target-i386/kvm.c

   1 /*
   2  * QEMU KVM support
   3  *
   4  * Copyright (C) 2006-2008 Qumranet Technologies
   5  * Copyright IBM, Corp. 2008
   6  *
   7  * Authors:
   8  *  Anthony Liguori   <aliguori@us.ibm.com>
   9  *
  10  * This work is licensed under the terms of the GNU GPL, version 2 or later.
  11  * See the COPYING file in the top-level directory.
  12  *
  13  */
  14
  15 #include <sys/types.h>
  16 #include <sys/ioctl.h>
  17 #include <sys/mman.h>
  18 #include <sys/utsname.h>
  19
  20 #include <linux/kvm.h>
  21
  22 #include "qemu-common.h"
  23 #include "sysemu.h"
  24 #include "kvm.h"
  25 #include "cpu.h"
  26 #include "gdbstub.h"
  27 #include "host-utils.h"
  28 #include "hw/pc.h"
  29 #include "hw/apic.h"
  30 #include "ioport.h"
  31 #include "kvm_x86.h"
  32
  33 #ifdef CONFIG_KVM_PARA
  34 #include <linux/kvm_para.h>
  35 #endif
  36 //
  37 //#define DEBUG_KVM
  38
  39 #ifdef DEBUG_KVM
  40 #define DPRINTF(fmt, ...) \
  41     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  42 #else
  43 #define DPRINTF(fmt, ...) \
  44     do { } while (0)
  45 #endif
  46
  47 #define MSR_KVM_WALL_CLOCK  0x11
  48 #define MSR_KVM_SYSTEM_TIME 0x12
  49
  50 #ifndef BUS_MCEERR_AR
  51 #define BUS_MCEERR_AR 4
  52 #endif
  53 #ifndef BUS_MCEERR_AO
  54 #define BUS_MCEERR_AO 5
  55 #endif
  56
  57 static int lm_capable_kernel;
  58
  59 #ifdef KVM_CAP_EXT_CPUID
  60
  61 static struct kvm_cpuid2 *try_get_cpuid(KVMState *s, int max)
  62 {
  63     struct kvm_cpuid2 *cpuid;
  64     int r, size;
  65
  66     size = sizeof(*cpuid) + max * sizeof(*cpuid->entries);
  67     cpuid = (struct kvm_cpuid2 *)qemu_mallocz(size);
  68     cpuid->nent = max;
  69     r = kvm_ioctl(s, KVM_GET_SUPPORTED_CPUID, cpuid);
  70     if (r == 0 && cpuid->nent >= max) {
  71         r = -E2BIG;
  72     }
  73     if (r < 0) {
  74         if (r == -E2BIG) {
  75             qemu_free(cpuid);
  76             return NULL;
  77         } else {
  78             fprintf(stderr, "KVM_GET_SUPPORTED_CPUID failed: %s\n",
  79                     strerror(-r));
  80             exit(1);
  81         }
  82     }
  83     return cpuid;
  84 }
  85
  86 uint32_t kvm_arch_get_supported_cpuid(CPUState *env, uint32_t function,
  87                                       uint32_t index, int reg)
  88 {
  89     struct kvm_cpuid2 *cpuid;
  90     int i, max;
  91     uint32_t ret = 0;
  92     uint32_t cpuid_1_edx;
  93
  94     if (!kvm_check_extension(env->kvm_state, KVM_CAP_EXT_CPUID)) {
  95         return -1U;
  96     }
  97
  98     max = 1;
  99     while ((cpuid = try_get_cpuid(env->kvm_state, max)) == NULL) {
 100         max *= 2;
 101     }
 102
 103     for (i = 0; i < cpuid->nent; ++i) {
 104         if (cpuid->entries[i].function == function &&
 105             cpuid->entries[i].index == index) {
 106             switch (reg) {
 107             case R_EAX:
 108                 ret = cpuid->entries[i].eax;
 109                 break;
 110             case R_EBX:
 111                 ret = cpuid->entries[i].ebx;
 112                 break;
 113             case R_ECX:
 114                 ret = cpuid->entries[i].ecx;
 115                 break;
 116             case R_EDX:
 117                 ret = cpuid->entries[i].edx;
 118                 switch (function) {
 119                 case 1:
 120                     /* KVM before 2.6.30 misreports the following features */
 121                     ret |= CPUID_MTRR | CPUID_PAT | CPUID_MCE | CPUID_MCA;
 122                     break;
 123                 case 0x80000001:
 124                     /* On Intel, kvm returns cpuid according to the Intel spec,
 125                      * so add missing bits according to the AMD spec:
 126                      */
 127                     cpuid_1_edx = kvm_arch_get_supported_cpuid(env, 1, 0, R_EDX);
 128                     ret |= cpuid_1_edx & 0x183f7ff;
 129                     break;
 130                 }
 131                 break;
 132             }
 133         }
 134     }
 135
 136     qemu_free(cpuid);
 137
 138     return ret;
 139 }
 140
 141 #else
 142
 143 uint32_t kvm_arch_get_supported_cpuid(CPUState *env, uint32_t function,
 144                                       uint32_t index, int reg)
 145 {
 146     return -1U;
 147 }
 148
 149 #endif
 150
 151 #ifdef CONFIG_KVM_PARA
 152 struct kvm_para_features {
 153     int cap;
 154     int feature;
 155 } para_features[] = {
 156 #ifdef KVM_CAP_CLOCKSOURCE
 157     { KVM_CAP_CLOCKSOURCE, KVM_FEATURE_CLOCKSOURCE },
 158 #endif
 159 #ifdef KVM_CAP_NOP_IO_DELAY
 160     { KVM_CAP_NOP_IO_DELAY, KVM_FEATURE_NOP_IO_DELAY },
 161 #endif
 162 #ifdef KVM_CAP_PV_MMU
 163     { KVM_CAP_PV_MMU, KVM_FEATURE_MMU_OP },
 164 #endif
 165 #ifdef KVM_CAP_ASYNC_PF
 166     { KVM_CAP_ASYNC_PF, KVM_FEATURE_ASYNC_PF },
 167 #endif
 168     { -1, -1 }
 169 };
 170
 171 static int get_para_features(CPUState *env)
 172 {
 173     int i, features = 0;
 174
 175     for (i = 0; i < ARRAY_SIZE(para_features) - 1; i++) {
 176         if (kvm_check_extension(env->kvm_state, para_features[i].cap)) {
 177             features |= (1 << para_features[i].feature);
 178         }
 179     }
 180     return features;
 181 }
 182 #endif
 183
 184 #ifdef KVM_CAP_MCE
 185 static int kvm_get_mce_cap_supported(KVMState *s, uint64_t *mce_cap,
 186                                      int *max_banks)
 187 {
 188     int r;
 189
 190     r = kvm_check_extension(s, KVM_CAP_MCE);
 191     if (r > 0) {
 192         *max_banks = r;
 193         return kvm_ioctl(s, KVM_X86_GET_MCE_CAP_SUPPORTED, mce_cap);
 194     }
 195     return -ENOSYS;
 196 }
 197
 198 static int kvm_setup_mce(CPUState *env, uint64_t *mcg_cap)
 199 {
 200     return kvm_vcpu_ioctl(env, KVM_X86_SETUP_MCE, mcg_cap);
 201 }
 202
 203 static int kvm_set_mce(CPUState *env, struct kvm_x86_mce *m)
 204 {
 205     return kvm_vcpu_ioctl(env, KVM_X86_SET_MCE, m);
 206 }
 207
 208 static int kvm_get_msr(CPUState *env, struct kvm_msr_entry *msrs, int n)
 209 {
 210     struct kvm_msrs *kmsrs = qemu_malloc(sizeof *kmsrs + n * sizeof *msrs);
 211     int r;
 212
 213     kmsrs->nmsrs = n;
 214     memcpy(kmsrs->entries, msrs, n * sizeof *msrs);
 215     r = kvm_vcpu_ioctl(env, KVM_GET_MSRS, kmsrs);
 216     memcpy(msrs, kmsrs->entries, n * sizeof *msrs);
 217     free(kmsrs);
 218     return r;
 219 }
 220
 221 /* FIXME: kill this and kvm_get_msr, use env->mcg_status instead */
 222 static int kvm_mce_in_progress(CPUState *env)
 223 {
 224     struct kvm_msr_entry msr_mcg_status = {
 225         .index = MSR_MCG_STATUS,
 226     };
 227     int r;
 228
 229     r = kvm_get_msr(env, &msr_mcg_status, 1);
 230     if (r == -1 || r == 0) {
 231         fprintf(stderr, "Failed to get MCE status\n");
 232         return 0;
 233     }
 234     return !!(msr_mcg_status.data & MCG_STATUS_MCIP);
 235 }
 236
 237 struct kvm_x86_mce_data
 238 {
 239     CPUState *env;
 240     struct kvm_x86_mce *mce;
 241     int abort_on_error;
 242 };
 243
 244 static void kvm_do_inject_x86_mce(void *_data)
 245 {
 246     struct kvm_x86_mce_data *data = _data;
 247     int r;
 248
 249     /* If there is an MCE exception being processed, ignore this SRAO MCE */
 250     if ((data->env->mcg_cap & MCG_SER_P) &&
 251         !(data->mce->status & MCI_STATUS_AR)) {
 252         if (kvm_mce_in_progress(data->env)) {
 253             return;
 254         }
 255     }
 256
 257     r = kvm_set_mce(data->env, data->mce);
 258     if (r < 0) {
 259         perror("kvm_set_mce FAILED");
 260         if (data->abort_on_error) {
 261             abort();
 262         }
 263     }
 264 }
 265
 266 static void kvm_inject_x86_mce_on(CPUState *env, struct kvm_x86_mce *mce,
 267                                   int flag)
 268 {
 269     struct kvm_x86_mce_data data = {
 270         .env = env,
 271         .mce = mce,
 272         .abort_on_error = (flag & ABORT_ON_ERROR),
 273     };
 274
 275     if (!env->mcg_cap) {
 276         fprintf(stderr, "MCE support is not enabled!\n");
 277         return;
 278     }
 279
 280     run_on_cpu(env, kvm_do_inject_x86_mce, &data);
 281 }
 282
 283 static void kvm_mce_broadcast_rest(CPUState *env);
 284 #endif
 285
 286 void kvm_inject_x86_mce(CPUState *cenv, int bank, uint64_t status,
 287                         uint64_t mcg_status, uint64_t addr, uint64_t misc,
 288                         int flag)
 289 {
 290 #ifdef KVM_CAP_MCE
 291     struct kvm_x86_mce mce = {
 292         .bank = bank,
 293         .status = status,
 294         .mcg_status = mcg_status,
 295         .addr = addr,
 296         .misc = misc,
 297     };
 298
 299     if (flag & MCE_BROADCAST) {
 300         kvm_mce_broadcast_rest(cenv);
 301     }
 302
 303     kvm_inject_x86_mce_on(cenv, &mce, flag);
 304 #else
 305     if (flag & ABORT_ON_ERROR) {
 306         abort();
 307     }
 308 #endif
 309 }
 310
 311 int kvm_arch_init_vcpu(CPUState *env)
 312 {
 313     struct {
 314         struct kvm_cpuid2 cpuid;
 315         struct kvm_cpuid_entry2 entries[100];
 316     } __attribute__((packed)) cpuid_data;
 317     uint32_t limit, i, j, cpuid_i;
 318     uint32_t unused;
 319     struct kvm_cpuid_entry2 *c;
 320 #ifdef KVM_CPUID_SIGNATURE
 321     uint32_t signature[3];
 322 #endif
 323
 324     env->mp_state = KVM_MP_STATE_RUNNABLE;
 325
 326     env->cpuid_features &= kvm_arch_get_supported_cpuid(env, 1, 0, R_EDX);
 327
 328     i = env->cpuid_ext_features & CPUID_EXT_HYPERVISOR;
 329     env->cpuid_ext_features &= kvm_arch_get_supported_cpuid(env, 1, 0, R_ECX);
 330     env->cpuid_ext_features |= i;
 331
 332     env->cpuid_ext2_features &= kvm_arch_get_supported_cpuid(env, 0x80000001,
 333                                                              0, R_EDX);
 334     env->cpuid_ext3_features &= kvm_arch_get_supported_cpuid(env, 0x80000001,
 335                                                              0, R_ECX);
 336     env->cpuid_svm_features  &= kvm_arch_get_supported_cpuid(env, 0x8000000A,
 337                                                              0, R_EDX);
 338
 339
 340     cpuid_i = 0;
 341
 342 #ifdef CONFIG_KVM_PARA
 343     /* Paravirtualization CPUIDs */
 344     memcpy(signature, "KVMKVMKVM\0\0\0", 12);
 345     c = &cpuid_data.entries[cpuid_i++];
 346     memset(c, 0, sizeof(*c));
 347     c->function = KVM_CPUID_SIGNATURE;
 348     c->eax = 0;
 349     c->ebx = signature[0];
 350     c->ecx = signature[1];
 351     c->edx = signature[2];
 352
 353     c = &cpuid_data.entries[cpuid_i++];
 354     memset(c, 0, sizeof(*c));
 355     c->function = KVM_CPUID_FEATURES;
 356     c->eax = env->cpuid_kvm_features & get_para_features(env);
 357 #endif
 358
 359     cpu_x86_cpuid(env, 0, 0, &limit, &unused, &unused, &unused);
 360
 361     for (i = 0; i <= limit; i++) {
 362         c = &cpuid_data.entries[cpuid_i++];
 363
 364         switch (i) {
 365         case 2: {
 366             /* Keep reading function 2 till all the input is received */
 367             int times;
 368
 369             c->function = i;
 370             c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC |
 371                        KVM_CPUID_FLAG_STATE_READ_NEXT;
 372             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 373             times = c->eax & 0xff;
 374
 375             for (j = 1; j < times; ++j) {
 376                 c = &cpuid_data.entries[cpuid_i++];
 377                 c->function = i;
 378                 c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC;
 379                 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 380             }
 381             break;
 382         }
 383         case 4:
 384         case 0xb:
 385         case 0xd:
 386             for (j = 0; ; j++) {
 387                 c->function = i;
 388                 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 389                 c->index = j;
 390                 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
 391
 392                 if (i == 4 && c->eax == 0) {
 393                     break;
 394                 }
 395                 if (i == 0xb && !(c->ecx & 0xff00)) {
 396                     break;
 397                 }
 398                 if (i == 0xd && c->eax == 0) {
 399                     break;
 400                 }
 401                 c = &cpuid_data.entries[cpuid_i++];
 402             }
 403             break;
 404         default:
 405             c->function = i;
 406             c->flags = 0;
 407             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 408             break;
 409         }
 410     }
 411     cpu_x86_cpuid(env, 0x80000000, 0, &limit, &unused, &unused, &unused);
 412
 413     for (i = 0x80000000; i <= limit; i++) {
 414         c = &cpuid_data.entries[cpuid_i++];
 415
 416         c->function = i;
 417         c->flags = 0;
 418         cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 419     }
 420
 421     cpuid_data.cpuid.nent = cpuid_i;
 422
 423 #ifdef KVM_CAP_MCE
 424     if (((env->cpuid_version >> 8)&0xF) >= 6
 425         && (env->cpuid_features&(CPUID_MCE|CPUID_MCA)) == (CPUID_MCE|CPUID_MCA)
 426         && kvm_check_extension(env->kvm_state, KVM_CAP_MCE) > 0) {
 427         uint64_t mcg_cap;
 428         int banks;
 429
 430         if (kvm_get_mce_cap_supported(env->kvm_state, &mcg_cap, &banks)) {
 431             perror("kvm_get_mce_cap_supported FAILED");
 432         } else {
 433             if (banks > MCE_BANKS_DEF)
 434                 banks = MCE_BANKS_DEF;
 435             mcg_cap &= MCE_CAP_DEF;
 436             mcg_cap |= banks;
 437             if (kvm_setup_mce(env, &mcg_cap)) {
 438                 perror("kvm_setup_mce FAILED");
 439             } else {
 440                 env->mcg_cap = mcg_cap;
 441             }
 442         }
 443     }
 444 #endif
 445
 446     return kvm_vcpu_ioctl(env, KVM_SET_CPUID2, &cpuid_data);
 447 }
 448
 449 void kvm_arch_reset_vcpu(CPUState *env)
 450 {
 451     env->exception_injected = -1;
 452     env->interrupt_injected = -1;
 453     env->nmi_injected = 0;
 454     env->nmi_pending = 0;
 455     if (kvm_irqchip_in_kernel()) {
 456         env->mp_state = cpu_is_bsp(env) ? KVM_MP_STATE_RUNNABLE :
 457                                           KVM_MP_STATE_UNINITIALIZED;
 458     } else {
 459         env->mp_state = KVM_MP_STATE_RUNNABLE;
 460     }
 461 }
 462
 463 int has_msr_star;
 464 int has_msr_hsave_pa;
 465
 466 static void kvm_supported_msrs(CPUState *env)
 467 {
 468     static int kvm_supported_msrs;
 469     int ret;
 470
 471     /* first time */
 472     if (kvm_supported_msrs == 0) {
 473         struct kvm_msr_list msr_list, *kvm_msr_list;
 474
 475         kvm_supported_msrs = -1;
 476
 477         /* Obtain MSR list from KVM.  These are the MSRs that we must
 478          * save/restore */
 479         msr_list.nmsrs = 0;
 480         ret = kvm_ioctl(env->kvm_state, KVM_GET_MSR_INDEX_LIST, &msr_list);
 481         if (ret < 0 && ret != -E2BIG) {
 482             return;
 483         }
 484         /* Old kernel modules had a bug and could write beyond the provided
 485            memory. Allocate at least a safe amount of 1K. */
 486         kvm_msr_list = qemu_mallocz(MAX(1024, sizeof(msr_list) +
 487                                               msr_list.nmsrs *
 488                                               sizeof(msr_list.indices[0])));
 489
 490         kvm_msr_list->nmsrs = msr_list.nmsrs;
 491         ret = kvm_ioctl(env->kvm_state, KVM_GET_MSR_INDEX_LIST, kvm_msr_list);
 492         if (ret >= 0) {
 493             int i;
 494
 495             for (i = 0; i < kvm_msr_list->nmsrs; i++) {
 496                 if (kvm_msr_list->indices[i] == MSR_STAR) {
 497                     has_msr_star = 1;
 498                     continue;
 499                 }
 500                 if (kvm_msr_list->indices[i] == MSR_VM_HSAVE_PA) {
 501                     has_msr_hsave_pa = 1;
 502                     continue;
 503                 }
 504             }
 505         }
 506
 507         free(kvm_msr_list);
 508     }
 509
 510     return;
 511 }
 512
 513 static int kvm_has_msr_hsave_pa(CPUState *env)
 514 {
 515     kvm_supported_msrs(env);
 516     return has_msr_hsave_pa;
 517 }
 518
 519 static int kvm_has_msr_star(CPUState *env)
 520 {
 521     kvm_supported_msrs(env);
 522     return has_msr_star;
 523 }
 524
 525 static int kvm_init_identity_map_page(KVMState *s)
 526 {
 527 #ifdef KVM_CAP_SET_IDENTITY_MAP_ADDR
 528     int ret;
 529     uint64_t addr = 0xfffbc000;
 530
 531     if (!kvm_check_extension(s, KVM_CAP_SET_IDENTITY_MAP_ADDR)) {
 532         return 0;
 533     }
 534
 535     ret = kvm_vm_ioctl(s, KVM_SET_IDENTITY_MAP_ADDR, &addr);
 536     if (ret < 0) {
 537         fprintf(stderr, "kvm_set_identity_map_addr: %s\n", strerror(ret));
 538         return ret;
 539     }
 540 #endif
 541     return 0;
 542 }
 543
 544 int kvm_arch_init(KVMState *s, int smp_cpus)
 545 {
 546     int ret;
 547
 548     struct utsname utsname;
 549
 550     uname(&utsname);
 551     lm_capable_kernel = strcmp(utsname.machine, "x86_64") == 0;
 552
 553     /* create vm86 tss.  KVM uses vm86 mode to emulate 16-bit code
 554      * directly.  In order to use vm86 mode, a TSS is needed.  Since this
 555      * must be part of guest physical memory, we need to allocate it.  Older
 556      * versions of KVM just assumed that it would be at the end of physical
 557      * memory but that doesn't work with more than 4GB of memory.  We simply
 558      * refuse to work with those older versions of KVM. */
 559     ret = kvm_check_extension(s, KVM_CAP_SET_TSS_ADDR);
 560     if (ret <= 0) {
 561         fprintf(stderr, "kvm does not support KVM_CAP_SET_TSS_ADDR\n");
 562         return ret;
 563     }
 564
 565     /* this address is 3 pages before the bios, and the bios should present
 566      * as unavaible memory.  FIXME, need to ensure the e820 map deals with
 567      * this?
 568      */
 569     /*
 570      * Tell fw_cfg to notify the BIOS to reserve the range.
 571      */
 572     if (e820_add_entry(0xfffbc000, 0x4000, E820_RESERVED) < 0) {
 573         perror("e820_add_entry() table is full");
 574         exit(1);
 575     }
 576     ret = kvm_vm_ioctl(s, KVM_SET_TSS_ADDR, 0xfffbd000);
 577     if (ret < 0) {
 578         return ret;
 579     }
 580
 581     return kvm_init_identity_map_page(s);
 582 }
 583
 584 static void set_v8086_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
 585 {
 586     lhs->selector = rhs->selector;
 587     lhs->base = rhs->base;
 588     lhs->limit = rhs->limit;
 589     lhs->type = 3;
 590     lhs->present = 1;
 591     lhs->dpl = 3;
 592     lhs->db = 0;
 593     lhs->s = 1;
 594     lhs->l = 0;
 595     lhs->g = 0;
 596     lhs->avl = 0;
 597     lhs->unusable = 0;
 598 }
 599
 600 static void set_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
 601 {
 602     unsigned flags = rhs->flags;
 603     lhs->selector = rhs->selector;
 604     lhs->base = rhs->base;
 605     lhs->limit = rhs->limit;
 606     lhs->type = (flags >> DESC_TYPE_SHIFT) & 15;
 607     lhs->present = (flags & DESC_P_MASK) != 0;
 608     lhs->dpl = (flags >> DESC_DPL_SHIFT) & 3;
 609     lhs->db = (flags >> DESC_B_SHIFT) & 1;
 610     lhs->s = (flags & DESC_S_MASK) != 0;
 611     lhs->l = (flags >> DESC_L_SHIFT) & 1;
 612     lhs->g = (flags & DESC_G_MASK) != 0;
 613     lhs->avl = (flags & DESC_AVL_MASK) != 0;
 614     lhs->unusable = 0;
 615 }
 616
 617 static void get_seg(SegmentCache *lhs, const struct kvm_segment *rhs)
 618 {
 619     lhs->selector = rhs->selector;
 620     lhs->base = rhs->base;
 621     lhs->limit = rhs->limit;
 622     lhs->flags = (rhs->type << DESC_TYPE_SHIFT) |
 623                  (rhs->present * DESC_P_MASK) |
 624                  (rhs->dpl << DESC_DPL_SHIFT) |
 625                  (rhs->db << DESC_B_SHIFT) |
 626                  (rhs->s * DESC_S_MASK) |
 627                  (rhs->l << DESC_L_SHIFT) |
 628                  (rhs->g * DESC_G_MASK) |
 629                  (rhs->avl * DESC_AVL_MASK);
 630 }
 631
 632 static void kvm_getput_reg(__u64 *kvm_reg, target_ulong *qemu_reg, int set)
 633 {
 634     if (set) {
 635         *kvm_reg = *qemu_reg;
 636     } else {
 637         *qemu_reg = *kvm_reg;
 638     }
 639 }
 640
 641 static int kvm_getput_regs(CPUState *env, int set)
 642 {
 643     struct kvm_regs regs;
 644     int ret = 0;
 645
 646     if (!set) {
 647         ret = kvm_vcpu_ioctl(env, KVM_GET_REGS, &regs);
 648         if (ret < 0) {
 649             return ret;
 650         }
 651     }
 652
 653     kvm_getput_reg(&regs.rax, &env->regs[R_EAX], set);
 654     kvm_getput_reg(&regs.rbx, &env->regs[R_EBX], set);
 655     kvm_getput_reg(&regs.rcx, &env->regs[R_ECX], set);
 656     kvm_getput_reg(&regs.rdx, &env->regs[R_EDX], set);
 657     kvm_getput_reg(&regs.rsi, &env->regs[R_ESI], set);
 658     kvm_getput_reg(&regs.rdi, &env->regs[R_EDI], set);
 659     kvm_getput_reg(&regs.rsp, &env->regs[R_ESP], set);
 660     kvm_getput_reg(&regs.rbp, &env->regs[R_EBP], set);
 661 #ifdef TARGET_X86_64
 662     kvm_getput_reg(&regs.r8, &env->regs[8], set);
 663     kvm_getput_reg(&regs.r9, &env->regs[9], set);
 664     kvm_getput_reg(&regs.r10, &env->regs[10], set);
 665     kvm_getput_reg(&regs.r11, &env->regs[11], set);
 666     kvm_getput_reg(&regs.r12, &env->regs[12], set);
 667     kvm_getput_reg(&regs.r13, &env->regs[13], set);
 668     kvm_getput_reg(&regs.r14, &env->regs[14], set);
 669     kvm_getput_reg(&regs.r15, &env->regs[15], set);
 670 #endif
 671
 672     kvm_getput_reg(&regs.rflags, &env->eflags, set);
 673     kvm_getput_reg(&regs.rip, &env->eip, set);
 674
 675     if (set) {
 676         ret = kvm_vcpu_ioctl(env, KVM_SET_REGS, &regs);
 677     }
 678
 679     return ret;
 680 }
 681
 682 static int kvm_put_fpu(CPUState *env)
 683 {
 684     struct kvm_fpu fpu;
 685     int i;
 686
 687     memset(&fpu, 0, sizeof fpu);
 688     fpu.fsw = env->fpus & ~(7 << 11);
 689     fpu.fsw |= (env->fpstt & 7) << 11;
 690     fpu.fcw = env->fpuc;
 691     for (i = 0; i < 8; ++i) {
 692         fpu.ftwx |= (!env->fptags[i]) << i;
 693     }
 694     memcpy(fpu.fpr, env->fpregs, sizeof env->fpregs);
 695     memcpy(fpu.xmm, env->xmm_regs, sizeof env->xmm_regs);
 696     fpu.mxcsr = env->mxcsr;
 697
 698     return kvm_vcpu_ioctl(env, KVM_SET_FPU, &fpu);
 699 }
 700
 701 #ifdef KVM_CAP_XSAVE
 702 #define XSAVE_CWD_RIP     2
 703 #define XSAVE_CWD_RDP     4
 704 #define XSAVE_MXCSR       6
 705 #define XSAVE_ST_SPACE    8
 706 #define XSAVE_XMM_SPACE   40
 707 #define XSAVE_XSTATE_BV   128
 708 #define XSAVE_YMMH_SPACE  144
 709 #endif
 710
 711 static int kvm_put_xsave(CPUState *env)
 712 {
 713 #ifdef KVM_CAP_XSAVE
 714     int i, r;
 715     struct kvm_xsave* xsave;
 716     uint16_t cwd, swd, twd, fop;
 717
 718     if (!kvm_has_xsave()) {
 719         return kvm_put_fpu(env);
 720     }
 721
 722     xsave = qemu_memalign(4096, sizeof(struct kvm_xsave));
 723     memset(xsave, 0, sizeof(struct kvm_xsave));
 724     cwd = swd = twd = fop = 0;
 725     swd = env->fpus & ~(7 << 11);
 726     swd |= (env->fpstt & 7) << 11;
 727     cwd = env->fpuc;
 728     for (i = 0; i < 8; ++i) {
 729         twd |= (!env->fptags[i]) << i;
 730     }
 731     xsave->region[0] = (uint32_t)(swd << 16) + cwd;
 732     xsave->region[1] = (uint32_t)(fop << 16) + twd;
 733     memcpy(&xsave->region[XSAVE_ST_SPACE], env->fpregs,
 734             sizeof env->fpregs);
 735     memcpy(&xsave->region[XSAVE_XMM_SPACE], env->xmm_regs,
 736             sizeof env->xmm_regs);
 737     xsave->region[XSAVE_MXCSR] = env->mxcsr;
 738     *(uint64_t *)&xsave->region[XSAVE_XSTATE_BV] = env->xstate_bv;
 739     memcpy(&xsave->region[XSAVE_YMMH_SPACE], env->ymmh_regs,
 740             sizeof env->ymmh_regs);
 741     r = kvm_vcpu_ioctl(env, KVM_SET_XSAVE, xsave);
 742     qemu_free(xsave);
 743     return r;
 744 #else
 745     return kvm_put_fpu(env);
 746 #endif
 747 }
 748
 749 static int kvm_put_xcrs(CPUState *env)
 750 {
 751 #ifdef KVM_CAP_XCRS
 752     struct kvm_xcrs xcrs;
 753
 754     if (!kvm_has_xcrs()) {
 755         return 0;
 756     }
 757
 758     xcrs.nr_xcrs = 1;
 759     xcrs.flags = 0;
 760     xcrs.xcrs[0].xcr = 0;
 761     xcrs.xcrs[0].value = env->xcr0;
 762     return kvm_vcpu_ioctl(env, KVM_SET_XCRS, &xcrs);
 763 #else
 764     return 0;
 765 #endif
 766 }
 767
 768 static int kvm_put_sregs(CPUState *env)
 769 {
 770     struct kvm_sregs sregs;
 771
 772     memset(sregs.interrupt_bitmap, 0, sizeof(sregs.interrupt_bitmap));
 773     if (env->interrupt_injected >= 0) {
 774         sregs.interrupt_bitmap[env->interrupt_injected / 64] |=
 775                 (uint64_t)1 << (env->interrupt_injected % 64);
 776     }
 777
 778     if ((env->eflags & VM_MASK)) {
 779         set_v8086_seg(&sregs.cs, &env->segs[R_CS]);
 780         set_v8086_seg(&sregs.ds, &env->segs[R_DS]);
 781         set_v8086_seg(&sregs.es, &env->segs[R_ES]);
 782         set_v8086_seg(&sregs.fs, &env->segs[R_FS]);
 783         set_v8086_seg(&sregs.gs, &env->segs[R_GS]);
 784         set_v8086_seg(&sregs.ss, &env->segs[R_SS]);
 785     } else {
 786         set_seg(&sregs.cs, &env->segs[R_CS]);
 787         set_seg(&sregs.ds, &env->segs[R_DS]);
 788         set_seg(&sregs.es, &env->segs[R_ES]);
 789         set_seg(&sregs.fs, &env->segs[R_FS]);
 790         set_seg(&sregs.gs, &env->segs[R_GS]);
 791         set_seg(&sregs.ss, &env->segs[R_SS]);
 792     }
 793
 794     set_seg(&sregs.tr, &env->tr);
 795     set_seg(&sregs.ldt, &env->ldt);
 796
 797     sregs.idt.limit = env->idt.limit;
 798     sregs.idt.base = env->idt.base;
 799     sregs.gdt.limit = env->gdt.limit;
 800     sregs.gdt.base = env->gdt.base;
 801
 802     sregs.cr0 = env->cr[0];
 803     sregs.cr2 = env->cr[2];
 804     sregs.cr3 = env->cr[3];
 805     sregs.cr4 = env->cr[4];
 806
 807     sregs.cr8 = cpu_get_apic_tpr(env->apic_state);
 808     sregs.apic_base = cpu_get_apic_base(env->apic_state);
 809
 810     sregs.efer = env->efer;
 811
 812     return kvm_vcpu_ioctl(env, KVM_SET_SREGS, &sregs);
 813 }
 814
 815 static void kvm_msr_entry_set(struct kvm_msr_entry *entry,
 816                               uint32_t index, uint64_t value)
 817 {
 818     entry->index = index;
 819     entry->data = value;
 820 }
 821
 822 static int kvm_put_msrs(CPUState *env, int level)
 823 {
 824     struct {
 825         struct kvm_msrs info;
 826         struct kvm_msr_entry entries[100];
 827     } msr_data;
 828     struct kvm_msr_entry *msrs = msr_data.entries;
 829     int n = 0;
 830
 831     kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_CS, env->sysenter_cs);
 832     kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_ESP, env->sysenter_esp);
 833     kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_EIP, env->sysenter_eip);
 834     if (kvm_has_msr_star(env)) {
 835         kvm_msr_entry_set(&msrs[n++], MSR_STAR, env->star);
 836     }
 837     if (kvm_has_msr_hsave_pa(env)) {
 838         kvm_msr_entry_set(&msrs[n++], MSR_VM_HSAVE_PA, env->vm_hsave);
 839     }
 840 #ifdef TARGET_X86_64
 841     if (lm_capable_kernel) {
 842         kvm_msr_entry_set(&msrs[n++], MSR_CSTAR, env->cstar);
 843         kvm_msr_entry_set(&msrs[n++], MSR_KERNELGSBASE, env->kernelgsbase);
 844         kvm_msr_entry_set(&msrs[n++], MSR_FMASK, env->fmask);
 845         kvm_msr_entry_set(&msrs[n++], MSR_LSTAR, env->lstar);
 846     }
 847 #endif
 848     if (level == KVM_PUT_FULL_STATE) {
 849         /*
 850          * KVM is yet unable to synchronize TSC values of multiple VCPUs on
 851          * writeback. Until this is fixed, we only write the offset to SMP
 852          * guests after migration, desynchronizing the VCPUs, but avoiding
 853          * huge jump-backs that would occur without any writeback at all.
 854          */
 855         if (smp_cpus == 1 || env->tsc != 0) {
 856             kvm_msr_entry_set(&msrs[n++], MSR_IA32_TSC, env->tsc);
 857         }
 858         kvm_msr_entry_set(&msrs[n++], MSR_KVM_SYSTEM_TIME,
 859                           env->system_time_msr);
 860         kvm_msr_entry_set(&msrs[n++], MSR_KVM_WALL_CLOCK, env->wall_clock_msr);
 861 #ifdef KVM_CAP_ASYNC_PF
 862         kvm_msr_entry_set(&msrs[n++], MSR_KVM_ASYNC_PF_EN, env->async_pf_en_msr);
 863 #endif
 864     }
 865 #ifdef KVM_CAP_MCE
 866     if (env->mcg_cap) {
 867         int i;
 868
 869         if (level == KVM_PUT_RESET_STATE) {
 870             kvm_msr_entry_set(&msrs[n++], MSR_MCG_STATUS, env->mcg_status);
 871         } else if (level == KVM_PUT_FULL_STATE) {
 872             kvm_msr_entry_set(&msrs[n++], MSR_MCG_STATUS, env->mcg_status);
 873             kvm_msr_entry_set(&msrs[n++], MSR_MCG_CTL, env->mcg_ctl);
 874             for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++) {
 875                 kvm_msr_entry_set(&msrs[n++], MSR_MC0_CTL + i, env->mce_banks[i]);
 876             }
 877         }
 878     }
 879 #endif
 880
 881     msr_data.info.nmsrs = n;
 882
 883     return kvm_vcpu_ioctl(env, KVM_SET_MSRS, &msr_data);
 884
 885 }
 886
 887
 888 static int kvm_get_fpu(CPUState *env)
 889 {
 890     struct kvm_fpu fpu;
 891     int i, ret;
 892
 893     ret = kvm_vcpu_ioctl(env, KVM_GET_FPU, &fpu);
 894     if (ret < 0) {
 895         return ret;
 896     }
 897
 898     env->fpstt = (fpu.fsw >> 11) & 7;
 899     env->fpus = fpu.fsw;
 900     env->fpuc = fpu.fcw;
 901     for (i = 0; i < 8; ++i) {
 902         env->fptags[i] = !((fpu.ftwx >> i) & 1);
 903     }
 904     memcpy(env->fpregs, fpu.fpr, sizeof env->fpregs);
 905     memcpy(env->xmm_regs, fpu.xmm, sizeof env->xmm_regs);
 906     env->mxcsr = fpu.mxcsr;
 907
 908     return 0;
 909 }
 910
 911 static int kvm_get_xsave(CPUState *env)
 912 {
 913 #ifdef KVM_CAP_XSAVE
 914     struct kvm_xsave* xsave;
 915     int ret, i;
 916     uint16_t cwd, swd, twd, fop;
 917
 918     if (!kvm_has_xsave()) {
 919         return kvm_get_fpu(env);
 920     }
 921
 922     xsave = qemu_memalign(4096, sizeof(struct kvm_xsave));
 923     ret = kvm_vcpu_ioctl(env, KVM_GET_XSAVE, xsave);
 924     if (ret < 0) {
 925         qemu_free(xsave);
 926         return ret;
 927     }
 928
 929     cwd = (uint16_t)xsave->region[0];
 930     swd = (uint16_t)(xsave->region[0] >> 16);
 931     twd = (uint16_t)xsave->region[1];
 932     fop = (uint16_t)(xsave->region[1] >> 16);
 933     env->fpstt = (swd >> 11) & 7;
 934     env->fpus = swd;
 935     env->fpuc = cwd;
 936     for (i = 0; i < 8; ++i) {
 937         env->fptags[i] = !((twd >> i) & 1);
 938     }
 939     env->mxcsr = xsave->region[XSAVE_MXCSR];
 940     memcpy(env->fpregs, &xsave->region[XSAVE_ST_SPACE],
 941             sizeof env->fpregs);
 942     memcpy(env->xmm_regs, &xsave->region[XSAVE_XMM_SPACE],
 943             sizeof env->xmm_regs);
 944     env->xstate_bv = *(uint64_t *)&xsave->region[XSAVE_XSTATE_BV];
 945     memcpy(env->ymmh_regs, &xsave->region[XSAVE_YMMH_SPACE],
 946             sizeof env->ymmh_regs);
 947     qemu_free(xsave);
 948     return 0;
 949 #else
 950     return kvm_get_fpu(env);
 951 #endif
 952 }
 953
 954 static int kvm_get_xcrs(CPUState *env)
 955 {
 956 #ifdef KVM_CAP_XCRS
 957     int i, ret;
 958     struct kvm_xcrs xcrs;
 959
 960     if (!kvm_has_xcrs()) {
 961         return 0;
 962     }
 963
 964     ret = kvm_vcpu_ioctl(env, KVM_GET_XCRS, &xcrs);
 965     if (ret < 0) {
 966         return ret;
 967     }
 968
 969     for (i = 0; i < xcrs.nr_xcrs; i++) {
 970         /* Only support xcr0 now */
 971         if (xcrs.xcrs[0].xcr == 0) {
 972             env->xcr0 = xcrs.xcrs[0].value;
 973             break;
 974         }
 975     }
 976     return 0;
 977 #else
 978     return 0;
 979 #endif
 980 }
 981
 982 static int kvm_get_sregs(CPUState *env)
 983 {
 984     struct kvm_sregs sregs;
 985     uint32_t hflags;
 986     int bit, i, ret;
 987
 988     ret = kvm_vcpu_ioctl(env, KVM_GET_SREGS, &sregs);
 989     if (ret < 0) {
 990         return ret;
 991     }
 992
 993     /* There can only be one pending IRQ set in the bitmap at a time, so try
 994        to find it and save its number instead (-1 for none). */
 995     env->interrupt_injected = -1;
 996     for (i = 0; i < ARRAY_SIZE(sregs.interrupt_bitmap); i++) {
 997         if (sregs.interrupt_bitmap[i]) {
 998             bit = ctz64(sregs.interrupt_bitmap[i]);
 999             env->interrupt_injected = i * 64 + bit;
1000             break;
1001         }
1002     }
1003
1004     get_seg(&env->segs[R_CS], &sregs.cs);
1005     get_seg(&env->segs[R_DS], &sregs.ds);
1006     get_seg(&env->segs[R_ES], &sregs.es);
1007     get_seg(&env->segs[R_FS], &sregs.fs);
1008     get_seg(&env->segs[R_GS], &sregs.gs);
1009     get_seg(&env->segs[R_SS], &sregs.ss);
1010
1011     get_seg(&env->tr, &sregs.tr);
1012     get_seg(&env->ldt, &sregs.ldt);
1013
1014     env->idt.limit = sregs.idt.limit;
1015     env->idt.base = sregs.idt.base;
1016     env->gdt.limit = sregs.gdt.limit;
1017     env->gdt.base = sregs.gdt.base;
1018
1019     env->cr[0] = sregs.cr0;
1020     env->cr[2] = sregs.cr2;
1021     env->cr[3] = sregs.cr3;
1022     env->cr[4] = sregs.cr4;
1023
1024     cpu_set_apic_base(env->apic_state, sregs.apic_base);
1025
1026     env->efer = sregs.efer;
1027     //cpu_set_apic_tpr(env->apic_state, sregs.cr8);
1028
1029 #define HFLAG_COPY_MASK \
1030     ~( HF_CPL_MASK | HF_PE_MASK | HF_MP_MASK | HF_EM_MASK | \
1031        HF_TS_MASK | HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK | \
1032        HF_OSFXSR_MASK | HF_LMA_MASK | HF_CS32_MASK | \
1033        HF_SS32_MASK | HF_CS64_MASK | HF_ADDSEG_MASK)
1034
1035     hflags = (env->segs[R_CS].flags >> DESC_DPL_SHIFT) & HF_CPL_MASK;
1036     hflags |= (env->cr[0] & CR0_PE_MASK) << (HF_PE_SHIFT - CR0_PE_SHIFT);
1037     hflags |= (env->cr[0] << (HF_MP_SHIFT - CR0_MP_SHIFT)) &
1038                 (HF_MP_MASK | HF_EM_MASK | HF_TS_MASK);
1039     hflags |= (env->eflags & (HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK));
1040     hflags |= (env->cr[4] & CR4_OSFXSR_MASK) <<
1041                 (HF_OSFXSR_SHIFT - CR4_OSFXSR_SHIFT);
1042
1043     if (env->efer & MSR_EFER_LMA) {
1044         hflags |= HF_LMA_MASK;
1045     }
1046
1047     if ((hflags & HF_LMA_MASK) && (env->segs[R_CS].flags & DESC_L_MASK)) {
1048         hflags |= HF_CS32_MASK | HF_SS32_MASK | HF_CS64_MASK;
1049     } else {
1050         hflags |= (env->segs[R_CS].flags & DESC_B_MASK) >>
1051                     (DESC_B_SHIFT - HF_CS32_SHIFT);
1052         hflags |= (env->segs[R_SS].flags & DESC_B_MASK) >>
1053                     (DESC_B_SHIFT - HF_SS32_SHIFT);
1054         if (!(env->cr[0] & CR0_PE_MASK) || (env->eflags & VM_MASK) ||
1055             !(hflags & HF_CS32_MASK)) {
1056             hflags |= HF_ADDSEG_MASK;
1057         } else {
1058             hflags |= ((env->segs[R_DS].base | env->segs[R_ES].base |
1059                         env->segs[R_SS].base) != 0) << HF_ADDSEG_SHIFT;
1060         }
1061     }
1062     env->hflags = (env->hflags & HFLAG_COPY_MASK) | hflags;
1063
1064     return 0;
1065 }
1066
1067 static int kvm_get_msrs(CPUState *env)
1068 {
1069     struct {
1070         struct kvm_msrs info;
1071         struct kvm_msr_entry entries[100];
1072     } msr_data;
1073     struct kvm_msr_entry *msrs = msr_data.entries;
1074     int ret, i, n;
1075
1076     n = 0;
1077     msrs[n++].index = MSR_IA32_SYSENTER_CS;
1078     msrs[n++].index = MSR_IA32_SYSENTER_ESP;
1079     msrs[n++].index = MSR_IA32_SYSENTER_EIP;
1080     if (kvm_has_msr_star(env)) {
1081         msrs[n++].index = MSR_STAR;
1082     }
1083     if (kvm_has_msr_hsave_pa(env)) {
1084         msrs[n++].index = MSR_VM_HSAVE_PA;
1085     }
1086     msrs[n++].index = MSR_IA32_TSC;
1087 #ifdef TARGET_X86_64
1088     if (lm_capable_kernel) {
1089         msrs[n++].index = MSR_CSTAR;
1090         msrs[n++].index = MSR_KERNELGSBASE;
1091         msrs[n++].index = MSR_FMASK;
1092         msrs[n++].index = MSR_LSTAR;
1093     }
1094 #endif
1095     msrs[n++].index = MSR_KVM_SYSTEM_TIME;
1096     msrs[n++].index = MSR_KVM_WALL_CLOCK;
1097 #ifdef KVM_CAP_ASYNC_PF
1098     msrs[n++].index = MSR_KVM_ASYNC_PF_EN;
1099 #endif
1100
1101 #ifdef KVM_CAP_MCE
1102     if (env->mcg_cap) {
1103         msrs[n++].index = MSR_MCG_STATUS;
1104         msrs[n++].index = MSR_MCG_CTL;
1105         for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++) {
1106             msrs[n++].index = MSR_MC0_CTL + i;
1107         }
1108     }
1109 #endif
1110
1111     msr_data.info.nmsrs = n;
1112     ret = kvm_vcpu_ioctl(env, KVM_GET_MSRS, &msr_data);
1113     if (ret < 0) {
1114         return ret;
1115     }
1116
1117     for (i = 0; i < ret; i++) {
1118         switch (msrs[i].index) {
1119         case MSR_IA32_SYSENTER_CS:
1120             env->sysenter_cs = msrs[i].data;
1121             break;
1122         case MSR_IA32_SYSENTER_ESP:
1123             env->sysenter_esp = msrs[i].data;
1124             break;
1125         case MSR_IA32_SYSENTER_EIP:
1126             env->sysenter_eip = msrs[i].data;
1127             break;
1128         case MSR_STAR:
1129             env->star = msrs[i].data;
1130             break;
1131 #ifdef TARGET_X86_64
1132         case MSR_CSTAR:
1133             env->cstar = msrs[i].data;
1134             break;
1135         case MSR_KERNELGSBASE:
1136             env->kernelgsbase = msrs[i].data;
1137             break;
1138         case MSR_FMASK:
1139             env->fmask = msrs[i].data;
1140             break;
1141         case MSR_LSTAR:
1142             env->lstar = msrs[i].data;
1143             break;
1144 #endif
1145         case MSR_IA32_TSC:
1146             env->tsc = msrs[i].data;
1147             break;
1148         case MSR_VM_HSAVE_PA:
1149             env->vm_hsave = msrs[i].data;
1150             break;
1151         case MSR_KVM_SYSTEM_TIME:
1152             env->system_time_msr = msrs[i].data;
1153             break;
1154         case MSR_KVM_WALL_CLOCK:
1155             env->wall_clock_msr = msrs[i].data;
1156             break;
1157 #ifdef KVM_CAP_MCE
1158         case MSR_MCG_STATUS:
1159             env->mcg_status = msrs[i].data;
1160             break;
1161         case MSR_MCG_CTL:
1162             env->mcg_ctl = msrs[i].data;
1163             break;
1164 #endif
1165         default:
1166 #ifdef KVM_CAP_MCE
1167             if (msrs[i].index >= MSR_MC0_CTL &&
1168                 msrs[i].index < MSR_MC0_CTL + (env->mcg_cap & 0xff) * 4) {
1169                 env->mce_banks[msrs[i].index - MSR_MC0_CTL] = msrs[i].data;
1170             }
1171 #endif
1172             break;
1173 #ifdef KVM_CAP_ASYNC_PF
1174         case MSR_KVM_ASYNC_PF_EN:
1175             env->async_pf_en_msr = msrs[i].data;
1176             break;
1177 #endif
1178         }
1179     }
1180
1181     return 0;
1182 }
1183
1184 static int kvm_put_mp_state(CPUState *env)
1185 {
1186     struct kvm_mp_state mp_state = { .mp_state = env->mp_state };
1187
1188     return kvm_vcpu_ioctl(env, KVM_SET_MP_STATE, &mp_state);
1189 }
1190
1191 static int kvm_get_mp_state(CPUState *env)
1192 {
1193     struct kvm_mp_state mp_state;
1194     int ret;
1195
1196     ret = kvm_vcpu_ioctl(env, KVM_GET_MP_STATE, &mp_state);
1197     if (ret < 0) {
1198         return ret;
1199     }
1200     env->mp_state = mp_state.mp_state;
1201     return 0;
1202 }
1203
1204 static int kvm_put_vcpu_events(CPUState *env, int level)
1205 {
1206 #ifdef KVM_CAP_VCPU_EVENTS
1207     struct kvm_vcpu_events events;
1208
1209     if (!kvm_has_vcpu_events()) {
1210         return 0;
1211     }
1212
1213     events.exception.injected = (env->exception_injected >= 0);
1214     events.exception.nr = env->exception_injected;
1215     events.exception.has_error_code = env->has_error_code;
1216     events.exception.error_code = env->error_code;
1217
1218     events.interrupt.injected = (env->interrupt_injected >= 0);
1219     events.interrupt.nr = env->interrupt_injected;
1220     events.interrupt.soft = env->soft_interrupt;
1221
1222     events.nmi.injected = env->nmi_injected;
1223     events.nmi.pending = env->nmi_pending;
1224     events.nmi.masked = !!(env->hflags2 & HF2_NMI_MASK);
1225
1226     events.sipi_vector = env->sipi_vector;
1227
1228     events.flags = 0;
1229     if (level >= KVM_PUT_RESET_STATE) {
1230         events.flags |=
1231             KVM_VCPUEVENT_VALID_NMI_PENDING | KVM_VCPUEVENT_VALID_SIPI_VECTOR;
1232     }
1233
1234     return kvm_vcpu_ioctl(env, KVM_SET_VCPU_EVENTS, &events);
1235 #else
1236     return 0;
1237 #endif
1238 }
1239
1240 static int kvm_get_vcpu_events(CPUState *env)
1241 {
1242 #ifdef KVM_CAP_VCPU_EVENTS
1243     struct kvm_vcpu_events events;
1244     int ret;
1245
1246     if (!kvm_has_vcpu_events()) {
1247         return 0;
1248     }
1249
1250     ret = kvm_vcpu_ioctl(env, KVM_GET_VCPU_EVENTS, &events);
1251     if (ret < 0) {
1252        return ret;
1253     }
1254     env->exception_injected =
1255        events.exception.injected ? events.exception.nr : -1;
1256     env->has_error_code = events.exception.has_error_code;
1257     env->error_code = events.exception.error_code;
1258
1259     env->interrupt_injected =
1260         events.interrupt.injected ? events.interrupt.nr : -1;
1261     env->soft_interrupt = events.interrupt.soft;
1262
1263     env->nmi_injected = events.nmi.injected;
1264     env->nmi_pending = events.nmi.pending;
1265     if (events.nmi.masked) {
1266         env->hflags2 |= HF2_NMI_MASK;
1267     } else {
1268         env->hflags2 &= ~HF2_NMI_MASK;
1269     }
1270
1271     env->sipi_vector = events.sipi_vector;
1272 #endif
1273
1274     return 0;
1275 }
1276
1277 static int kvm_guest_debug_workarounds(CPUState *env)
1278 {
1279     int ret = 0;
1280 #ifdef KVM_CAP_SET_GUEST_DEBUG
1281     unsigned long reinject_trap = 0;
1282
1283     if (!kvm_has_vcpu_events()) {
1284         if (env->exception_injected == 1) {
1285             reinject_trap = KVM_GUESTDBG_INJECT_DB;
1286         } else if (env->exception_injected == 3) {
1287             reinject_trap = KVM_GUESTDBG_INJECT_BP;
1288         }
1289         env->exception_injected = -1;
1290     }
1291
1292     /*
1293      * Kernels before KVM_CAP_X86_ROBUST_SINGLESTEP overwrote flags.TF
1294      * injected via SET_GUEST_DEBUG while updating GP regs. Work around this
1295      * by updating the debug state once again if single-stepping is on.
1296      * Another reason to call kvm_update_guest_debug here is a pending debug
1297      * trap raise by the guest. On kernels without SET_VCPU_EVENTS we have to
1298      * reinject them via SET_GUEST_DEBUG.
1299      */
1300     if (reinject_trap ||
1301         (!kvm_has_robust_singlestep() && env->singlestep_enabled)) {
1302         ret = kvm_update_guest_debug(env, reinject_trap);
1303     }
1304 #endif /* KVM_CAP_SET_GUEST_DEBUG */
1305     return ret;
1306 }
1307
1308 static int kvm_put_debugregs(CPUState *env)
1309 {
1310 #ifdef KVM_CAP_DEBUGREGS
1311     struct kvm_debugregs dbgregs;
1312     int i;
1313
1314     if (!kvm_has_debugregs()) {
1315         return 0;
1316     }
1317
1318     for (i = 0; i < 4; i++) {
1319         dbgregs.db[i] = env->dr[i];
1320     }
1321     dbgregs.dr6 = env->dr[6];
1322     dbgregs.dr7 = env->dr[7];
1323     dbgregs.flags = 0;
1324
1325     return kvm_vcpu_ioctl(env, KVM_SET_DEBUGREGS, &dbgregs);
1326 #else
1327     return 0;
1328 #endif
1329 }
1330
1331 static int kvm_get_debugregs(CPUState *env)
1332 {
1333 #ifdef KVM_CAP_DEBUGREGS
1334     struct kvm_debugregs dbgregs;
1335     int i, ret;
1336
1337     if (!kvm_has_debugregs()) {
1338         return 0;
1339     }
1340
1341     ret = kvm_vcpu_ioctl(env, KVM_GET_DEBUGREGS, &dbgregs);
1342     if (ret < 0) {
1343         return ret;
1344     }
1345     for (i = 0; i < 4; i++) {
1346         env->dr[i] = dbgregs.db[i];
1347     }
1348     env->dr[4] = env->dr[6] = dbgregs.dr6;
1349     env->dr[5] = env->dr[7] = dbgregs.dr7;
1350 #endif
1351
1352     return 0;
1353 }
1354
1355 int kvm_arch_put_registers(CPUState *env, int level)
1356 {
1357     int ret;
1358
1359     assert(cpu_is_stopped(env) || qemu_cpu_self(env));
1360
1361     ret = kvm_getput_regs(env, 1);
1362     if (ret < 0) {
1363         return ret;
1364     }
1365     ret = kvm_put_xsave(env);
1366     if (ret < 0) {
1367         return ret;
1368     }
1369     ret = kvm_put_xcrs(env);
1370     if (ret < 0) {
1371         return ret;
1372     }
1373     ret = kvm_put_sregs(env);
1374     if (ret < 0) {
1375         return ret;
1376     }
1377     ret = kvm_put_msrs(env, level);
1378     if (ret < 0) {
1379         return ret;
1380     }
1381     if (level >= KVM_PUT_RESET_STATE) {
1382         ret = kvm_put_mp_state(env);
1383         if (ret < 0) {
1384             return ret;
1385         }
1386     }
1387     ret = kvm_put_vcpu_events(env, level);
1388     if (ret < 0) {
1389         return ret;
1390     }
1391     /* must be last */
1392     ret = kvm_guest_debug_workarounds(env);
1393     if (ret < 0) {
1394         return ret;
1395     }
1396     ret = kvm_put_debugregs(env);
1397     if (ret < 0) {
1398         return ret;
1399     }
1400     return 0;
1401 }
1402
1403 int kvm_arch_get_registers(CPUState *env)
1404 {
1405     int ret;
1406
1407     assert(cpu_is_stopped(env) || qemu_cpu_self(env));
1408
1409     ret = kvm_getput_regs(env, 0);
1410     if (ret < 0) {
1411         return ret;
1412     }
1413     ret = kvm_get_xsave(env);
1414     if (ret < 0) {
1415         return ret;
1416     }
1417     ret = kvm_get_xcrs(env);
1418     if (ret < 0) {
1419         return ret;
1420     }
1421     ret = kvm_get_sregs(env);
1422     if (ret < 0) {
1423         return ret;
1424     }
1425     ret = kvm_get_msrs(env);
1426     if (ret < 0) {
1427         return ret;
1428     }
1429     ret = kvm_get_mp_state(env);
1430     if (ret < 0) {
1431         return ret;
1432     }
1433     ret = kvm_get_vcpu_events(env);
1434     if (ret < 0) {
1435         return ret;
1436     }
1437     ret = kvm_get_debugregs(env);
1438     if (ret < 0) {
1439         return ret;
1440     }
1441     return 0;
1442 }
1443
1444 int kvm_arch_pre_run(CPUState *env, struct kvm_run *run)
1445 {
1446     /* Inject NMI */
1447     if (env->interrupt_request & CPU_INTERRUPT_NMI) {
1448         env->interrupt_request &= ~CPU_INTERRUPT_NMI;
1449         DPRINTF("injected NMI\n");
1450         kvm_vcpu_ioctl(env, KVM_NMI);
1451     }
1452
1453     /* Try to inject an interrupt if the guest can accept it */
1454     if (run->ready_for_interrupt_injection &&
1455         (env->interrupt_request & CPU_INTERRUPT_HARD) &&
1456         (env->eflags & IF_MASK)) {
1457         int irq;
1458
1459         env->interrupt_request &= ~CPU_INTERRUPT_HARD;
1460         irq = cpu_get_pic_interrupt(env);
1461         if (irq >= 0) {
1462             struct kvm_interrupt intr;
1463             intr.irq = irq;
1464             /* FIXME: errors */
1465             DPRINTF("injected interrupt %d\n", irq);
1466             kvm_vcpu_ioctl(env, KVM_INTERRUPT, &intr);
1467         }
1468     }
1469
1470     /* If we have an interrupt but the guest is not ready to receive an
1471      * interrupt, request an interrupt window exit.  This will
1472      * cause a return to userspace as soon as the guest is ready to
1473      * receive interrupts. */
1474     if ((env->interrupt_request & CPU_INTERRUPT_HARD)) {
1475         run->request_interrupt_window = 1;
1476     } else {
1477         run->request_interrupt_window = 0;
1478     }
1479
1480     DPRINTF("setting tpr\n");
1481     run->cr8 = cpu_get_apic_tpr(env->apic_state);
1482
1483     return 0;
1484 }
1485
1486 int kvm_arch_post_run(CPUState *env, struct kvm_run *run)
1487 {
1488     if (run->if_flag) {
1489         env->eflags |= IF_MASK;
1490     } else {
1491         env->eflags &= ~IF_MASK;
1492     }
1493     cpu_set_apic_tpr(env->apic_state, run->cr8);
1494     cpu_set_apic_base(env->apic_state, run->apic_base);
1495
1496     return 0;
1497 }
1498
1499 int kvm_arch_process_irqchip_events(CPUState *env)
1500 {
1501     if (env->interrupt_request & CPU_INTERRUPT_INIT) {
1502         kvm_cpu_synchronize_state(env);
1503         do_cpu_init(env);
1504         env->exception_index = EXCP_HALTED;
1505     }
1506
1507     if (env->interrupt_request & CPU_INTERRUPT_SIPI) {
1508         kvm_cpu_synchronize_state(env);
1509         do_cpu_sipi(env);
1510     }
1511
1512     return env->halted;
1513 }
1514
1515 static int kvm_handle_halt(CPUState *env)
1516 {
1517     if (!((env->interrupt_request & CPU_INTERRUPT_HARD) &&
1518           (env->eflags & IF_MASK)) &&
1519         !(env->interrupt_request & CPU_INTERRUPT_NMI)) {
1520         env->halted = 1;
1521         env->exception_index = EXCP_HLT;
1522         return 0;
1523     }
1524
1525     return 1;
1526 }
1527
1528 static bool host_supports_vmx(void)
1529 {
1530     uint32_t ecx, unused;
1531
1532     host_cpuid(1, 0, &unused, &unused, &ecx, &unused);
1533     return ecx & CPUID_EXT_VMX;
1534 }
1535
1536 #define VMX_INVALID_GUEST_STATE 0x80000021
1537
1538 int kvm_arch_handle_exit(CPUState *env, struct kvm_run *run)
1539 {
1540     uint64_t code;
1541     int ret = 0;
1542
1543     switch (run->exit_reason) {
1544     case KVM_EXIT_HLT:
1545         DPRINTF("handle_hlt\n");
1546         ret = kvm_handle_halt(env);
1547         break;
1548     case KVM_EXIT_SET_TPR:
1549         ret = 1;
1550         break;
1551     case KVM_EXIT_FAIL_ENTRY:
1552         code = run->fail_entry.hardware_entry_failure_reason;
1553         fprintf(stderr, "KVM: entry failed, hardware error 0x%" PRIx64 "\n",
1554                 code);
1555         if (host_supports_vmx() && code == VMX_INVALID_GUEST_STATE) {
1556             fprintf(stderr,
1557                     "\nIf you're runnning a guest on an Intel machine without "
1558                         "unrestricted mode\n"
1559                     "support, the failure can be most likely due to the guest "
1560                         "entering an invalid\n"
1561                     "state for Intel VT. For example, the guest maybe running "
1562                         "in big real mode\n"
1563                     "which is not supported on less recent Intel processors."
1564                         "\n\n");
1565         }
1566         ret = -1;
1567         break;
1568     case KVM_EXIT_EXCEPTION:
1569         fprintf(stderr, "KVM: exception %d exit (error code 0x%x)\n",
1570                 run->ex.exception, run->ex.error_code);
1571         ret = -1;
1572         break;
1573     default:
1574         fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
1575         ret = -1;
1576         break;
1577     }
1578
1579     return ret;
1580 }
1581
1582 #ifdef KVM_CAP_SET_GUEST_DEBUG
1583 int kvm_arch_insert_sw_breakpoint(CPUState *env, struct kvm_sw_breakpoint *bp)
1584 {
1585     static const uint8_t int3 = 0xcc;
1586
1587     if (cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&bp->saved_insn, 1, 0) ||
1588         cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&int3, 1, 1)) {
1589         return -EINVAL;
1590     }
1591     return 0;
1592 }
1593
1594 int kvm_arch_remove_sw_breakpoint(CPUState *env, struct kvm_sw_breakpoint *bp)
1595 {
1596     uint8_t int3;
1597
1598     if (cpu_memory_rw_debug(env, bp->pc, &int3, 1, 0) || int3 != 0xcc ||
1599         cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&bp->saved_insn, 1, 1)) {
1600         return -EINVAL;
1601     }
1602     return 0;
1603 }
1604
1605 static struct {
1606     target_ulong addr;
1607     int len;
1608     int type;
1609 } hw_breakpoint[4];
1610
1611 static int nb_hw_breakpoint;
1612
1613 static int find_hw_breakpoint(target_ulong addr, int len, int type)
1614 {
1615     int n;
1616
1617     for (n = 0; n < nb_hw_breakpoint; n++) {
1618         if (hw_breakpoint[n].addr == addr && hw_breakpoint[n].type == type &&
1619             (hw_breakpoint[n].len == len || len == -1)) {
1620             return n;
1621         }
1622     }
1623     return -1;
1624 }
1625
1626 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
1627                                   target_ulong len, int type)
1628 {
1629     switch (type) {
1630     case GDB_BREAKPOINT_HW:
1631         len = 1;
1632         break;
1633     case GDB_WATCHPOINT_WRITE:
1634     case GDB_WATCHPOINT_ACCESS:
1635         switch (len) {
1636         case 1:
1637             break;
1638         case 2:
1639         case 4:
1640         case 8:
1641             if (addr & (len - 1)) {
1642                 return -EINVAL;
1643             }
1644             break;
1645         default:
1646             return -EINVAL;
1647         }
1648         break;
1649     default:
1650         return -ENOSYS;
1651     }
1652
1653     if (nb_hw_breakpoint == 4) {
1654         return -ENOBUFS;
1655     }
1656     if (find_hw_breakpoint(addr, len, type) >= 0) {
1657         return -EEXIST;
1658     }
1659     hw_breakpoint[nb_hw_breakpoint].addr = addr;
1660     hw_breakpoint[nb_hw_breakpoint].len = len;
1661     hw_breakpoint[nb_hw_breakpoint].type = type;
1662     nb_hw_breakpoint++;
1663
1664     return 0;
1665 }
1666
1667 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
1668                                   target_ulong len, int type)
1669 {
1670     int n;
1671
1672     n = find_hw_breakpoint(addr, (type == GDB_BREAKPOINT_HW) ? 1 : len, type);
1673     if (n < 0) {
1674         return -ENOENT;
1675     }
1676     nb_hw_breakpoint--;
1677     hw_breakpoint[n] = hw_breakpoint[nb_hw_breakpoint];
1678
1679     return 0;
1680 }
1681
1682 void kvm_arch_remove_all_hw_breakpoints(void)
1683 {
1684     nb_hw_breakpoint = 0;
1685 }
1686
1687 static CPUWatchpoint hw_watchpoint;
1688
1689 int kvm_arch_debug(struct kvm_debug_exit_arch *arch_info)
1690 {
1691     int handle = 0;
1692     int n;
1693
1694     if (arch_info->exception == 1) {
1695         if (arch_info->dr6 & (1 << 14)) {
1696             if (cpu_single_env->singlestep_enabled) {
1697                 handle = 1;
1698             }
1699         } else {
1700             for (n = 0; n < 4; n++) {
1701                 if (arch_info->dr6 & (1 << n)) {
1702                     switch ((arch_info->dr7 >> (16 + n*4)) & 0x3) {
1703                     case 0x0:
1704                         handle = 1;
1705                         break;
1706                     case 0x1:
1707                         handle = 1;
1708                         cpu_single_env->watchpoint_hit = &hw_watchpoint;
1709                         hw_watchpoint.vaddr = hw_breakpoint[n].addr;
1710                         hw_watchpoint.flags = BP_MEM_WRITE;
1711                         break;
1712                     case 0x3:
1713                         handle = 1;
1714                         cpu_single_env->watchpoint_hit = &hw_watchpoint;
1715                         hw_watchpoint.vaddr = hw_breakpoint[n].addr;
1716                         hw_watchpoint.flags = BP_MEM_ACCESS;
1717                         break;
1718                     }
1719                 }
1720             }
1721         }
1722     } else if (kvm_find_sw_breakpoint(cpu_single_env, arch_info->pc)) {
1723         handle = 1;
1724     }
1725     if (!handle) {
1726         cpu_synchronize_state(cpu_single_env);
1727         assert(cpu_single_env->exception_injected == -1);
1728
1729         cpu_single_env->exception_injected = arch_info->exception;
1730         cpu_single_env->has_error_code = 0;
1731     }
1732
1733     return handle;
1734 }
1735
1736 void kvm_arch_update_guest_debug(CPUState *env, struct kvm_guest_debug *dbg)
1737 {
1738     const uint8_t type_code[] = {
1739         [GDB_BREAKPOINT_HW] = 0x0,
1740         [GDB_WATCHPOINT_WRITE] = 0x1,
1741         [GDB_WATCHPOINT_ACCESS] = 0x3
1742     };
1743     const uint8_t len_code[] = {
1744         [1] = 0x0, [2] = 0x1, [4] = 0x3, [8] = 0x2
1745     };
1746     int n;
1747
1748     if (kvm_sw_breakpoints_active(env)) {
1749         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
1750     }
1751     if (nb_hw_breakpoint > 0) {
1752         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
1753         dbg->arch.debugreg[7] = 0x0600;
1754         for (n = 0; n < nb_hw_breakpoint; n++) {
1755             dbg->arch.debugreg[n] = hw_breakpoint[n].addr;
1756             dbg->arch.debugreg[7] |= (2 << (n * 2)) |
1757                 (type_code[hw_breakpoint[n].type] << (16 + n*4)) |
1758                 ((uint32_t)len_code[hw_breakpoint[n].len] << (18 + n*4));
1759         }
1760     }
1761     /* Legal xcr0 for loading */
1762     env->xcr0 = 1;
1763 }
1764 #endif /* KVM_CAP_SET_GUEST_DEBUG */
1765
1766 bool kvm_arch_stop_on_emulation_error(CPUState *env)
1767 {
1768     return !(env->cr[0] & CR0_PE_MASK) ||
1769            ((env->segs[R_CS].selector  & 3) != 3);
1770 }
1771
1772 static void hardware_memory_error(void)
1773 {
1774     fprintf(stderr, "Hardware memory error!\n");
1775     exit(1);
1776 }
1777
1778 #ifdef KVM_CAP_MCE
1779 static void kvm_mce_broadcast_rest(CPUState *env)
1780 {
1781     struct kvm_x86_mce mce = {
1782         .bank = 1,
1783         .status = MCI_STATUS_VAL | MCI_STATUS_UC,
1784         .mcg_status = MCG_STATUS_MCIP | MCG_STATUS_RIPV,
1785         .addr = 0,
1786         .misc = 0,
1787     };
1788     CPUState *cenv;
1789
1790     /* Broadcast MCA signal for processor version 06H_EH and above */
1791     if (cpu_x86_support_mca_broadcast(env)) {
1792         for (cenv = first_cpu; cenv != NULL; cenv = cenv->next_cpu) {
1793             if (cenv == env) {
1794                 continue;
1795             }
1796             kvm_inject_x86_mce_on(cenv, &mce, ABORT_ON_ERROR);
1797         }
1798     }
1799 }
1800
1801 static void kvm_mce_inj_srar_dataload(CPUState *env, target_phys_addr_t paddr)
1802 {
1803     struct kvm_x86_mce mce = {
1804         .bank = 9,
1805         .status = MCI_STATUS_VAL | MCI_STATUS_UC | MCI_STATUS_EN
1806                   | MCI_STATUS_MISCV | MCI_STATUS_ADDRV | MCI_STATUS_S
1807                   | MCI_STATUS_AR | 0x134,
1808         .mcg_status = MCG_STATUS_MCIP | MCG_STATUS_EIPV,
1809         .addr = paddr,
1810         .misc = (MCM_ADDR_PHYS << 6) | 0xc,
1811     };
1812     int r;
1813
1814     r = kvm_set_mce(env, &mce);
1815     if (r < 0) {
1816         fprintf(stderr, "kvm_set_mce: %s\n", strerror(errno));
1817         abort();
1818     }
1819     kvm_mce_broadcast_rest(env);
1820 }
1821
1822 static void kvm_mce_inj_srao_memscrub(CPUState *env, target_phys_addr_t paddr)
1823 {
1824     struct kvm_x86_mce mce = {
1825         .bank = 9,
1826         .status = MCI_STATUS_VAL | MCI_STATUS_UC | MCI_STATUS_EN
1827                   | MCI_STATUS_MISCV | MCI_STATUS_ADDRV | MCI_STATUS_S
1828                   | 0xc0,
1829         .mcg_status = MCG_STATUS_MCIP | MCG_STATUS_RIPV,
1830         .addr = paddr,
1831         .misc = (MCM_ADDR_PHYS << 6) | 0xc,
1832     };
1833     int r;
1834
1835     r = kvm_set_mce(env, &mce);
1836     if (r < 0) {
1837         fprintf(stderr, "kvm_set_mce: %s\n", strerror(errno));
1838         abort();
1839     }
1840     kvm_mce_broadcast_rest(env);
1841 }
1842
1843 static void kvm_mce_inj_srao_memscrub2(CPUState *env, target_phys_addr_t paddr)
1844 {
1845     struct kvm_x86_mce mce = {
1846         .bank = 9,
1847         .status = MCI_STATUS_VAL | MCI_STATUS_UC | MCI_STATUS_EN
1848                   | MCI_STATUS_MISCV | MCI_STATUS_ADDRV | MCI_STATUS_S
1849                   | 0xc0,
1850         .mcg_status = MCG_STATUS_MCIP | MCG_STATUS_RIPV,
1851         .addr = paddr,
1852         .misc = (MCM_ADDR_PHYS << 6) | 0xc,
1853     };
1854
1855     kvm_inject_x86_mce_on(env, &mce, ABORT_ON_ERROR);
1856     kvm_mce_broadcast_rest(env);
1857 }
1858
1859 #endif
1860
1861 int kvm_on_sigbus_vcpu(CPUState *env, int code, void *addr)
1862 {
1863 #if defined(KVM_CAP_MCE)
1864     void *vaddr;
1865     ram_addr_t ram_addr;
1866     target_phys_addr_t paddr;
1867
1868     if ((env->mcg_cap & MCG_SER_P) && addr
1869         && (code == BUS_MCEERR_AR
1870             || code == BUS_MCEERR_AO)) {
1871         vaddr = (void *)addr;
1872         if (qemu_ram_addr_from_host(vaddr, &ram_addr) ||
1873             !kvm_physical_memory_addr_from_ram(env->kvm_state, ram_addr, &paddr)) {
1874             fprintf(stderr, "Hardware memory error for memory used by "
1875                     "QEMU itself instead of guest system!\n");
1876             /* Hope we are lucky for AO MCE */
1877             if (code == BUS_MCEERR_AO) {
1878                 return 0;
1879             } else {
1880                 hardware_memory_error();
1881             }
1882         }
1883
1884         if (code == BUS_MCEERR_AR) {
1885             /* Fake an Intel architectural Data Load SRAR UCR */
1886             kvm_mce_inj_srar_dataload(env, paddr);
1887         } else {
1888             /*
1889              * If there is an MCE excpetion being processed, ignore
1890              * this SRAO MCE
1891              */
1892             if (!kvm_mce_in_progress(env)) {
1893                 /* Fake an Intel architectural Memory scrubbing UCR */
1894                 kvm_mce_inj_srao_memscrub(env, paddr);
1895             }
1896         }
1897     } else
1898 #endif
1899     {
1900         if (code == BUS_MCEERR_AO) {
1901             return 0;
1902         } else if (code == BUS_MCEERR_AR) {
1903             hardware_memory_error();
1904         } else {
1905             return 1;
1906         }
1907     }
1908     return 0;
1909 }
1910
1911 int kvm_on_sigbus(int code, void *addr)
1912 {
1913 #if defined(KVM_CAP_MCE)
1914     if ((first_cpu->mcg_cap & MCG_SER_P) && addr && code == BUS_MCEERR_AO) {
1915         void *vaddr;
1916         ram_addr_t ram_addr;
1917         target_phys_addr_t paddr;
1918
1919         /* Hope we are lucky for AO MCE */
1920         vaddr = addr;
1921         if (qemu_ram_addr_from_host(vaddr, &ram_addr) ||
1922             !kvm_physical_memory_addr_from_ram(first_cpu->kvm_state, ram_addr, &paddr)) {
1923             fprintf(stderr, "Hardware memory error for memory used by "
1924                     "QEMU itself instead of guest system!: %p\n", addr);
1925             return 0;
1926         }
1927         kvm_mce_inj_srao_memscrub2(first_cpu, paddr);
1928     } else
1929 #endif
1930     {
1931         if (code == BUS_MCEERR_AO) {
1932             return 0;
1933         } else if (code == BUS_MCEERR_AR) {
1934             hardware_memory_error();
1935         } else {
1936             return 1;
1937         }
1938     }
1939     return 0;
1940 }