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