target/i386/machine.c

   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 #include "hw/xen/xen.h"
  10
  11 #include "sysemu/kvm.h"
  12 #include "sysemu/kvm_xen.h"
  13 #include "sysemu/tcg.h"
  14
  15 #include "qemu/error-report.h"
  16
  17 static const VMStateDescription vmstate_segment = {
  18     .name = "segment",
  19     .version_id = 1,
  20     .minimum_version_id = 1,
  21     .fields = (VMStateField[]) {
  22         VMSTATE_UINT32(selector, SegmentCache),
  23         VMSTATE_UINTTL(base, SegmentCache),
  24         VMSTATE_UINT32(limit, SegmentCache),
  25         VMSTATE_UINT32(flags, SegmentCache),
  26         VMSTATE_END_OF_LIST()
  27     }
  28 };
  29
  30 #define VMSTATE_SEGMENT(_field, _state) {                            \
  31     .name       = (stringify(_field)),                               \
  32     .size       = sizeof(SegmentCache),                              \
  33     .vmsd       = &vmstate_segment,                                  \
  34     .flags      = VMS_STRUCT,                                        \
  35     .offset     = offsetof(_state, _field)                           \
  36             + type_check(SegmentCache,typeof_field(_state, _field))  \
  37 }
  38
  39 #define VMSTATE_SEGMENT_ARRAY(_field, _state, _n)                    \
  40     VMSTATE_STRUCT_ARRAY(_field, _state, _n, 0, vmstate_segment, SegmentCache)
  41
  42 static const VMStateDescription vmstate_xmm_reg = {
  43     .name = "xmm_reg",
  44     .version_id = 1,
  45     .minimum_version_id = 1,
  46     .fields = (VMStateField[]) {
  47         VMSTATE_UINT64(ZMM_Q(0), ZMMReg),
  48         VMSTATE_UINT64(ZMM_Q(1), ZMMReg),
  49         VMSTATE_END_OF_LIST()
  50     }
  51 };
  52
  53 #define VMSTATE_XMM_REGS(_field, _state, _start)                         \
  54     VMSTATE_STRUCT_SUB_ARRAY(_field, _state, _start, CPU_NB_REGS, 0,     \
  55                              vmstate_xmm_reg, ZMMReg)
  56
  57 /* YMMH format is the same as XMM, but for bits 128-255 */
  58 static const VMStateDescription vmstate_ymmh_reg = {
  59     .name = "ymmh_reg",
  60     .version_id = 1,
  61     .minimum_version_id = 1,
  62     .fields = (VMStateField[]) {
  63         VMSTATE_UINT64(ZMM_Q(2), ZMMReg),
  64         VMSTATE_UINT64(ZMM_Q(3), ZMMReg),
  65         VMSTATE_END_OF_LIST()
  66     }
  67 };
  68
  69 #define VMSTATE_YMMH_REGS_VARS(_field, _state, _start, _v)               \
  70     VMSTATE_STRUCT_SUB_ARRAY(_field, _state, _start, CPU_NB_REGS, _v,    \
  71                              vmstate_ymmh_reg, ZMMReg)
  72
  73 static const VMStateDescription vmstate_zmmh_reg = {
  74     .name = "zmmh_reg",
  75     .version_id = 1,
  76     .minimum_version_id = 1,
  77     .fields = (VMStateField[]) {
  78         VMSTATE_UINT64(ZMM_Q(4), ZMMReg),
  79         VMSTATE_UINT64(ZMM_Q(5), ZMMReg),
  80         VMSTATE_UINT64(ZMM_Q(6), ZMMReg),
  81         VMSTATE_UINT64(ZMM_Q(7), ZMMReg),
  82         VMSTATE_END_OF_LIST()
  83     }
  84 };
  85
  86 #define VMSTATE_ZMMH_REGS_VARS(_field, _state, _start)                   \
  87     VMSTATE_STRUCT_SUB_ARRAY(_field, _state, _start, CPU_NB_REGS, 0,     \
  88                              vmstate_zmmh_reg, ZMMReg)
  89
  90 #ifdef TARGET_X86_64
  91 static const VMStateDescription vmstate_hi16_zmm_reg = {
  92     .name = "hi16_zmm_reg",
  93     .version_id = 1,
  94     .minimum_version_id = 1,
  95     .fields = (VMStateField[]) {
  96         VMSTATE_UINT64(ZMM_Q(0), ZMMReg),
  97         VMSTATE_UINT64(ZMM_Q(1), ZMMReg),
  98         VMSTATE_UINT64(ZMM_Q(2), ZMMReg),
  99         VMSTATE_UINT64(ZMM_Q(3), ZMMReg),
 100         VMSTATE_UINT64(ZMM_Q(4), ZMMReg),
 101         VMSTATE_UINT64(ZMM_Q(5), ZMMReg),
 102         VMSTATE_UINT64(ZMM_Q(6), ZMMReg),
 103         VMSTATE_UINT64(ZMM_Q(7), ZMMReg),
 104         VMSTATE_END_OF_LIST()
 105     }
 106 };
 107
 108 #define VMSTATE_Hi16_ZMM_REGS_VARS(_field, _state, _start)               \
 109     VMSTATE_STRUCT_SUB_ARRAY(_field, _state, _start, CPU_NB_REGS, 0,     \
 110                              vmstate_hi16_zmm_reg, ZMMReg)
 111 #endif
 112
 113 static const VMStateDescription vmstate_bnd_regs = {
 114     .name = "bnd_regs",
 115     .version_id = 1,
 116     .minimum_version_id = 1,
 117     .fields = (VMStateField[]) {
 118         VMSTATE_UINT64(lb, BNDReg),
 119         VMSTATE_UINT64(ub, BNDReg),
 120         VMSTATE_END_OF_LIST()
 121     }
 122 };
 123
 124 #define VMSTATE_BND_REGS(_field, _state, _n)          \
 125     VMSTATE_STRUCT_ARRAY(_field, _state, _n, 0, vmstate_bnd_regs, BNDReg)
 126
 127 static const VMStateDescription vmstate_mtrr_var = {
 128     .name = "mtrr_var",
 129     .version_id = 1,
 130     .minimum_version_id = 1,
 131     .fields = (VMStateField[]) {
 132         VMSTATE_UINT64(base, MTRRVar),
 133         VMSTATE_UINT64(mask, MTRRVar),
 134         VMSTATE_END_OF_LIST()
 135     }
 136 };
 137
 138 #define VMSTATE_MTRR_VARS(_field, _state, _n, _v)                    \
 139     VMSTATE_STRUCT_ARRAY(_field, _state, _n, _v, vmstate_mtrr_var, MTRRVar)
 140
 141 static const VMStateDescription vmstate_lbr_records_var = {
 142     .name = "lbr_records_var",
 143     .version_id = 1,
 144     .minimum_version_id = 1,
 145     .fields = (VMStateField[]) {
 146         VMSTATE_UINT64(from, LBREntry),
 147         VMSTATE_UINT64(to, LBREntry),
 148         VMSTATE_UINT64(info, LBREntry),
 149         VMSTATE_END_OF_LIST()
 150     }
 151 };
 152
 153 #define VMSTATE_LBR_VARS(_field, _state, _n, _v)                    \
 154     VMSTATE_STRUCT_ARRAY(_field, _state, _n, _v, vmstate_lbr_records_var, \
 155                          LBREntry)
 156
 157 typedef struct x86_FPReg_tmp {
 158     FPReg *parent;
 159     uint64_t tmp_mant;
 160     uint16_t tmp_exp;
 161 } x86_FPReg_tmp;
 162
 163 static void cpu_get_fp80(uint64_t *pmant, uint16_t *pexp, floatx80 f)
 164 {
 165     CPU_LDoubleU temp;
 166
 167     temp.d = f;
 168     *pmant = temp.l.lower;
 169     *pexp = temp.l.upper;
 170 }
 171
 172 static floatx80 cpu_set_fp80(uint64_t mant, uint16_t upper)
 173 {
 174     CPU_LDoubleU temp;
 175
 176     temp.l.upper = upper;
 177     temp.l.lower = mant;
 178     return temp.d;
 179 }
 180
 181 static int fpreg_pre_save(void *opaque)
 182 {
 183     x86_FPReg_tmp *tmp = opaque;
 184
 185     /* we save the real CPU data (in case of MMX usage only 'mant'
 186        contains the MMX register */
 187     cpu_get_fp80(&tmp->tmp_mant, &tmp->tmp_exp, tmp->parent->d);
 188
 189     return 0;
 190 }
 191
 192 static int fpreg_post_load(void *opaque, int version)
 193 {
 194     x86_FPReg_tmp *tmp = opaque;
 195
 196     tmp->parent->d = cpu_set_fp80(tmp->tmp_mant, tmp->tmp_exp);
 197     return 0;
 198 }
 199
 200 static const VMStateDescription vmstate_fpreg_tmp = {
 201     .name = "fpreg_tmp",
 202     .post_load = fpreg_post_load,
 203     .pre_save  = fpreg_pre_save,
 204     .fields = (VMStateField[]) {
 205         VMSTATE_UINT64(tmp_mant, x86_FPReg_tmp),
 206         VMSTATE_UINT16(tmp_exp, x86_FPReg_tmp),
 207         VMSTATE_END_OF_LIST()
 208     }
 209 };
 210
 211 static const VMStateDescription vmstate_fpreg = {
 212     .name = "fpreg",
 213     .fields = (VMStateField[]) {
 214         VMSTATE_WITH_TMP(FPReg, x86_FPReg_tmp, vmstate_fpreg_tmp),
 215         VMSTATE_END_OF_LIST()
 216     }
 217 };
 218
 219 static int cpu_pre_save(void *opaque)
 220 {
 221     X86CPU *cpu = opaque;
 222     CPUX86State *env = &cpu->env;
 223     int i;
 224     env->v_tpr = env->int_ctl & V_TPR_MASK;
 225     /* FPU */
 226     env->fpus_vmstate = (env->fpus & ~0x3800) | (env->fpstt & 0x7) << 11;
 227     env->fptag_vmstate = 0;
 228     for(i = 0; i < 8; i++) {
 229         env->fptag_vmstate |= ((!env->fptags[i]) << i);
 230     }
 231
 232     env->fpregs_format_vmstate = 0;
 233
 234     /*
 235      * Real mode guest segments register DPL should be zero.
 236      * Older KVM version were setting it wrongly.
 237      * Fixing it will allow live migration to host with unrestricted guest
 238      * support (otherwise the migration will fail with invalid guest state
 239      * error).
 240      */
 241     if (!(env->cr[0] & CR0_PE_MASK) &&
 242         (env->segs[R_CS].flags >> DESC_DPL_SHIFT & 3) != 0) {
 243         env->segs[R_CS].flags &= ~(env->segs[R_CS].flags & DESC_DPL_MASK);
 244         env->segs[R_DS].flags &= ~(env->segs[R_DS].flags & DESC_DPL_MASK);
 245         env->segs[R_ES].flags &= ~(env->segs[R_ES].flags & DESC_DPL_MASK);
 246         env->segs[R_FS].flags &= ~(env->segs[R_FS].flags & DESC_DPL_MASK);
 247         env->segs[R_GS].flags &= ~(env->segs[R_GS].flags & DESC_DPL_MASK);
 248         env->segs[R_SS].flags &= ~(env->segs[R_SS].flags & DESC_DPL_MASK);
 249     }
 250
 251 #ifdef CONFIG_KVM
 252     /*
 253      * In case vCPU may have enabled VMX, we need to make sure kernel have
 254      * required capabilities in order to perform migration correctly:
 255      *
 256      * 1) We must be able to extract vCPU nested-state from KVM.
 257      *
 258      * 2) In case vCPU is running in guest-mode and it has a pending exception,
 259      * we must be able to determine if it's in a pending or injected state.
 260      * Note that in case KVM don't have required capability to do so,
 261      * a pending/injected exception will always appear as an
 262      * injected exception.
 263      */
 264     if (kvm_enabled() && cpu_vmx_maybe_enabled(env) &&
 265         (!env->nested_state ||
 266          (!kvm_has_exception_payload() && (env->hflags & HF_GUEST_MASK) &&
 267           env->exception_injected))) {
 268         error_report("Guest maybe enabled nested virtualization but kernel "
 269                 "does not support required capabilities to save vCPU "
 270                 "nested state");
 271         return -EINVAL;
 272     }
 273 #endif
 274
 275     /*
 276      * When vCPU is running L2 and exception is still pending,
 277      * it can potentially be intercepted by L1 hypervisor.
 278      * In contrast to an injected exception which cannot be
 279      * intercepted anymore.
 280      *
 281      * Furthermore, when a L2 exception is intercepted by L1
 282      * hypervisor, its exception payload (CR2/DR6 on #PF/#DB)
 283      * should not be set yet in the respective vCPU register.
 284      * Thus, in case an exception is pending, it is
 285      * important to save the exception payload seperately.
 286      *
 287      * Therefore, if an exception is not in a pending state
 288      * or vCPU is not in guest-mode, it is not important to
 289      * distinguish between a pending and injected exception
 290      * and we don't need to store seperately the exception payload.
 291      *
 292      * In order to preserve better backwards-compatible migration,
 293      * convert a pending exception to an injected exception in
 294      * case it is not important to distinguish between them
 295      * as described above.
 296      */
 297     if (env->exception_pending && !(env->hflags & HF_GUEST_MASK)) {
 298         env->exception_pending = 0;
 299         env->exception_injected = 1;
 300
 301         if (env->exception_has_payload) {
 302             if (env->exception_nr == EXCP01_DB) {
 303                 env->dr[6] = env->exception_payload;
 304             } else if (env->exception_nr == EXCP0E_PAGE) {
 305                 env->cr[2] = env->exception_payload;
 306             }
 307         }
 308     }
 309
 310     return 0;
 311 }
 312
 313 static int cpu_post_load(void *opaque, int version_id)
 314 {
 315     X86CPU *cpu = opaque;
 316     CPUState *cs = CPU(cpu);
 317     CPUX86State *env = &cpu->env;
 318     int i;
 319
 320     if (env->tsc_khz && env->user_tsc_khz &&
 321         env->tsc_khz != env->user_tsc_khz) {
 322         error_report("Mismatch between user-specified TSC frequency and "
 323                      "migrated TSC frequency");
 324         return -EINVAL;
 325     }
 326
 327     if (env->fpregs_format_vmstate) {
 328         error_report("Unsupported old non-softfloat CPU state");
 329         return -EINVAL;
 330     }
 331     /*
 332      * Real mode guest segments register DPL should be zero.
 333      * Older KVM version were setting it wrongly.
 334      * Fixing it will allow live migration from such host that don't have
 335      * restricted guest support to a host with unrestricted guest support
 336      * (otherwise the migration will fail with invalid guest state
 337      * error).
 338      */
 339     if (!(env->cr[0] & CR0_PE_MASK) &&
 340         (env->segs[R_CS].flags >> DESC_DPL_SHIFT & 3) != 0) {
 341         env->segs[R_CS].flags &= ~(env->segs[R_CS].flags & DESC_DPL_MASK);
 342         env->segs[R_DS].flags &= ~(env->segs[R_DS].flags & DESC_DPL_MASK);
 343         env->segs[R_ES].flags &= ~(env->segs[R_ES].flags & DESC_DPL_MASK);
 344         env->segs[R_FS].flags &= ~(env->segs[R_FS].flags & DESC_DPL_MASK);
 345         env->segs[R_GS].flags &= ~(env->segs[R_GS].flags & DESC_DPL_MASK);
 346         env->segs[R_SS].flags &= ~(env->segs[R_SS].flags & DESC_DPL_MASK);
 347     }
 348
 349     /* Older versions of QEMU incorrectly used CS.DPL as the CPL when
 350      * running under KVM.  This is wrong for conforming code segments.
 351      * Luckily, in our implementation the CPL field of hflags is redundant
 352      * and we can get the right value from the SS descriptor privilege level.
 353      */
 354     env->hflags &= ~HF_CPL_MASK;
 355     env->hflags |= (env->segs[R_SS].flags >> DESC_DPL_SHIFT) & HF_CPL_MASK;
 356
 357 #ifdef CONFIG_KVM
 358     if ((env->hflags & HF_GUEST_MASK) &&
 359         (!env->nested_state ||
 360         !(env->nested_state->flags & KVM_STATE_NESTED_GUEST_MODE))) {
 361         error_report("vCPU set in guest-mode inconsistent with "
 362                      "migrated kernel nested state");
 363         return -EINVAL;
 364     }
 365 #endif
 366
 367     /*
 368      * There are cases that we can get valid exception_nr with both
 369      * exception_pending and exception_injected being cleared.
 370      * This can happen in one of the following scenarios:
 371      * 1) Source is older QEMU without KVM_CAP_EXCEPTION_PAYLOAD support.
 372      * 2) Source is running on kernel without KVM_CAP_EXCEPTION_PAYLOAD support.
 373      * 3) "cpu/exception_info" subsection not sent because there is no exception
 374      *    pending or guest wasn't running L2 (See comment in cpu_pre_save()).
 375      *
 376      * In those cases, we can just deduce that a valid exception_nr means
 377      * we can treat the exception as already injected.
 378      */
 379     if ((env->exception_nr != -1) &&
 380         !env->exception_pending && !env->exception_injected) {
 381         env->exception_injected = 1;
 382     }
 383
 384     env->fpstt = (env->fpus_vmstate >> 11) & 7;
 385     env->fpus = env->fpus_vmstate & ~0x3800;
 386     env->fptag_vmstate ^= 0xff;
 387     for(i = 0; i < 8; i++) {
 388         env->fptags[i] = (env->fptag_vmstate >> i) & 1;
 389     }
 390     if (tcg_enabled()) {
 391         target_ulong dr7;
 392         update_fp_status(env);
 393         update_mxcsr_status(env);
 394
 395         cpu_breakpoint_remove_all(cs, BP_CPU);
 396         cpu_watchpoint_remove_all(cs, BP_CPU);
 397
 398         /* Indicate all breakpoints disabled, as they are, then
 399            let the helper re-enable them.  */
 400         dr7 = env->dr[7];
 401         env->dr[7] = dr7 & ~(DR7_GLOBAL_BP_MASK | DR7_LOCAL_BP_MASK);
 402         cpu_x86_update_dr7(env, dr7);
 403     }
 404     tlb_flush(cs);
 405     return 0;
 406 }
 407
 408 static bool async_pf_msr_needed(void *opaque)
 409 {
 410     X86CPU *cpu = opaque;
 411
 412     return cpu->env.async_pf_en_msr != 0;
 413 }
 414
 415 static bool async_pf_int_msr_needed(void *opaque)
 416 {
 417     X86CPU *cpu = opaque;
 418
 419     return cpu->env.async_pf_int_msr != 0;
 420 }
 421
 422 static bool pv_eoi_msr_needed(void *opaque)
 423 {
 424     X86CPU *cpu = opaque;
 425
 426     return cpu->env.pv_eoi_en_msr != 0;
 427 }
 428
 429 static bool steal_time_msr_needed(void *opaque)
 430 {
 431     X86CPU *cpu = opaque;
 432
 433     return cpu->env.steal_time_msr != 0;
 434 }
 435
 436 static bool exception_info_needed(void *opaque)
 437 {
 438     X86CPU *cpu = opaque;
 439     CPUX86State *env = &cpu->env;
 440
 441     /*
 442      * It is important to save exception-info only in case
 443      * we need to distinguish between a pending and injected
 444      * exception. Which is only required in case there is a
 445      * pending exception and vCPU is running L2.
 446      * For more info, refer to comment in cpu_pre_save().
 447      */
 448     return env->exception_pending && (env->hflags & HF_GUEST_MASK);
 449 }
 450
 451 static const VMStateDescription vmstate_exception_info = {
 452     .name = "cpu/exception_info",
 453     .version_id = 1,
 454     .minimum_version_id = 1,
 455     .needed = exception_info_needed,
 456     .fields = (VMStateField[]) {
 457         VMSTATE_UINT8(env.exception_pending, X86CPU),
 458         VMSTATE_UINT8(env.exception_injected, X86CPU),
 459         VMSTATE_UINT8(env.exception_has_payload, X86CPU),
 460         VMSTATE_UINT64(env.exception_payload, X86CPU),
 461         VMSTATE_END_OF_LIST()
 462     }
 463 };
 464
 465 /* Poll control MSR enabled by default */
 466 static bool poll_control_msr_needed(void *opaque)
 467 {
 468     X86CPU *cpu = opaque;
 469
 470     return cpu->env.poll_control_msr != 1;
 471 }
 472
 473 static const VMStateDescription vmstate_steal_time_msr = {
 474     .name = "cpu/steal_time_msr",
 475     .version_id = 1,
 476     .minimum_version_id = 1,
 477     .needed = steal_time_msr_needed,
 478     .fields = (VMStateField[]) {
 479         VMSTATE_UINT64(env.steal_time_msr, X86CPU),
 480         VMSTATE_END_OF_LIST()
 481     }
 482 };
 483
 484 static const VMStateDescription vmstate_async_pf_msr = {
 485     .name = "cpu/async_pf_msr",
 486     .version_id = 1,
 487     .minimum_version_id = 1,
 488     .needed = async_pf_msr_needed,
 489     .fields = (VMStateField[]) {
 490         VMSTATE_UINT64(env.async_pf_en_msr, X86CPU),
 491         VMSTATE_END_OF_LIST()
 492     }
 493 };
 494
 495 static const VMStateDescription vmstate_async_pf_int_msr = {
 496     .name = "cpu/async_pf_int_msr",
 497     .version_id = 1,
 498     .minimum_version_id = 1,
 499     .needed = async_pf_int_msr_needed,
 500     .fields = (VMStateField[]) {
 501         VMSTATE_UINT64(env.async_pf_int_msr, X86CPU),
 502         VMSTATE_END_OF_LIST()
 503     }
 504 };
 505
 506 static const VMStateDescription vmstate_pv_eoi_msr = {
 507     .name = "cpu/async_pv_eoi_msr",
 508     .version_id = 1,
 509     .minimum_version_id = 1,
 510     .needed = pv_eoi_msr_needed,
 511     .fields = (VMStateField[]) {
 512         VMSTATE_UINT64(env.pv_eoi_en_msr, X86CPU),
 513         VMSTATE_END_OF_LIST()
 514     }
 515 };
 516
 517 static const VMStateDescription vmstate_poll_control_msr = {
 518     .name = "cpu/poll_control_msr",
 519     .version_id = 1,
 520     .minimum_version_id = 1,
 521     .needed = poll_control_msr_needed,
 522     .fields = (VMStateField[]) {
 523         VMSTATE_UINT64(env.poll_control_msr, X86CPU),
 524         VMSTATE_END_OF_LIST()
 525     }
 526 };
 527
 528 static bool fpop_ip_dp_needed(void *opaque)
 529 {
 530     X86CPU *cpu = opaque;
 531     CPUX86State *env = &cpu->env;
 532
 533     return env->fpop != 0 || env->fpip != 0 || env->fpdp != 0;
 534 }
 535
 536 static const VMStateDescription vmstate_fpop_ip_dp = {
 537     .name = "cpu/fpop_ip_dp",
 538     .version_id = 1,
 539     .minimum_version_id = 1,
 540     .needed = fpop_ip_dp_needed,
 541     .fields = (VMStateField[]) {
 542         VMSTATE_UINT16(env.fpop, X86CPU),
 543         VMSTATE_UINT64(env.fpip, X86CPU),
 544         VMSTATE_UINT64(env.fpdp, X86CPU),
 545         VMSTATE_END_OF_LIST()
 546     }
 547 };
 548
 549 static bool tsc_adjust_needed(void *opaque)
 550 {
 551     X86CPU *cpu = opaque;
 552     CPUX86State *env = &cpu->env;
 553
 554     return env->tsc_adjust != 0;
 555 }
 556
 557 static const VMStateDescription vmstate_msr_tsc_adjust = {
 558     .name = "cpu/msr_tsc_adjust",
 559     .version_id = 1,
 560     .minimum_version_id = 1,
 561     .needed = tsc_adjust_needed,
 562     .fields = (VMStateField[]) {
 563         VMSTATE_UINT64(env.tsc_adjust, X86CPU),
 564         VMSTATE_END_OF_LIST()
 565     }
 566 };
 567
 568 static bool msr_smi_count_needed(void *opaque)
 569 {
 570     X86CPU *cpu = opaque;
 571     CPUX86State *env = &cpu->env;
 572
 573     return cpu->migrate_smi_count && env->msr_smi_count != 0;
 574 }
 575
 576 static const VMStateDescription vmstate_msr_smi_count = {
 577     .name = "cpu/msr_smi_count",
 578     .version_id = 1,
 579     .minimum_version_id = 1,
 580     .needed = msr_smi_count_needed,
 581     .fields = (VMStateField[]) {
 582         VMSTATE_UINT64(env.msr_smi_count, X86CPU),
 583         VMSTATE_END_OF_LIST()
 584     }
 585 };
 586
 587 static bool tscdeadline_needed(void *opaque)
 588 {
 589     X86CPU *cpu = opaque;
 590     CPUX86State *env = &cpu->env;
 591
 592     return env->tsc_deadline != 0;
 593 }
 594
 595 static const VMStateDescription vmstate_msr_tscdeadline = {
 596     .name = "cpu/msr_tscdeadline",
 597     .version_id = 1,
 598     .minimum_version_id = 1,
 599     .needed = tscdeadline_needed,
 600     .fields = (VMStateField[]) {
 601         VMSTATE_UINT64(env.tsc_deadline, X86CPU),
 602         VMSTATE_END_OF_LIST()
 603     }
 604 };
 605
 606 static bool misc_enable_needed(void *opaque)
 607 {
 608     X86CPU *cpu = opaque;
 609     CPUX86State *env = &cpu->env;
 610
 611     return env->msr_ia32_misc_enable != MSR_IA32_MISC_ENABLE_DEFAULT;
 612 }
 613
 614 static bool feature_control_needed(void *opaque)
 615 {
 616     X86CPU *cpu = opaque;
 617     CPUX86State *env = &cpu->env;
 618
 619     return env->msr_ia32_feature_control != 0;
 620 }
 621
 622 static const VMStateDescription vmstate_msr_ia32_misc_enable = {
 623     .name = "cpu/msr_ia32_misc_enable",
 624     .version_id = 1,
 625     .minimum_version_id = 1,
 626     .needed = misc_enable_needed,
 627     .fields = (VMStateField[]) {
 628         VMSTATE_UINT64(env.msr_ia32_misc_enable, X86CPU),
 629         VMSTATE_END_OF_LIST()
 630     }
 631 };
 632
 633 static const VMStateDescription vmstate_msr_ia32_feature_control = {
 634     .name = "cpu/msr_ia32_feature_control",
 635     .version_id = 1,
 636     .minimum_version_id = 1,
 637     .needed = feature_control_needed,
 638     .fields = (VMStateField[]) {
 639         VMSTATE_UINT64(env.msr_ia32_feature_control, X86CPU),
 640         VMSTATE_END_OF_LIST()
 641     }
 642 };
 643
 644 static bool pmu_enable_needed(void *opaque)
 645 {
 646     X86CPU *cpu = opaque;
 647     CPUX86State *env = &cpu->env;
 648     int i;
 649
 650     if (env->msr_fixed_ctr_ctrl || env->msr_global_ctrl ||
 651         env->msr_global_status || env->msr_global_ovf_ctrl) {
 652         return true;
 653     }
 654     for (i = 0; i < MAX_FIXED_COUNTERS; i++) {
 655         if (env->msr_fixed_counters[i]) {
 656             return true;
 657         }
 658     }
 659     for (i = 0; i < MAX_GP_COUNTERS; i++) {
 660         if (env->msr_gp_counters[i] || env->msr_gp_evtsel[i]) {
 661             return true;
 662         }
 663     }
 664
 665     return false;
 666 }
 667
 668 static const VMStateDescription vmstate_msr_architectural_pmu = {
 669     .name = "cpu/msr_architectural_pmu",
 670     .version_id = 1,
 671     .minimum_version_id = 1,
 672     .needed = pmu_enable_needed,
 673     .fields = (VMStateField[]) {
 674         VMSTATE_UINT64(env.msr_fixed_ctr_ctrl, X86CPU),
 675         VMSTATE_UINT64(env.msr_global_ctrl, X86CPU),
 676         VMSTATE_UINT64(env.msr_global_status, X86CPU),
 677         VMSTATE_UINT64(env.msr_global_ovf_ctrl, X86CPU),
 678         VMSTATE_UINT64_ARRAY(env.msr_fixed_counters, X86CPU, MAX_FIXED_COUNTERS),
 679         VMSTATE_UINT64_ARRAY(env.msr_gp_counters, X86CPU, MAX_GP_COUNTERS),
 680         VMSTATE_UINT64_ARRAY(env.msr_gp_evtsel, X86CPU, MAX_GP_COUNTERS),
 681         VMSTATE_END_OF_LIST()
 682     }
 683 };
 684
 685 static bool mpx_needed(void *opaque)
 686 {
 687     X86CPU *cpu = opaque;
 688     CPUX86State *env = &cpu->env;
 689     unsigned int i;
 690
 691     for (i = 0; i < 4; i++) {
 692         if (env->bnd_regs[i].lb || env->bnd_regs[i].ub) {
 693             return true;
 694         }
 695     }
 696
 697     if (env->bndcs_regs.cfgu || env->bndcs_regs.sts) {
 698         return true;
 699     }
 700
 701     return !!env->msr_bndcfgs;
 702 }
 703
 704 static const VMStateDescription vmstate_mpx = {
 705     .name = "cpu/mpx",
 706     .version_id = 1,
 707     .minimum_version_id = 1,
 708     .needed = mpx_needed,
 709     .fields = (VMStateField[]) {
 710         VMSTATE_BND_REGS(env.bnd_regs, X86CPU, 4),
 711         VMSTATE_UINT64(env.bndcs_regs.cfgu, X86CPU),
 712         VMSTATE_UINT64(env.bndcs_regs.sts, X86CPU),
 713         VMSTATE_UINT64(env.msr_bndcfgs, X86CPU),
 714         VMSTATE_END_OF_LIST()
 715     }
 716 };
 717
 718 static bool hyperv_hypercall_enable_needed(void *opaque)
 719 {
 720     X86CPU *cpu = opaque;
 721     CPUX86State *env = &cpu->env;
 722
 723     return env->msr_hv_hypercall != 0 || env->msr_hv_guest_os_id != 0;
 724 }
 725
 726 static const VMStateDescription vmstate_msr_hyperv_hypercall = {
 727     .name = "cpu/msr_hyperv_hypercall",
 728     .version_id = 1,
 729     .minimum_version_id = 1,
 730     .needed = hyperv_hypercall_enable_needed,
 731     .fields = (VMStateField[]) {
 732         VMSTATE_UINT64(env.msr_hv_guest_os_id, X86CPU),
 733         VMSTATE_UINT64(env.msr_hv_hypercall, X86CPU),
 734         VMSTATE_END_OF_LIST()
 735     }
 736 };
 737
 738 static bool hyperv_vapic_enable_needed(void *opaque)
 739 {
 740     X86CPU *cpu = opaque;
 741     CPUX86State *env = &cpu->env;
 742
 743     return env->msr_hv_vapic != 0;
 744 }
 745
 746 static const VMStateDescription vmstate_msr_hyperv_vapic = {
 747     .name = "cpu/msr_hyperv_vapic",
 748     .version_id = 1,
 749     .minimum_version_id = 1,
 750     .needed = hyperv_vapic_enable_needed,
 751     .fields = (VMStateField[]) {
 752         VMSTATE_UINT64(env.msr_hv_vapic, X86CPU),
 753         VMSTATE_END_OF_LIST()
 754     }
 755 };
 756
 757 static bool hyperv_time_enable_needed(void *opaque)
 758 {
 759     X86CPU *cpu = opaque;
 760     CPUX86State *env = &cpu->env;
 761
 762     return env->msr_hv_tsc != 0;
 763 }
 764
 765 static const VMStateDescription vmstate_msr_hyperv_time = {
 766     .name = "cpu/msr_hyperv_time",
 767     .version_id = 1,
 768     .minimum_version_id = 1,
 769     .needed = hyperv_time_enable_needed,
 770     .fields = (VMStateField[]) {
 771         VMSTATE_UINT64(env.msr_hv_tsc, X86CPU),
 772         VMSTATE_END_OF_LIST()
 773     }
 774 };
 775
 776 static bool hyperv_crash_enable_needed(void *opaque)
 777 {
 778     X86CPU *cpu = opaque;
 779     CPUX86State *env = &cpu->env;
 780     int i;
 781
 782     for (i = 0; i < HV_CRASH_PARAMS; i++) {
 783         if (env->msr_hv_crash_params[i]) {
 784             return true;
 785         }
 786     }
 787     return false;
 788 }
 789
 790 static const VMStateDescription vmstate_msr_hyperv_crash = {
 791     .name = "cpu/msr_hyperv_crash",
 792     .version_id = 1,
 793     .minimum_version_id = 1,
 794     .needed = hyperv_crash_enable_needed,
 795     .fields = (VMStateField[]) {
 796         VMSTATE_UINT64_ARRAY(env.msr_hv_crash_params, X86CPU, HV_CRASH_PARAMS),
 797         VMSTATE_END_OF_LIST()
 798     }
 799 };
 800
 801 static bool hyperv_runtime_enable_needed(void *opaque)
 802 {
 803     X86CPU *cpu = opaque;
 804     CPUX86State *env = &cpu->env;
 805
 806     if (!hyperv_feat_enabled(cpu, HYPERV_FEAT_RUNTIME)) {
 807         return false;
 808     }
 809
 810     return env->msr_hv_runtime != 0;
 811 }
 812
 813 static const VMStateDescription vmstate_msr_hyperv_runtime = {
 814     .name = "cpu/msr_hyperv_runtime",
 815     .version_id = 1,
 816     .minimum_version_id = 1,
 817     .needed = hyperv_runtime_enable_needed,
 818     .fields = (VMStateField[]) {
 819         VMSTATE_UINT64(env.msr_hv_runtime, X86CPU),
 820         VMSTATE_END_OF_LIST()
 821     }
 822 };
 823
 824 static bool hyperv_synic_enable_needed(void *opaque)
 825 {
 826     X86CPU *cpu = opaque;
 827     CPUX86State *env = &cpu->env;
 828     int i;
 829
 830     if (env->msr_hv_synic_control != 0 ||
 831         env->msr_hv_synic_evt_page != 0 ||
 832         env->msr_hv_synic_msg_page != 0) {
 833         return true;
 834     }
 835
 836     for (i = 0; i < ARRAY_SIZE(env->msr_hv_synic_sint); i++) {
 837         if (env->msr_hv_synic_sint[i] != 0) {
 838             return true;
 839         }
 840     }
 841
 842     return false;
 843 }
 844
 845 static int hyperv_synic_post_load(void *opaque, int version_id)
 846 {
 847     X86CPU *cpu = opaque;
 848     hyperv_x86_synic_update(cpu);
 849     return 0;
 850 }
 851
 852 static const VMStateDescription vmstate_msr_hyperv_synic = {
 853     .name = "cpu/msr_hyperv_synic",
 854     .version_id = 1,
 855     .minimum_version_id = 1,
 856     .needed = hyperv_synic_enable_needed,
 857     .post_load = hyperv_synic_post_load,
 858     .fields = (VMStateField[]) {
 859         VMSTATE_UINT64(env.msr_hv_synic_control, X86CPU),
 860         VMSTATE_UINT64(env.msr_hv_synic_evt_page, X86CPU),
 861         VMSTATE_UINT64(env.msr_hv_synic_msg_page, X86CPU),
 862         VMSTATE_UINT64_ARRAY(env.msr_hv_synic_sint, X86CPU, HV_SINT_COUNT),
 863         VMSTATE_END_OF_LIST()
 864     }
 865 };
 866
 867 static bool hyperv_stimer_enable_needed(void *opaque)
 868 {
 869     X86CPU *cpu = opaque;
 870     CPUX86State *env = &cpu->env;
 871     int i;
 872
 873     for (i = 0; i < ARRAY_SIZE(env->msr_hv_stimer_config); i++) {
 874         if (env->msr_hv_stimer_config[i] || env->msr_hv_stimer_count[i]) {
 875             return true;
 876         }
 877     }
 878     return false;
 879 }
 880
 881 static const VMStateDescription vmstate_msr_hyperv_stimer = {
 882     .name = "cpu/msr_hyperv_stimer",
 883     .version_id = 1,
 884     .minimum_version_id = 1,
 885     .needed = hyperv_stimer_enable_needed,
 886     .fields = (VMStateField[]) {
 887         VMSTATE_UINT64_ARRAY(env.msr_hv_stimer_config, X86CPU,
 888                              HV_STIMER_COUNT),
 889         VMSTATE_UINT64_ARRAY(env.msr_hv_stimer_count, X86CPU, HV_STIMER_COUNT),
 890         VMSTATE_END_OF_LIST()
 891     }
 892 };
 893
 894 static bool hyperv_reenlightenment_enable_needed(void *opaque)
 895 {
 896     X86CPU *cpu = opaque;
 897     CPUX86State *env = &cpu->env;
 898
 899     return env->msr_hv_reenlightenment_control != 0 ||
 900         env->msr_hv_tsc_emulation_control != 0 ||
 901         env->msr_hv_tsc_emulation_status != 0;
 902 }
 903
 904 static int hyperv_reenlightenment_post_load(void *opaque, int version_id)
 905 {
 906     X86CPU *cpu = opaque;
 907     CPUX86State *env = &cpu->env;
 908
 909     /*
 910      * KVM doesn't fully support re-enlightenment notifications so we need to
 911      * make sure TSC frequency doesn't change upon migration.
 912      */
 913     if ((env->msr_hv_reenlightenment_control & HV_REENLIGHTENMENT_ENABLE_BIT) &&
 914         !env->user_tsc_khz) {
 915         error_report("Guest enabled re-enlightenment notifications, "
 916                      "'tsc-frequency=' has to be specified");
 917         return -EINVAL;
 918     }
 919
 920     return 0;
 921 }
 922
 923 static const VMStateDescription vmstate_msr_hyperv_reenlightenment = {
 924     .name = "cpu/msr_hyperv_reenlightenment",
 925     .version_id = 1,
 926     .minimum_version_id = 1,
 927     .needed = hyperv_reenlightenment_enable_needed,
 928     .post_load = hyperv_reenlightenment_post_load,
 929     .fields = (VMStateField[]) {
 930         VMSTATE_UINT64(env.msr_hv_reenlightenment_control, X86CPU),
 931         VMSTATE_UINT64(env.msr_hv_tsc_emulation_control, X86CPU),
 932         VMSTATE_UINT64(env.msr_hv_tsc_emulation_status, X86CPU),
 933         VMSTATE_END_OF_LIST()
 934     }
 935 };
 936
 937 static bool avx512_needed(void *opaque)
 938 {
 939     X86CPU *cpu = opaque;
 940     CPUX86State *env = &cpu->env;
 941     unsigned int i;
 942
 943     for (i = 0; i < NB_OPMASK_REGS; i++) {
 944         if (env->opmask_regs[i]) {
 945             return true;
 946         }
 947     }
 948
 949     for (i = 0; i < CPU_NB_REGS; i++) {
 950 #define ENV_XMM(reg, field) (env->xmm_regs[reg].ZMM_Q(field))
 951         if (ENV_XMM(i, 4) || ENV_XMM(i, 6) ||
 952             ENV_XMM(i, 5) || ENV_XMM(i, 7)) {
 953             return true;
 954         }
 955 #ifdef TARGET_X86_64
 956         if (ENV_XMM(i+16, 0) || ENV_XMM(i+16, 1) ||
 957             ENV_XMM(i+16, 2) || ENV_XMM(i+16, 3) ||
 958             ENV_XMM(i+16, 4) || ENV_XMM(i+16, 5) ||
 959             ENV_XMM(i+16, 6) || ENV_XMM(i+16, 7)) {
 960             return true;
 961         }
 962 #endif
 963     }
 964
 965     return false;
 966 }
 967
 968 static const VMStateDescription vmstate_avx512 = {
 969     .name = "cpu/avx512",
 970     .version_id = 1,
 971     .minimum_version_id = 1,
 972     .needed = avx512_needed,
 973     .fields = (VMStateField[]) {
 974         VMSTATE_UINT64_ARRAY(env.opmask_regs, X86CPU, NB_OPMASK_REGS),
 975         VMSTATE_ZMMH_REGS_VARS(env.xmm_regs, X86CPU, 0),
 976 #ifdef TARGET_X86_64
 977         VMSTATE_Hi16_ZMM_REGS_VARS(env.xmm_regs, X86CPU, 16),
 978 #endif
 979         VMSTATE_END_OF_LIST()
 980     }
 981 };
 982
 983 static bool xss_needed(void *opaque)
 984 {
 985     X86CPU *cpu = opaque;
 986     CPUX86State *env = &cpu->env;
 987
 988     return env->xss != 0;
 989 }
 990
 991 static const VMStateDescription vmstate_xss = {
 992     .name = "cpu/xss",
 993     .version_id = 1,
 994     .minimum_version_id = 1,
 995     .needed = xss_needed,
 996     .fields = (VMStateField[]) {
 997         VMSTATE_UINT64(env.xss, X86CPU),
 998         VMSTATE_END_OF_LIST()
 999     }
1000 };
1001
1002 static bool umwait_needed(void *opaque)
1003 {
1004     X86CPU *cpu = opaque;
1005     CPUX86State *env = &cpu->env;
1006
1007     return env->umwait != 0;
1008 }
1009
1010 static const VMStateDescription vmstate_umwait = {
1011     .name = "cpu/umwait",
1012     .version_id = 1,
1013     .minimum_version_id = 1,
1014     .needed = umwait_needed,
1015     .fields = (VMStateField[]) {
1016         VMSTATE_UINT32(env.umwait, X86CPU),
1017         VMSTATE_END_OF_LIST()
1018     }
1019 };
1020
1021 static bool pkru_needed(void *opaque)
1022 {
1023     X86CPU *cpu = opaque;
1024     CPUX86State *env = &cpu->env;
1025
1026     return env->pkru != 0;
1027 }
1028
1029 static const VMStateDescription vmstate_pkru = {
1030     .name = "cpu/pkru",
1031     .version_id = 1,
1032     .minimum_version_id = 1,
1033     .needed = pkru_needed,
1034     .fields = (VMStateField[]){
1035         VMSTATE_UINT32(env.pkru, X86CPU),
1036         VMSTATE_END_OF_LIST()
1037     }
1038 };
1039
1040 static bool pkrs_needed(void *opaque)
1041 {
1042     X86CPU *cpu = opaque;
1043     CPUX86State *env = &cpu->env;
1044
1045     return env->pkrs != 0;
1046 }
1047
1048 static const VMStateDescription vmstate_pkrs = {
1049     .name = "cpu/pkrs",
1050     .version_id = 1,
1051     .minimum_version_id = 1,
1052     .needed = pkrs_needed,
1053     .fields = (VMStateField[]){
1054         VMSTATE_UINT32(env.pkrs, X86CPU),
1055         VMSTATE_END_OF_LIST()
1056     }
1057 };
1058
1059 static bool tsc_khz_needed(void *opaque)
1060 {
1061     X86CPU *cpu = opaque;
1062     CPUX86State *env = &cpu->env;
1063     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
1064     X86MachineClass *x86mc = X86_MACHINE_CLASS(mc);
1065     return env->tsc_khz && x86mc->save_tsc_khz;
1066 }
1067
1068 static const VMStateDescription vmstate_tsc_khz = {
1069     .name = "cpu/tsc_khz",
1070     .version_id = 1,
1071     .minimum_version_id = 1,
1072     .needed = tsc_khz_needed,
1073     .fields = (VMStateField[]) {
1074         VMSTATE_INT64(env.tsc_khz, X86CPU),
1075         VMSTATE_END_OF_LIST()
1076     }
1077 };
1078
1079 #ifdef CONFIG_KVM
1080
1081 static bool vmx_vmcs12_needed(void *opaque)
1082 {
1083     struct kvm_nested_state *nested_state = opaque;
1084     return (nested_state->size >
1085             offsetof(struct kvm_nested_state, data.vmx[0].vmcs12));
1086 }
1087
1088 static const VMStateDescription vmstate_vmx_vmcs12 = {
1089     .name = "cpu/kvm_nested_state/vmx/vmcs12",
1090     .version_id = 1,
1091     .minimum_version_id = 1,
1092     .needed = vmx_vmcs12_needed,
1093     .fields = (VMStateField[]) {
1094         VMSTATE_UINT8_ARRAY(data.vmx[0].vmcs12,
1095                             struct kvm_nested_state,
1096                             KVM_STATE_NESTED_VMX_VMCS_SIZE),
1097         VMSTATE_END_OF_LIST()
1098     }
1099 };
1100
1101 static bool vmx_shadow_vmcs12_needed(void *opaque)
1102 {
1103     struct kvm_nested_state *nested_state = opaque;
1104     return (nested_state->size >
1105             offsetof(struct kvm_nested_state, data.vmx[0].shadow_vmcs12));
1106 }
1107
1108 static const VMStateDescription vmstate_vmx_shadow_vmcs12 = {
1109     .name = "cpu/kvm_nested_state/vmx/shadow_vmcs12",
1110     .version_id = 1,
1111     .minimum_version_id = 1,
1112     .needed = vmx_shadow_vmcs12_needed,
1113     .fields = (VMStateField[]) {
1114         VMSTATE_UINT8_ARRAY(data.vmx[0].shadow_vmcs12,
1115                             struct kvm_nested_state,
1116                             KVM_STATE_NESTED_VMX_VMCS_SIZE),
1117         VMSTATE_END_OF_LIST()
1118     }
1119 };
1120
1121 static bool vmx_nested_state_needed(void *opaque)
1122 {
1123     struct kvm_nested_state *nested_state = opaque;
1124
1125     return (nested_state->format == KVM_STATE_NESTED_FORMAT_VMX &&
1126             nested_state->hdr.vmx.vmxon_pa != -1ull);
1127 }
1128
1129 static const VMStateDescription vmstate_vmx_nested_state = {
1130     .name = "cpu/kvm_nested_state/vmx",
1131     .version_id = 1,
1132     .minimum_version_id = 1,
1133     .needed = vmx_nested_state_needed,
1134     .fields = (VMStateField[]) {
1135         VMSTATE_U64(hdr.vmx.vmxon_pa, struct kvm_nested_state),
1136         VMSTATE_U64(hdr.vmx.vmcs12_pa, struct kvm_nested_state),
1137         VMSTATE_U16(hdr.vmx.smm.flags, struct kvm_nested_state),
1138         VMSTATE_END_OF_LIST()
1139     },
1140     .subsections = (const VMStateDescription*[]) {
1141         &vmstate_vmx_vmcs12,
1142         &vmstate_vmx_shadow_vmcs12,
1143         NULL,
1144     }
1145 };
1146
1147 static bool svm_nested_state_needed(void *opaque)
1148 {
1149     struct kvm_nested_state *nested_state = opaque;
1150
1151     /*
1152      * HF_GUEST_MASK and HF2_GIF_MASK are already serialized
1153      * via hflags and hflags2, all that's left is the opaque
1154      * nested state blob.
1155      */
1156     return (nested_state->format == KVM_STATE_NESTED_FORMAT_SVM &&
1157             nested_state->size > offsetof(struct kvm_nested_state, data));
1158 }
1159
1160 static const VMStateDescription vmstate_svm_nested_state = {
1161     .name = "cpu/kvm_nested_state/svm",
1162     .version_id = 1,
1163     .minimum_version_id = 1,
1164     .needed = svm_nested_state_needed,
1165     .fields = (VMStateField[]) {
1166         VMSTATE_U64(hdr.svm.vmcb_pa, struct kvm_nested_state),
1167         VMSTATE_UINT8_ARRAY(data.svm[0].vmcb12,
1168                             struct kvm_nested_state,
1169                             KVM_STATE_NESTED_SVM_VMCB_SIZE),
1170         VMSTATE_END_OF_LIST()
1171     }
1172 };
1173
1174 static bool nested_state_needed(void *opaque)
1175 {
1176     X86CPU *cpu = opaque;
1177     CPUX86State *env = &cpu->env;
1178
1179     return (env->nested_state &&
1180             (vmx_nested_state_needed(env->nested_state) ||
1181              svm_nested_state_needed(env->nested_state)));
1182 }
1183
1184 static int nested_state_post_load(void *opaque, int version_id)
1185 {
1186     X86CPU *cpu = opaque;
1187     CPUX86State *env = &cpu->env;
1188     struct kvm_nested_state *nested_state = env->nested_state;
1189     int min_nested_state_len = offsetof(struct kvm_nested_state, data);
1190     int max_nested_state_len = kvm_max_nested_state_length();
1191
1192     /*
1193      * If our kernel don't support setting nested state
1194      * and we have received nested state from migration stream,
1195      * we need to fail migration
1196      */
1197     if (max_nested_state_len <= 0) {
1198         error_report("Received nested state when kernel cannot restore it");
1199         return -EINVAL;
1200     }
1201
1202     /*
1203      * Verify that the size of received nested_state struct
1204      * at least cover required header and is not larger
1205      * than the max size that our kernel support
1206      */
1207     if (nested_state->size < min_nested_state_len) {
1208         error_report("Received nested state size less than min: "
1209                      "len=%d, min=%d",
1210                      nested_state->size, min_nested_state_len);
1211         return -EINVAL;
1212     }
1213     if (nested_state->size > max_nested_state_len) {
1214         error_report("Received unsupported nested state size: "
1215                      "nested_state->size=%d, max=%d",
1216                      nested_state->size, max_nested_state_len);
1217         return -EINVAL;
1218     }
1219
1220     /* Verify format is valid */
1221     if ((nested_state->format != KVM_STATE_NESTED_FORMAT_VMX) &&
1222         (nested_state->format != KVM_STATE_NESTED_FORMAT_SVM)) {
1223         error_report("Received invalid nested state format: %d",
1224                      nested_state->format);
1225         return -EINVAL;
1226     }
1227
1228     return 0;
1229 }
1230
1231 static const VMStateDescription vmstate_kvm_nested_state = {
1232     .name = "cpu/kvm_nested_state",
1233     .version_id = 1,
1234     .minimum_version_id = 1,
1235     .fields = (VMStateField[]) {
1236         VMSTATE_U16(flags, struct kvm_nested_state),
1237         VMSTATE_U16(format, struct kvm_nested_state),
1238         VMSTATE_U32(size, struct kvm_nested_state),
1239         VMSTATE_END_OF_LIST()
1240     },
1241     .subsections = (const VMStateDescription*[]) {
1242         &vmstate_vmx_nested_state,
1243         &vmstate_svm_nested_state,
1244         NULL
1245     }
1246 };
1247
1248 static const VMStateDescription vmstate_nested_state = {
1249     .name = "cpu/nested_state",
1250     .version_id = 1,
1251     .minimum_version_id = 1,
1252     .needed = nested_state_needed,
1253     .post_load = nested_state_post_load,
1254     .fields = (VMStateField[]) {
1255         VMSTATE_STRUCT_POINTER(env.nested_state, X86CPU,
1256                 vmstate_kvm_nested_state,
1257                 struct kvm_nested_state),
1258         VMSTATE_END_OF_LIST()
1259     }
1260 };
1261
1262 static bool xen_vcpu_needed(void *opaque)
1263 {
1264     return (xen_mode == XEN_EMULATE);
1265 }
1266
1267 static const VMStateDescription vmstate_xen_vcpu = {
1268     .name = "cpu/xen_vcpu",
1269     .version_id = 1,
1270     .minimum_version_id = 1,
1271     .needed = xen_vcpu_needed,
1272     .fields = (VMStateField[]) {
1273         VMSTATE_UINT64(env.xen_vcpu_info_gpa, X86CPU),
1274         VMSTATE_UINT64(env.xen_vcpu_info_default_gpa, X86CPU),
1275         VMSTATE_UINT64(env.xen_vcpu_time_info_gpa, X86CPU),
1276         VMSTATE_UINT64(env.xen_vcpu_runstate_gpa, X86CPU),
1277         VMSTATE_UINT8(env.xen_vcpu_callback_vector, X86CPU),
1278         VMSTATE_UINT16_ARRAY(env.xen_virq, X86CPU, XEN_NR_VIRQS),
1279         VMSTATE_UINT64(env.xen_singleshot_timer_ns, X86CPU),
1280         VMSTATE_UINT64(env.xen_periodic_timer_period, X86CPU),
1281         VMSTATE_END_OF_LIST()
1282     }
1283 };
1284 #endif
1285
1286 static bool mcg_ext_ctl_needed(void *opaque)
1287 {
1288     X86CPU *cpu = opaque;
1289     CPUX86State *env = &cpu->env;
1290     return cpu->enable_lmce && env->mcg_ext_ctl;
1291 }
1292
1293 static const VMStateDescription vmstate_mcg_ext_ctl = {
1294     .name = "cpu/mcg_ext_ctl",
1295     .version_id = 1,
1296     .minimum_version_id = 1,
1297     .needed = mcg_ext_ctl_needed,
1298     .fields = (VMStateField[]) {
1299         VMSTATE_UINT64(env.mcg_ext_ctl, X86CPU),
1300         VMSTATE_END_OF_LIST()
1301     }
1302 };
1303
1304 static bool spec_ctrl_needed(void *opaque)
1305 {
1306     X86CPU *cpu = opaque;
1307     CPUX86State *env = &cpu->env;
1308
1309     return env->spec_ctrl != 0;
1310 }
1311
1312 static const VMStateDescription vmstate_spec_ctrl = {
1313     .name = "cpu/spec_ctrl",
1314     .version_id = 1,
1315     .minimum_version_id = 1,
1316     .needed = spec_ctrl_needed,
1317     .fields = (VMStateField[]){
1318         VMSTATE_UINT64(env.spec_ctrl, X86CPU),
1319         VMSTATE_END_OF_LIST()
1320     }
1321 };
1322
1323
1324 static bool amd_tsc_scale_msr_needed(void *opaque)
1325 {
1326     X86CPU *cpu = opaque;
1327     CPUX86State *env = &cpu->env;
1328
1329     return (env->features[FEAT_SVM] & CPUID_SVM_TSCSCALE);
1330 }
1331
1332 static const VMStateDescription amd_tsc_scale_msr_ctrl = {
1333     .name = "cpu/amd_tsc_scale_msr",
1334     .version_id = 1,
1335     .minimum_version_id = 1,
1336     .needed = amd_tsc_scale_msr_needed,
1337     .fields = (VMStateField[]){
1338         VMSTATE_UINT64(env.amd_tsc_scale_msr, X86CPU),
1339         VMSTATE_END_OF_LIST()
1340     }
1341 };
1342
1343
1344 static bool intel_pt_enable_needed(void *opaque)
1345 {
1346     X86CPU *cpu = opaque;
1347     CPUX86State *env = &cpu->env;
1348     int i;
1349
1350     if (env->msr_rtit_ctrl || env->msr_rtit_status ||
1351         env->msr_rtit_output_base || env->msr_rtit_output_mask ||
1352         env->msr_rtit_cr3_match) {
1353         return true;
1354     }
1355
1356     for (i = 0; i < MAX_RTIT_ADDRS; i++) {
1357         if (env->msr_rtit_addrs[i]) {
1358             return true;
1359         }
1360     }
1361
1362     return false;
1363 }
1364
1365 static const VMStateDescription vmstate_msr_intel_pt = {
1366     .name = "cpu/intel_pt",
1367     .version_id = 1,
1368     .minimum_version_id = 1,
1369     .needed = intel_pt_enable_needed,
1370     .fields = (VMStateField[]) {
1371         VMSTATE_UINT64(env.msr_rtit_ctrl, X86CPU),
1372         VMSTATE_UINT64(env.msr_rtit_status, X86CPU),
1373         VMSTATE_UINT64(env.msr_rtit_output_base, X86CPU),
1374         VMSTATE_UINT64(env.msr_rtit_output_mask, X86CPU),
1375         VMSTATE_UINT64(env.msr_rtit_cr3_match, X86CPU),
1376         VMSTATE_UINT64_ARRAY(env.msr_rtit_addrs, X86CPU, MAX_RTIT_ADDRS),
1377         VMSTATE_END_OF_LIST()
1378     }
1379 };
1380
1381 static bool virt_ssbd_needed(void *opaque)
1382 {
1383     X86CPU *cpu = opaque;
1384     CPUX86State *env = &cpu->env;
1385
1386     return env->virt_ssbd != 0;
1387 }
1388
1389 static const VMStateDescription vmstate_msr_virt_ssbd = {
1390     .name = "cpu/virt_ssbd",
1391     .version_id = 1,
1392     .minimum_version_id = 1,
1393     .needed = virt_ssbd_needed,
1394     .fields = (VMStateField[]){
1395         VMSTATE_UINT64(env.virt_ssbd, X86CPU),
1396         VMSTATE_END_OF_LIST()
1397     }
1398 };
1399
1400 static bool svm_npt_needed(void *opaque)
1401 {
1402     X86CPU *cpu = opaque;
1403     CPUX86State *env = &cpu->env;
1404
1405     return !!(env->hflags2 & HF2_NPT_MASK);
1406 }
1407
1408 static const VMStateDescription vmstate_svm_npt = {
1409     .name = "cpu/svn_npt",
1410     .version_id = 1,
1411     .minimum_version_id = 1,
1412     .needed = svm_npt_needed,
1413     .fields = (VMStateField[]){
1414         VMSTATE_UINT64(env.nested_cr3, X86CPU),
1415         VMSTATE_UINT32(env.nested_pg_mode, X86CPU),
1416         VMSTATE_END_OF_LIST()
1417     }
1418 };
1419
1420 static bool svm_guest_needed(void *opaque)
1421 {
1422     X86CPU *cpu = opaque;
1423     CPUX86State *env = &cpu->env;
1424
1425     return tcg_enabled() && env->int_ctl;
1426 }
1427
1428 static const VMStateDescription vmstate_svm_guest = {
1429     .name = "cpu/svm_guest",
1430     .version_id = 1,
1431     .minimum_version_id = 1,
1432     .needed = svm_guest_needed,
1433     .fields = (VMStateField[]){
1434         VMSTATE_UINT32(env.int_ctl, X86CPU),
1435         VMSTATE_END_OF_LIST()
1436     }
1437 };
1438
1439 #ifndef TARGET_X86_64
1440 static bool intel_efer32_needed(void *opaque)
1441 {
1442     X86CPU *cpu = opaque;
1443     CPUX86State *env = &cpu->env;
1444
1445     return env->efer != 0;
1446 }
1447
1448 static const VMStateDescription vmstate_efer32 = {
1449     .name = "cpu/efer32",
1450     .version_id = 1,
1451     .minimum_version_id = 1,
1452     .needed = intel_efer32_needed,
1453     .fields = (VMStateField[]) {
1454         VMSTATE_UINT64(env.efer, X86CPU),
1455         VMSTATE_END_OF_LIST()
1456     }
1457 };
1458 #endif
1459
1460 static bool msr_tsx_ctrl_needed(void *opaque)
1461 {
1462     X86CPU *cpu = opaque;
1463     CPUX86State *env = &cpu->env;
1464
1465     return env->features[FEAT_ARCH_CAPABILITIES] & ARCH_CAP_TSX_CTRL_MSR;
1466 }
1467
1468 static const VMStateDescription vmstate_msr_tsx_ctrl = {
1469     .name = "cpu/msr_tsx_ctrl",
1470     .version_id = 1,
1471     .minimum_version_id = 1,
1472     .needed = msr_tsx_ctrl_needed,
1473     .fields = (VMStateField[]) {
1474         VMSTATE_UINT32(env.tsx_ctrl, X86CPU),
1475         VMSTATE_END_OF_LIST()
1476     }
1477 };
1478
1479 static bool intel_sgx_msrs_needed(void *opaque)
1480 {
1481     X86CPU *cpu = opaque;
1482     CPUX86State *env = &cpu->env;
1483
1484     return !!(env->features[FEAT_7_0_ECX] & CPUID_7_0_ECX_SGX_LC);
1485 }
1486
1487 static const VMStateDescription vmstate_msr_intel_sgx = {
1488     .name = "cpu/intel_sgx",
1489     .version_id = 1,
1490     .minimum_version_id = 1,
1491     .needed = intel_sgx_msrs_needed,
1492     .fields = (VMStateField[]) {
1493         VMSTATE_UINT64_ARRAY(env.msr_ia32_sgxlepubkeyhash, X86CPU, 4),
1494                 VMSTATE_END_OF_LIST()
1495             }
1496         };
1497
1498 static bool pdptrs_needed(void *opaque)
1499 {
1500     X86CPU *cpu = opaque;
1501     CPUX86State *env = &cpu->env;
1502     return env->pdptrs_valid;
1503 }
1504
1505 static int pdptrs_post_load(void *opaque, int version_id)
1506 {
1507     X86CPU *cpu = opaque;
1508     CPUX86State *env = &cpu->env;
1509     env->pdptrs_valid = true;
1510     return 0;
1511 }
1512
1513
1514 static const VMStateDescription vmstate_pdptrs = {
1515     .name = "cpu/pdptrs",
1516     .version_id = 1,
1517     .minimum_version_id = 1,
1518     .needed = pdptrs_needed,
1519     .post_load = pdptrs_post_load,
1520     .fields = (VMStateField[]) {
1521         VMSTATE_UINT64_ARRAY(env.pdptrs, X86CPU, 4),
1522         VMSTATE_END_OF_LIST()
1523     }
1524 };
1525
1526 static bool xfd_msrs_needed(void *opaque)
1527 {
1528     X86CPU *cpu = opaque;
1529     CPUX86State *env = &cpu->env;
1530
1531     return !!(env->features[FEAT_XSAVE] & CPUID_D_1_EAX_XFD);
1532 }
1533
1534 static const VMStateDescription vmstate_msr_xfd = {
1535     .name = "cpu/msr_xfd",
1536     .version_id = 1,
1537     .minimum_version_id = 1,
1538     .needed = xfd_msrs_needed,
1539     .fields = (VMStateField[]) {
1540         VMSTATE_UINT64(env.msr_xfd, X86CPU),
1541         VMSTATE_UINT64(env.msr_xfd_err, X86CPU),
1542         VMSTATE_END_OF_LIST()
1543     }
1544 };
1545
1546 #ifdef TARGET_X86_64
1547 static bool amx_xtile_needed(void *opaque)
1548 {
1549     X86CPU *cpu = opaque;
1550     CPUX86State *env = &cpu->env;
1551
1552     return !!(env->features[FEAT_7_0_EDX] & CPUID_7_0_EDX_AMX_TILE);
1553 }
1554
1555 static const VMStateDescription vmstate_amx_xtile = {
1556     .name = "cpu/intel_amx_xtile",
1557     .version_id = 1,
1558     .minimum_version_id = 1,
1559     .needed = amx_xtile_needed,
1560     .fields = (VMStateField[]) {
1561         VMSTATE_UINT8_ARRAY(env.xtilecfg, X86CPU, 64),
1562         VMSTATE_UINT8_ARRAY(env.xtiledata, X86CPU, 8192),
1563         VMSTATE_END_OF_LIST()
1564     }
1565 };
1566 #endif
1567
1568 static bool arch_lbr_needed(void *opaque)
1569 {
1570     X86CPU *cpu = opaque;
1571     CPUX86State *env = &cpu->env;
1572
1573     return !!(env->features[FEAT_7_0_EDX] & CPUID_7_0_EDX_ARCH_LBR);
1574 }
1575
1576 static const VMStateDescription vmstate_arch_lbr = {
1577     .name = "cpu/arch_lbr",
1578     .version_id = 1,
1579     .minimum_version_id = 1,
1580     .needed = arch_lbr_needed,
1581     .fields = (VMStateField[]) {
1582         VMSTATE_UINT64(env.msr_lbr_ctl, X86CPU),
1583         VMSTATE_UINT64(env.msr_lbr_depth, X86CPU),
1584         VMSTATE_LBR_VARS(env.lbr_records, X86CPU, ARCH_LBR_NR_ENTRIES, 1),
1585         VMSTATE_END_OF_LIST()
1586     }
1587 };
1588
1589 static bool triple_fault_needed(void *opaque)
1590 {
1591     X86CPU *cpu = opaque;
1592     CPUX86State *env = &cpu->env;
1593
1594     return env->triple_fault_pending;
1595 }
1596
1597 static const VMStateDescription vmstate_triple_fault = {
1598     .name = "cpu/triple_fault",
1599     .version_id = 1,
1600     .minimum_version_id = 1,
1601     .needed = triple_fault_needed,
1602     .fields = (VMStateField[]) {
1603         VMSTATE_UINT8(env.triple_fault_pending, X86CPU),
1604         VMSTATE_END_OF_LIST()
1605     }
1606 };
1607
1608 const VMStateDescription vmstate_x86_cpu = {
1609     .name = "cpu",
1610     .version_id = 12,
1611     .minimum_version_id = 11,
1612     .pre_save = cpu_pre_save,
1613     .post_load = cpu_post_load,
1614     .fields = (VMStateField[]) {
1615         VMSTATE_UINTTL_ARRAY(env.regs, X86CPU, CPU_NB_REGS),
1616         VMSTATE_UINTTL(env.eip, X86CPU),
1617         VMSTATE_UINTTL(env.eflags, X86CPU),
1618         VMSTATE_UINT32(env.hflags, X86CPU),
1619         /* FPU */
1620         VMSTATE_UINT16(env.fpuc, X86CPU),
1621         VMSTATE_UINT16(env.fpus_vmstate, X86CPU),
1622         VMSTATE_UINT16(env.fptag_vmstate, X86CPU),
1623         VMSTATE_UINT16(env.fpregs_format_vmstate, X86CPU),
1624
1625         VMSTATE_STRUCT_ARRAY(env.fpregs, X86CPU, 8, 0, vmstate_fpreg, FPReg),
1626
1627         VMSTATE_SEGMENT_ARRAY(env.segs, X86CPU, 6),
1628         VMSTATE_SEGMENT(env.ldt, X86CPU),
1629         VMSTATE_SEGMENT(env.tr, X86CPU),
1630         VMSTATE_SEGMENT(env.gdt, X86CPU),
1631         VMSTATE_SEGMENT(env.idt, X86CPU),
1632
1633         VMSTATE_UINT32(env.sysenter_cs, X86CPU),
1634         VMSTATE_UINTTL(env.sysenter_esp, X86CPU),
1635         VMSTATE_UINTTL(env.sysenter_eip, X86CPU),
1636
1637         VMSTATE_UINTTL(env.cr[0], X86CPU),
1638         VMSTATE_UINTTL(env.cr[2], X86CPU),
1639         VMSTATE_UINTTL(env.cr[3], X86CPU),
1640         VMSTATE_UINTTL(env.cr[4], X86CPU),
1641         VMSTATE_UINTTL_ARRAY(env.dr, X86CPU, 8),
1642         /* MMU */
1643         VMSTATE_INT32(env.a20_mask, X86CPU),
1644         /* XMM */
1645         VMSTATE_UINT32(env.mxcsr, X86CPU),
1646         VMSTATE_XMM_REGS(env.xmm_regs, X86CPU, 0),
1647
1648 #ifdef TARGET_X86_64
1649         VMSTATE_UINT64(env.efer, X86CPU),
1650         VMSTATE_UINT64(env.star, X86CPU),
1651         VMSTATE_UINT64(env.lstar, X86CPU),
1652         VMSTATE_UINT64(env.cstar, X86CPU),
1653         VMSTATE_UINT64(env.fmask, X86CPU),
1654         VMSTATE_UINT64(env.kernelgsbase, X86CPU),
1655 #endif
1656         VMSTATE_UINT32(env.smbase, X86CPU),
1657
1658         VMSTATE_UINT64(env.pat, X86CPU),
1659         VMSTATE_UINT32(env.hflags2, X86CPU),
1660
1661         VMSTATE_UINT64(env.vm_hsave, X86CPU),
1662         VMSTATE_UINT64(env.vm_vmcb, X86CPU),
1663         VMSTATE_UINT64(env.tsc_offset, X86CPU),
1664         VMSTATE_UINT64(env.intercept, X86CPU),
1665         VMSTATE_UINT16(env.intercept_cr_read, X86CPU),
1666         VMSTATE_UINT16(env.intercept_cr_write, X86CPU),
1667         VMSTATE_UINT16(env.intercept_dr_read, X86CPU),
1668         VMSTATE_UINT16(env.intercept_dr_write, X86CPU),
1669         VMSTATE_UINT32(env.intercept_exceptions, X86CPU),
1670         VMSTATE_UINT8(env.v_tpr, X86CPU),
1671         /* MTRRs */
1672         VMSTATE_UINT64_ARRAY(env.mtrr_fixed, X86CPU, 11),
1673         VMSTATE_UINT64(env.mtrr_deftype, X86CPU),
1674         VMSTATE_MTRR_VARS(env.mtrr_var, X86CPU, MSR_MTRRcap_VCNT, 8),
1675         /* KVM-related states */
1676         VMSTATE_INT32(env.interrupt_injected, X86CPU),
1677         VMSTATE_UINT32(env.mp_state, X86CPU),
1678         VMSTATE_UINT64(env.tsc, X86CPU),
1679         VMSTATE_INT32(env.exception_nr, X86CPU),
1680         VMSTATE_UINT8(env.soft_interrupt, X86CPU),
1681         VMSTATE_UINT8(env.nmi_injected, X86CPU),
1682         VMSTATE_UINT8(env.nmi_pending, X86CPU),
1683         VMSTATE_UINT8(env.has_error_code, X86CPU),
1684         VMSTATE_UINT32(env.sipi_vector, X86CPU),
1685         /* MCE */
1686         VMSTATE_UINT64(env.mcg_cap, X86CPU),
1687         VMSTATE_UINT64(env.mcg_status, X86CPU),
1688         VMSTATE_UINT64(env.mcg_ctl, X86CPU),
1689         VMSTATE_UINT64_ARRAY(env.mce_banks, X86CPU, MCE_BANKS_DEF * 4),
1690         /* rdtscp */
1691         VMSTATE_UINT64(env.tsc_aux, X86CPU),
1692         /* KVM pvclock msr */
1693         VMSTATE_UINT64(env.system_time_msr, X86CPU),
1694         VMSTATE_UINT64(env.wall_clock_msr, X86CPU),
1695         /* XSAVE related fields */
1696         VMSTATE_UINT64_V(env.xcr0, X86CPU, 12),
1697         VMSTATE_UINT64_V(env.xstate_bv, X86CPU, 12),
1698         VMSTATE_YMMH_REGS_VARS(env.xmm_regs, X86CPU, 0, 12),
1699         VMSTATE_END_OF_LIST()
1700         /* The above list is not sorted /wrt version numbers, watch out! */
1701     },
1702     .subsections = (const VMStateDescription*[]) {
1703         &vmstate_exception_info,
1704         &vmstate_async_pf_msr,
1705         &vmstate_async_pf_int_msr,
1706         &vmstate_pv_eoi_msr,
1707         &vmstate_steal_time_msr,
1708         &vmstate_poll_control_msr,
1709         &vmstate_fpop_ip_dp,
1710         &vmstate_msr_tsc_adjust,
1711         &vmstate_msr_tscdeadline,
1712         &vmstate_msr_ia32_misc_enable,
1713         &vmstate_msr_ia32_feature_control,
1714         &vmstate_msr_architectural_pmu,
1715         &vmstate_mpx,
1716         &vmstate_msr_hyperv_hypercall,
1717         &vmstate_msr_hyperv_vapic,
1718         &vmstate_msr_hyperv_time,
1719         &vmstate_msr_hyperv_crash,
1720         &vmstate_msr_hyperv_runtime,
1721         &vmstate_msr_hyperv_synic,
1722         &vmstate_msr_hyperv_stimer,
1723         &vmstate_msr_hyperv_reenlightenment,
1724         &vmstate_avx512,
1725         &vmstate_xss,
1726         &vmstate_umwait,
1727         &vmstate_tsc_khz,
1728         &vmstate_msr_smi_count,
1729         &vmstate_pkru,
1730         &vmstate_pkrs,
1731         &vmstate_spec_ctrl,
1732         &amd_tsc_scale_msr_ctrl,
1733         &vmstate_mcg_ext_ctl,
1734         &vmstate_msr_intel_pt,
1735         &vmstate_msr_virt_ssbd,
1736         &vmstate_svm_npt,
1737         &vmstate_svm_guest,
1738 #ifndef TARGET_X86_64
1739         &vmstate_efer32,
1740 #endif
1741 #ifdef CONFIG_KVM
1742         &vmstate_nested_state,
1743         &vmstate_xen_vcpu,
1744 #endif
1745         &vmstate_msr_tsx_ctrl,
1746         &vmstate_msr_intel_sgx,
1747         &vmstate_pdptrs,
1748         &vmstate_msr_xfd,
1749 #ifdef TARGET_X86_64
1750         &vmstate_amx_xtile,
1751 #endif
1752         &vmstate_arch_lbr,
1753         &vmstate_triple_fault,
1754         NULL
1755     }
1756 };