qemu/target/i386/machine.c
<<
>>
Prefs
   1#include "qemu/osdep.h"
   2#include "cpu.h"
   3#include "exec/exec-all.h"
   4#include "hw/isa/isa.h"
   5#include "migration/cpu.h"
   6#include "kvm/hyperv.h"
   7#include "hw/i386/x86.h"
   8#include "kvm/kvm_i386.h"
   9
  10#include "sysemu/kvm.h"
  11#include "sysemu/tcg.h"
  12
  13#include "qemu/error-report.h"
  14
  15static const VMStateDescription vmstate_segment = {
  16    .name = "segment",
  17    .version_id = 1,
  18    .minimum_version_id = 1,
  19    .fields = (VMStateField[]) {
  20        VMSTATE_UINT32(selector, SegmentCache),
  21        VMSTATE_UINTTL(base, SegmentCache),
  22        VMSTATE_UINT32(limit, SegmentCache),
  23        VMSTATE_UINT32(flags, SegmentCache),
  24        VMSTATE_END_OF_LIST()
  25    }
  26};
  27
  28#define VMSTATE_SEGMENT(_field, _state) {                            \
  29    .name       = (stringify(_field)),                               \
  30    .size       = sizeof(SegmentCache),                              \
  31    .vmsd       = &vmstate_segment,                                  \
  32    .flags      = VMS_STRUCT,                                        \
  33    .offset     = offsetof(_state, _field)                           \
  34            + type_check(SegmentCache,typeof_field(_state, _field))  \
  35}
  36
  37#define VMSTATE_SEGMENT_ARRAY(_field, _state, _n)                    \
  38    VMSTATE_STRUCT_ARRAY(_field, _state, _n, 0, vmstate_segment, SegmentCache)
  39
  40static const VMStateDescription vmstate_xmm_reg = {
  41    .name = "xmm_reg",
  42    .version_id = 1,
  43    .minimum_version_id = 1,
  44    .fields = (VMStateField[]) {
  45        VMSTATE_UINT64(ZMM_Q(0), ZMMReg),
  46        VMSTATE_UINT64(ZMM_Q(1), ZMMReg),
  47        VMSTATE_END_OF_LIST()
  48    }
  49};
  50
  51#define VMSTATE_XMM_REGS(_field, _state, _start)                         \
  52    VMSTATE_STRUCT_SUB_ARRAY(_field, _state, _start, CPU_NB_REGS, 0,     \
  53                             vmstate_xmm_reg, ZMMReg)
  54
  55/* YMMH format is the same as XMM, but for bits 128-255 */
  56static const VMStateDescription vmstate_ymmh_reg = {
  57    .name = "ymmh_reg",
  58    .version_id = 1,
  59    .minimum_version_id = 1,
  60    .fields = (VMStateField[]) {
  61        VMSTATE_UINT64(ZMM_Q(2), ZMMReg),
  62        VMSTATE_UINT64(ZMM_Q(3), ZMMReg),
  63        VMSTATE_END_OF_LIST()
  64    }
  65};
  66
  67#define VMSTATE_YMMH_REGS_VARS(_field, _state, _start, _v)               \
  68    VMSTATE_STRUCT_SUB_ARRAY(_field, _state, _start, CPU_NB_REGS, _v,    \
  69                             vmstate_ymmh_reg, ZMMReg)
  70
  71static const VMStateDescription vmstate_zmmh_reg = {
  72    .name = "zmmh_reg",
  73    .version_id = 1,
  74    .minimum_version_id = 1,
  75    .fields = (VMStateField[]) {
  76        VMSTATE_UINT64(ZMM_Q(4), ZMMReg),
  77        VMSTATE_UINT64(ZMM_Q(5), ZMMReg),
  78        VMSTATE_UINT64(ZMM_Q(6), ZMMReg),
  79        VMSTATE_UINT64(ZMM_Q(7), ZMMReg),
  80        VMSTATE_END_OF_LIST()
  81    }
  82};
  83
  84#define VMSTATE_ZMMH_REGS_VARS(_field, _state, _start)                   \
  85    VMSTATE_STRUCT_SUB_ARRAY(_field, _state, _start, CPU_NB_REGS, 0,     \
  86                             vmstate_zmmh_reg, ZMMReg)
  87
  88#ifdef TARGET_X86_64
  89static const VMStateDescription vmstate_hi16_zmm_reg = {
  90    .name = "hi16_zmm_reg",
  91    .version_id = 1,
  92    .minimum_version_id = 1,
  93    .fields = (VMStateField[]) {
  94        VMSTATE_UINT64(ZMM_Q(0), ZMMReg),
  95        VMSTATE_UINT64(ZMM_Q(1), ZMMReg),
  96        VMSTATE_UINT64(ZMM_Q(2), ZMMReg),
  97        VMSTATE_UINT64(ZMM_Q(3), ZMMReg),
  98        VMSTATE_UINT64(ZMM_Q(4), ZMMReg),
  99        VMSTATE_UINT64(ZMM_Q(5), ZMMReg),
 100        VMSTATE_UINT64(ZMM_Q(6), ZMMReg),
 101        VMSTATE_UINT64(ZMM_Q(7), ZMMReg),
 102        VMSTATE_END_OF_LIST()
 103    }
 104};
 105
 106#define VMSTATE_Hi16_ZMM_REGS_VARS(_field, _state, _start)               \
 107    VMSTATE_STRUCT_SUB_ARRAY(_field, _state, _start, CPU_NB_REGS, 0,     \
 108                             vmstate_hi16_zmm_reg, ZMMReg)
 109#endif
 110
 111static const VMStateDescription vmstate_bnd_regs = {
 112    .name = "bnd_regs",
 113    .version_id = 1,
 114    .minimum_version_id = 1,
 115    .fields = (VMStateField[]) {
 116        VMSTATE_UINT64(lb, BNDReg),
 117        VMSTATE_UINT64(ub, BNDReg),
 118        VMSTATE_END_OF_LIST()
 119    }
 120};
 121
 122#define VMSTATE_BND_REGS(_field, _state, _n)          \
 123    VMSTATE_STRUCT_ARRAY(_field, _state, _n, 0, vmstate_bnd_regs, BNDReg)
 124
 125static const VMStateDescription vmstate_mtrr_var = {
 126    .name = "mtrr_var",
 127    .version_id = 1,
 128    .minimum_version_id = 1,
 129    .fields = (VMStateField[]) {
 130        VMSTATE_UINT64(base, MTRRVar),
 131        VMSTATE_UINT64(mask, MTRRVar),
 132        VMSTATE_END_OF_LIST()
 133    }
 134};
 135
 136#define VMSTATE_MTRR_VARS(_field, _state, _n, _v)                    \
 137    VMSTATE_STRUCT_ARRAY(_field, _state, _n, _v, vmstate_mtrr_var, MTRRVar)
 138
 139typedef struct x86_FPReg_tmp {
 140    FPReg *parent;
 141    uint64_t tmp_mant;
 142    uint16_t tmp_exp;
 143} x86_FPReg_tmp;
 144
 145static void cpu_get_fp80(uint64_t *pmant, uint16_t *pexp, floatx80 f)
 146{
 147    CPU_LDoubleU temp;
 148
 149    temp.d = f;
 150    *pmant = temp.l.lower;
 151    *pexp = temp.l.upper;
 152}
 153
 154static floatx80 cpu_set_fp80(uint64_t mant, uint16_t upper)
 155{
 156    CPU_LDoubleU temp;
 157
 158    temp.l.upper = upper;
 159    temp.l.lower = mant;
 160    return temp.d;
 161}
 162
 163static int fpreg_pre_save(void *opaque)
 164{
 165    x86_FPReg_tmp *tmp = opaque;
 166
 167    /* we save the real CPU data (in case of MMX usage only 'mant'
 168       contains the MMX register */
 169    cpu_get_fp80(&tmp->tmp_mant, &tmp->tmp_exp, tmp->parent->d);
 170
 171    return 0;
 172}
 173
 174static int fpreg_post_load(void *opaque, int version)
 175{
 176    x86_FPReg_tmp *tmp = opaque;
 177
 178    tmp->parent->d = cpu_set_fp80(tmp->tmp_mant, tmp->tmp_exp);
 179    return 0;
 180}
 181
 182static const VMStateDescription vmstate_fpreg_tmp = {
 183    .name = "fpreg_tmp",
 184    .post_load = fpreg_post_load,
 185    .pre_save  = fpreg_pre_save,
 186    .fields = (VMStateField[]) {
 187        VMSTATE_UINT64(tmp_mant, x86_FPReg_tmp),
 188        VMSTATE_UINT16(tmp_exp, x86_FPReg_tmp),
 189        VMSTATE_END_OF_LIST()
 190    }
 191};
 192
 193static const VMStateDescription vmstate_fpreg = {
 194    .name = "fpreg",
 195    .fields = (VMStateField[]) {
 196        VMSTATE_WITH_TMP(FPReg, x86_FPReg_tmp, vmstate_fpreg_tmp),
 197        VMSTATE_END_OF_LIST()
 198    }
 199};
 200
 201static int cpu_pre_save(void *opaque)
 202{
 203    X86CPU *cpu = opaque;
 204    CPUX86State *env = &cpu->env;
 205    int i;
 206
 207    /* FPU */
 208    env->fpus_vmstate = (env->fpus & ~0x3800) | (env->fpstt & 0x7) << 11;
 209    env->fptag_vmstate = 0;
 210    for(i = 0; i < 8; i++) {
 211        env->fptag_vmstate |= ((!env->fptags[i]) << i);
 212    }
 213
 214    env->fpregs_format_vmstate = 0;
 215
 216    /*
 217     * Real mode guest segments register DPL should be zero.
 218     * Older KVM version were setting it wrongly.
 219     * Fixing it will allow live migration to host with unrestricted guest
 220     * support (otherwise the migration will fail with invalid guest state
 221     * error).
 222     */
 223    if (!(env->cr[0] & CR0_PE_MASK) &&
 224        (env->segs[R_CS].flags >> DESC_DPL_SHIFT & 3) != 0) {
 225        env->segs[R_CS].flags &= ~(env->segs[R_CS].flags & DESC_DPL_MASK);
 226        env->segs[R_DS].flags &= ~(env->segs[R_DS].flags & DESC_DPL_MASK);
 227        env->segs[R_ES].flags &= ~(env->segs[R_ES].flags & DESC_DPL_MASK);
 228        env->segs[R_FS].flags &= ~(env->segs[R_FS].flags & DESC_DPL_MASK);
 229        env->segs[R_GS].flags &= ~(env->segs[R_GS].flags & DESC_DPL_MASK);
 230        env->segs[R_SS].flags &= ~(env->segs[R_SS].flags & DESC_DPL_MASK);
 231    }
 232
 233#ifdef CONFIG_KVM
 234    /*
 235     * In case vCPU may have enabled VMX, we need to make sure kernel have
 236     * required capabilities in order to perform migration correctly:
 237     *
 238     * 1) We must be able to extract vCPU nested-state from KVM.
 239     *
 240     * 2) In case vCPU is running in guest-mode and it has a pending exception,
 241     * we must be able to determine if it's in a pending or injected state.
 242     * Note that in case KVM don't have required capability to do so,
 243     * a pending/injected exception will always appear as an
 244     * injected exception.
 245     */
 246    if (kvm_enabled() && cpu_vmx_maybe_enabled(env) &&
 247        (!env->nested_state ||
 248         (!kvm_has_exception_payload() && (env->hflags & HF_GUEST_MASK) &&
 249          env->exception_injected))) {
 250        error_report("Guest maybe enabled nested virtualization but kernel "
 251                "does not support required capabilities to save vCPU "
 252                "nested state");
 253        return -EINVAL;
 254    }
 255#endif
 256
 257    /*
 258     * When vCPU is running L2 and exception is still pending,
 259     * it can potentially be intercepted by L1 hypervisor.
 260     * In contrast to an injected exception which cannot be
 261     * intercepted anymore.
 262     *
 263     * Furthermore, when a L2 exception is intercepted by L1
 264     * hypervisor, its exception payload (CR2/DR6 on #PF/#DB)
 265     * should not be set yet in the respective vCPU register.
 266     * Thus, in case an exception is pending, it is
 267     * important to save the exception payload seperately.
 268     *
 269     * Therefore, if an exception is not in a pending state
 270     * or vCPU is not in guest-mode, it is not important to
 271     * distinguish between a pending and injected exception
 272     * and we don't need to store seperately the exception payload.
 273     *
 274     * In order to preserve better backwards-compatible migration,
 275     * convert a pending exception to an injected exception in
 276     * case it is not important to distinguish between them
 277     * as described above.
 278     */
 279    if (env->exception_pending && !(env->hflags & HF_GUEST_MASK)) {
 280        env->exception_pending = 0;
 281        env->exception_injected = 1;
 282
 283        if (env->exception_has_payload) {
 284            if (env->exception_nr == EXCP01_DB) {
 285                env->dr[6] = env->exception_payload;
 286            } else if (env->exception_nr == EXCP0E_PAGE) {
 287                env->cr[2] = env->exception_payload;
 288            }
 289        }
 290    }
 291
 292    return 0;
 293}
 294
 295static int cpu_post_load(void *opaque, int version_id)
 296{
 297    X86CPU *cpu = opaque;
 298    CPUState *cs = CPU(cpu);
 299    CPUX86State *env = &cpu->env;
 300    int i;
 301
 302    if (env->tsc_khz && env->user_tsc_khz &&
 303        env->tsc_khz != env->user_tsc_khz) {
 304        error_report("Mismatch between user-specified TSC frequency and "
 305                     "migrated TSC frequency");
 306        return -EINVAL;
 307    }
 308
 309    if (env->fpregs_format_vmstate) {
 310        error_report("Unsupported old non-softfloat CPU state");
 311        return -EINVAL;
 312    }
 313    /*
 314     * Real mode guest segments register DPL should be zero.
 315     * Older KVM version were setting it wrongly.
 316     * Fixing it will allow live migration from such host that don't have
 317     * restricted guest support to a host with unrestricted guest support
 318     * (otherwise the migration will fail with invalid guest state
 319     * error).
 320     */
 321    if (!(env->cr[0] & CR0_PE_MASK) &&
 322        (env->segs[R_CS].flags >> DESC_DPL_SHIFT & 3) != 0) {
 323        env->segs[R_CS].flags &= ~(env->segs[R_CS].flags & DESC_DPL_MASK);
 324        env->segs[R_DS].flags &= ~(env->segs[R_DS].flags & DESC_DPL_MASK);
 325        env->segs[R_ES].flags &= ~(env->segs[R_ES].flags & DESC_DPL_MASK);
 326        env->segs[R_FS].flags &= ~(env->segs[R_FS].flags & DESC_DPL_MASK);
 327        env->segs[R_GS].flags &= ~(env->segs[R_GS].flags & DESC_DPL_MASK);
 328        env->segs[R_SS].flags &= ~(env->segs[R_SS].flags & DESC_DPL_MASK);
 329    }
 330
 331    /* Older versions of QEMU incorrectly used CS.DPL as the CPL when
 332     * running under KVM.  This is wrong for conforming code segments.
 333     * Luckily, in our implementation the CPL field of hflags is redundant
 334     * and we can get the right value from the SS descriptor privilege level.
 335     */
 336    env->hflags &= ~HF_CPL_MASK;
 337    env->hflags |= (env->segs[R_SS].flags >> DESC_DPL_SHIFT) & HF_CPL_MASK;
 338
 339#ifdef CONFIG_KVM
 340    if ((env->hflags & HF_GUEST_MASK) &&
 341        (!env->nested_state ||
 342        !(env->nested_state->flags & KVM_STATE_NESTED_GUEST_MODE))) {
 343        error_report("vCPU set in guest-mode inconsistent with "
 344                     "migrated kernel nested state");
 345        return -EINVAL;
 346    }
 347#endif
 348
 349    /*
 350     * There are cases that we can get valid exception_nr with both
 351     * exception_pending and exception_injected being cleared.
 352     * This can happen in one of the following scenarios:
 353     * 1) Source is older QEMU without KVM_CAP_EXCEPTION_PAYLOAD support.
 354     * 2) Source is running on kernel without KVM_CAP_EXCEPTION_PAYLOAD support.
 355     * 3) "cpu/exception_info" subsection not sent because there is no exception
 356     *    pending or guest wasn't running L2 (See comment in cpu_pre_save()).
 357     *
 358     * In those cases, we can just deduce that a valid exception_nr means
 359     * we can treat the exception as already injected.
 360     */
 361    if ((env->exception_nr != -1) &&
 362        !env->exception_pending && !env->exception_injected) {
 363        env->exception_injected = 1;
 364    }
 365
 366    env->fpstt = (env->fpus_vmstate >> 11) & 7;
 367    env->fpus = env->fpus_vmstate & ~0x3800;
 368    env->fptag_vmstate ^= 0xff;
 369    for(i = 0; i < 8; i++) {
 370        env->fptags[i] = (env->fptag_vmstate >> i) & 1;
 371    }
 372    if (tcg_enabled()) {
 373        target_ulong dr7;
 374        update_fp_status(env);
 375        update_mxcsr_status(env);
 376
 377        cpu_breakpoint_remove_all(cs, BP_CPU);
 378        cpu_watchpoint_remove_all(cs, BP_CPU);
 379
 380        /* Indicate all breakpoints disabled, as they are, then
 381           let the helper re-enable them.  */
 382        dr7 = env->dr[7];
 383        env->dr[7] = dr7 & ~(DR7_GLOBAL_BP_MASK | DR7_LOCAL_BP_MASK);
 384        cpu_x86_update_dr7(env, dr7);
 385    }
 386    tlb_flush(cs);
 387    return 0;
 388}
 389
 390static bool async_pf_msr_needed(void *opaque)
 391{
 392    X86CPU *cpu = opaque;
 393
 394    return cpu->env.async_pf_en_msr != 0;
 395}
 396
 397static bool async_pf_int_msr_needed(void *opaque)
 398{
 399    X86CPU *cpu = opaque;
 400
 401    return cpu->env.async_pf_int_msr != 0;
 402}
 403
 404static bool pv_eoi_msr_needed(void *opaque)
 405{
 406    X86CPU *cpu = opaque;
 407
 408    return cpu->env.pv_eoi_en_msr != 0;
 409}
 410
 411static bool steal_time_msr_needed(void *opaque)
 412{
 413    X86CPU *cpu = opaque;
 414
 415    return cpu->env.steal_time_msr != 0;
 416}
 417
 418static bool exception_info_needed(void *opaque)
 419{
 420    X86CPU *cpu = opaque;
 421    CPUX86State *env = &cpu->env;
 422
 423    /*
 424     * It is important to save exception-info only in case
 425     * we need to distinguish between a pending and injected
 426     * exception. Which is only required in case there is a
 427     * pending exception and vCPU is running L2.
 428     * For more info, refer to comment in cpu_pre_save().
 429     */
 430    return env->exception_pending && (env->hflags & HF_GUEST_MASK);
 431}
 432
 433static const VMStateDescription vmstate_exception_info = {
 434    .name = "cpu/exception_info",
 435    .version_id = 1,
 436    .minimum_version_id = 1,
 437    .needed = exception_info_needed,
 438    .fields = (VMStateField[]) {
 439        VMSTATE_UINT8(env.exception_pending, X86CPU),
 440        VMSTATE_UINT8(env.exception_injected, X86CPU),
 441        VMSTATE_UINT8(env.exception_has_payload, X86CPU),
 442        VMSTATE_UINT64(env.exception_payload, X86CPU),
 443        VMSTATE_END_OF_LIST()
 444    }
 445};
 446
 447/* Poll control MSR enabled by default */
 448static bool poll_control_msr_needed(void *opaque)
 449{
 450    X86CPU *cpu = opaque;
 451
 452    return cpu->env.poll_control_msr != 1;
 453}
 454
 455static const VMStateDescription vmstate_steal_time_msr = {
 456    .name = "cpu/steal_time_msr",
 457    .version_id = 1,
 458    .minimum_version_id = 1,
 459    .needed = steal_time_msr_needed,
 460    .fields = (VMStateField[]) {
 461        VMSTATE_UINT64(env.steal_time_msr, X86CPU),
 462        VMSTATE_END_OF_LIST()
 463    }
 464};
 465
 466static const VMStateDescription vmstate_async_pf_msr = {
 467    .name = "cpu/async_pf_msr",
 468    .version_id = 1,
 469    .minimum_version_id = 1,
 470    .needed = async_pf_msr_needed,
 471    .fields = (VMStateField[]) {
 472        VMSTATE_UINT64(env.async_pf_en_msr, X86CPU),
 473        VMSTATE_END_OF_LIST()
 474    }
 475};
 476
 477static const VMStateDescription vmstate_async_pf_int_msr = {
 478    .name = "cpu/async_pf_int_msr",
 479    .version_id = 1,
 480    .minimum_version_id = 1,
 481    .needed = async_pf_int_msr_needed,
 482    .fields = (VMStateField[]) {
 483        VMSTATE_UINT64(env.async_pf_int_msr, X86CPU),
 484        VMSTATE_END_OF_LIST()
 485    }
 486};
 487
 488static const VMStateDescription vmstate_pv_eoi_msr = {
 489    .name = "cpu/async_pv_eoi_msr",
 490    .version_id = 1,
 491    .minimum_version_id = 1,
 492    .needed = pv_eoi_msr_needed,
 493    .fields = (VMStateField[]) {
 494        VMSTATE_UINT64(env.pv_eoi_en_msr, X86CPU),
 495        VMSTATE_END_OF_LIST()
 496    }
 497};
 498
 499static const VMStateDescription vmstate_poll_control_msr = {
 500    .name = "cpu/poll_control_msr",
 501    .version_id = 1,
 502    .minimum_version_id = 1,
 503    .needed = poll_control_msr_needed,
 504    .fields = (VMStateField[]) {
 505        VMSTATE_UINT64(env.poll_control_msr, X86CPU),
 506        VMSTATE_END_OF_LIST()
 507    }
 508};
 509
 510static bool fpop_ip_dp_needed(void *opaque)
 511{
 512    X86CPU *cpu = opaque;
 513    CPUX86State *env = &cpu->env;
 514
 515    return env->fpop != 0 || env->fpip != 0 || env->fpdp != 0;
 516}
 517
 518static const VMStateDescription vmstate_fpop_ip_dp = {
 519    .name = "cpu/fpop_ip_dp",
 520    .version_id = 1,
 521    .minimum_version_id = 1,
 522    .needed = fpop_ip_dp_needed,
 523    .fields = (VMStateField[]) {
 524        VMSTATE_UINT16(env.fpop, X86CPU),
 525        VMSTATE_UINT64(env.fpip, X86CPU),
 526        VMSTATE_UINT64(env.fpdp, X86CPU),
 527        VMSTATE_END_OF_LIST()
 528    }
 529};
 530
 531static bool tsc_adjust_needed(void *opaque)
 532{
 533    X86CPU *cpu = opaque;
 534    CPUX86State *env = &cpu->env;
 535
 536    return env->tsc_adjust != 0;
 537}
 538
 539static const VMStateDescription vmstate_msr_tsc_adjust = {
 540    .name = "cpu/msr_tsc_adjust",
 541    .version_id = 1,
 542    .minimum_version_id = 1,
 543    .needed = tsc_adjust_needed,
 544    .fields = (VMStateField[]) {
 545        VMSTATE_UINT64(env.tsc_adjust, X86CPU),
 546        VMSTATE_END_OF_LIST()
 547    }
 548};
 549
 550static bool msr_smi_count_needed(void *opaque)
 551{
 552    X86CPU *cpu = opaque;
 553    CPUX86State *env = &cpu->env;
 554
 555    return cpu->migrate_smi_count && env->msr_smi_count != 0;
 556}
 557
 558static const VMStateDescription vmstate_msr_smi_count = {
 559    .name = "cpu/msr_smi_count",
 560    .version_id = 1,
 561    .minimum_version_id = 1,
 562    .needed = msr_smi_count_needed,
 563    .fields = (VMStateField[]) {
 564        VMSTATE_UINT64(env.msr_smi_count, X86CPU),
 565        VMSTATE_END_OF_LIST()
 566    }
 567};
 568
 569static bool tscdeadline_needed(void *opaque)
 570{
 571    X86CPU *cpu = opaque;
 572    CPUX86State *env = &cpu->env;
 573
 574    return env->tsc_deadline != 0;
 575}
 576
 577static const VMStateDescription vmstate_msr_tscdeadline = {
 578    .name = "cpu/msr_tscdeadline",
 579    .version_id = 1,
 580    .minimum_version_id = 1,
 581    .needed = tscdeadline_needed,
 582    .fields = (VMStateField[]) {
 583        VMSTATE_UINT64(env.tsc_deadline, X86CPU),
 584        VMSTATE_END_OF_LIST()
 585    }
 586};
 587
 588static bool misc_enable_needed(void *opaque)
 589{
 590    X86CPU *cpu = opaque;
 591    CPUX86State *env = &cpu->env;
 592
 593    return env->msr_ia32_misc_enable != MSR_IA32_MISC_ENABLE_DEFAULT;
 594}
 595
 596static bool feature_control_needed(void *opaque)
 597{
 598    X86CPU *cpu = opaque;
 599    CPUX86State *env = &cpu->env;
 600
 601    return env->msr_ia32_feature_control != 0;
 602}
 603
 604static const VMStateDescription vmstate_msr_ia32_misc_enable = {
 605    .name = "cpu/msr_ia32_misc_enable",
 606    .version_id = 1,
 607    .minimum_version_id = 1,
 608    .needed = misc_enable_needed,
 609    .fields = (VMStateField[]) {
 610        VMSTATE_UINT64(env.msr_ia32_misc_enable, X86CPU),
 611        VMSTATE_END_OF_LIST()
 612    }
 613};
 614
 615static const VMStateDescription vmstate_msr_ia32_feature_control = {
 616    .name = "cpu/msr_ia32_feature_control",
 617    .version_id = 1,
 618    .minimum_version_id = 1,
 619    .needed = feature_control_needed,
 620    .fields = (VMStateField[]) {
 621        VMSTATE_UINT64(env.msr_ia32_feature_control, X86CPU),
 622        VMSTATE_END_OF_LIST()
 623    }
 624};
 625
 626static bool pmu_enable_needed(void *opaque)
 627{
 628    X86CPU *cpu = opaque;
 629    CPUX86State *env = &cpu->env;
 630    int i;
 631
 632    if (env->msr_fixed_ctr_ctrl || env->msr_global_ctrl ||
 633        env->msr_global_status || env->msr_global_ovf_ctrl) {
 634        return true;
 635    }
 636    for (i = 0; i < MAX_FIXED_COUNTERS; i++) {
 637        if (env->msr_fixed_counters[i]) {
 638            return true;
 639        }
 640    }
 641    for (i = 0; i < MAX_GP_COUNTERS; i++) {
 642        if (env->msr_gp_counters[i] || env->msr_gp_evtsel[i]) {
 643            return true;
 644        }
 645    }
 646
 647    return false;
 648}
 649
 650static const VMStateDescription vmstate_msr_architectural_pmu = {
 651    .name = "cpu/msr_architectural_pmu",
 652    .version_id = 1,
 653    .minimum_version_id = 1,
 654    .needed = pmu_enable_needed,
 655    .fields = (VMStateField[]) {
 656        VMSTATE_UINT64(env.msr_fixed_ctr_ctrl, X86CPU),
 657        VMSTATE_UINT64(env.msr_global_ctrl, X86CPU),
 658        VMSTATE_UINT64(env.msr_global_status, X86CPU),
 659        VMSTATE_UINT64(env.msr_global_ovf_ctrl, X86CPU),
 660        VMSTATE_UINT64_ARRAY(env.msr_fixed_counters, X86CPU, MAX_FIXED_COUNTERS),
 661        VMSTATE_UINT64_ARRAY(env.msr_gp_counters, X86CPU, MAX_GP_COUNTERS),
 662        VMSTATE_UINT64_ARRAY(env.msr_gp_evtsel, X86CPU, MAX_GP_COUNTERS),
 663        VMSTATE_END_OF_LIST()
 664    }
 665};
 666
 667static bool mpx_needed(void *opaque)
 668{
 669    X86CPU *cpu = opaque;
 670    CPUX86State *env = &cpu->env;
 671    unsigned int i;
 672
 673    for (i = 0; i < 4; i++) {
 674        if (env->bnd_regs[i].lb || env->bnd_regs[i].ub) {
 675            return true;
 676        }
 677    }
 678
 679    if (env->bndcs_regs.cfgu || env->bndcs_regs.sts) {
 680        return true;
 681    }
 682
 683    return !!env->msr_bndcfgs;
 684}
 685
 686static const VMStateDescription vmstate_mpx = {
 687    .name = "cpu/mpx",
 688    .version_id = 1,
 689    .minimum_version_id = 1,
 690    .needed = mpx_needed,
 691    .fields = (VMStateField[]) {
 692        VMSTATE_BND_REGS(env.bnd_regs, X86CPU, 4),
 693        VMSTATE_UINT64(env.bndcs_regs.cfgu, X86CPU),
 694        VMSTATE_UINT64(env.bndcs_regs.sts, X86CPU),
 695        VMSTATE_UINT64(env.msr_bndcfgs, X86CPU),
 696        VMSTATE_END_OF_LIST()
 697    }
 698};
 699
 700static bool hyperv_hypercall_enable_needed(void *opaque)
 701{
 702    X86CPU *cpu = opaque;
 703    CPUX86State *env = &cpu->env;
 704
 705    return env->msr_hv_hypercall != 0 || env->msr_hv_guest_os_id != 0;
 706}
 707
 708static const VMStateDescription vmstate_msr_hyperv_hypercall = {
 709    .name = "cpu/msr_hyperv_hypercall",
 710    .version_id = 1,
 711    .minimum_version_id = 1,
 712    .needed = hyperv_hypercall_enable_needed,
 713    .fields = (VMStateField[]) {
 714        VMSTATE_UINT64(env.msr_hv_guest_os_id, X86CPU),
 715        VMSTATE_UINT64(env.msr_hv_hypercall, X86CPU),
 716        VMSTATE_END_OF_LIST()
 717    }
 718};
 719
 720static bool hyperv_vapic_enable_needed(void *opaque)
 721{
 722    X86CPU *cpu = opaque;
 723    CPUX86State *env = &cpu->env;
 724
 725    return env->msr_hv_vapic != 0;
 726}
 727
 728static const VMStateDescription vmstate_msr_hyperv_vapic = {
 729    .name = "cpu/msr_hyperv_vapic",
 730    .version_id = 1,
 731    .minimum_version_id = 1,
 732    .needed = hyperv_vapic_enable_needed,
 733    .fields = (VMStateField[]) {
 734        VMSTATE_UINT64(env.msr_hv_vapic, X86CPU),
 735        VMSTATE_END_OF_LIST()
 736    }
 737};
 738
 739static bool hyperv_time_enable_needed(void *opaque)
 740{
 741    X86CPU *cpu = opaque;
 742    CPUX86State *env = &cpu->env;
 743
 744    return env->msr_hv_tsc != 0;
 745}
 746
 747static const VMStateDescription vmstate_msr_hyperv_time = {
 748    .name = "cpu/msr_hyperv_time",
 749    .version_id = 1,
 750    .minimum_version_id = 1,
 751    .needed = hyperv_time_enable_needed,
 752    .fields = (VMStateField[]) {
 753        VMSTATE_UINT64(env.msr_hv_tsc, X86CPU),
 754        VMSTATE_END_OF_LIST()
 755    }
 756};
 757
 758static bool hyperv_crash_enable_needed(void *opaque)
 759{
 760    X86CPU *cpu = opaque;
 761    CPUX86State *env = &cpu->env;
 762    int i;
 763
 764    for (i = 0; i < HV_CRASH_PARAMS; i++) {
 765        if (env->msr_hv_crash_params[i]) {
 766            return true;
 767        }
 768    }
 769    return false;
 770}
 771
 772static const VMStateDescription vmstate_msr_hyperv_crash = {
 773    .name = "cpu/msr_hyperv_crash",
 774    .version_id = 1,
 775    .minimum_version_id = 1,
 776    .needed = hyperv_crash_enable_needed,
 777    .fields = (VMStateField[]) {
 778        VMSTATE_UINT64_ARRAY(env.msr_hv_crash_params, X86CPU, HV_CRASH_PARAMS),
 779        VMSTATE_END_OF_LIST()
 780    }
 781};
 782
 783static bool hyperv_runtime_enable_needed(void *opaque)
 784{
 785    X86CPU *cpu = opaque;
 786    CPUX86State *env = &cpu->env;
 787
 788    if (!hyperv_feat_enabled(cpu, HYPERV_FEAT_RUNTIME)) {
 789        return false;
 790    }
 791
 792    return env->msr_hv_runtime != 0;
 793}
 794
 795static const VMStateDescription vmstate_msr_hyperv_runtime = {
 796    .name = "cpu/msr_hyperv_runtime",
 797    .version_id = 1,
 798    .minimum_version_id = 1,
 799    .needed = hyperv_runtime_enable_needed,
 800    .fields = (VMStateField[]) {
 801        VMSTATE_UINT64(env.msr_hv_runtime, X86CPU),
 802        VMSTATE_END_OF_LIST()
 803    }
 804};
 805
 806static bool hyperv_synic_enable_needed(void *opaque)
 807{
 808    X86CPU *cpu = opaque;
 809    CPUX86State *env = &cpu->env;
 810    int i;
 811
 812    if (env->msr_hv_synic_control != 0 ||
 813        env->msr_hv_synic_evt_page != 0 ||
 814        env->msr_hv_synic_msg_page != 0) {
 815        return true;
 816    }
 817
 818    for (i = 0; i < ARRAY_SIZE(env->msr_hv_synic_sint); i++) {
 819        if (env->msr_hv_synic_sint[i] != 0) {
 820            return true;
 821        }
 822    }
 823
 824    return false;
 825}
 826
 827static int hyperv_synic_post_load(void *opaque, int version_id)
 828{
 829    X86CPU *cpu = opaque;
 830    hyperv_x86_synic_update(cpu);
 831    return 0;
 832}
 833
 834static const VMStateDescription vmstate_msr_hyperv_synic = {
 835    .name = "cpu/msr_hyperv_synic",
 836    .version_id = 1,
 837    .minimum_version_id = 1,
 838    .needed = hyperv_synic_enable_needed,
 839    .post_load = hyperv_synic_post_load,
 840    .fields = (VMStateField[]) {
 841        VMSTATE_UINT64(env.msr_hv_synic_control, X86CPU),
 842        VMSTATE_UINT64(env.msr_hv_synic_evt_page, X86CPU),
 843        VMSTATE_UINT64(env.msr_hv_synic_msg_page, X86CPU),
 844        VMSTATE_UINT64_ARRAY(env.msr_hv_synic_sint, X86CPU, HV_SINT_COUNT),
 845        VMSTATE_END_OF_LIST()
 846    }
 847};
 848
 849static bool hyperv_stimer_enable_needed(void *opaque)
 850{
 851    X86CPU *cpu = opaque;
 852    CPUX86State *env = &cpu->env;
 853    int i;
 854
 855    for (i = 0; i < ARRAY_SIZE(env->msr_hv_stimer_config); i++) {
 856        if (env->msr_hv_stimer_config[i] || env->msr_hv_stimer_count[i]) {
 857            return true;
 858        }
 859    }
 860    return false;
 861}
 862
 863static const VMStateDescription vmstate_msr_hyperv_stimer = {
 864    .name = "cpu/msr_hyperv_stimer",
 865    .version_id = 1,
 866    .minimum_version_id = 1,
 867    .needed = hyperv_stimer_enable_needed,
 868    .fields = (VMStateField[]) {
 869        VMSTATE_UINT64_ARRAY(env.msr_hv_stimer_config, X86CPU,
 870                             HV_STIMER_COUNT),
 871        VMSTATE_UINT64_ARRAY(env.msr_hv_stimer_count, X86CPU, HV_STIMER_COUNT),
 872        VMSTATE_END_OF_LIST()
 873    }
 874};
 875
 876static bool hyperv_reenlightenment_enable_needed(void *opaque)
 877{
 878    X86CPU *cpu = opaque;
 879    CPUX86State *env = &cpu->env;
 880
 881    return env->msr_hv_reenlightenment_control != 0 ||
 882        env->msr_hv_tsc_emulation_control != 0 ||
 883        env->msr_hv_tsc_emulation_status != 0;
 884}
 885
 886static int hyperv_reenlightenment_post_load(void *opaque, int version_id)
 887{
 888    X86CPU *cpu = opaque;
 889    CPUX86State *env = &cpu->env;
 890
 891    /*
 892     * KVM doesn't fully support re-enlightenment notifications so we need to
 893     * make sure TSC frequency doesn't change upon migration.
 894     */
 895    if ((env->msr_hv_reenlightenment_control & HV_REENLIGHTENMENT_ENABLE_BIT) &&
 896        !env->user_tsc_khz) {
 897        error_report("Guest enabled re-enlightenment notifications, "
 898                     "'tsc-frequency=' has to be specified");
 899        return -EINVAL;
 900    }
 901
 902    return 0;
 903}
 904
 905static const VMStateDescription vmstate_msr_hyperv_reenlightenment = {
 906    .name = "cpu/msr_hyperv_reenlightenment",
 907    .version_id = 1,
 908    .minimum_version_id = 1,
 909    .needed = hyperv_reenlightenment_enable_needed,
 910    .post_load = hyperv_reenlightenment_post_load,
 911    .fields = (VMStateField[]) {
 912        VMSTATE_UINT64(env.msr_hv_reenlightenment_control, X86CPU),
 913        VMSTATE_UINT64(env.msr_hv_tsc_emulation_control, X86CPU),
 914        VMSTATE_UINT64(env.msr_hv_tsc_emulation_status, X86CPU),
 915        VMSTATE_END_OF_LIST()
 916    }
 917};
 918
 919static bool avx512_needed(void *opaque)
 920{
 921    X86CPU *cpu = opaque;
 922    CPUX86State *env = &cpu->env;
 923    unsigned int i;
 924
 925    for (i = 0; i < NB_OPMASK_REGS; i++) {
 926        if (env->opmask_regs[i]) {
 927            return true;
 928        }
 929    }
 930
 931    for (i = 0; i < CPU_NB_REGS; i++) {
 932#define ENV_XMM(reg, field) (env->xmm_regs[reg].ZMM_Q(field))
 933        if (ENV_XMM(i, 4) || ENV_XMM(i, 6) ||
 934            ENV_XMM(i, 5) || ENV_XMM(i, 7)) {
 935            return true;
 936        }
 937#ifdef TARGET_X86_64
 938        if (ENV_XMM(i+16, 0) || ENV_XMM(i+16, 1) ||
 939            ENV_XMM(i+16, 2) || ENV_XMM(i+16, 3) ||
 940            ENV_XMM(i+16, 4) || ENV_XMM(i+16, 5) ||
 941            ENV_XMM(i+16, 6) || ENV_XMM(i+16, 7)) {
 942            return true;
 943        }
 944#endif
 945    }
 946
 947    return false;
 948}
 949
 950static const VMStateDescription vmstate_avx512 = {
 951    .name = "cpu/avx512",
 952    .version_id = 1,
 953    .minimum_version_id = 1,
 954    .needed = avx512_needed,
 955    .fields = (VMStateField[]) {
 956        VMSTATE_UINT64_ARRAY(env.opmask_regs, X86CPU, NB_OPMASK_REGS),
 957        VMSTATE_ZMMH_REGS_VARS(env.xmm_regs, X86CPU, 0),
 958#ifdef TARGET_X86_64
 959        VMSTATE_Hi16_ZMM_REGS_VARS(env.xmm_regs, X86CPU, 16),
 960#endif
 961        VMSTATE_END_OF_LIST()
 962    }
 963};
 964
 965static bool xss_needed(void *opaque)
 966{
 967    X86CPU *cpu = opaque;
 968    CPUX86State *env = &cpu->env;
 969
 970    return env->xss != 0;
 971}
 972
 973static const VMStateDescription vmstate_xss = {
 974    .name = "cpu/xss",
 975    .version_id = 1,
 976    .minimum_version_id = 1,
 977    .needed = xss_needed,
 978    .fields = (VMStateField[]) {
 979        VMSTATE_UINT64(env.xss, X86CPU),
 980        VMSTATE_END_OF_LIST()
 981    }
 982};
 983
 984static bool umwait_needed(void *opaque)
 985{
 986    X86CPU *cpu = opaque;
 987    CPUX86State *env = &cpu->env;
 988
 989    return env->umwait != 0;
 990}
 991
 992static const VMStateDescription vmstate_umwait = {
 993    .name = "cpu/umwait",
 994    .version_id = 1,
 995    .minimum_version_id = 1,
 996    .needed = umwait_needed,
 997    .fields = (VMStateField[]) {
 998        VMSTATE_UINT32(env.umwait, X86CPU),
 999        VMSTATE_END_OF_LIST()
1000    }
1001};
1002
1003static bool pkru_needed(void *opaque)
1004{
1005    X86CPU *cpu = opaque;
1006    CPUX86State *env = &cpu->env;
1007
1008    return env->pkru != 0;
1009}
1010
1011static const VMStateDescription vmstate_pkru = {
1012    .name = "cpu/pkru",
1013    .version_id = 1,
1014    .minimum_version_id = 1,
1015    .needed = pkru_needed,
1016    .fields = (VMStateField[]){
1017        VMSTATE_UINT32(env.pkru, X86CPU),
1018        VMSTATE_END_OF_LIST()
1019    }
1020};
1021
1022static bool pkrs_needed(void *opaque)
1023{
1024    X86CPU *cpu = opaque;
1025    CPUX86State *env = &cpu->env;
1026
1027    return env->pkrs != 0;
1028}
1029
1030static const VMStateDescription vmstate_pkrs = {
1031    .name = "cpu/pkrs",
1032    .version_id = 1,
1033    .minimum_version_id = 1,
1034    .needed = pkrs_needed,
1035    .fields = (VMStateField[]){
1036        VMSTATE_UINT32(env.pkrs, X86CPU),
1037        VMSTATE_END_OF_LIST()
1038    }
1039};
1040
1041static bool tsc_khz_needed(void *opaque)
1042{
1043    X86CPU *cpu = opaque;
1044    CPUX86State *env = &cpu->env;
1045    MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
1046    X86MachineClass *x86mc = X86_MACHINE_CLASS(mc);
1047    return env->tsc_khz && x86mc->save_tsc_khz;
1048}
1049
1050static const VMStateDescription vmstate_tsc_khz = {
1051    .name = "cpu/tsc_khz",
1052    .version_id = 1,
1053    .minimum_version_id = 1,
1054    .needed = tsc_khz_needed,
1055    .fields = (VMStateField[]) {
1056        VMSTATE_INT64(env.tsc_khz, X86CPU),
1057        VMSTATE_END_OF_LIST()
1058    }
1059};
1060
1061#ifdef CONFIG_KVM
1062
1063static bool vmx_vmcs12_needed(void *opaque)
1064{
1065    struct kvm_nested_state *nested_state = opaque;
1066    return (nested_state->size >
1067            offsetof(struct kvm_nested_state, data.vmx[0].vmcs12));
1068}
1069
1070static const VMStateDescription vmstate_vmx_vmcs12 = {
1071    .name = "cpu/kvm_nested_state/vmx/vmcs12",
1072    .version_id = 1,
1073    .minimum_version_id = 1,
1074    .needed = vmx_vmcs12_needed,
1075    .fields = (VMStateField[]) {
1076        VMSTATE_UINT8_ARRAY(data.vmx[0].vmcs12,
1077                            struct kvm_nested_state,
1078                            KVM_STATE_NESTED_VMX_VMCS_SIZE),
1079        VMSTATE_END_OF_LIST()
1080    }
1081};
1082
1083static bool vmx_shadow_vmcs12_needed(void *opaque)
1084{
1085    struct kvm_nested_state *nested_state = opaque;
1086    return (nested_state->size >
1087            offsetof(struct kvm_nested_state, data.vmx[0].shadow_vmcs12));
1088}
1089
1090static const VMStateDescription vmstate_vmx_shadow_vmcs12 = {
1091    .name = "cpu/kvm_nested_state/vmx/shadow_vmcs12",
1092    .version_id = 1,
1093    .minimum_version_id = 1,
1094    .needed = vmx_shadow_vmcs12_needed,
1095    .fields = (VMStateField[]) {
1096        VMSTATE_UINT8_ARRAY(data.vmx[0].shadow_vmcs12,
1097                            struct kvm_nested_state,
1098                            KVM_STATE_NESTED_VMX_VMCS_SIZE),
1099        VMSTATE_END_OF_LIST()
1100    }
1101};
1102
1103static bool vmx_nested_state_needed(void *opaque)
1104{
1105    struct kvm_nested_state *nested_state = opaque;
1106
1107    return (nested_state->format == KVM_STATE_NESTED_FORMAT_VMX &&
1108            nested_state->hdr.vmx.vmxon_pa != -1ull);
1109}
1110
1111static const VMStateDescription vmstate_vmx_nested_state = {
1112    .name = "cpu/kvm_nested_state/vmx",
1113    .version_id = 1,
1114    .minimum_version_id = 1,
1115    .needed = vmx_nested_state_needed,
1116    .fields = (VMStateField[]) {
1117        VMSTATE_U64(hdr.vmx.vmxon_pa, struct kvm_nested_state),
1118        VMSTATE_U64(hdr.vmx.vmcs12_pa, struct kvm_nested_state),
1119        VMSTATE_U16(hdr.vmx.smm.flags, struct kvm_nested_state),
1120        VMSTATE_END_OF_LIST()
1121    },
1122    .subsections = (const VMStateDescription*[]) {
1123        &vmstate_vmx_vmcs12,
1124        &vmstate_vmx_shadow_vmcs12,
1125        NULL,
1126    }
1127};
1128
1129static bool svm_nested_state_needed(void *opaque)
1130{
1131    struct kvm_nested_state *nested_state = opaque;
1132
1133    /*
1134     * HF_GUEST_MASK and HF2_GIF_MASK are already serialized
1135     * via hflags and hflags2, all that's left is the opaque
1136     * nested state blob.
1137     */
1138    return (nested_state->format == KVM_STATE_NESTED_FORMAT_SVM &&
1139            nested_state->size > offsetof(struct kvm_nested_state, data));
1140}
1141
1142static const VMStateDescription vmstate_svm_nested_state = {
1143    .name = "cpu/kvm_nested_state/svm",
1144    .version_id = 1,
1145    .minimum_version_id = 1,
1146    .needed = svm_nested_state_needed,
1147    .fields = (VMStateField[]) {
1148        VMSTATE_U64(hdr.svm.vmcb_pa, struct kvm_nested_state),
1149        VMSTATE_UINT8_ARRAY(data.svm[0].vmcb12,
1150                            struct kvm_nested_state,
1151                            KVM_STATE_NESTED_SVM_VMCB_SIZE),
1152        VMSTATE_END_OF_LIST()
1153    }
1154};
1155
1156static bool nested_state_needed(void *opaque)
1157{
1158    X86CPU *cpu = opaque;
1159    CPUX86State *env = &cpu->env;
1160
1161    return (env->nested_state &&
1162            (vmx_nested_state_needed(env->nested_state) ||
1163             svm_nested_state_needed(env->nested_state)));
1164}
1165
1166static int nested_state_post_load(void *opaque, int version_id)
1167{
1168    X86CPU *cpu = opaque;
1169    CPUX86State *env = &cpu->env;
1170    struct kvm_nested_state *nested_state = env->nested_state;
1171    int min_nested_state_len = offsetof(struct kvm_nested_state, data);
1172    int max_nested_state_len = kvm_max_nested_state_length();
1173
1174    /*
1175     * If our kernel don't support setting nested state
1176     * and we have received nested state from migration stream,
1177     * we need to fail migration
1178     */
1179    if (max_nested_state_len <= 0) {
1180        error_report("Received nested state when kernel cannot restore it");
1181        return -EINVAL;
1182    }
1183
1184    /*
1185     * Verify that the size of received nested_state struct
1186     * at least cover required header and is not larger
1187     * than the max size that our kernel support
1188     */
1189    if (nested_state->size < min_nested_state_len) {
1190        error_report("Received nested state size less than min: "
1191                     "len=%d, min=%d",
1192                     nested_state->size, min_nested_state_len);
1193        return -EINVAL;
1194    }
1195    if (nested_state->size > max_nested_state_len) {
1196        error_report("Received unsupported nested state size: "
1197                     "nested_state->size=%d, max=%d",
1198                     nested_state->size, max_nested_state_len);
1199        return -EINVAL;
1200    }
1201
1202    /* Verify format is valid */
1203    if ((nested_state->format != KVM_STATE_NESTED_FORMAT_VMX) &&
1204        (nested_state->format != KVM_STATE_NESTED_FORMAT_SVM)) {
1205        error_report("Received invalid nested state format: %d",
1206                     nested_state->format);
1207        return -EINVAL;
1208    }
1209
1210    return 0;
1211}
1212
1213static const VMStateDescription vmstate_kvm_nested_state = {
1214    .name = "cpu/kvm_nested_state",
1215    .version_id = 1,
1216    .minimum_version_id = 1,
1217    .fields = (VMStateField[]) {
1218        VMSTATE_U16(flags, struct kvm_nested_state),
1219        VMSTATE_U16(format, struct kvm_nested_state),
1220        VMSTATE_U32(size, struct kvm_nested_state),
1221        VMSTATE_END_OF_LIST()
1222    },
1223    .subsections = (const VMStateDescription*[]) {
1224        &vmstate_vmx_nested_state,
1225        &vmstate_svm_nested_state,
1226        NULL
1227    }
1228};
1229
1230static const VMStateDescription vmstate_nested_state = {
1231    .name = "cpu/nested_state",
1232    .version_id = 1,
1233    .minimum_version_id = 1,
1234    .needed = nested_state_needed,
1235    .post_load = nested_state_post_load,
1236    .fields = (VMStateField[]) {
1237        VMSTATE_STRUCT_POINTER(env.nested_state, X86CPU,
1238                vmstate_kvm_nested_state,
1239                struct kvm_nested_state),
1240        VMSTATE_END_OF_LIST()
1241    }
1242};
1243
1244#endif
1245
1246static bool mcg_ext_ctl_needed(void *opaque)
1247{
1248    X86CPU *cpu = opaque;
1249    CPUX86State *env = &cpu->env;
1250    return cpu->enable_lmce && env->mcg_ext_ctl;
1251}
1252
1253static const VMStateDescription vmstate_mcg_ext_ctl = {
1254    .name = "cpu/mcg_ext_ctl",
1255    .version_id = 1,
1256    .minimum_version_id = 1,
1257    .needed = mcg_ext_ctl_needed,
1258    .fields = (VMStateField[]) {
1259        VMSTATE_UINT64(env.mcg_ext_ctl, X86CPU),
1260        VMSTATE_END_OF_LIST()
1261    }
1262};
1263
1264static bool spec_ctrl_needed(void *opaque)
1265{
1266    X86CPU *cpu = opaque;
1267    CPUX86State *env = &cpu->env;
1268
1269    return env->spec_ctrl != 0;
1270}
1271
1272static const VMStateDescription vmstate_spec_ctrl = {
1273    .name = "cpu/spec_ctrl",
1274    .version_id = 1,
1275    .minimum_version_id = 1,
1276    .needed = spec_ctrl_needed,
1277    .fields = (VMStateField[]){
1278        VMSTATE_UINT64(env.spec_ctrl, X86CPU),
1279        VMSTATE_END_OF_LIST()
1280    }
1281};
1282
1283static bool intel_pt_enable_needed(void *opaque)
1284{
1285    X86CPU *cpu = opaque;
1286    CPUX86State *env = &cpu->env;
1287    int i;
1288
1289    if (env->msr_rtit_ctrl || env->msr_rtit_status ||
1290        env->msr_rtit_output_base || env->msr_rtit_output_mask ||
1291        env->msr_rtit_cr3_match) {
1292        return true;
1293    }
1294
1295    for (i = 0; i < MAX_RTIT_ADDRS; i++) {
1296        if (env->msr_rtit_addrs[i]) {
1297            return true;
1298        }
1299    }
1300
1301    return false;
1302}
1303
1304static const VMStateDescription vmstate_msr_intel_pt = {
1305    .name = "cpu/intel_pt",
1306    .version_id = 1,
1307    .minimum_version_id = 1,
1308    .needed = intel_pt_enable_needed,
1309    .fields = (VMStateField[]) {
1310        VMSTATE_UINT64(env.msr_rtit_ctrl, X86CPU),
1311        VMSTATE_UINT64(env.msr_rtit_status, X86CPU),
1312        VMSTATE_UINT64(env.msr_rtit_output_base, X86CPU),
1313        VMSTATE_UINT64(env.msr_rtit_output_mask, X86CPU),
1314        VMSTATE_UINT64(env.msr_rtit_cr3_match, X86CPU),
1315        VMSTATE_UINT64_ARRAY(env.msr_rtit_addrs, X86CPU, MAX_RTIT_ADDRS),
1316        VMSTATE_END_OF_LIST()
1317    }
1318};
1319
1320static bool virt_ssbd_needed(void *opaque)
1321{
1322    X86CPU *cpu = opaque;
1323    CPUX86State *env = &cpu->env;
1324
1325    return env->virt_ssbd != 0;
1326}
1327
1328static const VMStateDescription vmstate_msr_virt_ssbd = {
1329    .name = "cpu/virt_ssbd",
1330    .version_id = 1,
1331    .minimum_version_id = 1,
1332    .needed = virt_ssbd_needed,
1333    .fields = (VMStateField[]){
1334        VMSTATE_UINT64(env.virt_ssbd, X86CPU),
1335        VMSTATE_END_OF_LIST()
1336    }
1337};
1338
1339static bool svm_npt_needed(void *opaque)
1340{
1341    X86CPU *cpu = opaque;
1342    CPUX86State *env = &cpu->env;
1343
1344    return !!(env->hflags2 & HF2_NPT_MASK);
1345}
1346
1347static const VMStateDescription vmstate_svm_npt = {
1348    .name = "cpu/svn_npt",
1349    .version_id = 1,
1350    .minimum_version_id = 1,
1351    .needed = svm_npt_needed,
1352    .fields = (VMStateField[]){
1353        VMSTATE_UINT64(env.nested_cr3, X86CPU),
1354        VMSTATE_UINT32(env.nested_pg_mode, X86CPU),
1355        VMSTATE_END_OF_LIST()
1356    }
1357};
1358
1359#ifndef TARGET_X86_64
1360static bool intel_efer32_needed(void *opaque)
1361{
1362    X86CPU *cpu = opaque;
1363    CPUX86State *env = &cpu->env;
1364
1365    return env->efer != 0;
1366}
1367
1368static const VMStateDescription vmstate_efer32 = {
1369    .name = "cpu/efer32",
1370    .version_id = 1,
1371    .minimum_version_id = 1,
1372    .needed = intel_efer32_needed,
1373    .fields = (VMStateField[]) {
1374        VMSTATE_UINT64(env.efer, X86CPU),
1375        VMSTATE_END_OF_LIST()
1376    }
1377};
1378#endif
1379
1380static bool msr_tsx_ctrl_needed(void *opaque)
1381{
1382    X86CPU *cpu = opaque;
1383    CPUX86State *env = &cpu->env;
1384
1385    return env->features[FEAT_ARCH_CAPABILITIES] & ARCH_CAP_TSX_CTRL_MSR;
1386}
1387
1388static const VMStateDescription vmstate_msr_tsx_ctrl = {
1389    .name = "cpu/msr_tsx_ctrl",
1390    .version_id = 1,
1391    .minimum_version_id = 1,
1392    .needed = msr_tsx_ctrl_needed,
1393    .fields = (VMStateField[]) {
1394        VMSTATE_UINT32(env.tsx_ctrl, X86CPU),
1395        VMSTATE_END_OF_LIST()
1396    }
1397};
1398
1399const VMStateDescription vmstate_x86_cpu = {
1400    .name = "cpu",
1401    .version_id = 12,
1402    .minimum_version_id = 11,
1403    .pre_save = cpu_pre_save,
1404    .post_load = cpu_post_load,
1405    .fields = (VMStateField[]) {
1406        VMSTATE_UINTTL_ARRAY(env.regs, X86CPU, CPU_NB_REGS),
1407        VMSTATE_UINTTL(env.eip, X86CPU),
1408        VMSTATE_UINTTL(env.eflags, X86CPU),
1409        VMSTATE_UINT32(env.hflags, X86CPU),
1410        /* FPU */
1411        VMSTATE_UINT16(env.fpuc, X86CPU),
1412        VMSTATE_UINT16(env.fpus_vmstate, X86CPU),
1413        VMSTATE_UINT16(env.fptag_vmstate, X86CPU),
1414        VMSTATE_UINT16(env.fpregs_format_vmstate, X86CPU),
1415
1416        VMSTATE_STRUCT_ARRAY(env.fpregs, X86CPU, 8, 0, vmstate_fpreg, FPReg),
1417
1418        VMSTATE_SEGMENT_ARRAY(env.segs, X86CPU, 6),
1419        VMSTATE_SEGMENT(env.ldt, X86CPU),
1420        VMSTATE_SEGMENT(env.tr, X86CPU),
1421        VMSTATE_SEGMENT(env.gdt, X86CPU),
1422        VMSTATE_SEGMENT(env.idt, X86CPU),
1423
1424        VMSTATE_UINT32(env.sysenter_cs, X86CPU),
1425        VMSTATE_UINTTL(env.sysenter_esp, X86CPU),
1426        VMSTATE_UINTTL(env.sysenter_eip, X86CPU),
1427
1428        VMSTATE_UINTTL(env.cr[0], X86CPU),
1429        VMSTATE_UINTTL(env.cr[2], X86CPU),
1430        VMSTATE_UINTTL(env.cr[3], X86CPU),
1431        VMSTATE_UINTTL(env.cr[4], X86CPU),
1432        VMSTATE_UINTTL_ARRAY(env.dr, X86CPU, 8),
1433        /* MMU */
1434        VMSTATE_INT32(env.a20_mask, X86CPU),
1435        /* XMM */
1436        VMSTATE_UINT32(env.mxcsr, X86CPU),
1437        VMSTATE_XMM_REGS(env.xmm_regs, X86CPU, 0),
1438
1439#ifdef TARGET_X86_64
1440        VMSTATE_UINT64(env.efer, X86CPU),
1441        VMSTATE_UINT64(env.star, X86CPU),
1442        VMSTATE_UINT64(env.lstar, X86CPU),
1443        VMSTATE_UINT64(env.cstar, X86CPU),
1444        VMSTATE_UINT64(env.fmask, X86CPU),
1445        VMSTATE_UINT64(env.kernelgsbase, X86CPU),
1446#endif
1447        VMSTATE_UINT32(env.smbase, X86CPU),
1448
1449        VMSTATE_UINT64(env.pat, X86CPU),
1450        VMSTATE_UINT32(env.hflags2, X86CPU),
1451
1452        VMSTATE_UINT64(env.vm_hsave, X86CPU),
1453        VMSTATE_UINT64(env.vm_vmcb, X86CPU),
1454        VMSTATE_UINT64(env.tsc_offset, X86CPU),
1455        VMSTATE_UINT64(env.intercept, X86CPU),
1456        VMSTATE_UINT16(env.intercept_cr_read, X86CPU),
1457        VMSTATE_UINT16(env.intercept_cr_write, X86CPU),
1458        VMSTATE_UINT16(env.intercept_dr_read, X86CPU),
1459        VMSTATE_UINT16(env.intercept_dr_write, X86CPU),
1460        VMSTATE_UINT32(env.intercept_exceptions, X86CPU),
1461        VMSTATE_UINT8(env.v_tpr, X86CPU),
1462        /* MTRRs */
1463        VMSTATE_UINT64_ARRAY(env.mtrr_fixed, X86CPU, 11),
1464        VMSTATE_UINT64(env.mtrr_deftype, X86CPU),
1465        VMSTATE_MTRR_VARS(env.mtrr_var, X86CPU, MSR_MTRRcap_VCNT, 8),
1466        /* KVM-related states */
1467        VMSTATE_INT32(env.interrupt_injected, X86CPU),
1468        VMSTATE_UINT32(env.mp_state, X86CPU),
1469        VMSTATE_UINT64(env.tsc, X86CPU),
1470        VMSTATE_INT32(env.exception_nr, X86CPU),
1471        VMSTATE_UINT8(env.soft_interrupt, X86CPU),
1472        VMSTATE_UINT8(env.nmi_injected, X86CPU),
1473        VMSTATE_UINT8(env.nmi_pending, X86CPU),
1474        VMSTATE_UINT8(env.has_error_code, X86CPU),
1475        VMSTATE_UINT32(env.sipi_vector, X86CPU),
1476        /* MCE */
1477        VMSTATE_UINT64(env.mcg_cap, X86CPU),
1478        VMSTATE_UINT64(env.mcg_status, X86CPU),
1479        VMSTATE_UINT64(env.mcg_ctl, X86CPU),
1480        VMSTATE_UINT64_ARRAY(env.mce_banks, X86CPU, MCE_BANKS_DEF * 4),
1481        /* rdtscp */
1482        VMSTATE_UINT64(env.tsc_aux, X86CPU),
1483        /* KVM pvclock msr */
1484        VMSTATE_UINT64(env.system_time_msr, X86CPU),
1485        VMSTATE_UINT64(env.wall_clock_msr, X86CPU),
1486        /* XSAVE related fields */
1487        VMSTATE_UINT64_V(env.xcr0, X86CPU, 12),
1488        VMSTATE_UINT64_V(env.xstate_bv, X86CPU, 12),
1489        VMSTATE_YMMH_REGS_VARS(env.xmm_regs, X86CPU, 0, 12),
1490        VMSTATE_END_OF_LIST()
1491        /* The above list is not sorted /wrt version numbers, watch out! */
1492    },
1493    .subsections = (const VMStateDescription*[]) {
1494        &vmstate_exception_info,
1495        &vmstate_async_pf_msr,
1496        &vmstate_async_pf_int_msr,
1497        &vmstate_pv_eoi_msr,
1498        &vmstate_steal_time_msr,
1499        &vmstate_poll_control_msr,
1500        &vmstate_fpop_ip_dp,
1501        &vmstate_msr_tsc_adjust,
1502        &vmstate_msr_tscdeadline,
1503        &vmstate_msr_ia32_misc_enable,
1504        &vmstate_msr_ia32_feature_control,
1505        &vmstate_msr_architectural_pmu,
1506        &vmstate_mpx,
1507        &vmstate_msr_hyperv_hypercall,
1508        &vmstate_msr_hyperv_vapic,
1509        &vmstate_msr_hyperv_time,
1510        &vmstate_msr_hyperv_crash,
1511        &vmstate_msr_hyperv_runtime,
1512        &vmstate_msr_hyperv_synic,
1513        &vmstate_msr_hyperv_stimer,
1514        &vmstate_msr_hyperv_reenlightenment,
1515        &vmstate_avx512,
1516        &vmstate_xss,
1517        &vmstate_umwait,
1518        &vmstate_tsc_khz,
1519        &vmstate_msr_smi_count,
1520        &vmstate_pkru,
1521        &vmstate_pkrs,
1522        &vmstate_spec_ctrl,
1523        &vmstate_mcg_ext_ctl,
1524        &vmstate_msr_intel_pt,
1525        &vmstate_msr_virt_ssbd,
1526        &vmstate_svm_npt,
1527#ifndef TARGET_X86_64
1528        &vmstate_efer32,
1529#endif
1530#ifdef CONFIG_KVM
1531        &vmstate_nested_state,
1532#endif
1533        &vmstate_msr_tsx_ctrl,
1534        NULL
1535    }
1536};
1537