr3: #GP, reserved bits\n");
 285                                 inject_gp(vcpu);
 286                                 return;
 287                         }
 288                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 289                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 290                                        "reserved bits\n");
 291                                 inject_gp(vcpu);
 292                                 return;
 293                         }
 294                 }
 295                 /*
 296                  * We don't check reserved bits in nonpae mode, because
 297                  * this isn't enforced, and VMware depends on this.
 298                  */
 299         }
 300
 301         mutex_lock(&vcpu->kvm->lock);
 302         /*
 303          * Does the new cr3 value map to physical memory? (Note, we
 304          * catch an invalid cr3 even in real-mode, because it would
 305          * cause trouble later on when we turn on paging anyway.)
 306          *
 307          * A real CPU would silently accept an invalid cr3 and would
 308          * attempt to use it - with largely undefined (and often hard
 309          * to debug) behavior on the guest side.
 310          */
 311         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 312                 inject_gp(vcpu);
 313         else {
 314                 vcpu->cr3 = cr3;
 315                 vcpu->mmu.new_cr3(vcpu);
 316         }
 317         mutex_unlock(&vcpu->kvm->lock);
 318 }
 319 EXPORT_SYMBOL_GPL(set_cr3);
 320
 321 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 322 {
 323         if (cr8 & CR8_RESERVED_BITS) {
 324                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 325                 inject_gp(vcpu);
 326                 return;
 327         }
 328         if (irqchip_in_kernel(vcpu->kvm))
 329                 kvm_lapic_set_tpr(vcpu, cr8);
 330         else
 331                 vcpu->cr8 = cr8;
 332 }
 333 EXPORT_SYMBOL_GPL(set_cr8);
 334
 335 unsigned long get_cr8(struct kvm_vcpu *vcpu)
 336 {
 337         if (irqchip_in_kernel(vcpu->kvm))
 338                 return kvm_lapic_get_cr8(vcpu);
 339         else
 340                 return vcpu->cr8;
 341 }
 342 EXPORT_SYMBOL_GPL(get_cr8);
 343
 344 /*
 345  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 346  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 347  *
 348  * This list is modified at module load time to reflect the
 349  * capabilities of the host cpu.
 350  */
 351 static u32 msrs_to_save[] = {
 352         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 353         MSR_K6_STAR,
 354 #ifdef CONFIG_X86_64
 355         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 356 #endif
 357         MSR_IA32_TIME_STAMP_COUNTER,
 358 };
 359
 360 static unsigned num_msrs_to_save;
 361
 362 static u32 emulated_msrs[] = {
 363         MSR_IA32_MISC_ENABLE,
 364 };
 365
 366 #ifdef CONFIG_X86_64
 367
 368 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 369 {
 370         if (efer & EFER_RESERVED_BITS) {
 371                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 372                        efer);
 373                 inject_gp(vcpu);
 374                 return;
 375         }
 376
 377         if (is_paging(vcpu)
 378             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 379                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 380                 inject_gp(vcpu);
 381                 return;
 382         }
 383
 384         kvm_x86_ops->set_efer(vcpu, efer);
 385
 386         efer &= ~EFER_LMA;
 387         efer |= vcpu->shadow_efer & EFER_LMA;
 388
 389         vcpu->shadow_efer = efer;
 390 }
 391
 392 #endif
 393
 394 /*
 395  * Writes msr value into into the appropriate "register".
 396  * Returns 0 on success, non-0 otherwise.
 397  * Assumes vcpu_load() was already called.
 398  */
 399 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 400 {
 401         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 402 }
 403
 404 /*
 405  * Adapt set_msr() to msr_io()'s calling convention
 406  */
 407 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 408 {
 409         return kvm_set_msr(vcpu, index, *data);
 410 }
 411
 412
 413 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 414 {
 415         switch (msr) {
 416 #ifdef CONFIG_X86_64
 417         case MSR_EFER:
 418                 set_efer(vcpu, data);
 419                 break;
 420 #endif
 421         case MSR_IA32_MC0_STATUS:
 422                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 423                        __FUNCTION__, data);
 424                 break;
 425         case MSR_IA32_MCG_STATUS:
 426                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 427                         __FUNCTION__, data);
 428                 break;
 429         case MSR_IA32_UCODE_REV:
 430         case MSR_IA32_UCODE_WRITE:
 431         case 0x200 ... 0x2ff: /* MTRRs */
 432                 break;
 433         case MSR_IA32_APICBASE:
 434                 kvm_set_apic_base(vcpu, data);
 435                 break;
 436         case MSR_IA32_MISC_ENABLE:
 437                 vcpu->ia32_misc_enable_msr = data;
 438                 break;
 439         default:
 440                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
 441                 return 1;
 442         }
 443         return 0;
 444 }
 445 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 446
 447
 448 /*
 449  * Reads an msr value (of 'msr_index') into 'pdata'.
 450  * Returns 0 on success, non-0 otherwise.
 451  * Assumes vcpu_load() was already called.
 452  */
 453 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 454 {
 455         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 456 }
 457
 458 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 459 {
 460         u64 data;
 461
 462         switch (msr) {
 463         case 0xc0010010: /* SYSCFG */
 464         case 0xc0010015: /* HWCR */
 465         case MSR_IA32_PLATFORM_ID:
 466         case MSR_IA32_P5_MC_ADDR:
 467         case MSR_IA32_P5_MC_TYPE:
 468         case MSR_IA32_MC0_CTL:
 469         case MSR_IA32_MCG_STATUS:
 470         case MSR_IA32_MCG_CAP:
 471         case MSR_IA32_MC0_MISC:
 472         case MSR_IA32_MC0_MISC+4:
 473         case MSR_IA32_MC0_MISC+8:
 474         case MSR_IA32_MC0_MISC+12:
 475         case MSR_IA32_MC0_MISC+16:
 476         case MSR_IA32_UCODE_REV:
 477         case MSR_IA32_PERF_STATUS:
 478         case MSR_IA32_EBL_CR_POWERON:
 479                 /* MTRR registers */
 480         case 0xfe:
 481         case 0x200 ... 0x2ff:
 482                 data = 0;
 483                 break;
 484         case 0xcd: /* fsb frequency */
 485                 data = 3;
 486                 break;
 487         case MSR_IA32_APICBASE:
 488                 data = kvm_get_apic_base(vcpu);
 489                 break;
 490         case MSR_IA32_MISC_ENABLE:
 491                 data = vcpu->ia32_misc_enable_msr;
 492                 break;
 493 #ifdef CONFIG_X86_64
 494         case MSR_EFER:
 495                 data = vcpu->shadow_efer;
 496                 break;
 497 #endif
 498         default:
 499                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 500                 return 1;
 501         }
 502         *pdata = data;
 503         return 0;
 504 }
 505 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 506
 507 /*
 508  * Read or write a bunch of msrs. All parameters are kernel addresses.
 509  *
 510  * @return number of msrs set successfully.
 511  */
 512 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 513                     struct kvm_msr_entry *entries,
 514                     int (*do_msr)(struct kvm_vcpu *vcpu,
 515                                   unsigned index, u64 *data))
 516 {
 517         int i;
 518
 519         vcpu_load(vcpu);
 520
 521         for (i = 0; i < msrs->nmsrs; ++i)
 522                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 523                         break;
 524
 525         vcpu_put(vcpu);
 526
 527         return i;
 528 }
 529
 530 /*
 531  * Read or write a bunch of msrs. Parameters are user addresses.
 532  *
 533  * @return number of msrs set successfully.
 534  */
 535 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 536                   int (*do_msr)(struct kvm_vcpu *vcpu,
 537                                 unsigned index, u64 *data),
 538                   int writeback)
 539 {
 540         struct kvm_msrs msrs;
 541         struct kvm_msr_entry *entries;
 542         int r, n;
 543         unsigned size;
 544
 545         r = -EFAULT;
 546         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
 547                 goto out;
 548
 549         r = -E2BIG;
 550         if (msrs.nmsrs >= MAX_IO_MSRS)
 551                 goto out;
 552
 553         r = -ENOMEM;
 554         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
 555         entries = vmalloc(size);
 556         if (!entries)
 557                 goto out;
 558
 559         r = -EFAULT;
 560         if (copy_from_user(entries, user_msrs->entries, size))
 561                 goto out_free;
 562
 563         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
 564         if (r < 0)
 565                 goto out_free;
 566
 567         r = -EFAULT;
 568         if (writeback && copy_to_user(user_msrs->entries, entries, size))
 569                 goto out_free;
 570
 571         r = n;
 572
 573 out_free:
 574         vfree(entries);
 575 out:
 576         return r;
 577 }
 578
 579 /*
 580  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
 581  * cached on it.
 582  */
 583 void decache_vcpus_on_cpu(int cpu)
 584 {
 585         struct kvm *vm;
 586         struct kvm_vcpu *vcpu;
 587         int i;
 588
 589         spin_lock(&kvm_lock);
 590         list_for_each_entry(vm, &vm_list, vm_list)
 591                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 592                         vcpu = vm->vcpus[i];
 593                         if (!vcpu)
 594                                 continue;
 595                         /*
 596                          * If the vcpu is locked, then it is running on some
 597                          * other cpu and therefore it is not cached on the
 598                          * cpu in question.
 599                          *
 600                          * If it's not locked, check the last cpu it executed
 601                          * on.
 602                          */
 603                         if (mutex_trylock(&vcpu->mutex)) {
 604                                 if (vcpu->cpu == cpu) {
 605                                         kvm_x86_ops->vcpu_decache(vcpu);
 606                                         vcpu->cpu = -1;
 607                                 }
 608                                 mutex_unlock(&vcpu->mutex);
 609                         }
 610                 }
 611         spin_unlock(&kvm_lock);
 612 }
 613
 614 int kvm_dev_ioctl_check_extension(long ext)
 615 {
 616         int r;
 617
 618         switch (ext) {
 619         case KVM_CAP_IRQCHIP:
 620         case KVM_CAP_HLT:
 621         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
 622         case KVM_CAP_USER_MEMORY:
 623         case KVM_CAP_SET_TSS_ADDR:
 624                 r = 1;
 625                 break;
 626         default:
 627                 r = 0;
 628                 break;
 629         }
 630         return r;
 631
 632 }
 633
 634 long kvm_arch_dev_ioctl(struct file *filp,
 635                         unsigned int ioctl, unsigned long arg)
 636 {
 637         void __user *argp = (void __user *)arg;
 638         long r;
 639
 640         switch (ioctl) {
 641         case KVM_GET_MSR_INDEX_LIST: {
 642                 struct kvm_msr_list __user *user_msr_list = argp;
 643                 struct kvm_msr_list msr_list;
 644                 unsigned n;
 645
 646                 r = -EFAULT;
 647                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
 648                         goto out;
 649                 n = msr_list.nmsrs;
 650                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
 651                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
 652                         goto out;
 653                 r = -E2BIG;
 654                 if (n < num_msrs_to_save)
 655                         goto out;
 656                 r = -EFAULT;
 657                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
 658                                  num_msrs_to_save * sizeof(u32)))
 659                         goto out;
 660                 if (copy_to_user(user_msr_list->indices
 661                                  + num_msrs_to_save * sizeof(u32),
 662                                  &emulated_msrs,
 663                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
 664                         goto out;
 665                 r = 0;
 666                 break;
 667         }
 668         default:
 669                 r = -EINVAL;
 670         }
 671 out:
 672         return r;
 673 }
 674
 675 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 676 {
 677         kvm_x86_ops->vcpu_load(vcpu, cpu);
 678 }
 679
 680 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 681 {
 682         kvm_x86_ops->vcpu_put(vcpu);
 683         kvm_put_guest_fpu(vcpu);
 684 }
 685
 686 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 687 {
 688         u64 efer;
 689         int i;
 690         struct kvm_cpuid_entry *e, *entry;
 691
 692         rdmsrl(MSR_EFER, efer);
 693         entry = NULL;
 694         for (i = 0; i < vcpu->cpuid_nent; ++i) {
 695                 e = &vcpu->cpuid_entries[i];
 696                 if (e->function == 0x80000001) {
 697                         entry = e;
 698                         break;
 699                 }
 700         }
 701         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
 702                 entry->edx &= ~(1 << 20);
 703                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 704         }
 705 }
 706
 707 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 708                                     struct kvm_cpuid *cpuid,
 709                                     struct kvm_cpuid_entry __user *entries)
 710 {
 711         int r;
 712
 713         r = -E2BIG;
 714         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 715                 goto out;
 716         r = -EFAULT;
 717         if (copy_from_user(&vcpu->cpuid_entries, entries,
 718                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 719                 goto out;
 720         vcpu->cpuid_nent = cpuid->nent;
 721         cpuid_fix_nx_cap(vcpu);
 722         return 0;
 723
 724 out:
 725         return r;
 726 }
 727
 728 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
 729                                     struct kvm_lapic_state *s)
 730 {
 731         vcpu_load(vcpu);
 732         memcpy(s->regs, vcpu->apic->regs, sizeof *s);
 733         vcpu_put(vcpu);
 734
 735         return 0;
 736 }
 737
 738 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
 739                                     struct kvm_lapic_state *s)
 740 {
 741         vcpu_load(vcpu);
 742         memcpy(vcpu->apic->regs, s->regs, sizeof *s);
 743         kvm_apic_post_state_restore(vcpu);
 744         vcpu_put(vcpu);
 745
 746         return 0;
 747 }
 748
 749 long kvm_arch_vcpu_ioctl(struct file *filp,
 750                          unsigned int ioctl, unsigned long arg)
 751 {
 752         struct kvm_vcpu *vcpu = filp->private_data;
 753         void __user *argp = (void __user *)arg;
 754         int r;
 755
 756         switch (ioctl) {
 757         case KVM_GET_LAPIC: {
 758                 struct kvm_lapic_state lapic;
 759
 760                 memset(&lapic, 0, sizeof lapic);
 761                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
 762                 if (r)
 763                         goto out;
 764                 r = -EFAULT;
 765                 if (copy_to_user(argp, &lapic, sizeof lapic))
 766                         goto out;
 767                 r = 0;
 768                 break;
 769         }
 770         case KVM_SET_LAPIC: {
 771                 struct kvm_lapic_state lapic;
 772
 773                 r = -EFAULT;
 774                 if (copy_from_user(&lapic, argp, sizeof lapic))
 775                         goto out;
 776                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
 777                 if (r)
 778                         goto out;
 779                 r = 0;
 780                 break;
 781         }
 782         case KVM_SET_CPUID: {
 783                 struct kvm_cpuid __user *cpuid_arg = argp;
 784                 struct kvm_cpuid cpuid;
 785
 786                 r = -EFAULT;
 787                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
 788                         goto out;
 789                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
 790                 if (r)
 791                         goto out;
 792                 break;
 793         }
 794         case KVM_GET_MSRS:
 795                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
 796                 break;
 797         case KVM_SET_MSRS:
 798                 r = msr_io(vcpu, argp, do_set_msr, 0);
 799                 break;
 800         default:
 801                 r = -EINVAL;
 802         }
 803 out:
 804         return r;
 805 }
 806
 807 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
 808 {
 809         int ret;
 810
 811         if (addr > (unsigned int)(-3 * PAGE_SIZE))
 812                 return -1;
 813         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
 814         return ret;
 815 }
 816
 817 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
 818                                           u32 kvm_nr_mmu_pages)
 819 {
 820         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
 821                 return -EINVAL;
 822
 823         mutex_lock(&kvm->lock);
 824
 825         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
 826         kvm->n_requested_mmu_pages = kvm_nr_mmu_pages;
 827
 828         mutex_unlock(&kvm->lock);
 829         return 0;
 830 }
 831
 832 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
 833 {
 834         return kvm->n_alloc_mmu_pages;
 835 }
 836
 837 /*
 838  * Set a new alias region.  Aliases map a portion of physical memory into
 839  * another portion.  This is useful for memory windows, for example the PC
 840  * VGA region.
 841  */
 842 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 843                                          struct kvm_memory_alias *alias)
 844 {
 845         int r, n;
 846         struct kvm_mem_alias *p;
 847
 848         r = -EINVAL;
 849         /* General sanity checks */
 850         if (alias->memory_size & (PAGE_SIZE - 1))
 851                 goto out;
 852         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 853                 goto out;
 854         if (alias->slot >= KVM_ALIAS_SLOTS)
 855                 goto out;
 856         if (alias->guest_phys_addr + alias->memory_size
 857             < alias->guest_phys_addr)
 858                 goto out;
 859         if (alias->target_phys_addr + alias->memory_size
 860             < alias->target_phys_addr)
 861                 goto out;
 862
 863         mutex_lock(&kvm->lock);
 864
 865         p = &kvm->aliases[alias->slot];
 866         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 867         p->npages = alias->memory_size >> PAGE_SHIFT;
 868         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 869
 870         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 871                 if (kvm->aliases[n - 1].npages)
 872                         break;
 873         kvm->naliases = n;
 874
 875         kvm_mmu_zap_all(kvm);
 876
 877         mutex_unlock(&kvm->lock);
 878
 879         return 0;
 880
 881 out:
 882         return r;
 883 }
 884
 885 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 886 {
 887         int r;
 888
 889         r = 0;
 890         switch (chip->chip_id) {
 891         case KVM_IRQCHIP_PIC_MASTER:
 892                 memcpy(&chip->chip.pic,
 893                         &pic_irqchip(kvm)->pics[0],
 894                         sizeof(struct kvm_pic_state));
 895                 break;
 896         case KVM_IRQCHIP_PIC_SLAVE:
 897                 memcpy(&chip->chip.pic,
 898                         &pic_irqchip(kvm)->pics[1],
 899                         sizeof(struct kvm_pic_state));
 900                 break;
 901         case KVM_IRQCHIP_IOAPIC:
 902                 memcpy(&chip->chip.ioapic,
 903                         ioapic_irqchip(kvm),
 904                         sizeof(struct kvm_ioapic_state));
 905                 break;
 906         default:
 907                 r = -EINVAL;
 908                 break;
 909         }
 910         return r;
 911 }
 912
 913 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 914 {
 915         int r;
 916
 917         r = 0;
 918         switch (chip->chip_id) {
 919         case KVM_IRQCHIP_PIC_MASTER:
 920                 memcpy(&pic_irqchip(kvm)->pics[0],
 921                         &chip->chip.pic,
 922                         sizeof(struct kvm_pic_state));
 923                 break;
 924         case KVM_IRQCHIP_PIC_SLAVE:
 925                 memcpy(&pic_irqchip(kvm)->pics[1],
 926                         &chip->chip.pic,
 927                         sizeof(struct kvm_pic_state));
 928                 break;
 929         case KVM_IRQCHIP_IOAPIC:
 930                 memcpy(ioapic_irqchip(kvm),
 931                         &chip->chip.ioapic,
 932                         sizeof(struct kvm_ioapic_state));
 933                 break;
 934         default:
 935                 r = -EINVAL;
 936                 break;
 937         }
 938         kvm_pic_update_irq(pic_irqchip(kvm));
 939         return r;
 940 }
 941
 942 /*
 943  * Get (and clear) the dirty memory log for a memory slot.
 944  */
 945 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 946                                       struct kvm_dirty_log *log)
 947 {
 948         int r;
 949         int n;
 950         struct kvm_memory_slot *memslot;
 951         int is_dirty = 0;
 952
 953         mutex_lock(&kvm->lock);
 954
 955         r = kvm_get_dirty_log(kvm, log, &is_dirty);
 956         if (r)
 957                 goto out;
 958
 959         /* If nothing is dirty, don't bother messing with page tables. */
 960         if (is_dirty) {
 961                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 962                 kvm_flush_remote_tlbs(kvm);
 963                 memslot = &kvm->memslots[log->slot];
 964                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 965                 memset(memslot->dirty_bitmap, 0, n);
 966         }
 967         r = 0;
 968 out:
 969         mutex_unlock(&kvm->lock);
 970         return r;
 971 }
 972
 973 long kvm_arch_vm_ioctl(struct file *filp,
 974                        unsigned int ioctl, unsigned long arg)
 975 {
 976         struct kvm *kvm = filp->private_data;
 977         void __user *argp = (void __user *)arg;
 978         int r = -EINVAL;
 979
 980         switch (ioctl) {
 981         case KVM_SET_TSS_ADDR:
 982                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
 983                 if (r < 0)
 984                         goto out;
 985                 break;
 986         case KVM_SET_MEMORY_REGION: {
 987                 struct kvm_memory_region kvm_mem;
 988                 struct kvm_userspace_memory_region kvm_userspace_mem;
 989
 990                 r = -EFAULT;
 991                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
 992                         goto out;
 993                 kvm_userspace_mem.slot = kvm_mem.slot;
 994                 kvm_userspace_mem.flags = kvm_mem.flags;
 995                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
 996                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
 997                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
 998                 if (r)
 999                         goto out;
1000                 break;
1001         }
1002         case KVM_SET_NR_MMU_PAGES:
1003                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1004                 if (r)
1005                         goto out;
1006                 break;
1007         case KVM_GET_NR_MMU_PAGES:
1008                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1009                 break;
1010         case KVM_SET_MEMORY_ALIAS: {
1011                 struct kvm_memory_alias alias;
1012
1013                 r = -EFAULT;
1014                 if (copy_from_user(&alias, argp, sizeof alias))
1015                         goto out;
1016                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
1017                 if (r)
1018                         goto out;
1019                 break;
1020         }
1021         case KVM_CREATE_IRQCHIP:
1022                 r = -ENOMEM;
1023                 kvm->vpic = kvm_create_pic(kvm);
1024                 if (kvm->vpic) {
1025                         r = kvm_ioapic_init(kvm);
1026                         if (r) {
1027                                 kfree(kvm->vpic);
1028                                 kvm->vpic = NULL;
1029                                 goto out;
1030                         }
1031                 } else
1032                         goto out;
1033                 break;
1034         case KVM_IRQ_LINE: {
1035                 struct kvm_irq_level irq_event;
1036
1037                 r = -EFAULT;
1038                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1039                         goto out;
1040                 if (irqchip_in_kernel(kvm)) {
1041                         mutex_lock(&kvm->lock);
1042                         if (irq_event.irq < 16)
1043                                 kvm_pic_set_irq(pic_irqchip(kvm),
1044                                         irq_event.irq,
1045                                         irq_event.level);
1046                         kvm_ioapic_set_irq(kvm->vioapic,
1047                                         irq_event.irq,
1048                                         irq_event.level);
1049                         mutex_unlock(&kvm->lock);
1050                         r = 0;
1051                 }
1052                 break;
1053         }
1054         case KVM_GET_IRQCHIP: {
1055                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1056                 struct kvm_irqchip chip;
1057
1058                 r = -EFAULT;
1059                 if (copy_from_user(&chip, argp, sizeof chip))
1060                         goto out;
1061                 r = -ENXIO;
1062                 if (!irqchip_in_kernel(kvm))
1063                         goto out;
1064                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1065                 if (r)
1066                         goto out;
1067                 r = -EFAULT;
1068                 if (copy_to_user(argp, &chip, sizeof chip))
1069                         goto out;
1070                 r = 0;
1071                 break;
1072         }
1073         case KVM_SET_IRQCHIP: {
1074                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1075                 struct kvm_irqchip chip;
1076
1077                 r = -EFAULT;
1078                 if (copy_from_user(&chip, argp, sizeof chip))
1079                         goto out;
1080                 r = -ENXIO;
1081                 if (!irqchip_in_kernel(kvm))
1082                         goto out;
1083                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1084                 if (r)
1085                         goto out;
1086                 r = 0;
1087                 break;
1088         }
1089         default:
1090                 ;
1091         }
1092 out:
1093         return r;
1094 }
1095
1096 static void kvm_init_msr_list(void)
1097 {
1098         u32 dummy[2];
1099         unsigned i, j;
1100
1101         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1102                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1103                         continue;
1104                 if (j < i)
1105                         msrs_to_save[j] = msrs_to_save[i];
1106                 j++;
1107         }
1108         num_msrs_to_save = j;
1109 }
1110
1111 /*
1112  * Only apic need an MMIO device hook, so shortcut now..
1113  */
1114 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1115                                                 gpa_t addr)
1116 {
1117         struct kvm_io_device *dev;
1118
1119         if (vcpu->apic) {
1120                 dev = &vcpu->apic->dev;
1121                 if (dev->in_range(dev, addr))
1122                         return dev;
1123         }
1124         return NULL;
1125 }
1126
1127
1128 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1129                                                 gpa_t addr)
1130 {
1131         struct kvm_io_device *dev;
1132
1133         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1134         if (dev == NULL)
1135                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1136         return dev;
1137 }
1138
1139 int emulator_read_std(unsigned long addr,
1140                              void *val,
1141                              unsigned int bytes,
1142                              struct kvm_vcpu *vcpu)
1143 {
1144         void *data = val;
1145
1146         while (bytes) {
1147                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1148                 unsigned offset = addr & (PAGE_SIZE-1);
1149                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1150                 int ret;
1151
1152                 if (gpa == UNMAPPED_GVA)
1153                         return X86EMUL_PROPAGATE_FAULT;
1154                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1155                 if (ret < 0)
1156                         return X86EMUL_UNHANDLEABLE;
1157
1158                 bytes -= tocopy;
1159                 data += tocopy;
1160                 addr += tocopy;
1161         }
1162
1163         return X86EMUL_CONTINUE;
1164 }
1165 EXPORT_SYMBOL_GPL(emulator_read_std);
1166
1167 static int emulator_read_emulated(unsigned long addr,
1168                                   void *val,
1169                                   unsigned int bytes,
1170                                   struct kvm_vcpu *vcpu)
1171 {
1172         struct kvm_io_device *mmio_dev;
1173         gpa_t                 gpa;
1174
1175         if (vcpu->mmio_read_completed) {
1176                 memcpy(val, vcpu->mmio_data, bytes);
1177                 vcpu->mmio_read_completed = 0;
1178                 return X86EMUL_CONTINUE;
1179         }
1180
1181         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1182
1183         /* For APIC access vmexit */
1184         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1185                 goto mmio;
1186
1187         if (emulator_read_std(addr, val, bytes, vcpu)
1188                         == X86EMUL_CONTINUE)
1189                 return X86EMUL_CONTINUE;
1190         if (gpa == UNMAPPED_GVA)
1191                 return X86EMUL_PROPAGATE_FAULT;
1192
1193 mmio:
1194         /*
1195          * Is this MMIO handled locally?
1196          */
1197         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1198         if (mmio_dev) {
1199                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1200                 return X86EMUL_CONTINUE;
1201         }
1202
1203         vcpu->mmio_needed = 1;
1204         vcpu->mmio_phys_addr = gpa;
1205         vcpu->mmio_size = bytes;
1206         vcpu->mmio_is_write = 0;
1207
1208         return X86EMUL_UNHANDLEABLE;
1209 }
1210
1211 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1212                                const void *val, int bytes)
1213 {
1214         int ret;
1215
1216         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1217         if (ret < 0)
1218                 return 0;
1219         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1220         return 1;
1221 }
1222
1223 static int emulator_write_emulated_onepage(unsigned long addr,
1224                                            const void *val,
1225                                            unsigned int bytes,
1226                                            struct kvm_vcpu *vcpu)
1227 {
1228         struct kvm_io_device *mmio_dev;
1229         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1230
1231         if (gpa == UNMAPPED_GVA) {
1232                 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1233                 return X86EMUL_PROPAGATE_FAULT;
1234         }
1235
1236         /* For APIC access vmexit */
1237         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1238                 goto mmio;
1239
1240         if (emulator_write_phys(vcpu, gpa, val, bytes))
1241                 return X86EMUL_CONTINUE;
1242
1243 mmio:
1244         /*
1245          * Is this MMIO handled locally?
1246          */
1247         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1248         if (mmio_dev) {
1249                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1250                 return X86EMUL_CONTINUE;
1251         }
1252
1253         vcpu->mmio_needed = 1;
1254         vcpu->mmio_phys_addr = gpa;
1255         vcpu->mmio_size = bytes;
1256         vcpu->mmio_is_write = 1;
1257         memcpy(vcpu->mmio_data, val, bytes);
1258
1259         return X86EMUL_CONTINUE;
1260 }
1261
1262 int emulator_write_emulated(unsigned long addr,
1263                                    const void *val,
1264                                    unsigned int bytes,
1265                                    struct kvm_vcpu *vcpu)
1266 {
1267         /* Crossing a page boundary? */
1268         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1269                 int rc, now;
1270
1271                 now = -addr & ~PAGE_MASK;
1272                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1273                 if (rc != X86EMUL_CONTINUE)
1274                         return rc;
1275                 addr += now;
1276                 val += now;
1277                 bytes -= now;
1278         }
1279         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1280 }
1281 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1282
1283 static int emulator_cmpxchg_emulated(unsigned long addr,
1284                                      const void *old,
1285                                      const void *new,
1286                                      unsigned int bytes,
1287                                      struct kvm_vcpu *vcpu)
1288 {
1289         static int reported;
1290
1291         if (!reported) {
1292                 reported = 1;
1293                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1294         }
1295         return emulator_write_emulated(addr, new, bytes, vcpu);
1296 }
1297
1298 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1299 {
1300         return kvm_x86_ops->get_segment_base(vcpu, seg);
1301 }
1302
1303 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1304 {
1305         return X86EMUL_CONTINUE;
1306 }
1307
1308 int emulate_clts(struct kvm_vcpu *vcpu)
1309 {
1310         kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1311         return X86EMUL_CONTINUE;
1312 }
1313
1314 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1315 {
1316         struct kvm_vcpu *vcpu = ctxt->vcpu;
1317
1318         switch (dr) {
1319         case 0 ... 3:
1320                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1321                 return X86EMUL_CONTINUE;
1322         default:
1323                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1324                 return X86EMUL_UNHANDLEABLE;
1325         }
1326 }
1327
1328 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1329 {
1330         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1331         int exception;
1332
1333         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1334         if (exception) {
1335                 /* FIXME: better handling */
1336                 return X86EMUL_UNHANDLEABLE;
1337         }
1338         return X86EMUL_CONTINUE;
1339 }
1340
1341 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1342 {
1343         static int reported;
1344         u8 opcodes[4];
1345         unsigned long rip = vcpu->rip;
1346         unsigned long rip_linear;
1347
1348         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1349
1350         if (reported)
1351                 return;
1352
1353         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1354
1355         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1356                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1357         reported = 1;
1358 }
1359 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1360
1361 struct x86_emulate_ops emulate_ops = {
1362         .read_std            = emulator_read_std,
1363         .read_emulated       = emulator_read_emulated,
1364         .write_emulated      = emulator_write_emulated,
1365         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1366 };
1367
1368 int emulate_instruction(struct kvm_vcpu *vcpu,
1369                         struct kvm_run *run,
1370                         unsigned long cr2,
1371                         u16 error_code,
1372                         int no_decode)
1373 {
1374         int r;
1375
1376         vcpu->mmio_fault_cr2 = cr2;
1377         kvm_x86_ops->cache_regs(vcpu);
1378
1379         vcpu->mmio_is_write = 0;
1380         vcpu->pio.string = 0;
1381
1382         if (!no_decode) {
1383                 int cs_db, cs_l;
1384                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1385
1386                 vcpu->emulate_ctxt.vcpu = vcpu;
1387                 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1388                 vcpu->emulate_ctxt.cr2 = cr2;
1389                 vcpu->emulate_ctxt.mode =
1390                         (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1391                         ? X86EMUL_MODE_REAL : cs_l
1392                         ? X86EMUL_MODE_PROT64 : cs_db
1393                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1394
1395                 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1396                         vcpu->emulate_ctxt.cs_base = 0;
1397                         vcpu->emulate_ctxt.ds_base = 0;
1398                         vcpu->emulate_ctxt.es_base = 0;
1399                         vcpu->emulate_ctxt.ss_base = 0;
1400                 } else {
1401                         vcpu->emulate_ctxt.cs_base =
1402                                         get_segment_base(vcpu, VCPU_SREG_CS);
1403                         vcpu->emulate_ctxt.ds_base =
1404                                         get_segment_base(vcpu, VCPU_SREG_DS);
1405                         vcpu->emulate_ctxt.es_base =
1406                                         get_segment_base(vcpu, VCPU_SREG_ES);
1407                         vcpu->emulate_ctxt.ss_base =
1408                                         get_segment_base(vcpu, VCPU_SREG_SS);
1409                 }
1410
1411                 vcpu->emulate_ctxt.gs_base =
1412                                         get_segment_base(vcpu, VCPU_SREG_GS);
1413                 vcpu->emulate_ctxt.fs_base =
1414                                         get_segment_base(vcpu, VCPU_SREG_FS);
1415
1416                 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
1417                 ++vcpu->stat.insn_emulation;
1418                 if (r)  {
1419                         ++vcpu->stat.insn_emulation_fail;
1420                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1421                                 return EMULATE_DONE;
1422                         return EMULATE_FAIL;
1423                 }
1424         }
1425
1426         r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1427
1428         if (vcpu->pio.string)
1429                 return EMULATE_DO_MMIO;
1430
1431         if ((r || vcpu->mmio_is_write) && run) {
1432                 run->exit_reason = KVM_EXIT_MMIO;
1433                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1434                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1435                 run->mmio.len = vcpu->mmio_size;
1436                 run->mmio.is_write = vcpu->mmio_is_write;
1437         }
1438
1439         if (r) {
1440                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1441                         return EMULATE_DONE;
1442                 if (!vcpu->mmio_needed) {
1443                         kvm_report_emulation_failure(vcpu, "mmio");
1444                         return EMULATE_FAIL;
1445                 }
1446                 return EMULATE_DO_MMIO;
1447         }
1448
1449         kvm_x86_ops->decache_regs(vcpu);
1450         kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1451
1452         if (vcpu->mmio_is_write) {
1453                 vcpu->mmio_needed = 0;
1454                 return EMULATE_DO_MMIO;
1455         }
1456
1457         return EMULATE_DONE;
1458 }
1459 EXPORT_SYMBOL_GPL(emulate_instruction);
1460
1461 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
1462 {
1463         int i;
1464
1465         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
1466                 if (vcpu->pio.guest_pages[i]) {
1467                         kvm_release_page_dirty(vcpu->pio.guest_pages[i]);
1468                         vcpu->pio.guest_pages[i] = NULL;
1469                 }
1470 }
1471
1472 static int pio_copy_data(struct kvm_vcpu *vcpu)
1473 {
1474         void *p = vcpu->pio_data;
1475         void *q;
1476         unsigned bytes;
1477         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1478
1479         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1480                  PAGE_KERNEL);
1481         if (!q) {
1482                 free_pio_guest_pages(vcpu);
1483                 return -ENOMEM;
1484         }
1485         q += vcpu->pio.guest_page_offset;
1486         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1487         if (vcpu->pio.in)
1488                 memcpy(q, p, bytes);
1489         else
1490                 memcpy(p, q, bytes);
1491         q -= vcpu->pio.guest_page_offset;
1492         vunmap(q);
1493         free_pio_guest_pages(vcpu);
1494         return 0;
1495 }
1496
1497 int complete_pio(struct kvm_vcpu *vcpu)
1498 {
1499         struct kvm_pio_request *io = &vcpu->pio;
1500         long delta;
1501         int r;
1502
1503         kvm_x86_ops->cache_regs(vcpu);
1504
1505         if (!io->string) {
1506                 if (io->in)
1507                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1508                                io->size);
1509         } else {
1510                 if (io->in) {
1511                         r = pio_copy_data(vcpu);
1512                         if (r) {
1513                                 kvm_x86_ops->cache_regs(vcpu);
1514                                 return r;
1515                         }
1516                 }
1517
1518                 delta = 1;
1519                 if (io->rep) {
1520                         delta *= io->cur_count;
1521                         /*
1522                          * The size of the register should really depend on
1523                          * current address size.
1524                          */
1525                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1526                 }
1527                 if (io->down)
1528                         delta = -delta;
1529                 delta *= io->size;
1530                 if (io->in)
1531                         vcpu->regs[VCPU_REGS_RDI] += delta;
1532                 else
1533                         vcpu->regs[VCPU_REGS_RSI] += delta;
1534         }
1535
1536         kvm_x86_ops->decache_regs(vcpu);
1537
1538         io->count -= io->cur_count;
1539         io->cur_count = 0;
1540
1541         return 0;
1542 }
1543
1544 static void kernel_pio(struct kvm_io_device *pio_dev,
1545                        struct kvm_vcpu *vcpu,
1546                        void *pd)
1547 {
1548         /* TODO: String I/O for in kernel device */
1549
1550         mutex_lock(&vcpu->kvm->lock);
1551         if (vcpu->pio.in)
1552                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1553                                   vcpu->pio.size,
1554                                   pd);
1555         else
1556                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1557                                    vcpu->pio.size,
1558                                    pd);
1559         mutex_unlock(&vcpu->kvm->lock);
1560 }
1561
1562 static void pio_string_write(struct kvm_io_device *pio_dev,
1563                              struct kvm_vcpu *vcpu)
1564 {
1565         struct kvm_pio_request *io = &vcpu->pio;
1566         void *pd = vcpu->pio_data;
1567         int i;
1568
1569         mutex_lock(&vcpu->kvm->lock);
1570         for (i = 0; i < io->cur_count; i++) {
1571                 kvm_iodevice_write(pio_dev, io->port,
1572                                    io->size,
1573                                    pd);
1574                 pd += io->size;
1575         }
1576         mutex_unlock(&vcpu->kvm->lock);
1577 }
1578
1579 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1580                                                gpa_t addr)
1581 {
1582         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1583 }
1584
1585 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1586                   int size, unsigned port)
1587 {
1588         struct kvm_io_device *pio_dev;
1589
1590         vcpu->run->exit_reason = KVM_EXIT_IO;
1591         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1592         vcpu->run->io.size = vcpu->pio.size = size;
1593         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1594         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1595         vcpu->run->io.port = vcpu->pio.port = port;
1596         vcpu->pio.in = in;
1597         vcpu->pio.string = 0;
1598         vcpu->pio.down = 0;
1599         vcpu->pio.guest_page_offset = 0;
1600         vcpu->pio.rep = 0;
1601
1602         kvm_x86_ops->cache_regs(vcpu);
1603         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1604         kvm_x86_ops->decache_regs(vcpu);
1605
1606         kvm_x86_ops->skip_emulated_instruction(vcpu);
1607
1608         pio_dev = vcpu_find_pio_dev(vcpu, port);
1609         if (pio_dev) {
1610                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1611                 complete_pio(vcpu);
1612                 return 1;
1613         }
1614         return 0;
1615 }
1616 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1617
1618 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1619                   int size, unsigned long count, int down,
1620                   gva_t address, int rep, unsigned port)
1621 {
1622         unsigned now, in_page;
1623         int i, ret = 0;
1624         int nr_pages = 1;
1625         struct page *page;
1626         struct kvm_io_device *pio_dev;
1627
1628         vcpu->run->exit_reason = KVM_EXIT_IO;
1629         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1630         vcpu->run->io.size = vcpu->pio.size = size;
1631         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1632         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1633         vcpu->run->io.port = vcpu->pio.port = port;
1634         vcpu->pio.in = in;
1635         vcpu->pio.string = 1;
1636         vcpu->pio.down = down;
1637         vcpu->pio.guest_page_offset = offset_in_page(address);
1638         vcpu->pio.rep = rep;
1639
1640         if (!count) {
1641                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1642                 return 1;
1643         }
1644
1645         if (!down)
1646                 in_page = PAGE_SIZE - offset_in_page(address);
1647         else
1648                 in_page = offset_in_page(address) + size;
1649         now = min(count, (unsigned long)in_page / size);
1650         if (!now) {
1651                 /*
1652                  * String I/O straddles page boundary.  Pin two guest pages
1653                  * so that we satisfy atomicity constraints.  Do just one
1654                  * transaction to avoid complexity.
1655                  */
1656                 nr_pages = 2;
1657                 now = 1;
1658         }
1659         if (down) {
1660                 /*
1661                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1662                  */
1663                 pr_unimpl(vcpu, "guest string pio down\n");
1664                 inject_gp(vcpu);
1665                 return 1;
1666         }
1667         vcpu->run->io.count = now;
1668         vcpu->pio.cur_count = now;
1669
1670         if (vcpu->pio.cur_count == vcpu->pio.count)
1671                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1672
1673         for (i = 0; i < nr_pages; ++i) {
1674                 mutex_lock(&vcpu->kvm->lock);
1675                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1676                 vcpu->pio.guest_pages[i] = page;
1677                 mutex_unlock(&vcpu->kvm->lock);
1678                 if (!page) {
1679                         inject_gp(vcpu);
1680                         free_pio_guest_pages(vcpu);
1681                         return 1;
1682                 }
1683         }
1684
1685         pio_dev = vcpu_find_pio_dev(vcpu, port);
1686         if (!vcpu->pio.in) {
1687                 /* string PIO write */
1688                 ret = pio_copy_data(vcpu);
1689                 if (ret >= 0 && pio_dev) {
1690                         pio_string_write(pio_dev, vcpu);
1691                         complete_pio(vcpu);
1692                         if (vcpu->pio.count == 0)
1693                                 ret = 1;
1694                 }
1695         } else if (pio_dev)
1696                 pr_unimpl(vcpu, "no string pio read support yet, "
1697                        "port %x size %d count %ld\n",
1698                         port, size, count);
1699
1700         return ret;
1701 }
1702 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1703
1704 int kvm_arch_init(void *opaque)
1705 {
1706         int r;
1707         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
1708
1709         r = kvm_mmu_module_init();
1710         if (r)
1711                 goto out_fail;
1712
1713         kvm_init_msr_list();
1714
1715         if (kvm_x86_ops) {
1716                 printk(KERN_ERR "kvm: already loaded the other module\n");
1717                 r = -EEXIST;
1718                 goto out;
1719         }
1720
1721         if (!ops->cpu_has_kvm_support()) {
1722                 printk(KERN_ERR "kvm: no hardware support\n");
1723                 r = -EOPNOTSUPP;
1724                 goto out;
1725         }
1726         if (ops->disabled_by_bios()) {
1727                 printk(KERN_ERR "kvm: disabled by bios\n");
1728                 r = -EOPNOTSUPP;
1729                 goto out;
1730         }
1731
1732         kvm_x86_ops = ops;
1733         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
1734         return 0;
1735
1736 out:
1737         kvm_mmu_module_exit();
1738 out_fail:
1739         return r;
1740 }
1741
1742 void kvm_arch_exit(void)
1743 {
1744         kvm_x86_ops = NULL;
1745         kvm_mmu_module_exit();
1746 }
1747
1748 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1749 {
1750         ++vcpu->stat.halt_exits;
1751         if (irqchip_in_kernel(vcpu->kvm)) {
1752                 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1753                 kvm_vcpu_block(vcpu);
1754                 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1755                         return -EINTR;
1756                 return 1;
1757         } else {
1758                 vcpu->run->exit_reason = KVM_EXIT_HLT;
1759                 return 0;
1760         }
1761 }
1762 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1763
1764 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
1765 {
1766         unsigned long nr, a0, a1, a2, a3, ret;
1767
1768         kvm_x86_ops->cache_regs(vcpu);
1769
1770         nr = vcpu->regs[VCPU_REGS_RAX];
1771         a0 = vcpu->regs[VCPU_REGS_RBX];
1772         a1 = vcpu->regs[VCPU_REGS_RCX];
1773         a2 = vcpu->regs[VCPU_REGS_RDX];
1774         a3 = vcpu->regs[VCPU_REGS_RSI];
1775
1776         if (!is_long_mode(vcpu)) {
1777                 nr &= 0xFFFFFFFF;
1778                 a0 &= 0xFFFFFFFF;
1779                 a1 &= 0xFFFFFFFF;
1780                 a2 &= 0xFFFFFFFF;
1781                 a3 &= 0xFFFFFFFF;
1782         }
1783
1784         switch (nr) {
1785         default:
1786                 ret = -KVM_ENOSYS;
1787                 break;
1788         }
1789         vcpu->regs[VCPU_REGS_RAX] = ret;
1790         kvm_x86_ops->decache_regs(vcpu);
1791         return 0;
1792 }
1793 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1794
1795 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1796 {
1797         char instruction[3];
1798         int ret = 0;
1799
1800         mutex_lock(&vcpu->kvm->lock);
1801
1802         /*
1803          * Blow out the MMU to ensure that no other VCPU has an active mapping
1804          * to ensure that the updated hypercall appears atomically across all
1805          * VCPUs.
1806          */
1807         kvm_mmu_zap_all(vcpu->kvm);
1808
1809         kvm_x86_ops->cache_regs(vcpu);
1810         kvm_x86_ops->patch_hypercall(vcpu, instruction);
1811         if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1812             != X86EMUL_CONTINUE)
1813                 ret = -EFAULT;
1814
1815         mutex_unlock(&vcpu->kvm->lock);
1816
1817         return ret;
1818 }
1819
1820 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1821 {
1822         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1823 }
1824
1825 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1826 {
1827         struct descriptor_table dt = { limit, base };
1828
1829         kvm_x86_ops->set_gdt(vcpu, &dt);
1830 }
1831
1832 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1833 {
1834         struct descriptor_table dt = { limit, base };
1835
1836         kvm_x86_ops->set_idt(vcpu, &dt);
1837 }
1838
1839 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1840                    unsigned long *rflags)
1841 {
1842         lmsw(vcpu, msw);
1843         *rflags = kvm_x86_ops->get_rflags(vcpu);
1844 }
1845
1846 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1847 {
1848         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1849         switch (cr) {
1850         case 0:
1851                 return vcpu->cr0;
1852         case 2:
1853                 return vcpu->cr2;
1854         case 3:
1855                 return vcpu->cr3;
1856         case 4:
1857                 return vcpu->cr4;
1858         default:
1859                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1860                 return 0;
1861         }
1862 }
1863
1864 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1865                      unsigned long *rflags)
1866 {
1867         switch (cr) {
1868         case 0:
1869                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1870                 *rflags = kvm_x86_ops->get_rflags(vcpu);
1871                 break;
1872         case 2:
1873                 vcpu->cr2 = val;
1874                 break;
1875         case 3:
1876                 set_cr3(vcpu, val);
1877                 break;
1878         case 4:
1879                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1880                 break;
1881         default:
1882                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1883         }
1884 }
1885
1886 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1887 {
1888         int i;
1889         u32 function;
1890         struct kvm_cpuid_entry *e, *best;
1891
1892         kvm_x86_ops->cache_regs(vcpu);
1893         function = vcpu->regs[VCPU_REGS_RAX];
1894         vcpu->regs[VCPU_REGS_RAX] = 0;
1895         vcpu->regs[VCPU_REGS_RBX] = 0;
1896         vcpu->regs[VCPU_REGS_RCX] = 0;
1897         vcpu->regs[VCPU_REGS_RDX] = 0;
1898         best = NULL;
1899         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1900                 e = &vcpu->cpuid_entries[i];
1901                 if (e->function == function) {
1902                         best = e;
1903                         break;
1904                 }
1905                 /*
1906                  * Both basic or both extended?
1907                  */
1908                 if (((e->function ^ function) & 0x80000000) == 0)
1909                         if (!best || e->function > best->function)
1910                                 best = e;
1911         }
1912         if (best) {
1913                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1914                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1915                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1916                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1917         }
1918         kvm_x86_ops->decache_regs(vcpu);
1919         kvm_x86_ops->skip_emulated_instruction(vcpu);
1920 }
1921 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1922
1923 /*
1924  * Check if userspace requested an interrupt window, and that the
1925  * interrupt window is open.
1926  *
1927  * No need to exit to userspace if we already have an interrupt queued.
1928  */
1929 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1930                                           struct kvm_run *kvm_run)
1931 {
1932         return (!vcpu->irq_summary &&
1933                 kvm_run->request_interrupt_window &&
1934                 vcpu->interrupt_window_open &&
1935                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
1936 }
1937
1938 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1939                               struct kvm_run *kvm_run)
1940 {
1941         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
1942         kvm_run->cr8 = get_cr8(vcpu);
1943         kvm_run->apic_base = kvm_get_apic_base(vcpu);
1944         if (irqchip_in_kernel(vcpu->kvm))
1945                 kvm_run->ready_for_interrupt_injection = 1;
1946         else
1947                 kvm_run->ready_for_interrupt_injection =
1948                                         (vcpu->interrupt_window_open &&
1949                                          vcpu->irq_summary == 0);
1950 }
1951
1952 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1953 {
1954         int r;
1955
1956         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
1957                 pr_debug("vcpu %d received sipi with vector # %x\n",
1958                        vcpu->vcpu_id, vcpu->sipi_vector);
1959                 kvm_lapic_reset(vcpu);
1960                 r = kvm_x86_ops->vcpu_reset(vcpu);
1961                 if (r)
1962                         return r;
1963                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
1964         }
1965
1966 preempted:
1967         if (vcpu->guest_debug.enabled)
1968                 kvm_x86_ops->guest_debug_pre(vcpu);
1969
1970 again:
1971         r = kvm_mmu_reload(vcpu);
1972         if (unlikely(r))
1973                 goto out;
1974
1975         kvm_inject_pending_timer_irqs(vcpu);
1976
1977         preempt_disable();
1978
1979         kvm_x86_ops->prepare_guest_switch(vcpu);
1980         kvm_load_guest_fpu(vcpu);
1981
1982         local_irq_disable();
1983
1984         if (signal_pending(current)) {
1985                 local_irq_enable();
1986                 preempt_enable();
1987                 r = -EINTR;
1988                 kvm_run->exit_reason = KVM_EXIT_INTR;
1989                 ++vcpu->stat.signal_exits;
1990                 goto out;
1991         }
1992
1993         if (irqchip_in_kernel(vcpu->kvm))
1994                 kvm_x86_ops->inject_pending_irq(vcpu);
1995         else if (!vcpu->mmio_read_completed)
1996                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
1997
1998         vcpu->guest_mode = 1;
1999         kvm_guest_enter();
2000
2001         if (vcpu->requests)
2002                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2003                         kvm_x86_ops->tlb_flush(vcpu);
2004
2005         kvm_x86_ops->run(vcpu, kvm_run);
2006
2007         vcpu->guest_mode = 0;
2008         local_irq_enable();
2009
2010         ++vcpu->stat.exits;
2011
2012         /*
2013          * We must have an instruction between local_irq_enable() and
2014          * kvm_guest_exit(), so the timer interrupt isn't delayed by
2015          * the interrupt shadow.  The stat.exits increment will do nicely.
2016          * But we need to prevent reordering, hence this barrier():
2017          */
2018         barrier();
2019
2020         kvm_guest_exit();
2021
2022         preempt_enable();
2023
2024         /*
2025          * Profile KVM exit RIPs:
2026          */
2027         if (unlikely(prof_on == KVM_PROFILING)) {
2028                 kvm_x86_ops->cache_regs(vcpu);
2029                 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
2030         }
2031
2032         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2033
2034         if (r > 0) {
2035                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2036                         r = -EINTR;
2037                         kvm_run->exit_reason = KVM_EXIT_INTR;
2038                         ++vcpu->stat.request_irq_exits;
2039                         goto out;
2040                 }
2041                 if (!need_resched())
2042                         goto again;
2043         }
2044
2045 out:
2046         if (r > 0) {
2047                 kvm_resched(vcpu);
2048                 goto preempted;
2049         }
2050
2051         post_kvm_run_save(vcpu, kvm_run);
2052
2053         return r;
2054 }
2055
2056 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2057 {
2058         int r;
2059         sigset_t sigsaved;
2060
2061         vcpu_load(vcpu);
2062
2063         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2064                 kvm_vcpu_block(vcpu);
2065                 vcpu_put(vcpu);
2066                 return -EAGAIN;
2067         }
2068
2069         if (vcpu->sigset_active)
2070                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2071
2072         /* re-sync apic's tpr */
2073         if (!irqchip_in_kernel(vcpu->kvm))
2074                 set_cr8(vcpu, kvm_run->cr8);
2075
2076         if (vcpu->pio.cur_count) {
2077                 r = complete_pio(vcpu);
2078                 if (r)
2079                         goto out;
2080         }
2081 #if CONFIG_HAS_IOMEM
2082         if (vcpu->mmio_needed) {
2083                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2084                 vcpu->mmio_read_completed = 1;
2085                 vcpu->mmio_needed = 0;
2086                 r = emulate_instruction(vcpu, kvm_run,
2087                                         vcpu->mmio_fault_cr2, 0, 1);
2088                 if (r == EMULATE_DO_MMIO) {
2089                         /*
2090                          * Read-modify-write.  Back to userspace.
2091                          */
2092                         r = 0;
2093                         goto out;
2094                 }
2095         }
2096 #endif
2097         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2098                 kvm_x86_ops->cache_regs(vcpu);
2099                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2100                 kvm_x86_ops->decache_regs(vcpu);
2101         }
2102
2103         r = __vcpu_run(vcpu, kvm_run);
2104
2105 out:
2106         if (vcpu->sigset_active)
2107                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2108
2109         vcpu_put(vcpu);
2110         return r;
2111 }
2112
2113 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2114 {
2115         vcpu_load(vcpu);
2116
2117         kvm_x86_ops->cache_regs(vcpu);
2118
2119         regs->rax = vcpu->regs[VCPU_REGS_RAX];
2120         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2121         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2122         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2123         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2124         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2125         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2126         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2127 #ifdef CONFIG_X86_64
2128         regs->r8 = vcpu->regs[VCPU_REGS_R8];
2129         regs->r9 = vcpu->regs[VCPU_REGS_R9];
2130         regs->r10 = vcpu->regs[VCPU_REGS_R10];
2131         regs->r11 = vcpu->regs[VCPU_REGS_R11];
2132         regs->r12 = vcpu->regs[VCPU_REGS_R12];
2133         regs->r13 = vcpu->regs[VCPU_REGS_R13];
2134         regs->r14 = vcpu->regs[VCPU_REGS_R14];
2135         regs->r15 = vcpu->regs[VCPU_REGS_R15];
2136 #endif
2137
2138         regs->rip = vcpu->rip;
2139         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2140
2141         /*
2142          * Don't leak debug flags in case they were set for guest debugging
2143          */
2144         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2145                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2146
2147         vcpu_put(vcpu);
2148
2149         return 0;
2150 }
2151
2152 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2153 {
2154         vcpu_load(vcpu);
2155
2156         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2157         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2158         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2159         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2160         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2161         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2162         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2163         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2164 #ifdef CONFIG_X86_64
2165         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2166         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2167         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2168         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2169         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2170         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2171         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2172         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2173 #endif
2174
2175         vcpu->rip = regs->rip;
2176         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2177
2178         kvm_x86_ops->decache_regs(vcpu);
2179
2180         vcpu_put(vcpu);
2181
2182         return 0;
2183 }
2184
2185 static void get_segment(struct kvm_vcpu *vcpu,
2186                         struct kvm_segment *var, int seg)
2187 {
2188         return kvm_x86_ops->get_segment(vcpu, var, seg);
2189 }
2190
2191 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2192 {
2193         struct kvm_segment cs;
2194
2195         get_segment(vcpu, &cs, VCPU_SREG_CS);
2196         *db = cs.db;
2197         *l = cs.l;
2198 }
2199 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2200
2201 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2202                                   struct kvm_sregs *sregs)
2203 {
2204         struct descriptor_table dt;
2205         int pending_vec;
2206
2207         vcpu_load(vcpu);
2208
2209         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2210         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2211         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2212         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2213         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2214         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2215
2216         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2217         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2218
2219         kvm_x86_ops->get_idt(vcpu, &dt);
2220         sregs->idt.limit = dt.limit;
2221         sregs->idt.base = dt.base;
2222         kvm_x86_ops->get_gdt(vcpu, &dt);
2223         sregs->gdt.limit = dt.limit;
2224         sregs->gdt.base = dt.base;
2225
2226         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2227         sregs->cr0 = vcpu->cr0;
2228         sregs->cr2 = vcpu->cr2;
2229         sregs->cr3 = vcpu->cr3;
2230         sregs->cr4 = vcpu->cr4;
2231         sregs->cr8 = get_cr8(vcpu);
2232         sregs->efer = vcpu->shadow_efer;
2233         sregs->apic_base = kvm_get_apic_base(vcpu);
2234
2235         if (irqchip_in_kernel(vcpu->kvm)) {
2236                 memset(sregs->interrupt_bitmap, 0,
2237                        sizeof sregs->interrupt_bitmap);
2238                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2239                 if (pending_vec >= 0)
2240                         set_bit(pending_vec,
2241                                 (unsigned long *)sregs->interrupt_bitmap);
2242         } else
2243                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2244                        sizeof sregs->interrupt_bitmap);
2245
2246         vcpu_put(vcpu);
2247
2248         return 0;
2249 }
2250
2251 static void set_segment(struct kvm_vcpu *vcpu,
2252                         struct kvm_segment *var, int seg)
2253 {
2254         return kvm_x86_ops->set_segment(vcpu, var, seg);
2255 }
2256
2257 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2258                                   struct kvm_sregs *sregs)
2259 {
2260         int mmu_reset_needed = 0;
2261         int i, pending_vec, max_bits;
2262         struct descriptor_table dt;
2263
2264         vcpu_load(vcpu);
2265
2266         dt.limit = sregs->idt.limit;
2267         dt.base = sregs->idt.base;
2268         kvm_x86_ops->set_idt(vcpu, &dt);
2269         dt.limit = sregs->gdt.limit;
2270         dt.base = sregs->gdt.base;
2271         kvm_x86_ops->set_gdt(vcpu, &dt);
2272
2273         vcpu->cr2 = sregs->cr2;
2274         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2275         vcpu->cr3 = sregs->cr3;
2276
2277         set_cr8(vcpu, sregs->cr8);
2278
2279         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2280 #ifdef CONFIG_X86_64
2281         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2282 #endif
2283         kvm_set_apic_base(vcpu, sregs->apic_base);
2284
2285         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2286
2287         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2288         vcpu->cr0 = sregs->cr0;
2289         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2290
2291         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2292         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2293         if (!is_long_mode(vcpu) && is_pae(vcpu))
2294                 load_pdptrs(vcpu, vcpu->cr3);
2295
2296         if (mmu_reset_needed)
2297                 kvm_mmu_reset_context(vcpu);
2298
2299         if (!irqchip_in_kernel(vcpu->kvm)) {
2300                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2301                        sizeof vcpu->irq_pending);
2302                 vcpu->irq_summary = 0;
2303                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2304                         if (vcpu->irq_pending[i])
2305                                 __set_bit(i, &vcpu->irq_summary);
2306         } else {
2307                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2308                 pending_vec = find_first_bit(
2309                         (const unsigned long *)sregs->interrupt_bitmap,
2310                         max_bits);
2311                 /* Only pending external irq is handled here */
2312                 if (pending_vec < max_bits) {
2313                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2314                         pr_debug("Set back pending irq %d\n",
2315                                  pending_vec);
2316                 }
2317         }
2318
2319         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2320         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2321         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2322         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2323         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2324         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2325
2326         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2327         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2328
2329         vcpu_put(vcpu);
2330
2331         return 0;
2332 }
2333
2334 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2335                                     struct kvm_debug_guest *dbg)
2336 {
2337         int r;
2338
2339         vcpu_load(vcpu);
2340
2341         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2342
2343         vcpu_put(vcpu);
2344
2345         return r;
2346 }
2347
2348 /*
2349  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2350  * we have asm/x86/processor.h
2351  */
2352 struct fxsave {
2353         u16     cwd;
2354         u16     swd;
2355         u16     twd;
2356         u16     fop;
2357         u64     rip;
2358         u64     rdp;
2359         u32     mxcsr;
2360         u32     mxcsr_mask;
2361         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2362 #ifdef CONFIG_X86_64
2363         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2364 #else
2365         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2366 #endif
2367 };
2368
2369 /*
2370  * Translate a guest virtual address to a guest physical address.
2371  */
2372 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2373                                     struct kvm_translation *tr)
2374 {
2375         unsigned long vaddr = tr->linear_address;
2376         gpa_t gpa;
2377
2378         vcpu_load(vcpu);
2379         mutex_lock(&vcpu->kvm->lock);
2380         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2381         tr->physical_address = gpa;
2382         tr->valid = gpa != UNMAPPED_GVA;
2383         tr->writeable = 1;
2384         tr->usermode = 0;
2385         mutex_unlock(&vcpu->kvm->lock);
2386         vcpu_put(vcpu);
2387
2388         return 0;
2389 }
2390
2391 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2392 {
2393         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2394
2395         vcpu_load(vcpu);
2396
2397         memcpy(fpu->fpr, fxsave->st_space, 128);
2398         fpu->fcw = fxsave->cwd;
2399         fpu->fsw = fxsave->swd;
2400         fpu->ftwx = fxsave->twd;
2401         fpu->last_opcode = fxsave->fop;
2402         fpu->last_ip = fxsave->rip;
2403         fpu->last_dp = fxsave->rdp;
2404         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2405
2406         vcpu_put(vcpu);
2407
2408         return 0;
2409 }
2410
2411 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2412 {
2413         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2414
2415         vcpu_load(vcpu);
2416
2417         memcpy(fxsave->st_space, fpu->fpr, 128);
2418         fxsave->cwd = fpu->fcw;
2419         fxsave->swd = fpu->fsw;
2420         fxsave->twd = fpu->ftwx;
2421         fxsave->fop = fpu->last_opcode;
2422         fxsave->rip = fpu->last_ip;
2423         fxsave->rdp = fpu->last_dp;
2424         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2425
2426         vcpu_put(vcpu);
2427
2428         return 0;
2429 }
2430
2431 void fx_init(struct kvm_vcpu *vcpu)
2432 {
2433         unsigned after_mxcsr_mask;
2434
2435         /* Initialize guest FPU by resetting ours and saving into guest's */
2436         preempt_disable();
2437         fx_save(&vcpu->host_fx_image);
2438         fpu_init();
2439         fx_save(&vcpu->guest_fx_image);
2440         fx_restore(&vcpu->host_fx_image);
2441         preempt_enable();
2442
2443         vcpu->cr0 |= X86_CR0_ET;
2444         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
2445         vcpu->guest_fx_image.mxcsr = 0x1f80;
2446         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
2447                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
2448 }
2449 EXPORT_SYMBOL_GPL(fx_init);
2450
2451 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
2452 {
2453         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
2454                 return;
2455
2456         vcpu->guest_fpu_loaded = 1;
2457         fx_save(&vcpu->host_fx_image);
2458         fx_restore(&vcpu->guest_fx_image);
2459 }
2460 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
2461
2462 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
2463 {
2464         if (!vcpu->guest_fpu_loaded)
2465                 return;
2466
2467         vcpu->guest_fpu_loaded = 0;
2468         fx_save(&vcpu->guest_fx_image);
2469         fx_restore(&vcpu->host_fx_image);
2470         ++vcpu->stat.fpu_reload;
2471 }
2472 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
2473
2474 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
2475 {
2476         kvm_x86_ops->vcpu_free(vcpu);
2477 }
2478
2479 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
2480                                                 unsigned int id)
2481 {
2482         return kvm_x86_ops->vcpu_create(kvm, id);
2483 }
2484
2485 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
2486 {
2487         int r;
2488
2489         /* We do fxsave: this must be aligned. */
2490         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2491
2492         vcpu_load(vcpu);
2493         r = kvm_arch_vcpu_reset(vcpu);
2494         if (r == 0)
2495                 r = kvm_mmu_setup(vcpu);
2496         vcpu_put(vcpu);
2497         if (r < 0)
2498                 goto free_vcpu;
2499
2500         return 0;
2501 free_vcpu:
2502         kvm_x86_ops->vcpu_free(vcpu);
2503         return r;
2504 }
2505
2506 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
2507 {
2508         vcpu_load(vcpu);
2509         kvm_mmu_unload(vcpu);
2510         vcpu_put(vcpu);
2511
2512         kvm_x86_ops->vcpu_free(vcpu);
2513 }
2514
2515 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
2516 {
2517         return kvm_x86_ops->vcpu_reset(vcpu);
2518 }
2519
2520 void kvm_arch_hardware_enable(void *garbage)
2521 {
2522         kvm_x86_ops->hardware_enable(garbage);
2523 }
2524
2525 void kvm_arch_hardware_disable(void *garbage)
2526 {
2527         kvm_x86_ops->hardware_disable(garbage);
2528 }
2529
2530 int kvm_arch_hardware_setup(void)
2531 {
2532         return kvm_x86_ops->hardware_setup();
2533 }
2534
2535 void kvm_arch_hardware_unsetup(void)
2536 {
2537         kvm_x86_ops->hardware_unsetup();
2538 }
2539
2540 void kvm_arch_check_processor_compat(void *rtn)
2541 {
2542         kvm_x86_ops->check_processor_compatibility(rtn);
2543 }
2544
2545 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
2546 {
2547         struct page *page;
2548         struct kvm *kvm;
2549         int r;
2550
2551         BUG_ON(vcpu->kvm == NULL);
2552         kvm = vcpu->kvm;
2553
2554         vcpu->mmu.root_hpa = INVALID_PAGE;
2555         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
2556                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2557         else
2558                 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
2559
2560         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2561         if (!page) {
2562                 r = -ENOMEM;
2563                 goto fail;
2564         }
2565         vcpu->pio_data = page_address(page);
2566
2567         r = kvm_mmu_create(vcpu);
2568         if (r < 0)
2569                 goto fail_free_pio_data;
2570
2571         if (irqchip_in_kernel(kvm)) {
2572                 r = kvm_create_lapic(vcpu);
2573                 if (r < 0)
2574                         goto fail_mmu_destroy;
2575         }
2576
2577         return 0;
2578
2579 fail_mmu_destroy:
2580         kvm_mmu_destroy(vcpu);
2581 fail_free_pio_data:
2582         free_page((unsigned long)vcpu->pio_data);
2583 fail:
2584         return r;
2585 }
2586
2587 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
2588 {
2589         kvm_free_lapic(vcpu);
2590         kvm_mmu_destroy(vcpu);
2591         free_page((unsigned long)vcpu->pio_data);
2592 }
2593
2594 struct  kvm *kvm_arch_create_vm(void)
2595 {
2596         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
2597
2598         if (!kvm)
2599                 return ERR_PTR(-ENOMEM);
2600
2601         INIT_LIST_HEAD(&kvm->active_mmu_pages);
2602
2603         return kvm;
2604 }
2605
2606 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
2607 {
2608         vcpu_load(vcpu);
2609         kvm_mmu_unload(vcpu);
2610         vcpu_put(vcpu);
2611 }
2612
2613 static void kvm_free_vcpus(struct kvm *kvm)
2614 {
2615         unsigned int i;
2616
2617         /*
2618          * Unpin any mmu pages first.
2619          */
2620         for (i = 0; i < KVM_MAX_VCPUS; ++i)
2621                 if (kvm->vcpus[i])
2622                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
2623         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2624                 if (kvm->vcpus[i]) {
2625                         kvm_arch_vcpu_free(kvm->vcpus[i]);
2626                         kvm->vcpus[i] = NULL;
2627                 }
2628         }
2629
2630 }
2631
2632 void kvm_arch_destroy_vm(struct kvm *kvm)
2633 {
2634         kfree(kvm->vpic);
2635         kfree(kvm->vioapic);
2636         kvm_free_vcpus(kvm);
2637         kvm_free_physmem(kvm);
2638         kfree(kvm);
2639 }
2640
2641 int kvm_arch_set_memory_region(struct kvm *kvm,
2642                                 struct kvm_userspace_memory_region *mem,
2643                                 struct kvm_memory_slot old,
2644                                 int user_alloc)
2645 {
2646         int npages = mem->memory_size >> PAGE_SHIFT;
2647         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
2648
2649         /*To keep backward compatibility with older userspace,
2650          *x86 needs to hanlde !user_alloc case.
2651          */
2652         if (!user_alloc) {
2653                 if (npages && !old.rmap) {
2654                         down_write(&current->mm->mmap_sem);
2655                         memslot->userspace_addr = do_mmap(NULL, 0,
2656                                                      npages * PAGE_SIZE,
2657                                                      PROT_READ | PROT_WRITE,
2658                                                      MAP_SHARED | MAP_ANONYMOUS,
2659                                                      0);
2660                         up_write(&current->mm->mmap_sem);
2661
2662                         if (IS_ERR((void *)memslot->userspace_addr))
2663                                 return PTR_ERR((void *)memslot->userspace_addr);
2664                 } else {
2665                         if (!old.user_alloc && old.rmap) {
2666                                 int ret;
2667
2668                                 down_write(&current->mm->mmap_sem);
2669                                 ret = do_munmap(current->mm, old.userspace_addr,
2670                                                 old.npages * PAGE_SIZE);
2671                                 up_write(&current->mm->mmap_sem);
2672                                 if (ret < 0)
2673                                         printk(KERN_WARNING
2674                                        "kvm_vm_ioctl_set_memory_region: "
2675                                        "failed to munmap memory\n");
2676                         }
2677                 }
2678         }
2679
2680         if (!kvm->n_requested_mmu_pages) {
2681                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
2682                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
2683         }
2684
2685         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
2686         kvm_flush_remote_tlbs(kvm);
2687
2688         return 0;
2689 }