xref: /openbmc/qemu/target/i386/machine.c (revision 2d7ac0af)
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 "hyperv.h"
7 #include "hw/i386/x86.h"
8 #include "kvm_i386.h"
9 
10 #include "sysemu/kvm.h"
11 #include "sysemu/tcg.h"
12 
13 #include "qemu/error-report.h"
14 
15 static 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 
40 static 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 */
56 static 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 
71 static 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
89 static 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 
111 static 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 
125 static 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 
139 typedef struct x86_FPReg_tmp {
140     FPReg *parent;
141     uint64_t tmp_mant;
142     uint16_t tmp_exp;
143 } x86_FPReg_tmp;
144 
145 static 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 
154 static 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 
163 static 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 
174 static 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 
182 static 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 
193 static 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 
201 static 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 
295 static 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 
390 static bool async_pf_msr_needed(void *opaque)
391 {
392     X86CPU *cpu = opaque;
393 
394     return cpu->env.async_pf_en_msr != 0;
395 }
396 
397 static bool pv_eoi_msr_needed(void *opaque)
398 {
399     X86CPU *cpu = opaque;
400 
401     return cpu->env.pv_eoi_en_msr != 0;
402 }
403 
404 static bool steal_time_msr_needed(void *opaque)
405 {
406     X86CPU *cpu = opaque;
407 
408     return cpu->env.steal_time_msr != 0;
409 }
410 
411 static bool exception_info_needed(void *opaque)
412 {
413     X86CPU *cpu = opaque;
414     CPUX86State *env = &cpu->env;
415 
416     /*
417      * It is important to save exception-info only in case
418      * we need to distinguish between a pending and injected
419      * exception. Which is only required in case there is a
420      * pending exception and vCPU is running L2.
421      * For more info, refer to comment in cpu_pre_save().
422      */
423     return env->exception_pending && (env->hflags & HF_GUEST_MASK);
424 }
425 
426 static const VMStateDescription vmstate_exception_info = {
427     .name = "cpu/exception_info",
428     .version_id = 1,
429     .minimum_version_id = 1,
430     .needed = exception_info_needed,
431     .fields = (VMStateField[]) {
432         VMSTATE_UINT8(env.exception_pending, X86CPU),
433         VMSTATE_UINT8(env.exception_injected, X86CPU),
434         VMSTATE_UINT8(env.exception_has_payload, X86CPU),
435         VMSTATE_UINT64(env.exception_payload, X86CPU),
436         VMSTATE_END_OF_LIST()
437     }
438 };
439 
440 /* Poll control MSR enabled by default */
441 static bool poll_control_msr_needed(void *opaque)
442 {
443     X86CPU *cpu = opaque;
444 
445     return cpu->env.poll_control_msr != 1;
446 }
447 
448 static const VMStateDescription vmstate_steal_time_msr = {
449     .name = "cpu/steal_time_msr",
450     .version_id = 1,
451     .minimum_version_id = 1,
452     .needed = steal_time_msr_needed,
453     .fields = (VMStateField[]) {
454         VMSTATE_UINT64(env.steal_time_msr, X86CPU),
455         VMSTATE_END_OF_LIST()
456     }
457 };
458 
459 static const VMStateDescription vmstate_async_pf_msr = {
460     .name = "cpu/async_pf_msr",
461     .version_id = 1,
462     .minimum_version_id = 1,
463     .needed = async_pf_msr_needed,
464     .fields = (VMStateField[]) {
465         VMSTATE_UINT64(env.async_pf_en_msr, X86CPU),
466         VMSTATE_END_OF_LIST()
467     }
468 };
469 
470 static const VMStateDescription vmstate_pv_eoi_msr = {
471     .name = "cpu/async_pv_eoi_msr",
472     .version_id = 1,
473     .minimum_version_id = 1,
474     .needed = pv_eoi_msr_needed,
475     .fields = (VMStateField[]) {
476         VMSTATE_UINT64(env.pv_eoi_en_msr, X86CPU),
477         VMSTATE_END_OF_LIST()
478     }
479 };
480 
481 static const VMStateDescription vmstate_poll_control_msr = {
482     .name = "cpu/poll_control_msr",
483     .version_id = 1,
484     .minimum_version_id = 1,
485     .needed = poll_control_msr_needed,
486     .fields = (VMStateField[]) {
487         VMSTATE_UINT64(env.poll_control_msr, X86CPU),
488         VMSTATE_END_OF_LIST()
489     }
490 };
491 
492 static bool fpop_ip_dp_needed(void *opaque)
493 {
494     X86CPU *cpu = opaque;
495     CPUX86State *env = &cpu->env;
496 
497     return env->fpop != 0 || env->fpip != 0 || env->fpdp != 0;
498 }
499 
500 static const VMStateDescription vmstate_fpop_ip_dp = {
501     .name = "cpu/fpop_ip_dp",
502     .version_id = 1,
503     .minimum_version_id = 1,
504     .needed = fpop_ip_dp_needed,
505     .fields = (VMStateField[]) {
506         VMSTATE_UINT16(env.fpop, X86CPU),
507         VMSTATE_UINT64(env.fpip, X86CPU),
508         VMSTATE_UINT64(env.fpdp, X86CPU),
509         VMSTATE_END_OF_LIST()
510     }
511 };
512 
513 static bool tsc_adjust_needed(void *opaque)
514 {
515     X86CPU *cpu = opaque;
516     CPUX86State *env = &cpu->env;
517 
518     return env->tsc_adjust != 0;
519 }
520 
521 static const VMStateDescription vmstate_msr_tsc_adjust = {
522     .name = "cpu/msr_tsc_adjust",
523     .version_id = 1,
524     .minimum_version_id = 1,
525     .needed = tsc_adjust_needed,
526     .fields = (VMStateField[]) {
527         VMSTATE_UINT64(env.tsc_adjust, X86CPU),
528         VMSTATE_END_OF_LIST()
529     }
530 };
531 
532 static bool msr_smi_count_needed(void *opaque)
533 {
534     X86CPU *cpu = opaque;
535     CPUX86State *env = &cpu->env;
536 
537     return cpu->migrate_smi_count && env->msr_smi_count != 0;
538 }
539 
540 static const VMStateDescription vmstate_msr_smi_count = {
541     .name = "cpu/msr_smi_count",
542     .version_id = 1,
543     .minimum_version_id = 1,
544     .needed = msr_smi_count_needed,
545     .fields = (VMStateField[]) {
546         VMSTATE_UINT64(env.msr_smi_count, X86CPU),
547         VMSTATE_END_OF_LIST()
548     }
549 };
550 
551 static bool tscdeadline_needed(void *opaque)
552 {
553     X86CPU *cpu = opaque;
554     CPUX86State *env = &cpu->env;
555 
556     return env->tsc_deadline != 0;
557 }
558 
559 static const VMStateDescription vmstate_msr_tscdeadline = {
560     .name = "cpu/msr_tscdeadline",
561     .version_id = 1,
562     .minimum_version_id = 1,
563     .needed = tscdeadline_needed,
564     .fields = (VMStateField[]) {
565         VMSTATE_UINT64(env.tsc_deadline, X86CPU),
566         VMSTATE_END_OF_LIST()
567     }
568 };
569 
570 static bool misc_enable_needed(void *opaque)
571 {
572     X86CPU *cpu = opaque;
573     CPUX86State *env = &cpu->env;
574 
575     return env->msr_ia32_misc_enable != MSR_IA32_MISC_ENABLE_DEFAULT;
576 }
577 
578 static bool feature_control_needed(void *opaque)
579 {
580     X86CPU *cpu = opaque;
581     CPUX86State *env = &cpu->env;
582 
583     return env->msr_ia32_feature_control != 0;
584 }
585 
586 static const VMStateDescription vmstate_msr_ia32_misc_enable = {
587     .name = "cpu/msr_ia32_misc_enable",
588     .version_id = 1,
589     .minimum_version_id = 1,
590     .needed = misc_enable_needed,
591     .fields = (VMStateField[]) {
592         VMSTATE_UINT64(env.msr_ia32_misc_enable, X86CPU),
593         VMSTATE_END_OF_LIST()
594     }
595 };
596 
597 static const VMStateDescription vmstate_msr_ia32_feature_control = {
598     .name = "cpu/msr_ia32_feature_control",
599     .version_id = 1,
600     .minimum_version_id = 1,
601     .needed = feature_control_needed,
602     .fields = (VMStateField[]) {
603         VMSTATE_UINT64(env.msr_ia32_feature_control, X86CPU),
604         VMSTATE_END_OF_LIST()
605     }
606 };
607 
608 static bool pmu_enable_needed(void *opaque)
609 {
610     X86CPU *cpu = opaque;
611     CPUX86State *env = &cpu->env;
612     int i;
613 
614     if (env->msr_fixed_ctr_ctrl || env->msr_global_ctrl ||
615         env->msr_global_status || env->msr_global_ovf_ctrl) {
616         return true;
617     }
618     for (i = 0; i < MAX_FIXED_COUNTERS; i++) {
619         if (env->msr_fixed_counters[i]) {
620             return true;
621         }
622     }
623     for (i = 0; i < MAX_GP_COUNTERS; i++) {
624         if (env->msr_gp_counters[i] || env->msr_gp_evtsel[i]) {
625             return true;
626         }
627     }
628 
629     return false;
630 }
631 
632 static const VMStateDescription vmstate_msr_architectural_pmu = {
633     .name = "cpu/msr_architectural_pmu",
634     .version_id = 1,
635     .minimum_version_id = 1,
636     .needed = pmu_enable_needed,
637     .fields = (VMStateField[]) {
638         VMSTATE_UINT64(env.msr_fixed_ctr_ctrl, X86CPU),
639         VMSTATE_UINT64(env.msr_global_ctrl, X86CPU),
640         VMSTATE_UINT64(env.msr_global_status, X86CPU),
641         VMSTATE_UINT64(env.msr_global_ovf_ctrl, X86CPU),
642         VMSTATE_UINT64_ARRAY(env.msr_fixed_counters, X86CPU, MAX_FIXED_COUNTERS),
643         VMSTATE_UINT64_ARRAY(env.msr_gp_counters, X86CPU, MAX_GP_COUNTERS),
644         VMSTATE_UINT64_ARRAY(env.msr_gp_evtsel, X86CPU, MAX_GP_COUNTERS),
645         VMSTATE_END_OF_LIST()
646     }
647 };
648 
649 static bool mpx_needed(void *opaque)
650 {
651     X86CPU *cpu = opaque;
652     CPUX86State *env = &cpu->env;
653     unsigned int i;
654 
655     for (i = 0; i < 4; i++) {
656         if (env->bnd_regs[i].lb || env->bnd_regs[i].ub) {
657             return true;
658         }
659     }
660 
661     if (env->bndcs_regs.cfgu || env->bndcs_regs.sts) {
662         return true;
663     }
664 
665     return !!env->msr_bndcfgs;
666 }
667 
668 static const VMStateDescription vmstate_mpx = {
669     .name = "cpu/mpx",
670     .version_id = 1,
671     .minimum_version_id = 1,
672     .needed = mpx_needed,
673     .fields = (VMStateField[]) {
674         VMSTATE_BND_REGS(env.bnd_regs, X86CPU, 4),
675         VMSTATE_UINT64(env.bndcs_regs.cfgu, X86CPU),
676         VMSTATE_UINT64(env.bndcs_regs.sts, X86CPU),
677         VMSTATE_UINT64(env.msr_bndcfgs, X86CPU),
678         VMSTATE_END_OF_LIST()
679     }
680 };
681 
682 static bool hyperv_hypercall_enable_needed(void *opaque)
683 {
684     X86CPU *cpu = opaque;
685     CPUX86State *env = &cpu->env;
686 
687     return env->msr_hv_hypercall != 0 || env->msr_hv_guest_os_id != 0;
688 }
689 
690 static const VMStateDescription vmstate_msr_hypercall_hypercall = {
691     .name = "cpu/msr_hyperv_hypercall",
692     .version_id = 1,
693     .minimum_version_id = 1,
694     .needed = hyperv_hypercall_enable_needed,
695     .fields = (VMStateField[]) {
696         VMSTATE_UINT64(env.msr_hv_guest_os_id, X86CPU),
697         VMSTATE_UINT64(env.msr_hv_hypercall, X86CPU),
698         VMSTATE_END_OF_LIST()
699     }
700 };
701 
702 static bool hyperv_vapic_enable_needed(void *opaque)
703 {
704     X86CPU *cpu = opaque;
705     CPUX86State *env = &cpu->env;
706 
707     return env->msr_hv_vapic != 0;
708 }
709 
710 static const VMStateDescription vmstate_msr_hyperv_vapic = {
711     .name = "cpu/msr_hyperv_vapic",
712     .version_id = 1,
713     .minimum_version_id = 1,
714     .needed = hyperv_vapic_enable_needed,
715     .fields = (VMStateField[]) {
716         VMSTATE_UINT64(env.msr_hv_vapic, X86CPU),
717         VMSTATE_END_OF_LIST()
718     }
719 };
720 
721 static bool hyperv_time_enable_needed(void *opaque)
722 {
723     X86CPU *cpu = opaque;
724     CPUX86State *env = &cpu->env;
725 
726     return env->msr_hv_tsc != 0;
727 }
728 
729 static const VMStateDescription vmstate_msr_hyperv_time = {
730     .name = "cpu/msr_hyperv_time",
731     .version_id = 1,
732     .minimum_version_id = 1,
733     .needed = hyperv_time_enable_needed,
734     .fields = (VMStateField[]) {
735         VMSTATE_UINT64(env.msr_hv_tsc, X86CPU),
736         VMSTATE_END_OF_LIST()
737     }
738 };
739 
740 static bool hyperv_crash_enable_needed(void *opaque)
741 {
742     X86CPU *cpu = opaque;
743     CPUX86State *env = &cpu->env;
744     int i;
745 
746     for (i = 0; i < HV_CRASH_PARAMS; i++) {
747         if (env->msr_hv_crash_params[i]) {
748             return true;
749         }
750     }
751     return false;
752 }
753 
754 static const VMStateDescription vmstate_msr_hyperv_crash = {
755     .name = "cpu/msr_hyperv_crash",
756     .version_id = 1,
757     .minimum_version_id = 1,
758     .needed = hyperv_crash_enable_needed,
759     .fields = (VMStateField[]) {
760         VMSTATE_UINT64_ARRAY(env.msr_hv_crash_params, X86CPU, HV_CRASH_PARAMS),
761         VMSTATE_END_OF_LIST()
762     }
763 };
764 
765 static bool hyperv_runtime_enable_needed(void *opaque)
766 {
767     X86CPU *cpu = opaque;
768     CPUX86State *env = &cpu->env;
769 
770     if (!hyperv_feat_enabled(cpu, HYPERV_FEAT_RUNTIME)) {
771         return false;
772     }
773 
774     return env->msr_hv_runtime != 0;
775 }
776 
777 static const VMStateDescription vmstate_msr_hyperv_runtime = {
778     .name = "cpu/msr_hyperv_runtime",
779     .version_id = 1,
780     .minimum_version_id = 1,
781     .needed = hyperv_runtime_enable_needed,
782     .fields = (VMStateField[]) {
783         VMSTATE_UINT64(env.msr_hv_runtime, X86CPU),
784         VMSTATE_END_OF_LIST()
785     }
786 };
787 
788 static bool hyperv_synic_enable_needed(void *opaque)
789 {
790     X86CPU *cpu = opaque;
791     CPUX86State *env = &cpu->env;
792     int i;
793 
794     if (env->msr_hv_synic_control != 0 ||
795         env->msr_hv_synic_evt_page != 0 ||
796         env->msr_hv_synic_msg_page != 0) {
797         return true;
798     }
799 
800     for (i = 0; i < ARRAY_SIZE(env->msr_hv_synic_sint); i++) {
801         if (env->msr_hv_synic_sint[i] != 0) {
802             return true;
803         }
804     }
805 
806     return false;
807 }
808 
809 static int hyperv_synic_post_load(void *opaque, int version_id)
810 {
811     X86CPU *cpu = opaque;
812     hyperv_x86_synic_update(cpu);
813     return 0;
814 }
815 
816 static const VMStateDescription vmstate_msr_hyperv_synic = {
817     .name = "cpu/msr_hyperv_synic",
818     .version_id = 1,
819     .minimum_version_id = 1,
820     .needed = hyperv_synic_enable_needed,
821     .post_load = hyperv_synic_post_load,
822     .fields = (VMStateField[]) {
823         VMSTATE_UINT64(env.msr_hv_synic_control, X86CPU),
824         VMSTATE_UINT64(env.msr_hv_synic_evt_page, X86CPU),
825         VMSTATE_UINT64(env.msr_hv_synic_msg_page, X86CPU),
826         VMSTATE_UINT64_ARRAY(env.msr_hv_synic_sint, X86CPU, HV_SINT_COUNT),
827         VMSTATE_END_OF_LIST()
828     }
829 };
830 
831 static bool hyperv_stimer_enable_needed(void *opaque)
832 {
833     X86CPU *cpu = opaque;
834     CPUX86State *env = &cpu->env;
835     int i;
836 
837     for (i = 0; i < ARRAY_SIZE(env->msr_hv_stimer_config); i++) {
838         if (env->msr_hv_stimer_config[i] || env->msr_hv_stimer_count[i]) {
839             return true;
840         }
841     }
842     return false;
843 }
844 
845 static const VMStateDescription vmstate_msr_hyperv_stimer = {
846     .name = "cpu/msr_hyperv_stimer",
847     .version_id = 1,
848     .minimum_version_id = 1,
849     .needed = hyperv_stimer_enable_needed,
850     .fields = (VMStateField[]) {
851         VMSTATE_UINT64_ARRAY(env.msr_hv_stimer_config, X86CPU,
852                              HV_STIMER_COUNT),
853         VMSTATE_UINT64_ARRAY(env.msr_hv_stimer_count, X86CPU, HV_STIMER_COUNT),
854         VMSTATE_END_OF_LIST()
855     }
856 };
857 
858 static bool hyperv_reenlightenment_enable_needed(void *opaque)
859 {
860     X86CPU *cpu = opaque;
861     CPUX86State *env = &cpu->env;
862 
863     return env->msr_hv_reenlightenment_control != 0 ||
864         env->msr_hv_tsc_emulation_control != 0 ||
865         env->msr_hv_tsc_emulation_status != 0;
866 }
867 
868 static const VMStateDescription vmstate_msr_hyperv_reenlightenment = {
869     .name = "cpu/msr_hyperv_reenlightenment",
870     .version_id = 1,
871     .minimum_version_id = 1,
872     .needed = hyperv_reenlightenment_enable_needed,
873     .fields = (VMStateField[]) {
874         VMSTATE_UINT64(env.msr_hv_reenlightenment_control, X86CPU),
875         VMSTATE_UINT64(env.msr_hv_tsc_emulation_control, X86CPU),
876         VMSTATE_UINT64(env.msr_hv_tsc_emulation_status, X86CPU),
877         VMSTATE_END_OF_LIST()
878     }
879 };
880 
881 static bool avx512_needed(void *opaque)
882 {
883     X86CPU *cpu = opaque;
884     CPUX86State *env = &cpu->env;
885     unsigned int i;
886 
887     for (i = 0; i < NB_OPMASK_REGS; i++) {
888         if (env->opmask_regs[i]) {
889             return true;
890         }
891     }
892 
893     for (i = 0; i < CPU_NB_REGS; i++) {
894 #define ENV_XMM(reg, field) (env->xmm_regs[reg].ZMM_Q(field))
895         if (ENV_XMM(i, 4) || ENV_XMM(i, 6) ||
896             ENV_XMM(i, 5) || ENV_XMM(i, 7)) {
897             return true;
898         }
899 #ifdef TARGET_X86_64
900         if (ENV_XMM(i+16, 0) || ENV_XMM(i+16, 1) ||
901             ENV_XMM(i+16, 2) || ENV_XMM(i+16, 3) ||
902             ENV_XMM(i+16, 4) || ENV_XMM(i+16, 5) ||
903             ENV_XMM(i+16, 6) || ENV_XMM(i+16, 7)) {
904             return true;
905         }
906 #endif
907     }
908 
909     return false;
910 }
911 
912 static const VMStateDescription vmstate_avx512 = {
913     .name = "cpu/avx512",
914     .version_id = 1,
915     .minimum_version_id = 1,
916     .needed = avx512_needed,
917     .fields = (VMStateField[]) {
918         VMSTATE_UINT64_ARRAY(env.opmask_regs, X86CPU, NB_OPMASK_REGS),
919         VMSTATE_ZMMH_REGS_VARS(env.xmm_regs, X86CPU, 0),
920 #ifdef TARGET_X86_64
921         VMSTATE_Hi16_ZMM_REGS_VARS(env.xmm_regs, X86CPU, 16),
922 #endif
923         VMSTATE_END_OF_LIST()
924     }
925 };
926 
927 static bool xss_needed(void *opaque)
928 {
929     X86CPU *cpu = opaque;
930     CPUX86State *env = &cpu->env;
931 
932     return env->xss != 0;
933 }
934 
935 static const VMStateDescription vmstate_xss = {
936     .name = "cpu/xss",
937     .version_id = 1,
938     .minimum_version_id = 1,
939     .needed = xss_needed,
940     .fields = (VMStateField[]) {
941         VMSTATE_UINT64(env.xss, X86CPU),
942         VMSTATE_END_OF_LIST()
943     }
944 };
945 
946 static bool umwait_needed(void *opaque)
947 {
948     X86CPU *cpu = opaque;
949     CPUX86State *env = &cpu->env;
950 
951     return env->umwait != 0;
952 }
953 
954 static const VMStateDescription vmstate_umwait = {
955     .name = "cpu/umwait",
956     .version_id = 1,
957     .minimum_version_id = 1,
958     .needed = umwait_needed,
959     .fields = (VMStateField[]) {
960         VMSTATE_UINT32(env.umwait, X86CPU),
961         VMSTATE_END_OF_LIST()
962     }
963 };
964 
965 #ifdef TARGET_X86_64
966 static bool pkru_needed(void *opaque)
967 {
968     X86CPU *cpu = opaque;
969     CPUX86State *env = &cpu->env;
970 
971     return env->pkru != 0;
972 }
973 
974 static const VMStateDescription vmstate_pkru = {
975     .name = "cpu/pkru",
976     .version_id = 1,
977     .minimum_version_id = 1,
978     .needed = pkru_needed,
979     .fields = (VMStateField[]){
980         VMSTATE_UINT32(env.pkru, X86CPU),
981         VMSTATE_END_OF_LIST()
982     }
983 };
984 #endif
985 
986 static bool tsc_khz_needed(void *opaque)
987 {
988     X86CPU *cpu = opaque;
989     CPUX86State *env = &cpu->env;
990     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
991     X86MachineClass *x86mc = X86_MACHINE_CLASS(mc);
992     return env->tsc_khz && x86mc->save_tsc_khz;
993 }
994 
995 static const VMStateDescription vmstate_tsc_khz = {
996     .name = "cpu/tsc_khz",
997     .version_id = 1,
998     .minimum_version_id = 1,
999     .needed = tsc_khz_needed,
1000     .fields = (VMStateField[]) {
1001         VMSTATE_INT64(env.tsc_khz, X86CPU),
1002         VMSTATE_END_OF_LIST()
1003     }
1004 };
1005 
1006 #ifdef CONFIG_KVM
1007 
1008 static bool vmx_vmcs12_needed(void *opaque)
1009 {
1010     struct kvm_nested_state *nested_state = opaque;
1011     return (nested_state->size >
1012             offsetof(struct kvm_nested_state, data.vmx[0].vmcs12));
1013 }
1014 
1015 static const VMStateDescription vmstate_vmx_vmcs12 = {
1016     .name = "cpu/kvm_nested_state/vmx/vmcs12",
1017     .version_id = 1,
1018     .minimum_version_id = 1,
1019     .needed = vmx_vmcs12_needed,
1020     .fields = (VMStateField[]) {
1021         VMSTATE_UINT8_ARRAY(data.vmx[0].vmcs12,
1022                             struct kvm_nested_state,
1023                             KVM_STATE_NESTED_VMX_VMCS_SIZE),
1024         VMSTATE_END_OF_LIST()
1025     }
1026 };
1027 
1028 static bool vmx_shadow_vmcs12_needed(void *opaque)
1029 {
1030     struct kvm_nested_state *nested_state = opaque;
1031     return (nested_state->size >
1032             offsetof(struct kvm_nested_state, data.vmx[0].shadow_vmcs12));
1033 }
1034 
1035 static const VMStateDescription vmstate_vmx_shadow_vmcs12 = {
1036     .name = "cpu/kvm_nested_state/vmx/shadow_vmcs12",
1037     .version_id = 1,
1038     .minimum_version_id = 1,
1039     .needed = vmx_shadow_vmcs12_needed,
1040     .fields = (VMStateField[]) {
1041         VMSTATE_UINT8_ARRAY(data.vmx[0].shadow_vmcs12,
1042                             struct kvm_nested_state,
1043                             KVM_STATE_NESTED_VMX_VMCS_SIZE),
1044         VMSTATE_END_OF_LIST()
1045     }
1046 };
1047 
1048 static bool vmx_nested_state_needed(void *opaque)
1049 {
1050     struct kvm_nested_state *nested_state = opaque;
1051 
1052     return (nested_state->format == KVM_STATE_NESTED_FORMAT_VMX &&
1053             nested_state->hdr.vmx.vmxon_pa != -1ull);
1054 }
1055 
1056 static const VMStateDescription vmstate_vmx_nested_state = {
1057     .name = "cpu/kvm_nested_state/vmx",
1058     .version_id = 1,
1059     .minimum_version_id = 1,
1060     .needed = vmx_nested_state_needed,
1061     .fields = (VMStateField[]) {
1062         VMSTATE_U64(hdr.vmx.vmxon_pa, struct kvm_nested_state),
1063         VMSTATE_U64(hdr.vmx.vmcs12_pa, struct kvm_nested_state),
1064         VMSTATE_U16(hdr.vmx.smm.flags, struct kvm_nested_state),
1065         VMSTATE_END_OF_LIST()
1066     },
1067     .subsections = (const VMStateDescription*[]) {
1068         &vmstate_vmx_vmcs12,
1069         &vmstate_vmx_shadow_vmcs12,
1070         NULL,
1071     }
1072 };
1073 
1074 static bool svm_nested_state_needed(void *opaque)
1075 {
1076     struct kvm_nested_state *nested_state = opaque;
1077 
1078     /*
1079      * HF_GUEST_MASK and HF2_GIF_MASK are already serialized
1080      * via hflags and hflags2, all that's left is the opaque
1081      * nested state blob.
1082      */
1083     return (nested_state->format == KVM_STATE_NESTED_FORMAT_SVM &&
1084             nested_state->size > offsetof(struct kvm_nested_state, data));
1085 }
1086 
1087 static const VMStateDescription vmstate_svm_nested_state = {
1088     .name = "cpu/kvm_nested_state/svm",
1089     .version_id = 1,
1090     .minimum_version_id = 1,
1091     .needed = svm_nested_state_needed,
1092     .fields = (VMStateField[]) {
1093         VMSTATE_U64(hdr.svm.vmcb_pa, struct kvm_nested_state),
1094         VMSTATE_UINT8_ARRAY(data.svm[0].vmcb12,
1095                             struct kvm_nested_state,
1096                             KVM_STATE_NESTED_SVM_VMCB_SIZE),
1097         VMSTATE_END_OF_LIST()
1098     }
1099 };
1100 
1101 static bool nested_state_needed(void *opaque)
1102 {
1103     X86CPU *cpu = opaque;
1104     CPUX86State *env = &cpu->env;
1105 
1106     return (env->nested_state &&
1107             (vmx_nested_state_needed(env->nested_state) ||
1108              svm_nested_state_needed(env->nested_state)));
1109 }
1110 
1111 static int nested_state_post_load(void *opaque, int version_id)
1112 {
1113     X86CPU *cpu = opaque;
1114     CPUX86State *env = &cpu->env;
1115     struct kvm_nested_state *nested_state = env->nested_state;
1116     int min_nested_state_len = offsetof(struct kvm_nested_state, data);
1117     int max_nested_state_len = kvm_max_nested_state_length();
1118 
1119     /*
1120      * If our kernel don't support setting nested state
1121      * and we have received nested state from migration stream,
1122      * we need to fail migration
1123      */
1124     if (max_nested_state_len <= 0) {
1125         error_report("Received nested state when kernel cannot restore it");
1126         return -EINVAL;
1127     }
1128 
1129     /*
1130      * Verify that the size of received nested_state struct
1131      * at least cover required header and is not larger
1132      * than the max size that our kernel support
1133      */
1134     if (nested_state->size < min_nested_state_len) {
1135         error_report("Received nested state size less than min: "
1136                      "len=%d, min=%d",
1137                      nested_state->size, min_nested_state_len);
1138         return -EINVAL;
1139     }
1140     if (nested_state->size > max_nested_state_len) {
1141         error_report("Recieved unsupported nested state size: "
1142                      "nested_state->size=%d, max=%d",
1143                      nested_state->size, max_nested_state_len);
1144         return -EINVAL;
1145     }
1146 
1147     /* Verify format is valid */
1148     if ((nested_state->format != KVM_STATE_NESTED_FORMAT_VMX) &&
1149         (nested_state->format != KVM_STATE_NESTED_FORMAT_SVM)) {
1150         error_report("Received invalid nested state format: %d",
1151                      nested_state->format);
1152         return -EINVAL;
1153     }
1154 
1155     return 0;
1156 }
1157 
1158 static const VMStateDescription vmstate_kvm_nested_state = {
1159     .name = "cpu/kvm_nested_state",
1160     .version_id = 1,
1161     .minimum_version_id = 1,
1162     .fields = (VMStateField[]) {
1163         VMSTATE_U16(flags, struct kvm_nested_state),
1164         VMSTATE_U16(format, struct kvm_nested_state),
1165         VMSTATE_U32(size, struct kvm_nested_state),
1166         VMSTATE_END_OF_LIST()
1167     },
1168     .subsections = (const VMStateDescription*[]) {
1169         &vmstate_vmx_nested_state,
1170         &vmstate_svm_nested_state,
1171         NULL
1172     }
1173 };
1174 
1175 static const VMStateDescription vmstate_nested_state = {
1176     .name = "cpu/nested_state",
1177     .version_id = 1,
1178     .minimum_version_id = 1,
1179     .needed = nested_state_needed,
1180     .post_load = nested_state_post_load,
1181     .fields = (VMStateField[]) {
1182         VMSTATE_STRUCT_POINTER(env.nested_state, X86CPU,
1183                 vmstate_kvm_nested_state,
1184                 struct kvm_nested_state),
1185         VMSTATE_END_OF_LIST()
1186     }
1187 };
1188 
1189 #endif
1190 
1191 static bool mcg_ext_ctl_needed(void *opaque)
1192 {
1193     X86CPU *cpu = opaque;
1194     CPUX86State *env = &cpu->env;
1195     return cpu->enable_lmce && env->mcg_ext_ctl;
1196 }
1197 
1198 static const VMStateDescription vmstate_mcg_ext_ctl = {
1199     .name = "cpu/mcg_ext_ctl",
1200     .version_id = 1,
1201     .minimum_version_id = 1,
1202     .needed = mcg_ext_ctl_needed,
1203     .fields = (VMStateField[]) {
1204         VMSTATE_UINT64(env.mcg_ext_ctl, X86CPU),
1205         VMSTATE_END_OF_LIST()
1206     }
1207 };
1208 
1209 static bool spec_ctrl_needed(void *opaque)
1210 {
1211     X86CPU *cpu = opaque;
1212     CPUX86State *env = &cpu->env;
1213 
1214     return env->spec_ctrl != 0;
1215 }
1216 
1217 static const VMStateDescription vmstate_spec_ctrl = {
1218     .name = "cpu/spec_ctrl",
1219     .version_id = 1,
1220     .minimum_version_id = 1,
1221     .needed = spec_ctrl_needed,
1222     .fields = (VMStateField[]){
1223         VMSTATE_UINT64(env.spec_ctrl, X86CPU),
1224         VMSTATE_END_OF_LIST()
1225     }
1226 };
1227 
1228 static bool intel_pt_enable_needed(void *opaque)
1229 {
1230     X86CPU *cpu = opaque;
1231     CPUX86State *env = &cpu->env;
1232     int i;
1233 
1234     if (env->msr_rtit_ctrl || env->msr_rtit_status ||
1235         env->msr_rtit_output_base || env->msr_rtit_output_mask ||
1236         env->msr_rtit_cr3_match) {
1237         return true;
1238     }
1239 
1240     for (i = 0; i < MAX_RTIT_ADDRS; i++) {
1241         if (env->msr_rtit_addrs[i]) {
1242             return true;
1243         }
1244     }
1245 
1246     return false;
1247 }
1248 
1249 static const VMStateDescription vmstate_msr_intel_pt = {
1250     .name = "cpu/intel_pt",
1251     .version_id = 1,
1252     .minimum_version_id = 1,
1253     .needed = intel_pt_enable_needed,
1254     .fields = (VMStateField[]) {
1255         VMSTATE_UINT64(env.msr_rtit_ctrl, X86CPU),
1256         VMSTATE_UINT64(env.msr_rtit_status, X86CPU),
1257         VMSTATE_UINT64(env.msr_rtit_output_base, X86CPU),
1258         VMSTATE_UINT64(env.msr_rtit_output_mask, X86CPU),
1259         VMSTATE_UINT64(env.msr_rtit_cr3_match, X86CPU),
1260         VMSTATE_UINT64_ARRAY(env.msr_rtit_addrs, X86CPU, MAX_RTIT_ADDRS),
1261         VMSTATE_END_OF_LIST()
1262     }
1263 };
1264 
1265 static bool virt_ssbd_needed(void *opaque)
1266 {
1267     X86CPU *cpu = opaque;
1268     CPUX86State *env = &cpu->env;
1269 
1270     return env->virt_ssbd != 0;
1271 }
1272 
1273 static const VMStateDescription vmstate_msr_virt_ssbd = {
1274     .name = "cpu/virt_ssbd",
1275     .version_id = 1,
1276     .minimum_version_id = 1,
1277     .needed = virt_ssbd_needed,
1278     .fields = (VMStateField[]){
1279         VMSTATE_UINT64(env.virt_ssbd, X86CPU),
1280         VMSTATE_END_OF_LIST()
1281     }
1282 };
1283 
1284 static bool svm_npt_needed(void *opaque)
1285 {
1286     X86CPU *cpu = opaque;
1287     CPUX86State *env = &cpu->env;
1288 
1289     return !!(env->hflags2 & HF2_NPT_MASK);
1290 }
1291 
1292 static const VMStateDescription vmstate_svm_npt = {
1293     .name = "cpu/svn_npt",
1294     .version_id = 1,
1295     .minimum_version_id = 1,
1296     .needed = svm_npt_needed,
1297     .fields = (VMStateField[]){
1298         VMSTATE_UINT64(env.nested_cr3, X86CPU),
1299         VMSTATE_UINT32(env.nested_pg_mode, X86CPU),
1300         VMSTATE_END_OF_LIST()
1301     }
1302 };
1303 
1304 #ifndef TARGET_X86_64
1305 static bool intel_efer32_needed(void *opaque)
1306 {
1307     X86CPU *cpu = opaque;
1308     CPUX86State *env = &cpu->env;
1309 
1310     return env->efer != 0;
1311 }
1312 
1313 static const VMStateDescription vmstate_efer32 = {
1314     .name = "cpu/efer32",
1315     .version_id = 1,
1316     .minimum_version_id = 1,
1317     .needed = intel_efer32_needed,
1318     .fields = (VMStateField[]) {
1319         VMSTATE_UINT64(env.efer, X86CPU),
1320         VMSTATE_END_OF_LIST()
1321     }
1322 };
1323 #endif
1324 
1325 static bool msr_tsx_ctrl_needed(void *opaque)
1326 {
1327     X86CPU *cpu = opaque;
1328     CPUX86State *env = &cpu->env;
1329 
1330     return env->features[FEAT_ARCH_CAPABILITIES] & ARCH_CAP_TSX_CTRL_MSR;
1331 }
1332 
1333 static const VMStateDescription vmstate_msr_tsx_ctrl = {
1334     .name = "cpu/msr_tsx_ctrl",
1335     .version_id = 1,
1336     .minimum_version_id = 1,
1337     .needed = msr_tsx_ctrl_needed,
1338     .fields = (VMStateField[]) {
1339         VMSTATE_UINT32(env.tsx_ctrl, X86CPU),
1340         VMSTATE_END_OF_LIST()
1341     }
1342 };
1343 
1344 VMStateDescription vmstate_x86_cpu = {
1345     .name = "cpu",
1346     .version_id = 12,
1347     .minimum_version_id = 11,
1348     .pre_save = cpu_pre_save,
1349     .post_load = cpu_post_load,
1350     .fields = (VMStateField[]) {
1351         VMSTATE_UINTTL_ARRAY(env.regs, X86CPU, CPU_NB_REGS),
1352         VMSTATE_UINTTL(env.eip, X86CPU),
1353         VMSTATE_UINTTL(env.eflags, X86CPU),
1354         VMSTATE_UINT32(env.hflags, X86CPU),
1355         /* FPU */
1356         VMSTATE_UINT16(env.fpuc, X86CPU),
1357         VMSTATE_UINT16(env.fpus_vmstate, X86CPU),
1358         VMSTATE_UINT16(env.fptag_vmstate, X86CPU),
1359         VMSTATE_UINT16(env.fpregs_format_vmstate, X86CPU),
1360 
1361         VMSTATE_STRUCT_ARRAY(env.fpregs, X86CPU, 8, 0, vmstate_fpreg, FPReg),
1362 
1363         VMSTATE_SEGMENT_ARRAY(env.segs, X86CPU, 6),
1364         VMSTATE_SEGMENT(env.ldt, X86CPU),
1365         VMSTATE_SEGMENT(env.tr, X86CPU),
1366         VMSTATE_SEGMENT(env.gdt, X86CPU),
1367         VMSTATE_SEGMENT(env.idt, X86CPU),
1368 
1369         VMSTATE_UINT32(env.sysenter_cs, X86CPU),
1370         VMSTATE_UINTTL(env.sysenter_esp, X86CPU),
1371         VMSTATE_UINTTL(env.sysenter_eip, X86CPU),
1372 
1373         VMSTATE_UINTTL(env.cr[0], X86CPU),
1374         VMSTATE_UINTTL(env.cr[2], X86CPU),
1375         VMSTATE_UINTTL(env.cr[3], X86CPU),
1376         VMSTATE_UINTTL(env.cr[4], X86CPU),
1377         VMSTATE_UINTTL_ARRAY(env.dr, X86CPU, 8),
1378         /* MMU */
1379         VMSTATE_INT32(env.a20_mask, X86CPU),
1380         /* XMM */
1381         VMSTATE_UINT32(env.mxcsr, X86CPU),
1382         VMSTATE_XMM_REGS(env.xmm_regs, X86CPU, 0),
1383 
1384 #ifdef TARGET_X86_64
1385         VMSTATE_UINT64(env.efer, X86CPU),
1386         VMSTATE_UINT64(env.star, X86CPU),
1387         VMSTATE_UINT64(env.lstar, X86CPU),
1388         VMSTATE_UINT64(env.cstar, X86CPU),
1389         VMSTATE_UINT64(env.fmask, X86CPU),
1390         VMSTATE_UINT64(env.kernelgsbase, X86CPU),
1391 #endif
1392         VMSTATE_UINT32(env.smbase, X86CPU),
1393 
1394         VMSTATE_UINT64(env.pat, X86CPU),
1395         VMSTATE_UINT32(env.hflags2, X86CPU),
1396 
1397         VMSTATE_UINT64(env.vm_hsave, X86CPU),
1398         VMSTATE_UINT64(env.vm_vmcb, X86CPU),
1399         VMSTATE_UINT64(env.tsc_offset, X86CPU),
1400         VMSTATE_UINT64(env.intercept, X86CPU),
1401         VMSTATE_UINT16(env.intercept_cr_read, X86CPU),
1402         VMSTATE_UINT16(env.intercept_cr_write, X86CPU),
1403         VMSTATE_UINT16(env.intercept_dr_read, X86CPU),
1404         VMSTATE_UINT16(env.intercept_dr_write, X86CPU),
1405         VMSTATE_UINT32(env.intercept_exceptions, X86CPU),
1406         VMSTATE_UINT8(env.v_tpr, X86CPU),
1407         /* MTRRs */
1408         VMSTATE_UINT64_ARRAY(env.mtrr_fixed, X86CPU, 11),
1409         VMSTATE_UINT64(env.mtrr_deftype, X86CPU),
1410         VMSTATE_MTRR_VARS(env.mtrr_var, X86CPU, MSR_MTRRcap_VCNT, 8),
1411         /* KVM-related states */
1412         VMSTATE_INT32(env.interrupt_injected, X86CPU),
1413         VMSTATE_UINT32(env.mp_state, X86CPU),
1414         VMSTATE_UINT64(env.tsc, X86CPU),
1415         VMSTATE_INT32(env.exception_nr, X86CPU),
1416         VMSTATE_UINT8(env.soft_interrupt, X86CPU),
1417         VMSTATE_UINT8(env.nmi_injected, X86CPU),
1418         VMSTATE_UINT8(env.nmi_pending, X86CPU),
1419         VMSTATE_UINT8(env.has_error_code, X86CPU),
1420         VMSTATE_UINT32(env.sipi_vector, X86CPU),
1421         /* MCE */
1422         VMSTATE_UINT64(env.mcg_cap, X86CPU),
1423         VMSTATE_UINT64(env.mcg_status, X86CPU),
1424         VMSTATE_UINT64(env.mcg_ctl, X86CPU),
1425         VMSTATE_UINT64_ARRAY(env.mce_banks, X86CPU, MCE_BANKS_DEF * 4),
1426         /* rdtscp */
1427         VMSTATE_UINT64(env.tsc_aux, X86CPU),
1428         /* KVM pvclock msr */
1429         VMSTATE_UINT64(env.system_time_msr, X86CPU),
1430         VMSTATE_UINT64(env.wall_clock_msr, X86CPU),
1431         /* XSAVE related fields */
1432         VMSTATE_UINT64_V(env.xcr0, X86CPU, 12),
1433         VMSTATE_UINT64_V(env.xstate_bv, X86CPU, 12),
1434         VMSTATE_YMMH_REGS_VARS(env.xmm_regs, X86CPU, 0, 12),
1435         VMSTATE_END_OF_LIST()
1436         /* The above list is not sorted /wrt version numbers, watch out! */
1437     },
1438     .subsections = (const VMStateDescription*[]) {
1439         &vmstate_exception_info,
1440         &vmstate_async_pf_msr,
1441         &vmstate_pv_eoi_msr,
1442         &vmstate_steal_time_msr,
1443         &vmstate_poll_control_msr,
1444         &vmstate_fpop_ip_dp,
1445         &vmstate_msr_tsc_adjust,
1446         &vmstate_msr_tscdeadline,
1447         &vmstate_msr_ia32_misc_enable,
1448         &vmstate_msr_ia32_feature_control,
1449         &vmstate_msr_architectural_pmu,
1450         &vmstate_mpx,
1451         &vmstate_msr_hypercall_hypercall,
1452         &vmstate_msr_hyperv_vapic,
1453         &vmstate_msr_hyperv_time,
1454         &vmstate_msr_hyperv_crash,
1455         &vmstate_msr_hyperv_runtime,
1456         &vmstate_msr_hyperv_synic,
1457         &vmstate_msr_hyperv_stimer,
1458         &vmstate_msr_hyperv_reenlightenment,
1459         &vmstate_avx512,
1460         &vmstate_xss,
1461         &vmstate_umwait,
1462         &vmstate_tsc_khz,
1463         &vmstate_msr_smi_count,
1464 #ifdef TARGET_X86_64
1465         &vmstate_pkru,
1466 #endif
1467         &vmstate_spec_ctrl,
1468         &vmstate_mcg_ext_ctl,
1469         &vmstate_msr_intel_pt,
1470         &vmstate_msr_virt_ssbd,
1471         &vmstate_svm_npt,
1472 #ifndef TARGET_X86_64
1473         &vmstate_efer32,
1474 #endif
1475 #ifdef CONFIG_KVM
1476         &vmstate_nested_state,
1477 #endif
1478         &vmstate_msr_tsx_ctrl,
1479         NULL
1480     }
1481 };
1482