xref: /openbmc/qemu/target/i386/kvm/kvm.c (revision 30b6852c)
1 /*
2  * QEMU KVM support
3  *
4  * Copyright (C) 2006-2008 Qumranet Technologies
5  * Copyright IBM, Corp. 2008
6  *
7  * Authors:
8  *  Anthony Liguori   <aliguori@us.ibm.com>
9  *
10  * This work is licensed under the terms of the GNU GPL, version 2 or later.
11  * See the COPYING file in the top-level directory.
12  *
13  */
14 
15 #include "qemu/osdep.h"
16 #include "qapi/qapi-events-run-state.h"
17 #include "qapi/error.h"
18 #include <sys/ioctl.h>
19 #include <sys/utsname.h>
20 
21 #include <linux/kvm.h>
22 #include "standard-headers/asm-x86/kvm_para.h"
23 
24 #include "cpu.h"
25 #include "host-cpu.h"
26 #include "sysemu/sysemu.h"
27 #include "sysemu/hw_accel.h"
28 #include "sysemu/kvm_int.h"
29 #include "sysemu/runstate.h"
30 #include "kvm_i386.h"
31 #include "sev_i386.h"
32 #include "hyperv.h"
33 #include "hyperv-proto.h"
34 
35 #include "exec/gdbstub.h"
36 #include "qemu/host-utils.h"
37 #include "qemu/main-loop.h"
38 #include "qemu/config-file.h"
39 #include "qemu/error-report.h"
40 #include "hw/i386/x86.h"
41 #include "hw/i386/apic.h"
42 #include "hw/i386/apic_internal.h"
43 #include "hw/i386/apic-msidef.h"
44 #include "hw/i386/intel_iommu.h"
45 #include "hw/i386/x86-iommu.h"
46 #include "hw/i386/e820_memory_layout.h"
47 #include "sysemu/sev.h"
48 
49 #include "hw/pci/pci.h"
50 #include "hw/pci/msi.h"
51 #include "hw/pci/msix.h"
52 #include "migration/blocker.h"
53 #include "exec/memattrs.h"
54 #include "trace.h"
55 
56 //#define DEBUG_KVM
57 
58 #ifdef DEBUG_KVM
59 #define DPRINTF(fmt, ...) \
60     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
61 #else
62 #define DPRINTF(fmt, ...) \
63     do { } while (0)
64 #endif
65 
66 /* From arch/x86/kvm/lapic.h */
67 #define KVM_APIC_BUS_CYCLE_NS       1
68 #define KVM_APIC_BUS_FREQUENCY      (1000000000ULL / KVM_APIC_BUS_CYCLE_NS)
69 
70 #define MSR_KVM_WALL_CLOCK  0x11
71 #define MSR_KVM_SYSTEM_TIME 0x12
72 
73 /* A 4096-byte buffer can hold the 8-byte kvm_msrs header, plus
74  * 255 kvm_msr_entry structs */
75 #define MSR_BUF_SIZE 4096
76 
77 static void kvm_init_msrs(X86CPU *cpu);
78 
79 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
80     KVM_CAP_INFO(SET_TSS_ADDR),
81     KVM_CAP_INFO(EXT_CPUID),
82     KVM_CAP_INFO(MP_STATE),
83     KVM_CAP_LAST_INFO
84 };
85 
86 static bool has_msr_star;
87 static bool has_msr_hsave_pa;
88 static bool has_msr_tsc_aux;
89 static bool has_msr_tsc_adjust;
90 static bool has_msr_tsc_deadline;
91 static bool has_msr_feature_control;
92 static bool has_msr_misc_enable;
93 static bool has_msr_smbase;
94 static bool has_msr_bndcfgs;
95 static int lm_capable_kernel;
96 static bool has_msr_hv_hypercall;
97 static bool has_msr_hv_crash;
98 static bool has_msr_hv_reset;
99 static bool has_msr_hv_vpindex;
100 static bool hv_vpindex_settable;
101 static bool has_msr_hv_runtime;
102 static bool has_msr_hv_synic;
103 static bool has_msr_hv_stimer;
104 static bool has_msr_hv_frequencies;
105 static bool has_msr_hv_reenlightenment;
106 static bool has_msr_xss;
107 static bool has_msr_umwait;
108 static bool has_msr_spec_ctrl;
109 static bool has_msr_tsx_ctrl;
110 static bool has_msr_virt_ssbd;
111 static bool has_msr_smi_count;
112 static bool has_msr_arch_capabs;
113 static bool has_msr_core_capabs;
114 static bool has_msr_vmx_vmfunc;
115 static bool has_msr_ucode_rev;
116 static bool has_msr_vmx_procbased_ctls2;
117 static bool has_msr_perf_capabs;
118 static bool has_msr_pkrs;
119 
120 static uint32_t has_architectural_pmu_version;
121 static uint32_t num_architectural_pmu_gp_counters;
122 static uint32_t num_architectural_pmu_fixed_counters;
123 
124 static int has_xsave;
125 static int has_xcrs;
126 static int has_pit_state2;
127 static int has_exception_payload;
128 
129 static bool has_msr_mcg_ext_ctl;
130 
131 static struct kvm_cpuid2 *cpuid_cache;
132 static struct kvm_cpuid2 *hv_cpuid_cache;
133 static struct kvm_msr_list *kvm_feature_msrs;
134 
135 #define BUS_LOCK_SLICE_TIME 1000000000ULL /* ns */
136 static RateLimit bus_lock_ratelimit_ctrl;
137 
138 int kvm_has_pit_state2(void)
139 {
140     return has_pit_state2;
141 }
142 
143 bool kvm_has_smm(void)
144 {
145     return kvm_vm_check_extension(kvm_state, KVM_CAP_X86_SMM);
146 }
147 
148 bool kvm_has_adjust_clock_stable(void)
149 {
150     int ret = kvm_check_extension(kvm_state, KVM_CAP_ADJUST_CLOCK);
151 
152     return (ret == KVM_CLOCK_TSC_STABLE);
153 }
154 
155 bool kvm_has_adjust_clock(void)
156 {
157     return kvm_check_extension(kvm_state, KVM_CAP_ADJUST_CLOCK);
158 }
159 
160 bool kvm_has_exception_payload(void)
161 {
162     return has_exception_payload;
163 }
164 
165 static bool kvm_x2apic_api_set_flags(uint64_t flags)
166 {
167     KVMState *s = KVM_STATE(current_accel());
168 
169     return !kvm_vm_enable_cap(s, KVM_CAP_X2APIC_API, 0, flags);
170 }
171 
172 #define MEMORIZE(fn, _result) \
173     ({ \
174         static bool _memorized; \
175         \
176         if (_memorized) { \
177             return _result; \
178         } \
179         _memorized = true; \
180         _result = fn; \
181     })
182 
183 static bool has_x2apic_api;
184 
185 bool kvm_has_x2apic_api(void)
186 {
187     return has_x2apic_api;
188 }
189 
190 bool kvm_enable_x2apic(void)
191 {
192     return MEMORIZE(
193              kvm_x2apic_api_set_flags(KVM_X2APIC_API_USE_32BIT_IDS |
194                                       KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK),
195              has_x2apic_api);
196 }
197 
198 bool kvm_hv_vpindex_settable(void)
199 {
200     return hv_vpindex_settable;
201 }
202 
203 static int kvm_get_tsc(CPUState *cs)
204 {
205     X86CPU *cpu = X86_CPU(cs);
206     CPUX86State *env = &cpu->env;
207     struct {
208         struct kvm_msrs info;
209         struct kvm_msr_entry entries[1];
210     } msr_data = {};
211     int ret;
212 
213     if (env->tsc_valid) {
214         return 0;
215     }
216 
217     memset(&msr_data, 0, sizeof(msr_data));
218     msr_data.info.nmsrs = 1;
219     msr_data.entries[0].index = MSR_IA32_TSC;
220     env->tsc_valid = !runstate_is_running();
221 
222     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MSRS, &msr_data);
223     if (ret < 0) {
224         return ret;
225     }
226 
227     assert(ret == 1);
228     env->tsc = msr_data.entries[0].data;
229     return 0;
230 }
231 
232 static inline void do_kvm_synchronize_tsc(CPUState *cpu, run_on_cpu_data arg)
233 {
234     kvm_get_tsc(cpu);
235 }
236 
237 void kvm_synchronize_all_tsc(void)
238 {
239     CPUState *cpu;
240 
241     if (kvm_enabled()) {
242         CPU_FOREACH(cpu) {
243             run_on_cpu(cpu, do_kvm_synchronize_tsc, RUN_ON_CPU_NULL);
244         }
245     }
246 }
247 
248 static struct kvm_cpuid2 *try_get_cpuid(KVMState *s, int max)
249 {
250     struct kvm_cpuid2 *cpuid;
251     int r, size;
252 
253     size = sizeof(*cpuid) + max * sizeof(*cpuid->entries);
254     cpuid = g_malloc0(size);
255     cpuid->nent = max;
256     r = kvm_ioctl(s, KVM_GET_SUPPORTED_CPUID, cpuid);
257     if (r == 0 && cpuid->nent >= max) {
258         r = -E2BIG;
259     }
260     if (r < 0) {
261         if (r == -E2BIG) {
262             g_free(cpuid);
263             return NULL;
264         } else {
265             fprintf(stderr, "KVM_GET_SUPPORTED_CPUID failed: %s\n",
266                     strerror(-r));
267             exit(1);
268         }
269     }
270     return cpuid;
271 }
272 
273 /* Run KVM_GET_SUPPORTED_CPUID ioctl(), allocating a buffer large enough
274  * for all entries.
275  */
276 static struct kvm_cpuid2 *get_supported_cpuid(KVMState *s)
277 {
278     struct kvm_cpuid2 *cpuid;
279     int max = 1;
280 
281     if (cpuid_cache != NULL) {
282         return cpuid_cache;
283     }
284     while ((cpuid = try_get_cpuid(s, max)) == NULL) {
285         max *= 2;
286     }
287     cpuid_cache = cpuid;
288     return cpuid;
289 }
290 
291 static bool host_tsx_broken(void)
292 {
293     int family, model, stepping;\
294     char vendor[CPUID_VENDOR_SZ + 1];
295 
296     host_cpu_vendor_fms(vendor, &family, &model, &stepping);
297 
298     /* Check if we are running on a Haswell host known to have broken TSX */
299     return !strcmp(vendor, CPUID_VENDOR_INTEL) &&
300            (family == 6) &&
301            ((model == 63 && stepping < 4) ||
302             model == 60 || model == 69 || model == 70);
303 }
304 
305 /* Returns the value for a specific register on the cpuid entry
306  */
307 static uint32_t cpuid_entry_get_reg(struct kvm_cpuid_entry2 *entry, int reg)
308 {
309     uint32_t ret = 0;
310     switch (reg) {
311     case R_EAX:
312         ret = entry->eax;
313         break;
314     case R_EBX:
315         ret = entry->ebx;
316         break;
317     case R_ECX:
318         ret = entry->ecx;
319         break;
320     case R_EDX:
321         ret = entry->edx;
322         break;
323     }
324     return ret;
325 }
326 
327 /* Find matching entry for function/index on kvm_cpuid2 struct
328  */
329 static struct kvm_cpuid_entry2 *cpuid_find_entry(struct kvm_cpuid2 *cpuid,
330                                                  uint32_t function,
331                                                  uint32_t index)
332 {
333     int i;
334     for (i = 0; i < cpuid->nent; ++i) {
335         if (cpuid->entries[i].function == function &&
336             cpuid->entries[i].index == index) {
337             return &cpuid->entries[i];
338         }
339     }
340     /* not found: */
341     return NULL;
342 }
343 
344 uint32_t kvm_arch_get_supported_cpuid(KVMState *s, uint32_t function,
345                                       uint32_t index, int reg)
346 {
347     struct kvm_cpuid2 *cpuid;
348     uint32_t ret = 0;
349     uint32_t cpuid_1_edx;
350 
351     cpuid = get_supported_cpuid(s);
352 
353     struct kvm_cpuid_entry2 *entry = cpuid_find_entry(cpuid, function, index);
354     if (entry) {
355         ret = cpuid_entry_get_reg(entry, reg);
356     }
357 
358     /* Fixups for the data returned by KVM, below */
359 
360     if (function == 1 && reg == R_EDX) {
361         /* KVM before 2.6.30 misreports the following features */
362         ret |= CPUID_MTRR | CPUID_PAT | CPUID_MCE | CPUID_MCA;
363     } else if (function == 1 && reg == R_ECX) {
364         /* We can set the hypervisor flag, even if KVM does not return it on
365          * GET_SUPPORTED_CPUID
366          */
367         ret |= CPUID_EXT_HYPERVISOR;
368         /* tsc-deadline flag is not returned by GET_SUPPORTED_CPUID, but it
369          * can be enabled if the kernel has KVM_CAP_TSC_DEADLINE_TIMER,
370          * and the irqchip is in the kernel.
371          */
372         if (kvm_irqchip_in_kernel() &&
373                 kvm_check_extension(s, KVM_CAP_TSC_DEADLINE_TIMER)) {
374             ret |= CPUID_EXT_TSC_DEADLINE_TIMER;
375         }
376 
377         /* x2apic is reported by GET_SUPPORTED_CPUID, but it can't be enabled
378          * without the in-kernel irqchip
379          */
380         if (!kvm_irqchip_in_kernel()) {
381             ret &= ~CPUID_EXT_X2APIC;
382         }
383 
384         if (enable_cpu_pm) {
385             int disable_exits = kvm_check_extension(s,
386                                                     KVM_CAP_X86_DISABLE_EXITS);
387 
388             if (disable_exits & KVM_X86_DISABLE_EXITS_MWAIT) {
389                 ret |= CPUID_EXT_MONITOR;
390             }
391         }
392     } else if (function == 6 && reg == R_EAX) {
393         ret |= CPUID_6_EAX_ARAT; /* safe to allow because of emulated APIC */
394     } else if (function == 7 && index == 0 && reg == R_EBX) {
395         if (host_tsx_broken()) {
396             ret &= ~(CPUID_7_0_EBX_RTM | CPUID_7_0_EBX_HLE);
397         }
398     } else if (function == 7 && index == 0 && reg == R_EDX) {
399         /*
400          * Linux v4.17-v4.20 incorrectly return ARCH_CAPABILITIES on SVM hosts.
401          * We can detect the bug by checking if MSR_IA32_ARCH_CAPABILITIES is
402          * returned by KVM_GET_MSR_INDEX_LIST.
403          */
404         if (!has_msr_arch_capabs) {
405             ret &= ~CPUID_7_0_EDX_ARCH_CAPABILITIES;
406         }
407     } else if (function == 0x80000001 && reg == R_ECX) {
408         /*
409          * It's safe to enable TOPOEXT even if it's not returned by
410          * GET_SUPPORTED_CPUID.  Unconditionally enabling TOPOEXT here allows
411          * us to keep CPU models including TOPOEXT runnable on older kernels.
412          */
413         ret |= CPUID_EXT3_TOPOEXT;
414     } else if (function == 0x80000001 && reg == R_EDX) {
415         /* On Intel, kvm returns cpuid according to the Intel spec,
416          * so add missing bits according to the AMD spec:
417          */
418         cpuid_1_edx = kvm_arch_get_supported_cpuid(s, 1, 0, R_EDX);
419         ret |= cpuid_1_edx & CPUID_EXT2_AMD_ALIASES;
420     } else if (function == KVM_CPUID_FEATURES && reg == R_EAX) {
421         /* kvm_pv_unhalt is reported by GET_SUPPORTED_CPUID, but it can't
422          * be enabled without the in-kernel irqchip
423          */
424         if (!kvm_irqchip_in_kernel()) {
425             ret &= ~(1U << KVM_FEATURE_PV_UNHALT);
426         }
427         if (kvm_irqchip_is_split()) {
428             ret |= 1U << KVM_FEATURE_MSI_EXT_DEST_ID;
429         }
430     } else if (function == KVM_CPUID_FEATURES && reg == R_EDX) {
431         ret |= 1U << KVM_HINTS_REALTIME;
432     }
433 
434     return ret;
435 }
436 
437 uint64_t kvm_arch_get_supported_msr_feature(KVMState *s, uint32_t index)
438 {
439     struct {
440         struct kvm_msrs info;
441         struct kvm_msr_entry entries[1];
442     } msr_data = {};
443     uint64_t value;
444     uint32_t ret, can_be_one, must_be_one;
445 
446     if (kvm_feature_msrs == NULL) { /* Host doesn't support feature MSRs */
447         return 0;
448     }
449 
450     /* Check if requested MSR is supported feature MSR */
451     int i;
452     for (i = 0; i < kvm_feature_msrs->nmsrs; i++)
453         if (kvm_feature_msrs->indices[i] == index) {
454             break;
455         }
456     if (i == kvm_feature_msrs->nmsrs) {
457         return 0; /* if the feature MSR is not supported, simply return 0 */
458     }
459 
460     msr_data.info.nmsrs = 1;
461     msr_data.entries[0].index = index;
462 
463     ret = kvm_ioctl(s, KVM_GET_MSRS, &msr_data);
464     if (ret != 1) {
465         error_report("KVM get MSR (index=0x%x) feature failed, %s",
466             index, strerror(-ret));
467         exit(1);
468     }
469 
470     value = msr_data.entries[0].data;
471     switch (index) {
472     case MSR_IA32_VMX_PROCBASED_CTLS2:
473         if (!has_msr_vmx_procbased_ctls2) {
474             /* KVM forgot to add these bits for some time, do this ourselves. */
475             if (kvm_arch_get_supported_cpuid(s, 0xD, 1, R_ECX) &
476                 CPUID_XSAVE_XSAVES) {
477                 value |= (uint64_t)VMX_SECONDARY_EXEC_XSAVES << 32;
478             }
479             if (kvm_arch_get_supported_cpuid(s, 1, 0, R_ECX) &
480                 CPUID_EXT_RDRAND) {
481                 value |= (uint64_t)VMX_SECONDARY_EXEC_RDRAND_EXITING << 32;
482             }
483             if (kvm_arch_get_supported_cpuid(s, 7, 0, R_EBX) &
484                 CPUID_7_0_EBX_INVPCID) {
485                 value |= (uint64_t)VMX_SECONDARY_EXEC_ENABLE_INVPCID << 32;
486             }
487             if (kvm_arch_get_supported_cpuid(s, 7, 0, R_EBX) &
488                 CPUID_7_0_EBX_RDSEED) {
489                 value |= (uint64_t)VMX_SECONDARY_EXEC_RDSEED_EXITING << 32;
490             }
491             if (kvm_arch_get_supported_cpuid(s, 0x80000001, 0, R_EDX) &
492                 CPUID_EXT2_RDTSCP) {
493                 value |= (uint64_t)VMX_SECONDARY_EXEC_RDTSCP << 32;
494             }
495         }
496         /* fall through */
497     case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
498     case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
499     case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
500     case MSR_IA32_VMX_TRUE_EXIT_CTLS:
501         /*
502          * Return true for bits that can be one, but do not have to be one.
503          * The SDM tells us which bits could have a "must be one" setting,
504          * so we can do the opposite transformation in make_vmx_msr_value.
505          */
506         must_be_one = (uint32_t)value;
507         can_be_one = (uint32_t)(value >> 32);
508         return can_be_one & ~must_be_one;
509 
510     default:
511         return value;
512     }
513 }
514 
515 static int kvm_get_mce_cap_supported(KVMState *s, uint64_t *mce_cap,
516                                      int *max_banks)
517 {
518     int r;
519 
520     r = kvm_check_extension(s, KVM_CAP_MCE);
521     if (r > 0) {
522         *max_banks = r;
523         return kvm_ioctl(s, KVM_X86_GET_MCE_CAP_SUPPORTED, mce_cap);
524     }
525     return -ENOSYS;
526 }
527 
528 static void kvm_mce_inject(X86CPU *cpu, hwaddr paddr, int code)
529 {
530     CPUState *cs = CPU(cpu);
531     CPUX86State *env = &cpu->env;
532     uint64_t status = MCI_STATUS_VAL | MCI_STATUS_UC | MCI_STATUS_EN |
533                       MCI_STATUS_MISCV | MCI_STATUS_ADDRV | MCI_STATUS_S;
534     uint64_t mcg_status = MCG_STATUS_MCIP;
535     int flags = 0;
536 
537     if (code == BUS_MCEERR_AR) {
538         status |= MCI_STATUS_AR | 0x134;
539         mcg_status |= MCG_STATUS_EIPV;
540     } else {
541         status |= 0xc0;
542         mcg_status |= MCG_STATUS_RIPV;
543     }
544 
545     flags = cpu_x86_support_mca_broadcast(env) ? MCE_INJECT_BROADCAST : 0;
546     /* We need to read back the value of MSR_EXT_MCG_CTL that was set by the
547      * guest kernel back into env->mcg_ext_ctl.
548      */
549     cpu_synchronize_state(cs);
550     if (env->mcg_ext_ctl & MCG_EXT_CTL_LMCE_EN) {
551         mcg_status |= MCG_STATUS_LMCE;
552         flags = 0;
553     }
554 
555     cpu_x86_inject_mce(NULL, cpu, 9, status, mcg_status, paddr,
556                        (MCM_ADDR_PHYS << 6) | 0xc, flags);
557 }
558 
559 static void emit_hypervisor_memory_failure(MemoryFailureAction action, bool ar)
560 {
561     MemoryFailureFlags mff = {.action_required = ar, .recursive = false};
562 
563     qapi_event_send_memory_failure(MEMORY_FAILURE_RECIPIENT_HYPERVISOR, action,
564                                    &mff);
565 }
566 
567 static void hardware_memory_error(void *host_addr)
568 {
569     emit_hypervisor_memory_failure(MEMORY_FAILURE_ACTION_FATAL, true);
570     error_report("QEMU got Hardware memory error at addr %p", host_addr);
571     exit(1);
572 }
573 
574 void kvm_arch_on_sigbus_vcpu(CPUState *c, int code, void *addr)
575 {
576     X86CPU *cpu = X86_CPU(c);
577     CPUX86State *env = &cpu->env;
578     ram_addr_t ram_addr;
579     hwaddr paddr;
580 
581     /* If we get an action required MCE, it has been injected by KVM
582      * while the VM was running.  An action optional MCE instead should
583      * be coming from the main thread, which qemu_init_sigbus identifies
584      * as the "early kill" thread.
585      */
586     assert(code == BUS_MCEERR_AR || code == BUS_MCEERR_AO);
587 
588     if ((env->mcg_cap & MCG_SER_P) && addr) {
589         ram_addr = qemu_ram_addr_from_host(addr);
590         if (ram_addr != RAM_ADDR_INVALID &&
591             kvm_physical_memory_addr_from_host(c->kvm_state, addr, &paddr)) {
592             kvm_hwpoison_page_add(ram_addr);
593             kvm_mce_inject(cpu, paddr, code);
594 
595             /*
596              * Use different logging severity based on error type.
597              * If there is additional MCE reporting on the hypervisor, QEMU VA
598              * could be another source to identify the PA and MCE details.
599              */
600             if (code == BUS_MCEERR_AR) {
601                 error_report("Guest MCE Memory Error at QEMU addr %p and "
602                     "GUEST addr 0x%" HWADDR_PRIx " of type %s injected",
603                     addr, paddr, "BUS_MCEERR_AR");
604             } else {
605                  warn_report("Guest MCE Memory Error at QEMU addr %p and "
606                      "GUEST addr 0x%" HWADDR_PRIx " of type %s injected",
607                      addr, paddr, "BUS_MCEERR_AO");
608             }
609 
610             return;
611         }
612 
613         if (code == BUS_MCEERR_AO) {
614             warn_report("Hardware memory error at addr %p of type %s "
615                 "for memory used by QEMU itself instead of guest system!",
616                  addr, "BUS_MCEERR_AO");
617         }
618     }
619 
620     if (code == BUS_MCEERR_AR) {
621         hardware_memory_error(addr);
622     }
623 
624     /* Hope we are lucky for AO MCE, just notify a event */
625     emit_hypervisor_memory_failure(MEMORY_FAILURE_ACTION_IGNORE, false);
626 }
627 
628 static void kvm_reset_exception(CPUX86State *env)
629 {
630     env->exception_nr = -1;
631     env->exception_pending = 0;
632     env->exception_injected = 0;
633     env->exception_has_payload = false;
634     env->exception_payload = 0;
635 }
636 
637 static void kvm_queue_exception(CPUX86State *env,
638                                 int32_t exception_nr,
639                                 uint8_t exception_has_payload,
640                                 uint64_t exception_payload)
641 {
642     assert(env->exception_nr == -1);
643     assert(!env->exception_pending);
644     assert(!env->exception_injected);
645     assert(!env->exception_has_payload);
646 
647     env->exception_nr = exception_nr;
648 
649     if (has_exception_payload) {
650         env->exception_pending = 1;
651 
652         env->exception_has_payload = exception_has_payload;
653         env->exception_payload = exception_payload;
654     } else {
655         env->exception_injected = 1;
656 
657         if (exception_nr == EXCP01_DB) {
658             assert(exception_has_payload);
659             env->dr[6] = exception_payload;
660         } else if (exception_nr == EXCP0E_PAGE) {
661             assert(exception_has_payload);
662             env->cr[2] = exception_payload;
663         } else {
664             assert(!exception_has_payload);
665         }
666     }
667 }
668 
669 static int kvm_inject_mce_oldstyle(X86CPU *cpu)
670 {
671     CPUX86State *env = &cpu->env;
672 
673     if (!kvm_has_vcpu_events() && env->exception_nr == EXCP12_MCHK) {
674         unsigned int bank, bank_num = env->mcg_cap & 0xff;
675         struct kvm_x86_mce mce;
676 
677         kvm_reset_exception(env);
678 
679         /*
680          * There must be at least one bank in use if an MCE is pending.
681          * Find it and use its values for the event injection.
682          */
683         for (bank = 0; bank < bank_num; bank++) {
684             if (env->mce_banks[bank * 4 + 1] & MCI_STATUS_VAL) {
685                 break;
686             }
687         }
688         assert(bank < bank_num);
689 
690         mce.bank = bank;
691         mce.status = env->mce_banks[bank * 4 + 1];
692         mce.mcg_status = env->mcg_status;
693         mce.addr = env->mce_banks[bank * 4 + 2];
694         mce.misc = env->mce_banks[bank * 4 + 3];
695 
696         return kvm_vcpu_ioctl(CPU(cpu), KVM_X86_SET_MCE, &mce);
697     }
698     return 0;
699 }
700 
701 static void cpu_update_state(void *opaque, bool running, RunState state)
702 {
703     CPUX86State *env = opaque;
704 
705     if (running) {
706         env->tsc_valid = false;
707     }
708 }
709 
710 unsigned long kvm_arch_vcpu_id(CPUState *cs)
711 {
712     X86CPU *cpu = X86_CPU(cs);
713     return cpu->apic_id;
714 }
715 
716 #ifndef KVM_CPUID_SIGNATURE_NEXT
717 #define KVM_CPUID_SIGNATURE_NEXT                0x40000100
718 #endif
719 
720 static bool hyperv_enabled(X86CPU *cpu)
721 {
722     return kvm_check_extension(kvm_state, KVM_CAP_HYPERV) > 0 &&
723         ((cpu->hyperv_spinlock_attempts != HYPERV_SPINLOCK_NEVER_NOTIFY) ||
724          cpu->hyperv_features || cpu->hyperv_passthrough);
725 }
726 
727 /*
728  * Check whether target_freq is within conservative
729  * ntp correctable bounds (250ppm) of freq
730  */
731 static inline bool freq_within_bounds(int freq, int target_freq)
732 {
733         int max_freq = freq + (freq * 250 / 1000000);
734         int min_freq = freq - (freq * 250 / 1000000);
735 
736         if (target_freq >= min_freq && target_freq <= max_freq) {
737                 return true;
738         }
739 
740         return false;
741 }
742 
743 static int kvm_arch_set_tsc_khz(CPUState *cs)
744 {
745     X86CPU *cpu = X86_CPU(cs);
746     CPUX86State *env = &cpu->env;
747     int r, cur_freq;
748     bool set_ioctl = false;
749 
750     if (!env->tsc_khz) {
751         return 0;
752     }
753 
754     cur_freq = kvm_check_extension(cs->kvm_state, KVM_CAP_GET_TSC_KHZ) ?
755                kvm_vcpu_ioctl(cs, KVM_GET_TSC_KHZ) : -ENOTSUP;
756 
757     /*
758      * If TSC scaling is supported, attempt to set TSC frequency.
759      */
760     if (kvm_check_extension(cs->kvm_state, KVM_CAP_TSC_CONTROL)) {
761         set_ioctl = true;
762     }
763 
764     /*
765      * If desired TSC frequency is within bounds of NTP correction,
766      * attempt to set TSC frequency.
767      */
768     if (cur_freq != -ENOTSUP && freq_within_bounds(cur_freq, env->tsc_khz)) {
769         set_ioctl = true;
770     }
771 
772     r = set_ioctl ?
773         kvm_vcpu_ioctl(cs, KVM_SET_TSC_KHZ, env->tsc_khz) :
774         -ENOTSUP;
775 
776     if (r < 0) {
777         /* When KVM_SET_TSC_KHZ fails, it's an error only if the current
778          * TSC frequency doesn't match the one we want.
779          */
780         cur_freq = kvm_check_extension(cs->kvm_state, KVM_CAP_GET_TSC_KHZ) ?
781                    kvm_vcpu_ioctl(cs, KVM_GET_TSC_KHZ) :
782                    -ENOTSUP;
783         if (cur_freq <= 0 || cur_freq != env->tsc_khz) {
784             warn_report("TSC frequency mismatch between "
785                         "VM (%" PRId64 " kHz) and host (%d kHz), "
786                         "and TSC scaling unavailable",
787                         env->tsc_khz, cur_freq);
788             return r;
789         }
790     }
791 
792     return 0;
793 }
794 
795 static bool tsc_is_stable_and_known(CPUX86State *env)
796 {
797     if (!env->tsc_khz) {
798         return false;
799     }
800     return (env->features[FEAT_8000_0007_EDX] & CPUID_APM_INVTSC)
801         || env->user_tsc_khz;
802 }
803 
804 static struct {
805     const char *desc;
806     struct {
807         uint32_t func;
808         int reg;
809         uint32_t bits;
810     } flags[2];
811     uint64_t dependencies;
812 } kvm_hyperv_properties[] = {
813     [HYPERV_FEAT_RELAXED] = {
814         .desc = "relaxed timing (hv-relaxed)",
815         .flags = {
816             {.func = HV_CPUID_ENLIGHTMENT_INFO, .reg = R_EAX,
817              .bits = HV_RELAXED_TIMING_RECOMMENDED}
818         }
819     },
820     [HYPERV_FEAT_VAPIC] = {
821         .desc = "virtual APIC (hv-vapic)",
822         .flags = {
823             {.func = HV_CPUID_FEATURES, .reg = R_EAX,
824              .bits = HV_APIC_ACCESS_AVAILABLE}
825         }
826     },
827     [HYPERV_FEAT_TIME] = {
828         .desc = "clocksources (hv-time)",
829         .flags = {
830             {.func = HV_CPUID_FEATURES, .reg = R_EAX,
831              .bits = HV_TIME_REF_COUNT_AVAILABLE | HV_REFERENCE_TSC_AVAILABLE}
832         }
833     },
834     [HYPERV_FEAT_CRASH] = {
835         .desc = "crash MSRs (hv-crash)",
836         .flags = {
837             {.func = HV_CPUID_FEATURES, .reg = R_EDX,
838              .bits = HV_GUEST_CRASH_MSR_AVAILABLE}
839         }
840     },
841     [HYPERV_FEAT_RESET] = {
842         .desc = "reset MSR (hv-reset)",
843         .flags = {
844             {.func = HV_CPUID_FEATURES, .reg = R_EAX,
845              .bits = HV_RESET_AVAILABLE}
846         }
847     },
848     [HYPERV_FEAT_VPINDEX] = {
849         .desc = "VP_INDEX MSR (hv-vpindex)",
850         .flags = {
851             {.func = HV_CPUID_FEATURES, .reg = R_EAX,
852              .bits = HV_VP_INDEX_AVAILABLE}
853         }
854     },
855     [HYPERV_FEAT_RUNTIME] = {
856         .desc = "VP_RUNTIME MSR (hv-runtime)",
857         .flags = {
858             {.func = HV_CPUID_FEATURES, .reg = R_EAX,
859              .bits = HV_VP_RUNTIME_AVAILABLE}
860         }
861     },
862     [HYPERV_FEAT_SYNIC] = {
863         .desc = "synthetic interrupt controller (hv-synic)",
864         .flags = {
865             {.func = HV_CPUID_FEATURES, .reg = R_EAX,
866              .bits = HV_SYNIC_AVAILABLE}
867         }
868     },
869     [HYPERV_FEAT_STIMER] = {
870         .desc = "synthetic timers (hv-stimer)",
871         .flags = {
872             {.func = HV_CPUID_FEATURES, .reg = R_EAX,
873              .bits = HV_SYNTIMERS_AVAILABLE}
874         },
875         .dependencies = BIT(HYPERV_FEAT_SYNIC) | BIT(HYPERV_FEAT_TIME)
876     },
877     [HYPERV_FEAT_FREQUENCIES] = {
878         .desc = "frequency MSRs (hv-frequencies)",
879         .flags = {
880             {.func = HV_CPUID_FEATURES, .reg = R_EAX,
881              .bits = HV_ACCESS_FREQUENCY_MSRS},
882             {.func = HV_CPUID_FEATURES, .reg = R_EDX,
883              .bits = HV_FREQUENCY_MSRS_AVAILABLE}
884         }
885     },
886     [HYPERV_FEAT_REENLIGHTENMENT] = {
887         .desc = "reenlightenment MSRs (hv-reenlightenment)",
888         .flags = {
889             {.func = HV_CPUID_FEATURES, .reg = R_EAX,
890              .bits = HV_ACCESS_REENLIGHTENMENTS_CONTROL}
891         }
892     },
893     [HYPERV_FEAT_TLBFLUSH] = {
894         .desc = "paravirtualized TLB flush (hv-tlbflush)",
895         .flags = {
896             {.func = HV_CPUID_ENLIGHTMENT_INFO, .reg = R_EAX,
897              .bits = HV_REMOTE_TLB_FLUSH_RECOMMENDED |
898              HV_EX_PROCESSOR_MASKS_RECOMMENDED}
899         },
900         .dependencies = BIT(HYPERV_FEAT_VPINDEX)
901     },
902     [HYPERV_FEAT_EVMCS] = {
903         .desc = "enlightened VMCS (hv-evmcs)",
904         .flags = {
905             {.func = HV_CPUID_ENLIGHTMENT_INFO, .reg = R_EAX,
906              .bits = HV_ENLIGHTENED_VMCS_RECOMMENDED}
907         },
908         .dependencies = BIT(HYPERV_FEAT_VAPIC)
909     },
910     [HYPERV_FEAT_IPI] = {
911         .desc = "paravirtualized IPI (hv-ipi)",
912         .flags = {
913             {.func = HV_CPUID_ENLIGHTMENT_INFO, .reg = R_EAX,
914              .bits = HV_CLUSTER_IPI_RECOMMENDED |
915              HV_EX_PROCESSOR_MASKS_RECOMMENDED}
916         },
917         .dependencies = BIT(HYPERV_FEAT_VPINDEX)
918     },
919     [HYPERV_FEAT_STIMER_DIRECT] = {
920         .desc = "direct mode synthetic timers (hv-stimer-direct)",
921         .flags = {
922             {.func = HV_CPUID_FEATURES, .reg = R_EDX,
923              .bits = HV_STIMER_DIRECT_MODE_AVAILABLE}
924         },
925         .dependencies = BIT(HYPERV_FEAT_STIMER)
926     },
927     [HYPERV_FEAT_AVIC] = {
928         .desc = "AVIC/APICv support (hv-avic/hv-apicv)",
929         .flags = {
930             {.func = HV_CPUID_ENLIGHTMENT_INFO, .reg = R_EAX,
931              .bits = HV_DEPRECATING_AEOI_RECOMMENDED}
932         }
933     },
934 };
935 
936 static struct kvm_cpuid2 *try_get_hv_cpuid(CPUState *cs, int max,
937                                            bool do_sys_ioctl)
938 {
939     struct kvm_cpuid2 *cpuid;
940     int r, size;
941 
942     size = sizeof(*cpuid) + max * sizeof(*cpuid->entries);
943     cpuid = g_malloc0(size);
944     cpuid->nent = max;
945 
946     if (do_sys_ioctl) {
947         r = kvm_ioctl(kvm_state, KVM_GET_SUPPORTED_HV_CPUID, cpuid);
948     } else {
949         r = kvm_vcpu_ioctl(cs, KVM_GET_SUPPORTED_HV_CPUID, cpuid);
950     }
951     if (r == 0 && cpuid->nent >= max) {
952         r = -E2BIG;
953     }
954     if (r < 0) {
955         if (r == -E2BIG) {
956             g_free(cpuid);
957             return NULL;
958         } else {
959             fprintf(stderr, "KVM_GET_SUPPORTED_HV_CPUID failed: %s\n",
960                     strerror(-r));
961             exit(1);
962         }
963     }
964     return cpuid;
965 }
966 
967 /*
968  * Run KVM_GET_SUPPORTED_HV_CPUID ioctl(), allocating a buffer large enough
969  * for all entries.
970  */
971 static struct kvm_cpuid2 *get_supported_hv_cpuid(CPUState *cs)
972 {
973     struct kvm_cpuid2 *cpuid;
974     /* 0x40000000..0x40000005, 0x4000000A, 0x40000080..0x40000080 leaves */
975     int max = 10;
976     int i;
977     bool do_sys_ioctl;
978 
979     do_sys_ioctl =
980         kvm_check_extension(kvm_state, KVM_CAP_SYS_HYPERV_CPUID) > 0;
981 
982     /*
983      * Non-empty KVM context is needed when KVM_CAP_SYS_HYPERV_CPUID is
984      * unsupported, kvm_hyperv_expand_features() checks for that.
985      */
986     assert(do_sys_ioctl || cs->kvm_state);
987 
988     /*
989      * When the buffer is too small, KVM_GET_SUPPORTED_HV_CPUID fails with
990      * -E2BIG, however, it doesn't report back the right size. Keep increasing
991      * it and re-trying until we succeed.
992      */
993     while ((cpuid = try_get_hv_cpuid(cs, max, do_sys_ioctl)) == NULL) {
994         max++;
995     }
996 
997     /*
998      * KVM_GET_SUPPORTED_HV_CPUID does not set EVMCS CPUID bit before
999      * KVM_CAP_HYPERV_ENLIGHTENED_VMCS is enabled but we want to get the
1000      * information early, just check for the capability and set the bit
1001      * manually.
1002      */
1003     if (!do_sys_ioctl && kvm_check_extension(cs->kvm_state,
1004                             KVM_CAP_HYPERV_ENLIGHTENED_VMCS) > 0) {
1005         for (i = 0; i < cpuid->nent; i++) {
1006             if (cpuid->entries[i].function == HV_CPUID_ENLIGHTMENT_INFO) {
1007                 cpuid->entries[i].eax |= HV_ENLIGHTENED_VMCS_RECOMMENDED;
1008             }
1009         }
1010     }
1011 
1012     return cpuid;
1013 }
1014 
1015 /*
1016  * When KVM_GET_SUPPORTED_HV_CPUID is not supported we fill CPUID feature
1017  * leaves from KVM_CAP_HYPERV* and present MSRs data.
1018  */
1019 static struct kvm_cpuid2 *get_supported_hv_cpuid_legacy(CPUState *cs)
1020 {
1021     X86CPU *cpu = X86_CPU(cs);
1022     struct kvm_cpuid2 *cpuid;
1023     struct kvm_cpuid_entry2 *entry_feat, *entry_recomm;
1024 
1025     /* HV_CPUID_FEATURES, HV_CPUID_ENLIGHTMENT_INFO */
1026     cpuid = g_malloc0(sizeof(*cpuid) + 2 * sizeof(*cpuid->entries));
1027     cpuid->nent = 2;
1028 
1029     /* HV_CPUID_VENDOR_AND_MAX_FUNCTIONS */
1030     entry_feat = &cpuid->entries[0];
1031     entry_feat->function = HV_CPUID_FEATURES;
1032 
1033     entry_recomm = &cpuid->entries[1];
1034     entry_recomm->function = HV_CPUID_ENLIGHTMENT_INFO;
1035     entry_recomm->ebx = cpu->hyperv_spinlock_attempts;
1036 
1037     if (kvm_check_extension(cs->kvm_state, KVM_CAP_HYPERV) > 0) {
1038         entry_feat->eax |= HV_HYPERCALL_AVAILABLE;
1039         entry_feat->eax |= HV_APIC_ACCESS_AVAILABLE;
1040         entry_feat->edx |= HV_CPU_DYNAMIC_PARTITIONING_AVAILABLE;
1041         entry_recomm->eax |= HV_RELAXED_TIMING_RECOMMENDED;
1042         entry_recomm->eax |= HV_APIC_ACCESS_RECOMMENDED;
1043     }
1044 
1045     if (kvm_check_extension(cs->kvm_state, KVM_CAP_HYPERV_TIME) > 0) {
1046         entry_feat->eax |= HV_TIME_REF_COUNT_AVAILABLE;
1047         entry_feat->eax |= HV_REFERENCE_TSC_AVAILABLE;
1048     }
1049 
1050     if (has_msr_hv_frequencies) {
1051         entry_feat->eax |= HV_ACCESS_FREQUENCY_MSRS;
1052         entry_feat->edx |= HV_FREQUENCY_MSRS_AVAILABLE;
1053     }
1054 
1055     if (has_msr_hv_crash) {
1056         entry_feat->edx |= HV_GUEST_CRASH_MSR_AVAILABLE;
1057     }
1058 
1059     if (has_msr_hv_reenlightenment) {
1060         entry_feat->eax |= HV_ACCESS_REENLIGHTENMENTS_CONTROL;
1061     }
1062 
1063     if (has_msr_hv_reset) {
1064         entry_feat->eax |= HV_RESET_AVAILABLE;
1065     }
1066 
1067     if (has_msr_hv_vpindex) {
1068         entry_feat->eax |= HV_VP_INDEX_AVAILABLE;
1069     }
1070 
1071     if (has_msr_hv_runtime) {
1072         entry_feat->eax |= HV_VP_RUNTIME_AVAILABLE;
1073     }
1074 
1075     if (has_msr_hv_synic) {
1076         unsigned int cap = cpu->hyperv_synic_kvm_only ?
1077             KVM_CAP_HYPERV_SYNIC : KVM_CAP_HYPERV_SYNIC2;
1078 
1079         if (kvm_check_extension(cs->kvm_state, cap) > 0) {
1080             entry_feat->eax |= HV_SYNIC_AVAILABLE;
1081         }
1082     }
1083 
1084     if (has_msr_hv_stimer) {
1085         entry_feat->eax |= HV_SYNTIMERS_AVAILABLE;
1086     }
1087 
1088     if (kvm_check_extension(cs->kvm_state,
1089                             KVM_CAP_HYPERV_TLBFLUSH) > 0) {
1090         entry_recomm->eax |= HV_REMOTE_TLB_FLUSH_RECOMMENDED;
1091         entry_recomm->eax |= HV_EX_PROCESSOR_MASKS_RECOMMENDED;
1092     }
1093 
1094     if (kvm_check_extension(cs->kvm_state,
1095                             KVM_CAP_HYPERV_ENLIGHTENED_VMCS) > 0) {
1096         entry_recomm->eax |= HV_ENLIGHTENED_VMCS_RECOMMENDED;
1097     }
1098 
1099     if (kvm_check_extension(cs->kvm_state,
1100                             KVM_CAP_HYPERV_SEND_IPI) > 0) {
1101         entry_recomm->eax |= HV_CLUSTER_IPI_RECOMMENDED;
1102         entry_recomm->eax |= HV_EX_PROCESSOR_MASKS_RECOMMENDED;
1103     }
1104 
1105     return cpuid;
1106 }
1107 
1108 static uint32_t hv_cpuid_get_host(CPUState *cs, uint32_t func, int reg)
1109 {
1110     struct kvm_cpuid_entry2 *entry;
1111     struct kvm_cpuid2 *cpuid;
1112 
1113     if (hv_cpuid_cache) {
1114         cpuid = hv_cpuid_cache;
1115     } else {
1116         if (kvm_check_extension(kvm_state, KVM_CAP_HYPERV_CPUID) > 0) {
1117             cpuid = get_supported_hv_cpuid(cs);
1118         } else {
1119             /*
1120              * 'cs->kvm_state' may be NULL when Hyper-V features are expanded
1121              * before KVM context is created but this is only done when
1122              * KVM_CAP_SYS_HYPERV_CPUID is supported and it implies
1123              * KVM_CAP_HYPERV_CPUID.
1124              */
1125             assert(cs->kvm_state);
1126 
1127             cpuid = get_supported_hv_cpuid_legacy(cs);
1128         }
1129         hv_cpuid_cache = cpuid;
1130     }
1131 
1132     if (!cpuid) {
1133         return 0;
1134     }
1135 
1136     entry = cpuid_find_entry(cpuid, func, 0);
1137     if (!entry) {
1138         return 0;
1139     }
1140 
1141     return cpuid_entry_get_reg(entry, reg);
1142 }
1143 
1144 static bool hyperv_feature_supported(CPUState *cs, int feature)
1145 {
1146     uint32_t func, bits;
1147     int i, reg;
1148 
1149     for (i = 0; i < ARRAY_SIZE(kvm_hyperv_properties[feature].flags); i++) {
1150 
1151         func = kvm_hyperv_properties[feature].flags[i].func;
1152         reg = kvm_hyperv_properties[feature].flags[i].reg;
1153         bits = kvm_hyperv_properties[feature].flags[i].bits;
1154 
1155         if (!func) {
1156             continue;
1157         }
1158 
1159         if ((hv_cpuid_get_host(cs, func, reg) & bits) != bits) {
1160             return false;
1161         }
1162     }
1163 
1164     return true;
1165 }
1166 
1167 /* Checks that all feature dependencies are enabled */
1168 static bool hv_feature_check_deps(X86CPU *cpu, int feature, Error **errp)
1169 {
1170     uint64_t deps;
1171     int dep_feat;
1172 
1173     deps = kvm_hyperv_properties[feature].dependencies;
1174     while (deps) {
1175         dep_feat = ctz64(deps);
1176         if (!(hyperv_feat_enabled(cpu, dep_feat))) {
1177             error_setg(errp, "Hyper-V %s requires Hyper-V %s",
1178                        kvm_hyperv_properties[feature].desc,
1179                        kvm_hyperv_properties[dep_feat].desc);
1180             return false;
1181         }
1182         deps &= ~(1ull << dep_feat);
1183     }
1184 
1185     return true;
1186 }
1187 
1188 static uint32_t hv_build_cpuid_leaf(CPUState *cs, uint32_t func, int reg)
1189 {
1190     X86CPU *cpu = X86_CPU(cs);
1191     uint32_t r = 0;
1192     int i, j;
1193 
1194     for (i = 0; i < ARRAY_SIZE(kvm_hyperv_properties); i++) {
1195         if (!hyperv_feat_enabled(cpu, i)) {
1196             continue;
1197         }
1198 
1199         for (j = 0; j < ARRAY_SIZE(kvm_hyperv_properties[i].flags); j++) {
1200             if (kvm_hyperv_properties[i].flags[j].func != func) {
1201                 continue;
1202             }
1203             if (kvm_hyperv_properties[i].flags[j].reg != reg) {
1204                 continue;
1205             }
1206 
1207             r |= kvm_hyperv_properties[i].flags[j].bits;
1208         }
1209     }
1210 
1211     return r;
1212 }
1213 
1214 /*
1215  * Expand Hyper-V CPU features. In partucular, check that all the requested
1216  * features are supported by the host and the sanity of the configuration
1217  * (that all the required dependencies are included). Also, this takes care
1218  * of 'hv_passthrough' mode and fills the environment with all supported
1219  * Hyper-V features.
1220  */
1221 bool kvm_hyperv_expand_features(X86CPU *cpu, Error **errp)
1222 {
1223     CPUState *cs = CPU(cpu);
1224     Error *local_err = NULL;
1225     int feat;
1226 
1227     if (!hyperv_enabled(cpu))
1228         return true;
1229 
1230     /*
1231      * When kvm_hyperv_expand_features is called at CPU feature expansion
1232      * time per-CPU kvm_state is not available yet so we can only proceed
1233      * when KVM_CAP_SYS_HYPERV_CPUID is supported.
1234      */
1235     if (!cs->kvm_state &&
1236         !kvm_check_extension(kvm_state, KVM_CAP_SYS_HYPERV_CPUID))
1237         return true;
1238 
1239     if (cpu->hyperv_passthrough) {
1240         cpu->hyperv_vendor_id[0] =
1241             hv_cpuid_get_host(cs, HV_CPUID_VENDOR_AND_MAX_FUNCTIONS, R_EBX);
1242         cpu->hyperv_vendor_id[1] =
1243             hv_cpuid_get_host(cs, HV_CPUID_VENDOR_AND_MAX_FUNCTIONS, R_ECX);
1244         cpu->hyperv_vendor_id[2] =
1245             hv_cpuid_get_host(cs, HV_CPUID_VENDOR_AND_MAX_FUNCTIONS, R_EDX);
1246         cpu->hyperv_vendor = g_realloc(cpu->hyperv_vendor,
1247                                        sizeof(cpu->hyperv_vendor_id) + 1);
1248         memcpy(cpu->hyperv_vendor, cpu->hyperv_vendor_id,
1249                sizeof(cpu->hyperv_vendor_id));
1250         cpu->hyperv_vendor[sizeof(cpu->hyperv_vendor_id)] = 0;
1251 
1252         cpu->hyperv_interface_id[0] =
1253             hv_cpuid_get_host(cs, HV_CPUID_INTERFACE, R_EAX);
1254         cpu->hyperv_interface_id[1] =
1255             hv_cpuid_get_host(cs, HV_CPUID_INTERFACE, R_EBX);
1256         cpu->hyperv_interface_id[2] =
1257             hv_cpuid_get_host(cs, HV_CPUID_INTERFACE, R_ECX);
1258         cpu->hyperv_interface_id[3] =
1259             hv_cpuid_get_host(cs, HV_CPUID_INTERFACE, R_EDX);
1260 
1261         cpu->hyperv_ver_id_build =
1262             hv_cpuid_get_host(cs, HV_CPUID_VERSION, R_EAX);
1263         cpu->hyperv_ver_id_major =
1264             hv_cpuid_get_host(cs, HV_CPUID_VERSION, R_EBX) >> 16;
1265         cpu->hyperv_ver_id_minor =
1266             hv_cpuid_get_host(cs, HV_CPUID_VERSION, R_EBX) & 0xffff;
1267         cpu->hyperv_ver_id_sp =
1268             hv_cpuid_get_host(cs, HV_CPUID_VERSION, R_ECX);
1269         cpu->hyperv_ver_id_sb =
1270             hv_cpuid_get_host(cs, HV_CPUID_VERSION, R_EDX) >> 24;
1271         cpu->hyperv_ver_id_sn =
1272             hv_cpuid_get_host(cs, HV_CPUID_VERSION, R_EDX) & 0xffffff;
1273 
1274         cpu->hv_max_vps = hv_cpuid_get_host(cs, HV_CPUID_IMPLEMENT_LIMITS,
1275                                             R_EAX);
1276         cpu->hyperv_limits[0] =
1277             hv_cpuid_get_host(cs, HV_CPUID_IMPLEMENT_LIMITS, R_EBX);
1278         cpu->hyperv_limits[1] =
1279             hv_cpuid_get_host(cs, HV_CPUID_IMPLEMENT_LIMITS, R_ECX);
1280         cpu->hyperv_limits[2] =
1281             hv_cpuid_get_host(cs, HV_CPUID_IMPLEMENT_LIMITS, R_EDX);
1282 
1283         cpu->hyperv_spinlock_attempts =
1284             hv_cpuid_get_host(cs, HV_CPUID_ENLIGHTMENT_INFO, R_EBX);
1285 
1286         /*
1287          * Mark feature as enabled in 'cpu->hyperv_features' as
1288          * hv_build_cpuid_leaf() uses this info to build guest CPUIDs.
1289          */
1290         for (feat = 0; feat < ARRAY_SIZE(kvm_hyperv_properties); feat++) {
1291             if (hyperv_feature_supported(cs, feat)) {
1292                 cpu->hyperv_features |= BIT(feat);
1293             }
1294         }
1295     } else {
1296         /* Check features availability and dependencies */
1297         for (feat = 0; feat < ARRAY_SIZE(kvm_hyperv_properties); feat++) {
1298             /* If the feature was not requested skip it. */
1299             if (!hyperv_feat_enabled(cpu, feat)) {
1300                 continue;
1301             }
1302 
1303             /* Check if the feature is supported by KVM */
1304             if (!hyperv_feature_supported(cs, feat)) {
1305                 error_setg(errp, "Hyper-V %s is not supported by kernel",
1306                            kvm_hyperv_properties[feat].desc);
1307                 return false;
1308             }
1309 
1310             /* Check dependencies */
1311             if (!hv_feature_check_deps(cpu, feat, &local_err)) {
1312                 error_propagate(errp, local_err);
1313                 return false;
1314             }
1315         }
1316     }
1317 
1318     /* Additional dependencies not covered by kvm_hyperv_properties[] */
1319     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_SYNIC) &&
1320         !cpu->hyperv_synic_kvm_only &&
1321         !hyperv_feat_enabled(cpu, HYPERV_FEAT_VPINDEX)) {
1322         error_setg(errp, "Hyper-V %s requires Hyper-V %s",
1323                    kvm_hyperv_properties[HYPERV_FEAT_SYNIC].desc,
1324                    kvm_hyperv_properties[HYPERV_FEAT_VPINDEX].desc);
1325         return false;
1326     }
1327 
1328     return true;
1329 }
1330 
1331 /*
1332  * Fill in Hyper-V CPUIDs. Returns the number of entries filled in cpuid_ent.
1333  */
1334 static int hyperv_fill_cpuids(CPUState *cs,
1335                               struct kvm_cpuid_entry2 *cpuid_ent)
1336 {
1337     X86CPU *cpu = X86_CPU(cs);
1338     struct kvm_cpuid_entry2 *c;
1339     uint32_t cpuid_i = 0;
1340 
1341     c = &cpuid_ent[cpuid_i++];
1342     c->function = HV_CPUID_VENDOR_AND_MAX_FUNCTIONS;
1343     c->eax = hyperv_feat_enabled(cpu, HYPERV_FEAT_EVMCS) ?
1344         HV_CPUID_NESTED_FEATURES : HV_CPUID_IMPLEMENT_LIMITS;
1345     c->ebx = cpu->hyperv_vendor_id[0];
1346     c->ecx = cpu->hyperv_vendor_id[1];
1347     c->edx = cpu->hyperv_vendor_id[2];
1348 
1349     c = &cpuid_ent[cpuid_i++];
1350     c->function = HV_CPUID_INTERFACE;
1351     c->eax = cpu->hyperv_interface_id[0];
1352     c->ebx = cpu->hyperv_interface_id[1];
1353     c->ecx = cpu->hyperv_interface_id[2];
1354     c->edx = cpu->hyperv_interface_id[3];
1355 
1356     c = &cpuid_ent[cpuid_i++];
1357     c->function = HV_CPUID_VERSION;
1358     c->eax = cpu->hyperv_ver_id_build;
1359     c->ebx = (uint32_t)cpu->hyperv_ver_id_major << 16 |
1360         cpu->hyperv_ver_id_minor;
1361     c->ecx = cpu->hyperv_ver_id_sp;
1362     c->edx = (uint32_t)cpu->hyperv_ver_id_sb << 24 |
1363         (cpu->hyperv_ver_id_sn & 0xffffff);
1364 
1365     c = &cpuid_ent[cpuid_i++];
1366     c->function = HV_CPUID_FEATURES;
1367     c->eax = hv_build_cpuid_leaf(cs, HV_CPUID_FEATURES, R_EAX);
1368     c->ebx = hv_build_cpuid_leaf(cs, HV_CPUID_FEATURES, R_EBX);
1369     c->edx = hv_build_cpuid_leaf(cs, HV_CPUID_FEATURES, R_EDX);
1370 
1371     /* Unconditionally required with any Hyper-V enlightenment */
1372     c->eax |= HV_HYPERCALL_AVAILABLE;
1373 
1374     /* SynIC and Vmbus devices require messages/signals hypercalls */
1375     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_SYNIC) &&
1376         !cpu->hyperv_synic_kvm_only) {
1377         c->ebx |= HV_POST_MESSAGES | HV_SIGNAL_EVENTS;
1378     }
1379 
1380 
1381     /* Not exposed by KVM but needed to make CPU hotplug in Windows work */
1382     c->edx |= HV_CPU_DYNAMIC_PARTITIONING_AVAILABLE;
1383 
1384     c = &cpuid_ent[cpuid_i++];
1385     c->function = HV_CPUID_ENLIGHTMENT_INFO;
1386     c->eax = hv_build_cpuid_leaf(cs, HV_CPUID_ENLIGHTMENT_INFO, R_EAX);
1387     c->ebx = cpu->hyperv_spinlock_attempts;
1388 
1389     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_VAPIC) &&
1390         !hyperv_feat_enabled(cpu, HYPERV_FEAT_AVIC)) {
1391         c->eax |= HV_APIC_ACCESS_RECOMMENDED;
1392     }
1393 
1394     if (cpu->hyperv_no_nonarch_cs == ON_OFF_AUTO_ON) {
1395         c->eax |= HV_NO_NONARCH_CORESHARING;
1396     } else if (cpu->hyperv_no_nonarch_cs == ON_OFF_AUTO_AUTO) {
1397         c->eax |= hv_cpuid_get_host(cs, HV_CPUID_ENLIGHTMENT_INFO, R_EAX) &
1398             HV_NO_NONARCH_CORESHARING;
1399     }
1400 
1401     c = &cpuid_ent[cpuid_i++];
1402     c->function = HV_CPUID_IMPLEMENT_LIMITS;
1403     c->eax = cpu->hv_max_vps;
1404     c->ebx = cpu->hyperv_limits[0];
1405     c->ecx = cpu->hyperv_limits[1];
1406     c->edx = cpu->hyperv_limits[2];
1407 
1408     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_EVMCS)) {
1409         __u32 function;
1410 
1411         /* Create zeroed 0x40000006..0x40000009 leaves */
1412         for (function = HV_CPUID_IMPLEMENT_LIMITS + 1;
1413              function < HV_CPUID_NESTED_FEATURES; function++) {
1414             c = &cpuid_ent[cpuid_i++];
1415             c->function = function;
1416         }
1417 
1418         c = &cpuid_ent[cpuid_i++];
1419         c->function = HV_CPUID_NESTED_FEATURES;
1420         c->eax = cpu->hyperv_nested[0];
1421     }
1422 
1423     return cpuid_i;
1424 }
1425 
1426 static Error *hv_passthrough_mig_blocker;
1427 static Error *hv_no_nonarch_cs_mig_blocker;
1428 
1429 /* Checks that the exposed eVMCS version range is supported by KVM */
1430 static bool evmcs_version_supported(uint16_t evmcs_version,
1431                                     uint16_t supported_evmcs_version)
1432 {
1433     uint8_t min_version = evmcs_version & 0xff;
1434     uint8_t max_version = evmcs_version >> 8;
1435     uint8_t min_supported_version = supported_evmcs_version & 0xff;
1436     uint8_t max_supported_version = supported_evmcs_version >> 8;
1437 
1438     return (min_version >= min_supported_version) &&
1439         (max_version <= max_supported_version);
1440 }
1441 
1442 #define DEFAULT_EVMCS_VERSION ((1 << 8) | 1)
1443 
1444 static int hyperv_init_vcpu(X86CPU *cpu)
1445 {
1446     CPUState *cs = CPU(cpu);
1447     Error *local_err = NULL;
1448     int ret;
1449 
1450     if (cpu->hyperv_passthrough && hv_passthrough_mig_blocker == NULL) {
1451         error_setg(&hv_passthrough_mig_blocker,
1452                    "'hv-passthrough' CPU flag prevents migration, use explicit"
1453                    " set of hv-* flags instead");
1454         ret = migrate_add_blocker(hv_passthrough_mig_blocker, &local_err);
1455         if (ret < 0) {
1456             error_report_err(local_err);
1457             return ret;
1458         }
1459     }
1460 
1461     if (cpu->hyperv_no_nonarch_cs == ON_OFF_AUTO_AUTO &&
1462         hv_no_nonarch_cs_mig_blocker == NULL) {
1463         error_setg(&hv_no_nonarch_cs_mig_blocker,
1464                    "'hv-no-nonarch-coresharing=auto' CPU flag prevents migration"
1465                    " use explicit 'hv-no-nonarch-coresharing=on' instead (but"
1466                    " make sure SMT is disabled and/or that vCPUs are properly"
1467                    " pinned)");
1468         ret = migrate_add_blocker(hv_no_nonarch_cs_mig_blocker, &local_err);
1469         if (ret < 0) {
1470             error_report_err(local_err);
1471             return ret;
1472         }
1473     }
1474 
1475     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_VPINDEX) && !hv_vpindex_settable) {
1476         /*
1477          * the kernel doesn't support setting vp_index; assert that its value
1478          * is in sync
1479          */
1480         struct {
1481             struct kvm_msrs info;
1482             struct kvm_msr_entry entries[1];
1483         } msr_data = {
1484             .info.nmsrs = 1,
1485             .entries[0].index = HV_X64_MSR_VP_INDEX,
1486         };
1487 
1488         ret = kvm_vcpu_ioctl(cs, KVM_GET_MSRS, &msr_data);
1489         if (ret < 0) {
1490             return ret;
1491         }
1492         assert(ret == 1);
1493 
1494         if (msr_data.entries[0].data != hyperv_vp_index(CPU(cpu))) {
1495             error_report("kernel's vp_index != QEMU's vp_index");
1496             return -ENXIO;
1497         }
1498     }
1499 
1500     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_SYNIC)) {
1501         uint32_t synic_cap = cpu->hyperv_synic_kvm_only ?
1502             KVM_CAP_HYPERV_SYNIC : KVM_CAP_HYPERV_SYNIC2;
1503         ret = kvm_vcpu_enable_cap(cs, synic_cap, 0);
1504         if (ret < 0) {
1505             error_report("failed to turn on HyperV SynIC in KVM: %s",
1506                          strerror(-ret));
1507             return ret;
1508         }
1509 
1510         if (!cpu->hyperv_synic_kvm_only) {
1511             ret = hyperv_x86_synic_add(cpu);
1512             if (ret < 0) {
1513                 error_report("failed to create HyperV SynIC: %s",
1514                              strerror(-ret));
1515                 return ret;
1516             }
1517         }
1518     }
1519 
1520     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_EVMCS)) {
1521         uint16_t evmcs_version = DEFAULT_EVMCS_VERSION;
1522         uint16_t supported_evmcs_version;
1523 
1524         ret = kvm_vcpu_enable_cap(cs, KVM_CAP_HYPERV_ENLIGHTENED_VMCS, 0,
1525                                   (uintptr_t)&supported_evmcs_version);
1526 
1527         /*
1528          * KVM is required to support EVMCS ver.1. as that's what 'hv-evmcs'
1529          * option sets. Note: we hardcode the maximum supported eVMCS version
1530          * to '1' as well so 'hv-evmcs' feature is migratable even when (and if)
1531          * ver.2 is implemented. A new option (e.g. 'hv-evmcs=2') will then have
1532          * to be added.
1533          */
1534         if (ret < 0) {
1535             error_report("Hyper-V %s is not supported by kernel",
1536                          kvm_hyperv_properties[HYPERV_FEAT_EVMCS].desc);
1537             return ret;
1538         }
1539 
1540         if (!evmcs_version_supported(evmcs_version, supported_evmcs_version)) {
1541             error_report("eVMCS version range [%d..%d] is not supported by "
1542                          "kernel (supported: [%d..%d])", evmcs_version & 0xff,
1543                          evmcs_version >> 8, supported_evmcs_version & 0xff,
1544                          supported_evmcs_version >> 8);
1545             return -ENOTSUP;
1546         }
1547 
1548         cpu->hyperv_nested[0] = evmcs_version;
1549     }
1550 
1551     if (cpu->hyperv_enforce_cpuid) {
1552         ret = kvm_vcpu_enable_cap(cs, KVM_CAP_HYPERV_ENFORCE_CPUID, 0, 1);
1553         if (ret < 0) {
1554             error_report("failed to enable KVM_CAP_HYPERV_ENFORCE_CPUID: %s",
1555                          strerror(-ret));
1556             return ret;
1557         }
1558     }
1559 
1560     return 0;
1561 }
1562 
1563 static Error *invtsc_mig_blocker;
1564 
1565 #define KVM_MAX_CPUID_ENTRIES  100
1566 
1567 int kvm_arch_init_vcpu(CPUState *cs)
1568 {
1569     struct {
1570         struct kvm_cpuid2 cpuid;
1571         struct kvm_cpuid_entry2 entries[KVM_MAX_CPUID_ENTRIES];
1572     } cpuid_data;
1573     /*
1574      * The kernel defines these structs with padding fields so there
1575      * should be no extra padding in our cpuid_data struct.
1576      */
1577     QEMU_BUILD_BUG_ON(sizeof(cpuid_data) !=
1578                       sizeof(struct kvm_cpuid2) +
1579                       sizeof(struct kvm_cpuid_entry2) * KVM_MAX_CPUID_ENTRIES);
1580 
1581     X86CPU *cpu = X86_CPU(cs);
1582     CPUX86State *env = &cpu->env;
1583     uint32_t limit, i, j, cpuid_i;
1584     uint32_t unused;
1585     struct kvm_cpuid_entry2 *c;
1586     uint32_t signature[3];
1587     int kvm_base = KVM_CPUID_SIGNATURE;
1588     int max_nested_state_len;
1589     int r;
1590     Error *local_err = NULL;
1591 
1592     memset(&cpuid_data, 0, sizeof(cpuid_data));
1593 
1594     cpuid_i = 0;
1595 
1596     r = kvm_arch_set_tsc_khz(cs);
1597     if (r < 0) {
1598         return r;
1599     }
1600 
1601     /* vcpu's TSC frequency is either specified by user, or following
1602      * the value used by KVM if the former is not present. In the
1603      * latter case, we query it from KVM and record in env->tsc_khz,
1604      * so that vcpu's TSC frequency can be migrated later via this field.
1605      */
1606     if (!env->tsc_khz) {
1607         r = kvm_check_extension(cs->kvm_state, KVM_CAP_GET_TSC_KHZ) ?
1608             kvm_vcpu_ioctl(cs, KVM_GET_TSC_KHZ) :
1609             -ENOTSUP;
1610         if (r > 0) {
1611             env->tsc_khz = r;
1612         }
1613     }
1614 
1615     env->apic_bus_freq = KVM_APIC_BUS_FREQUENCY;
1616 
1617     /*
1618      * kvm_hyperv_expand_features() is called here for the second time in case
1619      * KVM_CAP_SYS_HYPERV_CPUID is not supported. While we can't possibly handle
1620      * 'query-cpu-model-expansion' in this case as we don't have a KVM vCPU to
1621      * check which Hyper-V enlightenments are supported and which are not, we
1622      * can still proceed and check/expand Hyper-V enlightenments here so legacy
1623      * behavior is preserved.
1624      */
1625     if (!kvm_hyperv_expand_features(cpu, &local_err)) {
1626         error_report_err(local_err);
1627         return -ENOSYS;
1628     }
1629 
1630     if (hyperv_enabled(cpu)) {
1631         r = hyperv_init_vcpu(cpu);
1632         if (r) {
1633             return r;
1634         }
1635 
1636         cpuid_i = hyperv_fill_cpuids(cs, cpuid_data.entries);
1637         kvm_base = KVM_CPUID_SIGNATURE_NEXT;
1638         has_msr_hv_hypercall = true;
1639     }
1640 
1641     if (cpu->expose_kvm) {
1642         memcpy(signature, "KVMKVMKVM\0\0\0", 12);
1643         c = &cpuid_data.entries[cpuid_i++];
1644         c->function = KVM_CPUID_SIGNATURE | kvm_base;
1645         c->eax = KVM_CPUID_FEATURES | kvm_base;
1646         c->ebx = signature[0];
1647         c->ecx = signature[1];
1648         c->edx = signature[2];
1649 
1650         c = &cpuid_data.entries[cpuid_i++];
1651         c->function = KVM_CPUID_FEATURES | kvm_base;
1652         c->eax = env->features[FEAT_KVM];
1653         c->edx = env->features[FEAT_KVM_HINTS];
1654     }
1655 
1656     cpu_x86_cpuid(env, 0, 0, &limit, &unused, &unused, &unused);
1657 
1658     if (cpu->kvm_pv_enforce_cpuid) {
1659         r = kvm_vcpu_enable_cap(cs, KVM_CAP_ENFORCE_PV_FEATURE_CPUID, 0, 1);
1660         if (r < 0) {
1661             fprintf(stderr,
1662                     "failed to enable KVM_CAP_ENFORCE_PV_FEATURE_CPUID: %s",
1663                     strerror(-r));
1664             abort();
1665         }
1666     }
1667 
1668     for (i = 0; i <= limit; i++) {
1669         if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1670             fprintf(stderr, "unsupported level value: 0x%x\n", limit);
1671             abort();
1672         }
1673         c = &cpuid_data.entries[cpuid_i++];
1674 
1675         switch (i) {
1676         case 2: {
1677             /* Keep reading function 2 till all the input is received */
1678             int times;
1679 
1680             c->function = i;
1681             c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC |
1682                        KVM_CPUID_FLAG_STATE_READ_NEXT;
1683             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1684             times = c->eax & 0xff;
1685 
1686             for (j = 1; j < times; ++j) {
1687                 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1688                     fprintf(stderr, "cpuid_data is full, no space for "
1689                             "cpuid(eax:2):eax & 0xf = 0x%x\n", times);
1690                     abort();
1691                 }
1692                 c = &cpuid_data.entries[cpuid_i++];
1693                 c->function = i;
1694                 c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC;
1695                 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1696             }
1697             break;
1698         }
1699         case 0x1f:
1700             if (env->nr_dies < 2) {
1701                 break;
1702             }
1703             /* fallthrough */
1704         case 4:
1705         case 0xb:
1706         case 0xd:
1707             for (j = 0; ; j++) {
1708                 if (i == 0xd && j == 64) {
1709                     break;
1710                 }
1711 
1712                 if (i == 0x1f && j == 64) {
1713                     break;
1714                 }
1715 
1716                 c->function = i;
1717                 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1718                 c->index = j;
1719                 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
1720 
1721                 if (i == 4 && c->eax == 0) {
1722                     break;
1723                 }
1724                 if (i == 0xb && !(c->ecx & 0xff00)) {
1725                     break;
1726                 }
1727                 if (i == 0x1f && !(c->ecx & 0xff00)) {
1728                     break;
1729                 }
1730                 if (i == 0xd && c->eax == 0) {
1731                     continue;
1732                 }
1733                 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1734                     fprintf(stderr, "cpuid_data is full, no space for "
1735                             "cpuid(eax:0x%x,ecx:0x%x)\n", i, j);
1736                     abort();
1737                 }
1738                 c = &cpuid_data.entries[cpuid_i++];
1739             }
1740             break;
1741         case 0x7:
1742         case 0x12:
1743             for (j = 0; ; j++) {
1744                 c->function = i;
1745                 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1746                 c->index = j;
1747                 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
1748 
1749                 if (j > 1 && (c->eax & 0xf) != 1) {
1750                     break;
1751                 }
1752 
1753                 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1754                     fprintf(stderr, "cpuid_data is full, no space for "
1755                                 "cpuid(eax:0x12,ecx:0x%x)\n", j);
1756                     abort();
1757                 }
1758                 c = &cpuid_data.entries[cpuid_i++];
1759             }
1760             break;
1761         case 0x14: {
1762             uint32_t times;
1763 
1764             c->function = i;
1765             c->index = 0;
1766             c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1767             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1768             times = c->eax;
1769 
1770             for (j = 1; j <= times; ++j) {
1771                 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1772                     fprintf(stderr, "cpuid_data is full, no space for "
1773                                 "cpuid(eax:0x%x,ecx:0x%x)\n", i, j);
1774                     abort();
1775                 }
1776                 c = &cpuid_data.entries[cpuid_i++];
1777                 c->function = i;
1778                 c->index = j;
1779                 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1780                 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
1781             }
1782             break;
1783         }
1784         default:
1785             c->function = i;
1786             c->flags = 0;
1787             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1788             if (!c->eax && !c->ebx && !c->ecx && !c->edx) {
1789                 /*
1790                  * KVM already returns all zeroes if a CPUID entry is missing,
1791                  * so we can omit it and avoid hitting KVM's 80-entry limit.
1792                  */
1793                 cpuid_i--;
1794             }
1795             break;
1796         }
1797     }
1798 
1799     if (limit >= 0x0a) {
1800         uint32_t eax, edx;
1801 
1802         cpu_x86_cpuid(env, 0x0a, 0, &eax, &unused, &unused, &edx);
1803 
1804         has_architectural_pmu_version = eax & 0xff;
1805         if (has_architectural_pmu_version > 0) {
1806             num_architectural_pmu_gp_counters = (eax & 0xff00) >> 8;
1807 
1808             /* Shouldn't be more than 32, since that's the number of bits
1809              * available in EBX to tell us _which_ counters are available.
1810              * Play it safe.
1811              */
1812             if (num_architectural_pmu_gp_counters > MAX_GP_COUNTERS) {
1813                 num_architectural_pmu_gp_counters = MAX_GP_COUNTERS;
1814             }
1815 
1816             if (has_architectural_pmu_version > 1) {
1817                 num_architectural_pmu_fixed_counters = edx & 0x1f;
1818 
1819                 if (num_architectural_pmu_fixed_counters > MAX_FIXED_COUNTERS) {
1820                     num_architectural_pmu_fixed_counters = MAX_FIXED_COUNTERS;
1821                 }
1822             }
1823         }
1824     }
1825 
1826     cpu_x86_cpuid(env, 0x80000000, 0, &limit, &unused, &unused, &unused);
1827 
1828     for (i = 0x80000000; i <= limit; i++) {
1829         if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1830             fprintf(stderr, "unsupported xlevel value: 0x%x\n", limit);
1831             abort();
1832         }
1833         c = &cpuid_data.entries[cpuid_i++];
1834 
1835         switch (i) {
1836         case 0x8000001d:
1837             /* Query for all AMD cache information leaves */
1838             for (j = 0; ; j++) {
1839                 c->function = i;
1840                 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1841                 c->index = j;
1842                 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
1843 
1844                 if (c->eax == 0) {
1845                     break;
1846                 }
1847                 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1848                     fprintf(stderr, "cpuid_data is full, no space for "
1849                             "cpuid(eax:0x%x,ecx:0x%x)\n", i, j);
1850                     abort();
1851                 }
1852                 c = &cpuid_data.entries[cpuid_i++];
1853             }
1854             break;
1855         default:
1856             c->function = i;
1857             c->flags = 0;
1858             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1859             if (!c->eax && !c->ebx && !c->ecx && !c->edx) {
1860                 /*
1861                  * KVM already returns all zeroes if a CPUID entry is missing,
1862                  * so we can omit it and avoid hitting KVM's 80-entry limit.
1863                  */
1864                 cpuid_i--;
1865             }
1866             break;
1867         }
1868     }
1869 
1870     /* Call Centaur's CPUID instructions they are supported. */
1871     if (env->cpuid_xlevel2 > 0) {
1872         cpu_x86_cpuid(env, 0xC0000000, 0, &limit, &unused, &unused, &unused);
1873 
1874         for (i = 0xC0000000; i <= limit; i++) {
1875             if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1876                 fprintf(stderr, "unsupported xlevel2 value: 0x%x\n", limit);
1877                 abort();
1878             }
1879             c = &cpuid_data.entries[cpuid_i++];
1880 
1881             c->function = i;
1882             c->flags = 0;
1883             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1884         }
1885     }
1886 
1887     cpuid_data.cpuid.nent = cpuid_i;
1888 
1889     if (((env->cpuid_version >> 8)&0xF) >= 6
1890         && (env->features[FEAT_1_EDX] & (CPUID_MCE | CPUID_MCA)) ==
1891            (CPUID_MCE | CPUID_MCA)
1892         && kvm_check_extension(cs->kvm_state, KVM_CAP_MCE) > 0) {
1893         uint64_t mcg_cap, unsupported_caps;
1894         int banks;
1895         int ret;
1896 
1897         ret = kvm_get_mce_cap_supported(cs->kvm_state, &mcg_cap, &banks);
1898         if (ret < 0) {
1899             fprintf(stderr, "kvm_get_mce_cap_supported: %s", strerror(-ret));
1900             return ret;
1901         }
1902 
1903         if (banks < (env->mcg_cap & MCG_CAP_BANKS_MASK)) {
1904             error_report("kvm: Unsupported MCE bank count (QEMU = %d, KVM = %d)",
1905                          (int)(env->mcg_cap & MCG_CAP_BANKS_MASK), banks);
1906             return -ENOTSUP;
1907         }
1908 
1909         unsupported_caps = env->mcg_cap & ~(mcg_cap | MCG_CAP_BANKS_MASK);
1910         if (unsupported_caps) {
1911             if (unsupported_caps & MCG_LMCE_P) {
1912                 error_report("kvm: LMCE not supported");
1913                 return -ENOTSUP;
1914             }
1915             warn_report("Unsupported MCG_CAP bits: 0x%" PRIx64,
1916                         unsupported_caps);
1917         }
1918 
1919         env->mcg_cap &= mcg_cap | MCG_CAP_BANKS_MASK;
1920         ret = kvm_vcpu_ioctl(cs, KVM_X86_SETUP_MCE, &env->mcg_cap);
1921         if (ret < 0) {
1922             fprintf(stderr, "KVM_X86_SETUP_MCE: %s", strerror(-ret));
1923             return ret;
1924         }
1925     }
1926 
1927     cpu->vmsentry = qemu_add_vm_change_state_handler(cpu_update_state, env);
1928 
1929     c = cpuid_find_entry(&cpuid_data.cpuid, 1, 0);
1930     if (c) {
1931         has_msr_feature_control = !!(c->ecx & CPUID_EXT_VMX) ||
1932                                   !!(c->ecx & CPUID_EXT_SMX);
1933     }
1934 
1935     c = cpuid_find_entry(&cpuid_data.cpuid, 7, 0);
1936     if (c && (c->ebx & CPUID_7_0_EBX_SGX)) {
1937         has_msr_feature_control = true;
1938     }
1939 
1940     if (env->mcg_cap & MCG_LMCE_P) {
1941         has_msr_mcg_ext_ctl = has_msr_feature_control = true;
1942     }
1943 
1944     if (!env->user_tsc_khz) {
1945         if ((env->features[FEAT_8000_0007_EDX] & CPUID_APM_INVTSC) &&
1946             invtsc_mig_blocker == NULL) {
1947             error_setg(&invtsc_mig_blocker,
1948                        "State blocked by non-migratable CPU device"
1949                        " (invtsc flag)");
1950             r = migrate_add_blocker(invtsc_mig_blocker, &local_err);
1951             if (r < 0) {
1952                 error_report_err(local_err);
1953                 return r;
1954             }
1955         }
1956     }
1957 
1958     if (cpu->vmware_cpuid_freq
1959         /* Guests depend on 0x40000000 to detect this feature, so only expose
1960          * it if KVM exposes leaf 0x40000000. (Conflicts with Hyper-V) */
1961         && cpu->expose_kvm
1962         && kvm_base == KVM_CPUID_SIGNATURE
1963         /* TSC clock must be stable and known for this feature. */
1964         && tsc_is_stable_and_known(env)) {
1965 
1966         c = &cpuid_data.entries[cpuid_i++];
1967         c->function = KVM_CPUID_SIGNATURE | 0x10;
1968         c->eax = env->tsc_khz;
1969         c->ebx = env->apic_bus_freq / 1000; /* Hz to KHz */
1970         c->ecx = c->edx = 0;
1971 
1972         c = cpuid_find_entry(&cpuid_data.cpuid, kvm_base, 0);
1973         c->eax = MAX(c->eax, KVM_CPUID_SIGNATURE | 0x10);
1974     }
1975 
1976     cpuid_data.cpuid.nent = cpuid_i;
1977 
1978     cpuid_data.cpuid.padding = 0;
1979     r = kvm_vcpu_ioctl(cs, KVM_SET_CPUID2, &cpuid_data);
1980     if (r) {
1981         goto fail;
1982     }
1983 
1984     if (has_xsave) {
1985         env->xsave_buf_len = sizeof(struct kvm_xsave);
1986         env->xsave_buf = qemu_memalign(4096, env->xsave_buf_len);
1987         memset(env->xsave_buf, 0, env->xsave_buf_len);
1988 
1989         /*
1990          * The allocated storage must be large enough for all of the
1991          * possible XSAVE state components.
1992          */
1993         assert(kvm_arch_get_supported_cpuid(kvm_state, 0xd, 0, R_ECX)
1994                <= env->xsave_buf_len);
1995     }
1996 
1997     max_nested_state_len = kvm_max_nested_state_length();
1998     if (max_nested_state_len > 0) {
1999         assert(max_nested_state_len >= offsetof(struct kvm_nested_state, data));
2000 
2001         if (cpu_has_vmx(env) || cpu_has_svm(env)) {
2002             struct kvm_vmx_nested_state_hdr *vmx_hdr;
2003 
2004             env->nested_state = g_malloc0(max_nested_state_len);
2005             env->nested_state->size = max_nested_state_len;
2006 
2007             if (cpu_has_vmx(env)) {
2008                 env->nested_state->format = KVM_STATE_NESTED_FORMAT_VMX;
2009                 vmx_hdr = &env->nested_state->hdr.vmx;
2010                 vmx_hdr->vmxon_pa = -1ull;
2011                 vmx_hdr->vmcs12_pa = -1ull;
2012             } else {
2013                 env->nested_state->format = KVM_STATE_NESTED_FORMAT_SVM;
2014             }
2015         }
2016     }
2017 
2018     cpu->kvm_msr_buf = g_malloc0(MSR_BUF_SIZE);
2019 
2020     if (!(env->features[FEAT_8000_0001_EDX] & CPUID_EXT2_RDTSCP)) {
2021         has_msr_tsc_aux = false;
2022     }
2023 
2024     kvm_init_msrs(cpu);
2025 
2026     return 0;
2027 
2028  fail:
2029     migrate_del_blocker(invtsc_mig_blocker);
2030 
2031     return r;
2032 }
2033 
2034 int kvm_arch_destroy_vcpu(CPUState *cs)
2035 {
2036     X86CPU *cpu = X86_CPU(cs);
2037     CPUX86State *env = &cpu->env;
2038 
2039     if (cpu->kvm_msr_buf) {
2040         g_free(cpu->kvm_msr_buf);
2041         cpu->kvm_msr_buf = NULL;
2042     }
2043 
2044     if (env->nested_state) {
2045         g_free(env->nested_state);
2046         env->nested_state = NULL;
2047     }
2048 
2049     qemu_del_vm_change_state_handler(cpu->vmsentry);
2050 
2051     return 0;
2052 }
2053 
2054 void kvm_arch_reset_vcpu(X86CPU *cpu)
2055 {
2056     CPUX86State *env = &cpu->env;
2057 
2058     env->xcr0 = 1;
2059     if (kvm_irqchip_in_kernel()) {
2060         env->mp_state = cpu_is_bsp(cpu) ? KVM_MP_STATE_RUNNABLE :
2061                                           KVM_MP_STATE_UNINITIALIZED;
2062     } else {
2063         env->mp_state = KVM_MP_STATE_RUNNABLE;
2064     }
2065 
2066     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_SYNIC)) {
2067         int i;
2068         for (i = 0; i < ARRAY_SIZE(env->msr_hv_synic_sint); i++) {
2069             env->msr_hv_synic_sint[i] = HV_SINT_MASKED;
2070         }
2071 
2072         hyperv_x86_synic_reset(cpu);
2073     }
2074     /* enabled by default */
2075     env->poll_control_msr = 1;
2076 
2077     sev_es_set_reset_vector(CPU(cpu));
2078 }
2079 
2080 void kvm_arch_do_init_vcpu(X86CPU *cpu)
2081 {
2082     CPUX86State *env = &cpu->env;
2083 
2084     /* APs get directly into wait-for-SIPI state.  */
2085     if (env->mp_state == KVM_MP_STATE_UNINITIALIZED) {
2086         env->mp_state = KVM_MP_STATE_INIT_RECEIVED;
2087     }
2088 }
2089 
2090 static int kvm_get_supported_feature_msrs(KVMState *s)
2091 {
2092     int ret = 0;
2093 
2094     if (kvm_feature_msrs != NULL) {
2095         return 0;
2096     }
2097 
2098     if (!kvm_check_extension(s, KVM_CAP_GET_MSR_FEATURES)) {
2099         return 0;
2100     }
2101 
2102     struct kvm_msr_list msr_list;
2103 
2104     msr_list.nmsrs = 0;
2105     ret = kvm_ioctl(s, KVM_GET_MSR_FEATURE_INDEX_LIST, &msr_list);
2106     if (ret < 0 && ret != -E2BIG) {
2107         error_report("Fetch KVM feature MSR list failed: %s",
2108             strerror(-ret));
2109         return ret;
2110     }
2111 
2112     assert(msr_list.nmsrs > 0);
2113     kvm_feature_msrs = (struct kvm_msr_list *) \
2114         g_malloc0(sizeof(msr_list) +
2115                  msr_list.nmsrs * sizeof(msr_list.indices[0]));
2116 
2117     kvm_feature_msrs->nmsrs = msr_list.nmsrs;
2118     ret = kvm_ioctl(s, KVM_GET_MSR_FEATURE_INDEX_LIST, kvm_feature_msrs);
2119 
2120     if (ret < 0) {
2121         error_report("Fetch KVM feature MSR list failed: %s",
2122             strerror(-ret));
2123         g_free(kvm_feature_msrs);
2124         kvm_feature_msrs = NULL;
2125         return ret;
2126     }
2127 
2128     return 0;
2129 }
2130 
2131 static int kvm_get_supported_msrs(KVMState *s)
2132 {
2133     int ret = 0;
2134     struct kvm_msr_list msr_list, *kvm_msr_list;
2135 
2136     /*
2137      *  Obtain MSR list from KVM.  These are the MSRs that we must
2138      *  save/restore.
2139      */
2140     msr_list.nmsrs = 0;
2141     ret = kvm_ioctl(s, KVM_GET_MSR_INDEX_LIST, &msr_list);
2142     if (ret < 0 && ret != -E2BIG) {
2143         return ret;
2144     }
2145     /*
2146      * Old kernel modules had a bug and could write beyond the provided
2147      * memory. Allocate at least a safe amount of 1K.
2148      */
2149     kvm_msr_list = g_malloc0(MAX(1024, sizeof(msr_list) +
2150                                           msr_list.nmsrs *
2151                                           sizeof(msr_list.indices[0])));
2152 
2153     kvm_msr_list->nmsrs = msr_list.nmsrs;
2154     ret = kvm_ioctl(s, KVM_GET_MSR_INDEX_LIST, kvm_msr_list);
2155     if (ret >= 0) {
2156         int i;
2157 
2158         for (i = 0; i < kvm_msr_list->nmsrs; i++) {
2159             switch (kvm_msr_list->indices[i]) {
2160             case MSR_STAR:
2161                 has_msr_star = true;
2162                 break;
2163             case MSR_VM_HSAVE_PA:
2164                 has_msr_hsave_pa = true;
2165                 break;
2166             case MSR_TSC_AUX:
2167                 has_msr_tsc_aux = true;
2168                 break;
2169             case MSR_TSC_ADJUST:
2170                 has_msr_tsc_adjust = true;
2171                 break;
2172             case MSR_IA32_TSCDEADLINE:
2173                 has_msr_tsc_deadline = true;
2174                 break;
2175             case MSR_IA32_SMBASE:
2176                 has_msr_smbase = true;
2177                 break;
2178             case MSR_SMI_COUNT:
2179                 has_msr_smi_count = true;
2180                 break;
2181             case MSR_IA32_MISC_ENABLE:
2182                 has_msr_misc_enable = true;
2183                 break;
2184             case MSR_IA32_BNDCFGS:
2185                 has_msr_bndcfgs = true;
2186                 break;
2187             case MSR_IA32_XSS:
2188                 has_msr_xss = true;
2189                 break;
2190             case MSR_IA32_UMWAIT_CONTROL:
2191                 has_msr_umwait = true;
2192                 break;
2193             case HV_X64_MSR_CRASH_CTL:
2194                 has_msr_hv_crash = true;
2195                 break;
2196             case HV_X64_MSR_RESET:
2197                 has_msr_hv_reset = true;
2198                 break;
2199             case HV_X64_MSR_VP_INDEX:
2200                 has_msr_hv_vpindex = true;
2201                 break;
2202             case HV_X64_MSR_VP_RUNTIME:
2203                 has_msr_hv_runtime = true;
2204                 break;
2205             case HV_X64_MSR_SCONTROL:
2206                 has_msr_hv_synic = true;
2207                 break;
2208             case HV_X64_MSR_STIMER0_CONFIG:
2209                 has_msr_hv_stimer = true;
2210                 break;
2211             case HV_X64_MSR_TSC_FREQUENCY:
2212                 has_msr_hv_frequencies = true;
2213                 break;
2214             case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2215                 has_msr_hv_reenlightenment = true;
2216                 break;
2217             case MSR_IA32_SPEC_CTRL:
2218                 has_msr_spec_ctrl = true;
2219                 break;
2220             case MSR_IA32_TSX_CTRL:
2221                 has_msr_tsx_ctrl = true;
2222                 break;
2223             case MSR_VIRT_SSBD:
2224                 has_msr_virt_ssbd = true;
2225                 break;
2226             case MSR_IA32_ARCH_CAPABILITIES:
2227                 has_msr_arch_capabs = true;
2228                 break;
2229             case MSR_IA32_CORE_CAPABILITY:
2230                 has_msr_core_capabs = true;
2231                 break;
2232             case MSR_IA32_PERF_CAPABILITIES:
2233                 has_msr_perf_capabs = true;
2234                 break;
2235             case MSR_IA32_VMX_VMFUNC:
2236                 has_msr_vmx_vmfunc = true;
2237                 break;
2238             case MSR_IA32_UCODE_REV:
2239                 has_msr_ucode_rev = true;
2240                 break;
2241             case MSR_IA32_VMX_PROCBASED_CTLS2:
2242                 has_msr_vmx_procbased_ctls2 = true;
2243                 break;
2244             case MSR_IA32_PKRS:
2245                 has_msr_pkrs = true;
2246                 break;
2247             }
2248         }
2249     }
2250 
2251     g_free(kvm_msr_list);
2252 
2253     return ret;
2254 }
2255 
2256 static Notifier smram_machine_done;
2257 static KVMMemoryListener smram_listener;
2258 static AddressSpace smram_address_space;
2259 static MemoryRegion smram_as_root;
2260 static MemoryRegion smram_as_mem;
2261 
2262 static void register_smram_listener(Notifier *n, void *unused)
2263 {
2264     MemoryRegion *smram =
2265         (MemoryRegion *) object_resolve_path("/machine/smram", NULL);
2266 
2267     /* Outer container... */
2268     memory_region_init(&smram_as_root, OBJECT(kvm_state), "mem-container-smram", ~0ull);
2269     memory_region_set_enabled(&smram_as_root, true);
2270 
2271     /* ... with two regions inside: normal system memory with low
2272      * priority, and...
2273      */
2274     memory_region_init_alias(&smram_as_mem, OBJECT(kvm_state), "mem-smram",
2275                              get_system_memory(), 0, ~0ull);
2276     memory_region_add_subregion_overlap(&smram_as_root, 0, &smram_as_mem, 0);
2277     memory_region_set_enabled(&smram_as_mem, true);
2278 
2279     if (smram) {
2280         /* ... SMRAM with higher priority */
2281         memory_region_add_subregion_overlap(&smram_as_root, 0, smram, 10);
2282         memory_region_set_enabled(smram, true);
2283     }
2284 
2285     address_space_init(&smram_address_space, &smram_as_root, "KVM-SMRAM");
2286     kvm_memory_listener_register(kvm_state, &smram_listener,
2287                                  &smram_address_space, 1, "kvm-smram");
2288 }
2289 
2290 int kvm_arch_init(MachineState *ms, KVMState *s)
2291 {
2292     uint64_t identity_base = 0xfffbc000;
2293     uint64_t shadow_mem;
2294     int ret;
2295     struct utsname utsname;
2296     Error *local_err = NULL;
2297 
2298     /*
2299      * Initialize SEV context, if required
2300      *
2301      * If no memory encryption is requested (ms->cgs == NULL) this is
2302      * a no-op.
2303      *
2304      * It's also a no-op if a non-SEV confidential guest support
2305      * mechanism is selected.  SEV is the only mechanism available to
2306      * select on x86 at present, so this doesn't arise, but if new
2307      * mechanisms are supported in future (e.g. TDX), they'll need
2308      * their own initialization either here or elsewhere.
2309      */
2310     ret = sev_kvm_init(ms->cgs, &local_err);
2311     if (ret < 0) {
2312         error_report_err(local_err);
2313         return ret;
2314     }
2315 
2316     if (!kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
2317         error_report("kvm: KVM_CAP_IRQ_ROUTING not supported by KVM");
2318         return -ENOTSUP;
2319     }
2320 
2321     has_xsave = kvm_check_extension(s, KVM_CAP_XSAVE);
2322     has_xcrs = kvm_check_extension(s, KVM_CAP_XCRS);
2323     has_pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2);
2324 
2325     hv_vpindex_settable = kvm_check_extension(s, KVM_CAP_HYPERV_VP_INDEX);
2326 
2327     has_exception_payload = kvm_check_extension(s, KVM_CAP_EXCEPTION_PAYLOAD);
2328     if (has_exception_payload) {
2329         ret = kvm_vm_enable_cap(s, KVM_CAP_EXCEPTION_PAYLOAD, 0, true);
2330         if (ret < 0) {
2331             error_report("kvm: Failed to enable exception payload cap: %s",
2332                          strerror(-ret));
2333             return ret;
2334         }
2335     }
2336 
2337     ret = kvm_get_supported_msrs(s);
2338     if (ret < 0) {
2339         return ret;
2340     }
2341 
2342     kvm_get_supported_feature_msrs(s);
2343 
2344     uname(&utsname);
2345     lm_capable_kernel = strcmp(utsname.machine, "x86_64") == 0;
2346 
2347     /*
2348      * On older Intel CPUs, KVM uses vm86 mode to emulate 16-bit code directly.
2349      * In order to use vm86 mode, an EPT identity map and a TSS  are needed.
2350      * Since these must be part of guest physical memory, we need to allocate
2351      * them, both by setting their start addresses in the kernel and by
2352      * creating a corresponding e820 entry. We need 4 pages before the BIOS.
2353      *
2354      * Older KVM versions may not support setting the identity map base. In
2355      * that case we need to stick with the default, i.e. a 256K maximum BIOS
2356      * size.
2357      */
2358     if (kvm_check_extension(s, KVM_CAP_SET_IDENTITY_MAP_ADDR)) {
2359         /* Allows up to 16M BIOSes. */
2360         identity_base = 0xfeffc000;
2361 
2362         ret = kvm_vm_ioctl(s, KVM_SET_IDENTITY_MAP_ADDR, &identity_base);
2363         if (ret < 0) {
2364             return ret;
2365         }
2366     }
2367 
2368     /* Set TSS base one page after EPT identity map. */
2369     ret = kvm_vm_ioctl(s, KVM_SET_TSS_ADDR, identity_base + 0x1000);
2370     if (ret < 0) {
2371         return ret;
2372     }
2373 
2374     /* Tell fw_cfg to notify the BIOS to reserve the range. */
2375     ret = e820_add_entry(identity_base, 0x4000, E820_RESERVED);
2376     if (ret < 0) {
2377         fprintf(stderr, "e820_add_entry() table is full\n");
2378         return ret;
2379     }
2380 
2381     shadow_mem = object_property_get_int(OBJECT(s), "kvm-shadow-mem", &error_abort);
2382     if (shadow_mem != -1) {
2383         shadow_mem /= 4096;
2384         ret = kvm_vm_ioctl(s, KVM_SET_NR_MMU_PAGES, shadow_mem);
2385         if (ret < 0) {
2386             return ret;
2387         }
2388     }
2389 
2390     if (kvm_check_extension(s, KVM_CAP_X86_SMM) &&
2391         object_dynamic_cast(OBJECT(ms), TYPE_X86_MACHINE) &&
2392         x86_machine_is_smm_enabled(X86_MACHINE(ms))) {
2393         smram_machine_done.notify = register_smram_listener;
2394         qemu_add_machine_init_done_notifier(&smram_machine_done);
2395     }
2396 
2397     if (enable_cpu_pm) {
2398         int disable_exits = kvm_check_extension(s, KVM_CAP_X86_DISABLE_EXITS);
2399         int ret;
2400 
2401 /* Work around for kernel header with a typo. TODO: fix header and drop. */
2402 #if defined(KVM_X86_DISABLE_EXITS_HTL) && !defined(KVM_X86_DISABLE_EXITS_HLT)
2403 #define KVM_X86_DISABLE_EXITS_HLT KVM_X86_DISABLE_EXITS_HTL
2404 #endif
2405         if (disable_exits) {
2406             disable_exits &= (KVM_X86_DISABLE_EXITS_MWAIT |
2407                               KVM_X86_DISABLE_EXITS_HLT |
2408                               KVM_X86_DISABLE_EXITS_PAUSE |
2409                               KVM_X86_DISABLE_EXITS_CSTATE);
2410         }
2411 
2412         ret = kvm_vm_enable_cap(s, KVM_CAP_X86_DISABLE_EXITS, 0,
2413                                 disable_exits);
2414         if (ret < 0) {
2415             error_report("kvm: guest stopping CPU not supported: %s",
2416                          strerror(-ret));
2417         }
2418     }
2419 
2420     if (object_dynamic_cast(OBJECT(ms), TYPE_X86_MACHINE)) {
2421         X86MachineState *x86ms = X86_MACHINE(ms);
2422 
2423         if (x86ms->bus_lock_ratelimit > 0) {
2424             ret = kvm_check_extension(s, KVM_CAP_X86_BUS_LOCK_EXIT);
2425             if (!(ret & KVM_BUS_LOCK_DETECTION_EXIT)) {
2426                 error_report("kvm: bus lock detection unsupported");
2427                 return -ENOTSUP;
2428             }
2429             ret = kvm_vm_enable_cap(s, KVM_CAP_X86_BUS_LOCK_EXIT, 0,
2430                                     KVM_BUS_LOCK_DETECTION_EXIT);
2431             if (ret < 0) {
2432                 error_report("kvm: Failed to enable bus lock detection cap: %s",
2433                              strerror(-ret));
2434                 return ret;
2435             }
2436             ratelimit_init(&bus_lock_ratelimit_ctrl);
2437             ratelimit_set_speed(&bus_lock_ratelimit_ctrl,
2438                                 x86ms->bus_lock_ratelimit, BUS_LOCK_SLICE_TIME);
2439         }
2440     }
2441 
2442     return 0;
2443 }
2444 
2445 static void set_v8086_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
2446 {
2447     lhs->selector = rhs->selector;
2448     lhs->base = rhs->base;
2449     lhs->limit = rhs->limit;
2450     lhs->type = 3;
2451     lhs->present = 1;
2452     lhs->dpl = 3;
2453     lhs->db = 0;
2454     lhs->s = 1;
2455     lhs->l = 0;
2456     lhs->g = 0;
2457     lhs->avl = 0;
2458     lhs->unusable = 0;
2459 }
2460 
2461 static void set_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
2462 {
2463     unsigned flags = rhs->flags;
2464     lhs->selector = rhs->selector;
2465     lhs->base = rhs->base;
2466     lhs->limit = rhs->limit;
2467     lhs->type = (flags >> DESC_TYPE_SHIFT) & 15;
2468     lhs->present = (flags & DESC_P_MASK) != 0;
2469     lhs->dpl = (flags >> DESC_DPL_SHIFT) & 3;
2470     lhs->db = (flags >> DESC_B_SHIFT) & 1;
2471     lhs->s = (flags & DESC_S_MASK) != 0;
2472     lhs->l = (flags >> DESC_L_SHIFT) & 1;
2473     lhs->g = (flags & DESC_G_MASK) != 0;
2474     lhs->avl = (flags & DESC_AVL_MASK) != 0;
2475     lhs->unusable = !lhs->present;
2476     lhs->padding = 0;
2477 }
2478 
2479 static void get_seg(SegmentCache *lhs, const struct kvm_segment *rhs)
2480 {
2481     lhs->selector = rhs->selector;
2482     lhs->base = rhs->base;
2483     lhs->limit = rhs->limit;
2484     lhs->flags = (rhs->type << DESC_TYPE_SHIFT) |
2485                  ((rhs->present && !rhs->unusable) * DESC_P_MASK) |
2486                  (rhs->dpl << DESC_DPL_SHIFT) |
2487                  (rhs->db << DESC_B_SHIFT) |
2488                  (rhs->s * DESC_S_MASK) |
2489                  (rhs->l << DESC_L_SHIFT) |
2490                  (rhs->g * DESC_G_MASK) |
2491                  (rhs->avl * DESC_AVL_MASK);
2492 }
2493 
2494 static void kvm_getput_reg(__u64 *kvm_reg, target_ulong *qemu_reg, int set)
2495 {
2496     if (set) {
2497         *kvm_reg = *qemu_reg;
2498     } else {
2499         *qemu_reg = *kvm_reg;
2500     }
2501 }
2502 
2503 static int kvm_getput_regs(X86CPU *cpu, int set)
2504 {
2505     CPUX86State *env = &cpu->env;
2506     struct kvm_regs regs;
2507     int ret = 0;
2508 
2509     if (!set) {
2510         ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_REGS, &regs);
2511         if (ret < 0) {
2512             return ret;
2513         }
2514     }
2515 
2516     kvm_getput_reg(&regs.rax, &env->regs[R_EAX], set);
2517     kvm_getput_reg(&regs.rbx, &env->regs[R_EBX], set);
2518     kvm_getput_reg(&regs.rcx, &env->regs[R_ECX], set);
2519     kvm_getput_reg(&regs.rdx, &env->regs[R_EDX], set);
2520     kvm_getput_reg(&regs.rsi, &env->regs[R_ESI], set);
2521     kvm_getput_reg(&regs.rdi, &env->regs[R_EDI], set);
2522     kvm_getput_reg(&regs.rsp, &env->regs[R_ESP], set);
2523     kvm_getput_reg(&regs.rbp, &env->regs[R_EBP], set);
2524 #ifdef TARGET_X86_64
2525     kvm_getput_reg(&regs.r8, &env->regs[8], set);
2526     kvm_getput_reg(&regs.r9, &env->regs[9], set);
2527     kvm_getput_reg(&regs.r10, &env->regs[10], set);
2528     kvm_getput_reg(&regs.r11, &env->regs[11], set);
2529     kvm_getput_reg(&regs.r12, &env->regs[12], set);
2530     kvm_getput_reg(&regs.r13, &env->regs[13], set);
2531     kvm_getput_reg(&regs.r14, &env->regs[14], set);
2532     kvm_getput_reg(&regs.r15, &env->regs[15], set);
2533 #endif
2534 
2535     kvm_getput_reg(&regs.rflags, &env->eflags, set);
2536     kvm_getput_reg(&regs.rip, &env->eip, set);
2537 
2538     if (set) {
2539         ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_REGS, &regs);
2540     }
2541 
2542     return ret;
2543 }
2544 
2545 static int kvm_put_fpu(X86CPU *cpu)
2546 {
2547     CPUX86State *env = &cpu->env;
2548     struct kvm_fpu fpu;
2549     int i;
2550 
2551     memset(&fpu, 0, sizeof fpu);
2552     fpu.fsw = env->fpus & ~(7 << 11);
2553     fpu.fsw |= (env->fpstt & 7) << 11;
2554     fpu.fcw = env->fpuc;
2555     fpu.last_opcode = env->fpop;
2556     fpu.last_ip = env->fpip;
2557     fpu.last_dp = env->fpdp;
2558     for (i = 0; i < 8; ++i) {
2559         fpu.ftwx |= (!env->fptags[i]) << i;
2560     }
2561     memcpy(fpu.fpr, env->fpregs, sizeof env->fpregs);
2562     for (i = 0; i < CPU_NB_REGS; i++) {
2563         stq_p(&fpu.xmm[i][0], env->xmm_regs[i].ZMM_Q(0));
2564         stq_p(&fpu.xmm[i][8], env->xmm_regs[i].ZMM_Q(1));
2565     }
2566     fpu.mxcsr = env->mxcsr;
2567 
2568     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_FPU, &fpu);
2569 }
2570 
2571 static int kvm_put_xsave(X86CPU *cpu)
2572 {
2573     CPUX86State *env = &cpu->env;
2574     void *xsave = env->xsave_buf;
2575 
2576     if (!has_xsave) {
2577         return kvm_put_fpu(cpu);
2578     }
2579     x86_cpu_xsave_all_areas(cpu, xsave, env->xsave_buf_len);
2580 
2581     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_XSAVE, xsave);
2582 }
2583 
2584 static int kvm_put_xcrs(X86CPU *cpu)
2585 {
2586     CPUX86State *env = &cpu->env;
2587     struct kvm_xcrs xcrs = {};
2588 
2589     if (!has_xcrs) {
2590         return 0;
2591     }
2592 
2593     xcrs.nr_xcrs = 1;
2594     xcrs.flags = 0;
2595     xcrs.xcrs[0].xcr = 0;
2596     xcrs.xcrs[0].value = env->xcr0;
2597     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_XCRS, &xcrs);
2598 }
2599 
2600 static int kvm_put_sregs(X86CPU *cpu)
2601 {
2602     CPUX86State *env = &cpu->env;
2603     struct kvm_sregs sregs;
2604 
2605     memset(sregs.interrupt_bitmap, 0, sizeof(sregs.interrupt_bitmap));
2606     if (env->interrupt_injected >= 0) {
2607         sregs.interrupt_bitmap[env->interrupt_injected / 64] |=
2608                 (uint64_t)1 << (env->interrupt_injected % 64);
2609     }
2610 
2611     if ((env->eflags & VM_MASK)) {
2612         set_v8086_seg(&sregs.cs, &env->segs[R_CS]);
2613         set_v8086_seg(&sregs.ds, &env->segs[R_DS]);
2614         set_v8086_seg(&sregs.es, &env->segs[R_ES]);
2615         set_v8086_seg(&sregs.fs, &env->segs[R_FS]);
2616         set_v8086_seg(&sregs.gs, &env->segs[R_GS]);
2617         set_v8086_seg(&sregs.ss, &env->segs[R_SS]);
2618     } else {
2619         set_seg(&sregs.cs, &env->segs[R_CS]);
2620         set_seg(&sregs.ds, &env->segs[R_DS]);
2621         set_seg(&sregs.es, &env->segs[R_ES]);
2622         set_seg(&sregs.fs, &env->segs[R_FS]);
2623         set_seg(&sregs.gs, &env->segs[R_GS]);
2624         set_seg(&sregs.ss, &env->segs[R_SS]);
2625     }
2626 
2627     set_seg(&sregs.tr, &env->tr);
2628     set_seg(&sregs.ldt, &env->ldt);
2629 
2630     sregs.idt.limit = env->idt.limit;
2631     sregs.idt.base = env->idt.base;
2632     memset(sregs.idt.padding, 0, sizeof sregs.idt.padding);
2633     sregs.gdt.limit = env->gdt.limit;
2634     sregs.gdt.base = env->gdt.base;
2635     memset(sregs.gdt.padding, 0, sizeof sregs.gdt.padding);
2636 
2637     sregs.cr0 = env->cr[0];
2638     sregs.cr2 = env->cr[2];
2639     sregs.cr3 = env->cr[3];
2640     sregs.cr4 = env->cr[4];
2641 
2642     sregs.cr8 = cpu_get_apic_tpr(cpu->apic_state);
2643     sregs.apic_base = cpu_get_apic_base(cpu->apic_state);
2644 
2645     sregs.efer = env->efer;
2646 
2647     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_SREGS, &sregs);
2648 }
2649 
2650 static void kvm_msr_buf_reset(X86CPU *cpu)
2651 {
2652     memset(cpu->kvm_msr_buf, 0, MSR_BUF_SIZE);
2653 }
2654 
2655 static void kvm_msr_entry_add(X86CPU *cpu, uint32_t index, uint64_t value)
2656 {
2657     struct kvm_msrs *msrs = cpu->kvm_msr_buf;
2658     void *limit = ((void *)msrs) + MSR_BUF_SIZE;
2659     struct kvm_msr_entry *entry = &msrs->entries[msrs->nmsrs];
2660 
2661     assert((void *)(entry + 1) <= limit);
2662 
2663     entry->index = index;
2664     entry->reserved = 0;
2665     entry->data = value;
2666     msrs->nmsrs++;
2667 }
2668 
2669 static int kvm_put_one_msr(X86CPU *cpu, int index, uint64_t value)
2670 {
2671     kvm_msr_buf_reset(cpu);
2672     kvm_msr_entry_add(cpu, index, value);
2673 
2674     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MSRS, cpu->kvm_msr_buf);
2675 }
2676 
2677 void kvm_put_apicbase(X86CPU *cpu, uint64_t value)
2678 {
2679     int ret;
2680 
2681     ret = kvm_put_one_msr(cpu, MSR_IA32_APICBASE, value);
2682     assert(ret == 1);
2683 }
2684 
2685 static int kvm_put_tscdeadline_msr(X86CPU *cpu)
2686 {
2687     CPUX86State *env = &cpu->env;
2688     int ret;
2689 
2690     if (!has_msr_tsc_deadline) {
2691         return 0;
2692     }
2693 
2694     ret = kvm_put_one_msr(cpu, MSR_IA32_TSCDEADLINE, env->tsc_deadline);
2695     if (ret < 0) {
2696         return ret;
2697     }
2698 
2699     assert(ret == 1);
2700     return 0;
2701 }
2702 
2703 /*
2704  * Provide a separate write service for the feature control MSR in order to
2705  * kick the VCPU out of VMXON or even guest mode on reset. This has to be done
2706  * before writing any other state because forcibly leaving nested mode
2707  * invalidates the VCPU state.
2708  */
2709 static int kvm_put_msr_feature_control(X86CPU *cpu)
2710 {
2711     int ret;
2712 
2713     if (!has_msr_feature_control) {
2714         return 0;
2715     }
2716 
2717     ret = kvm_put_one_msr(cpu, MSR_IA32_FEATURE_CONTROL,
2718                           cpu->env.msr_ia32_feature_control);
2719     if (ret < 0) {
2720         return ret;
2721     }
2722 
2723     assert(ret == 1);
2724     return 0;
2725 }
2726 
2727 static uint64_t make_vmx_msr_value(uint32_t index, uint32_t features)
2728 {
2729     uint32_t default1, can_be_one, can_be_zero;
2730     uint32_t must_be_one;
2731 
2732     switch (index) {
2733     case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2734         default1 = 0x00000016;
2735         break;
2736     case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2737         default1 = 0x0401e172;
2738         break;
2739     case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2740         default1 = 0x000011ff;
2741         break;
2742     case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2743         default1 = 0x00036dff;
2744         break;
2745     case MSR_IA32_VMX_PROCBASED_CTLS2:
2746         default1 = 0;
2747         break;
2748     default:
2749         abort();
2750     }
2751 
2752     /* If a feature bit is set, the control can be either set or clear.
2753      * Otherwise the value is limited to either 0 or 1 by default1.
2754      */
2755     can_be_one = features | default1;
2756     can_be_zero = features | ~default1;
2757     must_be_one = ~can_be_zero;
2758 
2759     /*
2760      * Bit 0:31 -> 0 if the control bit can be zero (i.e. 1 if it must be one).
2761      * Bit 32:63 -> 1 if the control bit can be one.
2762      */
2763     return must_be_one | (((uint64_t)can_be_one) << 32);
2764 }
2765 
2766 static void kvm_msr_entry_add_vmx(X86CPU *cpu, FeatureWordArray f)
2767 {
2768     uint64_t kvm_vmx_basic =
2769         kvm_arch_get_supported_msr_feature(kvm_state,
2770                                            MSR_IA32_VMX_BASIC);
2771 
2772     if (!kvm_vmx_basic) {
2773         /* If the kernel doesn't support VMX feature (kvm_intel.nested=0),
2774          * then kvm_vmx_basic will be 0 and KVM_SET_MSR will fail.
2775          */
2776         return;
2777     }
2778 
2779     uint64_t kvm_vmx_misc =
2780         kvm_arch_get_supported_msr_feature(kvm_state,
2781                                            MSR_IA32_VMX_MISC);
2782     uint64_t kvm_vmx_ept_vpid =
2783         kvm_arch_get_supported_msr_feature(kvm_state,
2784                                            MSR_IA32_VMX_EPT_VPID_CAP);
2785 
2786     /*
2787      * If the guest is 64-bit, a value of 1 is allowed for the host address
2788      * space size vmexit control.
2789      */
2790     uint64_t fixed_vmx_exit = f[FEAT_8000_0001_EDX] & CPUID_EXT2_LM
2791         ? (uint64_t)VMX_VM_EXIT_HOST_ADDR_SPACE_SIZE << 32 : 0;
2792 
2793     /*
2794      * Bits 0-30, 32-44 and 50-53 come from the host.  KVM should
2795      * not change them for backwards compatibility.
2796      */
2797     uint64_t fixed_vmx_basic = kvm_vmx_basic &
2798         (MSR_VMX_BASIC_VMCS_REVISION_MASK |
2799          MSR_VMX_BASIC_VMXON_REGION_SIZE_MASK |
2800          MSR_VMX_BASIC_VMCS_MEM_TYPE_MASK);
2801 
2802     /*
2803      * Same for bits 0-4 and 25-27.  Bits 16-24 (CR3 target count) can
2804      * change in the future but are always zero for now, clear them to be
2805      * future proof.  Bits 32-63 in theory could change, though KVM does
2806      * not support dual-monitor treatment and probably never will; mask
2807      * them out as well.
2808      */
2809     uint64_t fixed_vmx_misc = kvm_vmx_misc &
2810         (MSR_VMX_MISC_PREEMPTION_TIMER_SHIFT_MASK |
2811          MSR_VMX_MISC_MAX_MSR_LIST_SIZE_MASK);
2812 
2813     /*
2814      * EPT memory types should not change either, so we do not bother
2815      * adding features for them.
2816      */
2817     uint64_t fixed_vmx_ept_mask =
2818             (f[FEAT_VMX_SECONDARY_CTLS] & VMX_SECONDARY_EXEC_ENABLE_EPT ?
2819              MSR_VMX_EPT_UC | MSR_VMX_EPT_WB : 0);
2820     uint64_t fixed_vmx_ept_vpid = kvm_vmx_ept_vpid & fixed_vmx_ept_mask;
2821 
2822     kvm_msr_entry_add(cpu, MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
2823                       make_vmx_msr_value(MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
2824                                          f[FEAT_VMX_PROCBASED_CTLS]));
2825     kvm_msr_entry_add(cpu, MSR_IA32_VMX_TRUE_PINBASED_CTLS,
2826                       make_vmx_msr_value(MSR_IA32_VMX_TRUE_PINBASED_CTLS,
2827                                          f[FEAT_VMX_PINBASED_CTLS]));
2828     kvm_msr_entry_add(cpu, MSR_IA32_VMX_TRUE_EXIT_CTLS,
2829                       make_vmx_msr_value(MSR_IA32_VMX_TRUE_EXIT_CTLS,
2830                                          f[FEAT_VMX_EXIT_CTLS]) | fixed_vmx_exit);
2831     kvm_msr_entry_add(cpu, MSR_IA32_VMX_TRUE_ENTRY_CTLS,
2832                       make_vmx_msr_value(MSR_IA32_VMX_TRUE_ENTRY_CTLS,
2833                                          f[FEAT_VMX_ENTRY_CTLS]));
2834     kvm_msr_entry_add(cpu, MSR_IA32_VMX_PROCBASED_CTLS2,
2835                       make_vmx_msr_value(MSR_IA32_VMX_PROCBASED_CTLS2,
2836                                          f[FEAT_VMX_SECONDARY_CTLS]));
2837     kvm_msr_entry_add(cpu, MSR_IA32_VMX_EPT_VPID_CAP,
2838                       f[FEAT_VMX_EPT_VPID_CAPS] | fixed_vmx_ept_vpid);
2839     kvm_msr_entry_add(cpu, MSR_IA32_VMX_BASIC,
2840                       f[FEAT_VMX_BASIC] | fixed_vmx_basic);
2841     kvm_msr_entry_add(cpu, MSR_IA32_VMX_MISC,
2842                       f[FEAT_VMX_MISC] | fixed_vmx_misc);
2843     if (has_msr_vmx_vmfunc) {
2844         kvm_msr_entry_add(cpu, MSR_IA32_VMX_VMFUNC, f[FEAT_VMX_VMFUNC]);
2845     }
2846 
2847     /*
2848      * Just to be safe, write these with constant values.  The CRn_FIXED1
2849      * MSRs are generated by KVM based on the vCPU's CPUID.
2850      */
2851     kvm_msr_entry_add(cpu, MSR_IA32_VMX_CR0_FIXED0,
2852                       CR0_PE_MASK | CR0_PG_MASK | CR0_NE_MASK);
2853     kvm_msr_entry_add(cpu, MSR_IA32_VMX_CR4_FIXED0,
2854                       CR4_VMXE_MASK);
2855 
2856     if (f[FEAT_VMX_SECONDARY_CTLS] & VMX_SECONDARY_EXEC_TSC_SCALING) {
2857         /* TSC multiplier (0x2032).  */
2858         kvm_msr_entry_add(cpu, MSR_IA32_VMX_VMCS_ENUM, 0x32);
2859     } else {
2860         /* Preemption timer (0x482E).  */
2861         kvm_msr_entry_add(cpu, MSR_IA32_VMX_VMCS_ENUM, 0x2E);
2862     }
2863 }
2864 
2865 static void kvm_msr_entry_add_perf(X86CPU *cpu, FeatureWordArray f)
2866 {
2867     uint64_t kvm_perf_cap =
2868         kvm_arch_get_supported_msr_feature(kvm_state,
2869                                            MSR_IA32_PERF_CAPABILITIES);
2870 
2871     if (kvm_perf_cap) {
2872         kvm_msr_entry_add(cpu, MSR_IA32_PERF_CAPABILITIES,
2873                         kvm_perf_cap & f[FEAT_PERF_CAPABILITIES]);
2874     }
2875 }
2876 
2877 static int kvm_buf_set_msrs(X86CPU *cpu)
2878 {
2879     int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MSRS, cpu->kvm_msr_buf);
2880     if (ret < 0) {
2881         return ret;
2882     }
2883 
2884     if (ret < cpu->kvm_msr_buf->nmsrs) {
2885         struct kvm_msr_entry *e = &cpu->kvm_msr_buf->entries[ret];
2886         error_report("error: failed to set MSR 0x%" PRIx32 " to 0x%" PRIx64,
2887                      (uint32_t)e->index, (uint64_t)e->data);
2888     }
2889 
2890     assert(ret == cpu->kvm_msr_buf->nmsrs);
2891     return 0;
2892 }
2893 
2894 static void kvm_init_msrs(X86CPU *cpu)
2895 {
2896     CPUX86State *env = &cpu->env;
2897 
2898     kvm_msr_buf_reset(cpu);
2899     if (has_msr_arch_capabs) {
2900         kvm_msr_entry_add(cpu, MSR_IA32_ARCH_CAPABILITIES,
2901                           env->features[FEAT_ARCH_CAPABILITIES]);
2902     }
2903 
2904     if (has_msr_core_capabs) {
2905         kvm_msr_entry_add(cpu, MSR_IA32_CORE_CAPABILITY,
2906                           env->features[FEAT_CORE_CAPABILITY]);
2907     }
2908 
2909     if (has_msr_perf_capabs && cpu->enable_pmu) {
2910         kvm_msr_entry_add_perf(cpu, env->features);
2911     }
2912 
2913     if (has_msr_ucode_rev) {
2914         kvm_msr_entry_add(cpu, MSR_IA32_UCODE_REV, cpu->ucode_rev);
2915     }
2916 
2917     /*
2918      * Older kernels do not include VMX MSRs in KVM_GET_MSR_INDEX_LIST, but
2919      * all kernels with MSR features should have them.
2920      */
2921     if (kvm_feature_msrs && cpu_has_vmx(env)) {
2922         kvm_msr_entry_add_vmx(cpu, env->features);
2923     }
2924 
2925     assert(kvm_buf_set_msrs(cpu) == 0);
2926 }
2927 
2928 static int kvm_put_msrs(X86CPU *cpu, int level)
2929 {
2930     CPUX86State *env = &cpu->env;
2931     int i;
2932 
2933     kvm_msr_buf_reset(cpu);
2934 
2935     kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_CS, env->sysenter_cs);
2936     kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_ESP, env->sysenter_esp);
2937     kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_EIP, env->sysenter_eip);
2938     kvm_msr_entry_add(cpu, MSR_PAT, env->pat);
2939     if (has_msr_star) {
2940         kvm_msr_entry_add(cpu, MSR_STAR, env->star);
2941     }
2942     if (has_msr_hsave_pa) {
2943         kvm_msr_entry_add(cpu, MSR_VM_HSAVE_PA, env->vm_hsave);
2944     }
2945     if (has_msr_tsc_aux) {
2946         kvm_msr_entry_add(cpu, MSR_TSC_AUX, env->tsc_aux);
2947     }
2948     if (has_msr_tsc_adjust) {
2949         kvm_msr_entry_add(cpu, MSR_TSC_ADJUST, env->tsc_adjust);
2950     }
2951     if (has_msr_misc_enable) {
2952         kvm_msr_entry_add(cpu, MSR_IA32_MISC_ENABLE,
2953                           env->msr_ia32_misc_enable);
2954     }
2955     if (has_msr_smbase) {
2956         kvm_msr_entry_add(cpu, MSR_IA32_SMBASE, env->smbase);
2957     }
2958     if (has_msr_smi_count) {
2959         kvm_msr_entry_add(cpu, MSR_SMI_COUNT, env->msr_smi_count);
2960     }
2961     if (has_msr_pkrs) {
2962         kvm_msr_entry_add(cpu, MSR_IA32_PKRS, env->pkrs);
2963     }
2964     if (has_msr_bndcfgs) {
2965         kvm_msr_entry_add(cpu, MSR_IA32_BNDCFGS, env->msr_bndcfgs);
2966     }
2967     if (has_msr_xss) {
2968         kvm_msr_entry_add(cpu, MSR_IA32_XSS, env->xss);
2969     }
2970     if (has_msr_umwait) {
2971         kvm_msr_entry_add(cpu, MSR_IA32_UMWAIT_CONTROL, env->umwait);
2972     }
2973     if (has_msr_spec_ctrl) {
2974         kvm_msr_entry_add(cpu, MSR_IA32_SPEC_CTRL, env->spec_ctrl);
2975     }
2976     if (has_msr_tsx_ctrl) {
2977         kvm_msr_entry_add(cpu, MSR_IA32_TSX_CTRL, env->tsx_ctrl);
2978     }
2979     if (has_msr_virt_ssbd) {
2980         kvm_msr_entry_add(cpu, MSR_VIRT_SSBD, env->virt_ssbd);
2981     }
2982 
2983 #ifdef TARGET_X86_64
2984     if (lm_capable_kernel) {
2985         kvm_msr_entry_add(cpu, MSR_CSTAR, env->cstar);
2986         kvm_msr_entry_add(cpu, MSR_KERNELGSBASE, env->kernelgsbase);
2987         kvm_msr_entry_add(cpu, MSR_FMASK, env->fmask);
2988         kvm_msr_entry_add(cpu, MSR_LSTAR, env->lstar);
2989     }
2990 #endif
2991 
2992     /*
2993      * The following MSRs have side effects on the guest or are too heavy
2994      * for normal writeback. Limit them to reset or full state updates.
2995      */
2996     if (level >= KVM_PUT_RESET_STATE) {
2997         kvm_msr_entry_add(cpu, MSR_IA32_TSC, env->tsc);
2998         kvm_msr_entry_add(cpu, MSR_KVM_SYSTEM_TIME, env->system_time_msr);
2999         kvm_msr_entry_add(cpu, MSR_KVM_WALL_CLOCK, env->wall_clock_msr);
3000         if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_ASYNC_PF_INT)) {
3001             kvm_msr_entry_add(cpu, MSR_KVM_ASYNC_PF_INT, env->async_pf_int_msr);
3002         }
3003         if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_ASYNC_PF)) {
3004             kvm_msr_entry_add(cpu, MSR_KVM_ASYNC_PF_EN, env->async_pf_en_msr);
3005         }
3006         if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_PV_EOI)) {
3007             kvm_msr_entry_add(cpu, MSR_KVM_PV_EOI_EN, env->pv_eoi_en_msr);
3008         }
3009         if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_STEAL_TIME)) {
3010             kvm_msr_entry_add(cpu, MSR_KVM_STEAL_TIME, env->steal_time_msr);
3011         }
3012 
3013         if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_POLL_CONTROL)) {
3014             kvm_msr_entry_add(cpu, MSR_KVM_POLL_CONTROL, env->poll_control_msr);
3015         }
3016 
3017         if (has_architectural_pmu_version > 0) {
3018             if (has_architectural_pmu_version > 1) {
3019                 /* Stop the counter.  */
3020                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR_CTRL, 0);
3021                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_CTRL, 0);
3022             }
3023 
3024             /* Set the counter values.  */
3025             for (i = 0; i < num_architectural_pmu_fixed_counters; i++) {
3026                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR0 + i,
3027                                   env->msr_fixed_counters[i]);
3028             }
3029             for (i = 0; i < num_architectural_pmu_gp_counters; i++) {
3030                 kvm_msr_entry_add(cpu, MSR_P6_PERFCTR0 + i,
3031                                   env->msr_gp_counters[i]);
3032                 kvm_msr_entry_add(cpu, MSR_P6_EVNTSEL0 + i,
3033                                   env->msr_gp_evtsel[i]);
3034             }
3035             if (has_architectural_pmu_version > 1) {
3036                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_STATUS,
3037                                   env->msr_global_status);
3038                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
3039                                   env->msr_global_ovf_ctrl);
3040 
3041                 /* Now start the PMU.  */
3042                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR_CTRL,
3043                                   env->msr_fixed_ctr_ctrl);
3044                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_CTRL,
3045                                   env->msr_global_ctrl);
3046             }
3047         }
3048         /*
3049          * Hyper-V partition-wide MSRs: to avoid clearing them on cpu hot-add,
3050          * only sync them to KVM on the first cpu
3051          */
3052         if (current_cpu == first_cpu) {
3053             if (has_msr_hv_hypercall) {
3054                 kvm_msr_entry_add(cpu, HV_X64_MSR_GUEST_OS_ID,
3055                                   env->msr_hv_guest_os_id);
3056                 kvm_msr_entry_add(cpu, HV_X64_MSR_HYPERCALL,
3057                                   env->msr_hv_hypercall);
3058             }
3059             if (hyperv_feat_enabled(cpu, HYPERV_FEAT_TIME)) {
3060                 kvm_msr_entry_add(cpu, HV_X64_MSR_REFERENCE_TSC,
3061                                   env->msr_hv_tsc);
3062             }
3063             if (hyperv_feat_enabled(cpu, HYPERV_FEAT_REENLIGHTENMENT)) {
3064                 kvm_msr_entry_add(cpu, HV_X64_MSR_REENLIGHTENMENT_CONTROL,
3065                                   env->msr_hv_reenlightenment_control);
3066                 kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_CONTROL,
3067                                   env->msr_hv_tsc_emulation_control);
3068                 kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_STATUS,
3069                                   env->msr_hv_tsc_emulation_status);
3070             }
3071         }
3072         if (hyperv_feat_enabled(cpu, HYPERV_FEAT_VAPIC)) {
3073             kvm_msr_entry_add(cpu, HV_X64_MSR_APIC_ASSIST_PAGE,
3074                               env->msr_hv_vapic);
3075         }
3076         if (has_msr_hv_crash) {
3077             int j;
3078 
3079             for (j = 0; j < HV_CRASH_PARAMS; j++)
3080                 kvm_msr_entry_add(cpu, HV_X64_MSR_CRASH_P0 + j,
3081                                   env->msr_hv_crash_params[j]);
3082 
3083             kvm_msr_entry_add(cpu, HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_NOTIFY);
3084         }
3085         if (has_msr_hv_runtime) {
3086             kvm_msr_entry_add(cpu, HV_X64_MSR_VP_RUNTIME, env->msr_hv_runtime);
3087         }
3088         if (hyperv_feat_enabled(cpu, HYPERV_FEAT_VPINDEX)
3089             && hv_vpindex_settable) {
3090             kvm_msr_entry_add(cpu, HV_X64_MSR_VP_INDEX,
3091                               hyperv_vp_index(CPU(cpu)));
3092         }
3093         if (hyperv_feat_enabled(cpu, HYPERV_FEAT_SYNIC)) {
3094             int j;
3095 
3096             kvm_msr_entry_add(cpu, HV_X64_MSR_SVERSION, HV_SYNIC_VERSION);
3097 
3098             kvm_msr_entry_add(cpu, HV_X64_MSR_SCONTROL,
3099                               env->msr_hv_synic_control);
3100             kvm_msr_entry_add(cpu, HV_X64_MSR_SIEFP,
3101                               env->msr_hv_synic_evt_page);
3102             kvm_msr_entry_add(cpu, HV_X64_MSR_SIMP,
3103                               env->msr_hv_synic_msg_page);
3104 
3105             for (j = 0; j < ARRAY_SIZE(env->msr_hv_synic_sint); j++) {
3106                 kvm_msr_entry_add(cpu, HV_X64_MSR_SINT0 + j,
3107                                   env->msr_hv_synic_sint[j]);
3108             }
3109         }
3110         if (has_msr_hv_stimer) {
3111             int j;
3112 
3113             for (j = 0; j < ARRAY_SIZE(env->msr_hv_stimer_config); j++) {
3114                 kvm_msr_entry_add(cpu, HV_X64_MSR_STIMER0_CONFIG + j * 2,
3115                                 env->msr_hv_stimer_config[j]);
3116             }
3117 
3118             for (j = 0; j < ARRAY_SIZE(env->msr_hv_stimer_count); j++) {
3119                 kvm_msr_entry_add(cpu, HV_X64_MSR_STIMER0_COUNT + j * 2,
3120                                 env->msr_hv_stimer_count[j]);
3121             }
3122         }
3123         if (env->features[FEAT_1_EDX] & CPUID_MTRR) {
3124             uint64_t phys_mask = MAKE_64BIT_MASK(0, cpu->phys_bits);
3125 
3126             kvm_msr_entry_add(cpu, MSR_MTRRdefType, env->mtrr_deftype);
3127             kvm_msr_entry_add(cpu, MSR_MTRRfix64K_00000, env->mtrr_fixed[0]);
3128             kvm_msr_entry_add(cpu, MSR_MTRRfix16K_80000, env->mtrr_fixed[1]);
3129             kvm_msr_entry_add(cpu, MSR_MTRRfix16K_A0000, env->mtrr_fixed[2]);
3130             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C0000, env->mtrr_fixed[3]);
3131             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C8000, env->mtrr_fixed[4]);
3132             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D0000, env->mtrr_fixed[5]);
3133             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D8000, env->mtrr_fixed[6]);
3134             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E0000, env->mtrr_fixed[7]);
3135             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E8000, env->mtrr_fixed[8]);
3136             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F0000, env->mtrr_fixed[9]);
3137             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F8000, env->mtrr_fixed[10]);
3138             for (i = 0; i < MSR_MTRRcap_VCNT; i++) {
3139                 /* The CPU GPs if we write to a bit above the physical limit of
3140                  * the host CPU (and KVM emulates that)
3141                  */
3142                 uint64_t mask = env->mtrr_var[i].mask;
3143                 mask &= phys_mask;
3144 
3145                 kvm_msr_entry_add(cpu, MSR_MTRRphysBase(i),
3146                                   env->mtrr_var[i].base);
3147                 kvm_msr_entry_add(cpu, MSR_MTRRphysMask(i), mask);
3148             }
3149         }
3150         if (env->features[FEAT_7_0_EBX] & CPUID_7_0_EBX_INTEL_PT) {
3151             int addr_num = kvm_arch_get_supported_cpuid(kvm_state,
3152                                                     0x14, 1, R_EAX) & 0x7;
3153 
3154             kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CTL,
3155                             env->msr_rtit_ctrl);
3156             kvm_msr_entry_add(cpu, MSR_IA32_RTIT_STATUS,
3157                             env->msr_rtit_status);
3158             kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_BASE,
3159                             env->msr_rtit_output_base);
3160             kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_MASK,
3161                             env->msr_rtit_output_mask);
3162             kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CR3_MATCH,
3163                             env->msr_rtit_cr3_match);
3164             for (i = 0; i < addr_num; i++) {
3165                 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_ADDR0_A + i,
3166                             env->msr_rtit_addrs[i]);
3167             }
3168         }
3169 
3170         if (env->features[FEAT_7_0_ECX] & CPUID_7_0_ECX_SGX_LC) {
3171             kvm_msr_entry_add(cpu, MSR_IA32_SGXLEPUBKEYHASH0,
3172                               env->msr_ia32_sgxlepubkeyhash[0]);
3173             kvm_msr_entry_add(cpu, MSR_IA32_SGXLEPUBKEYHASH1,
3174                               env->msr_ia32_sgxlepubkeyhash[1]);
3175             kvm_msr_entry_add(cpu, MSR_IA32_SGXLEPUBKEYHASH2,
3176                               env->msr_ia32_sgxlepubkeyhash[2]);
3177             kvm_msr_entry_add(cpu, MSR_IA32_SGXLEPUBKEYHASH3,
3178                               env->msr_ia32_sgxlepubkeyhash[3]);
3179         }
3180 
3181         /* Note: MSR_IA32_FEATURE_CONTROL is written separately, see
3182          *       kvm_put_msr_feature_control. */
3183     }
3184 
3185     if (env->mcg_cap) {
3186         int i;
3187 
3188         kvm_msr_entry_add(cpu, MSR_MCG_STATUS, env->mcg_status);
3189         kvm_msr_entry_add(cpu, MSR_MCG_CTL, env->mcg_ctl);
3190         if (has_msr_mcg_ext_ctl) {
3191             kvm_msr_entry_add(cpu, MSR_MCG_EXT_CTL, env->mcg_ext_ctl);
3192         }
3193         for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++) {
3194             kvm_msr_entry_add(cpu, MSR_MC0_CTL + i, env->mce_banks[i]);
3195         }
3196     }
3197 
3198     return kvm_buf_set_msrs(cpu);
3199 }
3200 
3201 
3202 static int kvm_get_fpu(X86CPU *cpu)
3203 {
3204     CPUX86State *env = &cpu->env;
3205     struct kvm_fpu fpu;
3206     int i, ret;
3207 
3208     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_FPU, &fpu);
3209     if (ret < 0) {
3210         return ret;
3211     }
3212 
3213     env->fpstt = (fpu.fsw >> 11) & 7;
3214     env->fpus = fpu.fsw;
3215     env->fpuc = fpu.fcw;
3216     env->fpop = fpu.last_opcode;
3217     env->fpip = fpu.last_ip;
3218     env->fpdp = fpu.last_dp;
3219     for (i = 0; i < 8; ++i) {
3220         env->fptags[i] = !((fpu.ftwx >> i) & 1);
3221     }
3222     memcpy(env->fpregs, fpu.fpr, sizeof env->fpregs);
3223     for (i = 0; i < CPU_NB_REGS; i++) {
3224         env->xmm_regs[i].ZMM_Q(0) = ldq_p(&fpu.xmm[i][0]);
3225         env->xmm_regs[i].ZMM_Q(1) = ldq_p(&fpu.xmm[i][8]);
3226     }
3227     env->mxcsr = fpu.mxcsr;
3228 
3229     return 0;
3230 }
3231 
3232 static int kvm_get_xsave(X86CPU *cpu)
3233 {
3234     CPUX86State *env = &cpu->env;
3235     void *xsave = env->xsave_buf;
3236     int ret;
3237 
3238     if (!has_xsave) {
3239         return kvm_get_fpu(cpu);
3240     }
3241 
3242     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_XSAVE, xsave);
3243     if (ret < 0) {
3244         return ret;
3245     }
3246     x86_cpu_xrstor_all_areas(cpu, xsave, env->xsave_buf_len);
3247 
3248     return 0;
3249 }
3250 
3251 static int kvm_get_xcrs(X86CPU *cpu)
3252 {
3253     CPUX86State *env = &cpu->env;
3254     int i, ret;
3255     struct kvm_xcrs xcrs;
3256 
3257     if (!has_xcrs) {
3258         return 0;
3259     }
3260 
3261     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_XCRS, &xcrs);
3262     if (ret < 0) {
3263         return ret;
3264     }
3265 
3266     for (i = 0; i < xcrs.nr_xcrs; i++) {
3267         /* Only support xcr0 now */
3268         if (xcrs.xcrs[i].xcr == 0) {
3269             env->xcr0 = xcrs.xcrs[i].value;
3270             break;
3271         }
3272     }
3273     return 0;
3274 }
3275 
3276 static int kvm_get_sregs(X86CPU *cpu)
3277 {
3278     CPUX86State *env = &cpu->env;
3279     struct kvm_sregs sregs;
3280     int bit, i, ret;
3281 
3282     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
3283     if (ret < 0) {
3284         return ret;
3285     }
3286 
3287     /* There can only be one pending IRQ set in the bitmap at a time, so try
3288        to find it and save its number instead (-1 for none). */
3289     env->interrupt_injected = -1;
3290     for (i = 0; i < ARRAY_SIZE(sregs.interrupt_bitmap); i++) {
3291         if (sregs.interrupt_bitmap[i]) {
3292             bit = ctz64(sregs.interrupt_bitmap[i]);
3293             env->interrupt_injected = i * 64 + bit;
3294             break;
3295         }
3296     }
3297 
3298     get_seg(&env->segs[R_CS], &sregs.cs);
3299     get_seg(&env->segs[R_DS], &sregs.ds);
3300     get_seg(&env->segs[R_ES], &sregs.es);
3301     get_seg(&env->segs[R_FS], &sregs.fs);
3302     get_seg(&env->segs[R_GS], &sregs.gs);
3303     get_seg(&env->segs[R_SS], &sregs.ss);
3304 
3305     get_seg(&env->tr, &sregs.tr);
3306     get_seg(&env->ldt, &sregs.ldt);
3307 
3308     env->idt.limit = sregs.idt.limit;
3309     env->idt.base = sregs.idt.base;
3310     env->gdt.limit = sregs.gdt.limit;
3311     env->gdt.base = sregs.gdt.base;
3312 
3313     env->cr[0] = sregs.cr0;
3314     env->cr[2] = sregs.cr2;
3315     env->cr[3] = sregs.cr3;
3316     env->cr[4] = sregs.cr4;
3317 
3318     env->efer = sregs.efer;
3319 
3320     /* changes to apic base and cr8/tpr are read back via kvm_arch_post_run */
3321     x86_update_hflags(env);
3322 
3323     return 0;
3324 }
3325 
3326 static int kvm_get_msrs(X86CPU *cpu)
3327 {
3328     CPUX86State *env = &cpu->env;
3329     struct kvm_msr_entry *msrs = cpu->kvm_msr_buf->entries;
3330     int ret, i;
3331     uint64_t mtrr_top_bits;
3332 
3333     kvm_msr_buf_reset(cpu);
3334 
3335     kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_CS, 0);
3336     kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_ESP, 0);
3337     kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_EIP, 0);
3338     kvm_msr_entry_add(cpu, MSR_PAT, 0);
3339     if (has_msr_star) {
3340         kvm_msr_entry_add(cpu, MSR_STAR, 0);
3341     }
3342     if (has_msr_hsave_pa) {
3343         kvm_msr_entry_add(cpu, MSR_VM_HSAVE_PA, 0);
3344     }
3345     if (has_msr_tsc_aux) {
3346         kvm_msr_entry_add(cpu, MSR_TSC_AUX, 0);
3347     }
3348     if (has_msr_tsc_adjust) {
3349         kvm_msr_entry_add(cpu, MSR_TSC_ADJUST, 0);
3350     }
3351     if (has_msr_tsc_deadline) {
3352         kvm_msr_entry_add(cpu, MSR_IA32_TSCDEADLINE, 0);
3353     }
3354     if (has_msr_misc_enable) {
3355         kvm_msr_entry_add(cpu, MSR_IA32_MISC_ENABLE, 0);
3356     }
3357     if (has_msr_smbase) {
3358         kvm_msr_entry_add(cpu, MSR_IA32_SMBASE, 0);
3359     }
3360     if (has_msr_smi_count) {
3361         kvm_msr_entry_add(cpu, MSR_SMI_COUNT, 0);
3362     }
3363     if (has_msr_feature_control) {
3364         kvm_msr_entry_add(cpu, MSR_IA32_FEATURE_CONTROL, 0);
3365     }
3366     if (has_msr_pkrs) {
3367         kvm_msr_entry_add(cpu, MSR_IA32_PKRS, 0);
3368     }
3369     if (has_msr_bndcfgs) {
3370         kvm_msr_entry_add(cpu, MSR_IA32_BNDCFGS, 0);
3371     }
3372     if (has_msr_xss) {
3373         kvm_msr_entry_add(cpu, MSR_IA32_XSS, 0);
3374     }
3375     if (has_msr_umwait) {
3376         kvm_msr_entry_add(cpu, MSR_IA32_UMWAIT_CONTROL, 0);
3377     }
3378     if (has_msr_spec_ctrl) {
3379         kvm_msr_entry_add(cpu, MSR_IA32_SPEC_CTRL, 0);
3380     }
3381     if (has_msr_tsx_ctrl) {
3382         kvm_msr_entry_add(cpu, MSR_IA32_TSX_CTRL, 0);
3383     }
3384     if (has_msr_virt_ssbd) {
3385         kvm_msr_entry_add(cpu, MSR_VIRT_SSBD, 0);
3386     }
3387     if (!env->tsc_valid) {
3388         kvm_msr_entry_add(cpu, MSR_IA32_TSC, 0);
3389         env->tsc_valid = !runstate_is_running();
3390     }
3391 
3392 #ifdef TARGET_X86_64
3393     if (lm_capable_kernel) {
3394         kvm_msr_entry_add(cpu, MSR_CSTAR, 0);
3395         kvm_msr_entry_add(cpu, MSR_KERNELGSBASE, 0);
3396         kvm_msr_entry_add(cpu, MSR_FMASK, 0);
3397         kvm_msr_entry_add(cpu, MSR_LSTAR, 0);
3398     }
3399 #endif
3400     kvm_msr_entry_add(cpu, MSR_KVM_SYSTEM_TIME, 0);
3401     kvm_msr_entry_add(cpu, MSR_KVM_WALL_CLOCK, 0);
3402     if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_ASYNC_PF_INT)) {
3403         kvm_msr_entry_add(cpu, MSR_KVM_ASYNC_PF_INT, 0);
3404     }
3405     if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_ASYNC_PF)) {
3406         kvm_msr_entry_add(cpu, MSR_KVM_ASYNC_PF_EN, 0);
3407     }
3408     if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_PV_EOI)) {
3409         kvm_msr_entry_add(cpu, MSR_KVM_PV_EOI_EN, 0);
3410     }
3411     if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_STEAL_TIME)) {
3412         kvm_msr_entry_add(cpu, MSR_KVM_STEAL_TIME, 0);
3413     }
3414     if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_POLL_CONTROL)) {
3415         kvm_msr_entry_add(cpu, MSR_KVM_POLL_CONTROL, 1);
3416     }
3417     if (has_architectural_pmu_version > 0) {
3418         if (has_architectural_pmu_version > 1) {
3419             kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR_CTRL, 0);
3420             kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_CTRL, 0);
3421             kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_STATUS, 0);
3422             kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_OVF_CTRL, 0);
3423         }
3424         for (i = 0; i < num_architectural_pmu_fixed_counters; i++) {
3425             kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR0 + i, 0);
3426         }
3427         for (i = 0; i < num_architectural_pmu_gp_counters; i++) {
3428             kvm_msr_entry_add(cpu, MSR_P6_PERFCTR0 + i, 0);
3429             kvm_msr_entry_add(cpu, MSR_P6_EVNTSEL0 + i, 0);
3430         }
3431     }
3432 
3433     if (env->mcg_cap) {
3434         kvm_msr_entry_add(cpu, MSR_MCG_STATUS, 0);
3435         kvm_msr_entry_add(cpu, MSR_MCG_CTL, 0);
3436         if (has_msr_mcg_ext_ctl) {
3437             kvm_msr_entry_add(cpu, MSR_MCG_EXT_CTL, 0);
3438         }
3439         for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++) {
3440             kvm_msr_entry_add(cpu, MSR_MC0_CTL + i, 0);
3441         }
3442     }
3443 
3444     if (has_msr_hv_hypercall) {
3445         kvm_msr_entry_add(cpu, HV_X64_MSR_HYPERCALL, 0);
3446         kvm_msr_entry_add(cpu, HV_X64_MSR_GUEST_OS_ID, 0);
3447     }
3448     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_VAPIC)) {
3449         kvm_msr_entry_add(cpu, HV_X64_MSR_APIC_ASSIST_PAGE, 0);
3450     }
3451     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_TIME)) {
3452         kvm_msr_entry_add(cpu, HV_X64_MSR_REFERENCE_TSC, 0);
3453     }
3454     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_REENLIGHTENMENT)) {
3455         kvm_msr_entry_add(cpu, HV_X64_MSR_REENLIGHTENMENT_CONTROL, 0);
3456         kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_CONTROL, 0);
3457         kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_STATUS, 0);
3458     }
3459     if (has_msr_hv_crash) {
3460         int j;
3461 
3462         for (j = 0; j < HV_CRASH_PARAMS; j++) {
3463             kvm_msr_entry_add(cpu, HV_X64_MSR_CRASH_P0 + j, 0);
3464         }
3465     }
3466     if (has_msr_hv_runtime) {
3467         kvm_msr_entry_add(cpu, HV_X64_MSR_VP_RUNTIME, 0);
3468     }
3469     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_SYNIC)) {
3470         uint32_t msr;
3471 
3472         kvm_msr_entry_add(cpu, HV_X64_MSR_SCONTROL, 0);
3473         kvm_msr_entry_add(cpu, HV_X64_MSR_SIEFP, 0);
3474         kvm_msr_entry_add(cpu, HV_X64_MSR_SIMP, 0);
3475         for (msr = HV_X64_MSR_SINT0; msr <= HV_X64_MSR_SINT15; msr++) {
3476             kvm_msr_entry_add(cpu, msr, 0);
3477         }
3478     }
3479     if (has_msr_hv_stimer) {
3480         uint32_t msr;
3481 
3482         for (msr = HV_X64_MSR_STIMER0_CONFIG; msr <= HV_X64_MSR_STIMER3_COUNT;
3483              msr++) {
3484             kvm_msr_entry_add(cpu, msr, 0);
3485         }
3486     }
3487     if (env->features[FEAT_1_EDX] & CPUID_MTRR) {
3488         kvm_msr_entry_add(cpu, MSR_MTRRdefType, 0);
3489         kvm_msr_entry_add(cpu, MSR_MTRRfix64K_00000, 0);
3490         kvm_msr_entry_add(cpu, MSR_MTRRfix16K_80000, 0);
3491         kvm_msr_entry_add(cpu, MSR_MTRRfix16K_A0000, 0);
3492         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C0000, 0);
3493         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C8000, 0);
3494         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D0000, 0);
3495         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D8000, 0);
3496         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E0000, 0);
3497         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E8000, 0);
3498         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F0000, 0);
3499         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F8000, 0);
3500         for (i = 0; i < MSR_MTRRcap_VCNT; i++) {
3501             kvm_msr_entry_add(cpu, MSR_MTRRphysBase(i), 0);
3502             kvm_msr_entry_add(cpu, MSR_MTRRphysMask(i), 0);
3503         }
3504     }
3505 
3506     if (env->features[FEAT_7_0_EBX] & CPUID_7_0_EBX_INTEL_PT) {
3507         int addr_num =
3508             kvm_arch_get_supported_cpuid(kvm_state, 0x14, 1, R_EAX) & 0x7;
3509 
3510         kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CTL, 0);
3511         kvm_msr_entry_add(cpu, MSR_IA32_RTIT_STATUS, 0);
3512         kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_BASE, 0);
3513         kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_MASK, 0);
3514         kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CR3_MATCH, 0);
3515         for (i = 0; i < addr_num; i++) {
3516             kvm_msr_entry_add(cpu, MSR_IA32_RTIT_ADDR0_A + i, 0);
3517         }
3518     }
3519 
3520     if (env->features[FEAT_7_0_ECX] & CPUID_7_0_ECX_SGX_LC) {
3521         kvm_msr_entry_add(cpu, MSR_IA32_SGXLEPUBKEYHASH0, 0);
3522         kvm_msr_entry_add(cpu, MSR_IA32_SGXLEPUBKEYHASH1, 0);
3523         kvm_msr_entry_add(cpu, MSR_IA32_SGXLEPUBKEYHASH2, 0);
3524         kvm_msr_entry_add(cpu, MSR_IA32_SGXLEPUBKEYHASH3, 0);
3525     }
3526 
3527     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MSRS, cpu->kvm_msr_buf);
3528     if (ret < 0) {
3529         return ret;
3530     }
3531 
3532     if (ret < cpu->kvm_msr_buf->nmsrs) {
3533         struct kvm_msr_entry *e = &cpu->kvm_msr_buf->entries[ret];
3534         error_report("error: failed to get MSR 0x%" PRIx32,
3535                      (uint32_t)e->index);
3536     }
3537 
3538     assert(ret == cpu->kvm_msr_buf->nmsrs);
3539     /*
3540      * MTRR masks: Each mask consists of 5 parts
3541      * a  10..0: must be zero
3542      * b  11   : valid bit
3543      * c n-1.12: actual mask bits
3544      * d  51..n: reserved must be zero
3545      * e  63.52: reserved must be zero
3546      *
3547      * 'n' is the number of physical bits supported by the CPU and is
3548      * apparently always <= 52.   We know our 'n' but don't know what
3549      * the destinations 'n' is; it might be smaller, in which case
3550      * it masks (c) on loading. It might be larger, in which case
3551      * we fill 'd' so that d..c is consistent irrespetive of the 'n'
3552      * we're migrating to.
3553      */
3554 
3555     if (cpu->fill_mtrr_mask) {
3556         QEMU_BUILD_BUG_ON(TARGET_PHYS_ADDR_SPACE_BITS > 52);
3557         assert(cpu->phys_bits <= TARGET_PHYS_ADDR_SPACE_BITS);
3558         mtrr_top_bits = MAKE_64BIT_MASK(cpu->phys_bits, 52 - cpu->phys_bits);
3559     } else {
3560         mtrr_top_bits = 0;
3561     }
3562 
3563     for (i = 0; i < ret; i++) {
3564         uint32_t index = msrs[i].index;
3565         switch (index) {
3566         case MSR_IA32_SYSENTER_CS:
3567             env->sysenter_cs = msrs[i].data;
3568             break;
3569         case MSR_IA32_SYSENTER_ESP:
3570             env->sysenter_esp = msrs[i].data;
3571             break;
3572         case MSR_IA32_SYSENTER_EIP:
3573             env->sysenter_eip = msrs[i].data;
3574             break;
3575         case MSR_PAT:
3576             env->pat = msrs[i].data;
3577             break;
3578         case MSR_STAR:
3579             env->star = msrs[i].data;
3580             break;
3581 #ifdef TARGET_X86_64
3582         case MSR_CSTAR:
3583             env->cstar = msrs[i].data;
3584             break;
3585         case MSR_KERNELGSBASE:
3586             env->kernelgsbase = msrs[i].data;
3587             break;
3588         case MSR_FMASK:
3589             env->fmask = msrs[i].data;
3590             break;
3591         case MSR_LSTAR:
3592             env->lstar = msrs[i].data;
3593             break;
3594 #endif
3595         case MSR_IA32_TSC:
3596             env->tsc = msrs[i].data;
3597             break;
3598         case MSR_TSC_AUX:
3599             env->tsc_aux = msrs[i].data;
3600             break;
3601         case MSR_TSC_ADJUST:
3602             env->tsc_adjust = msrs[i].data;
3603             break;
3604         case MSR_IA32_TSCDEADLINE:
3605             env->tsc_deadline = msrs[i].data;
3606             break;
3607         case MSR_VM_HSAVE_PA:
3608             env->vm_hsave = msrs[i].data;
3609             break;
3610         case MSR_KVM_SYSTEM_TIME:
3611             env->system_time_msr = msrs[i].data;
3612             break;
3613         case MSR_KVM_WALL_CLOCK:
3614             env->wall_clock_msr = msrs[i].data;
3615             break;
3616         case MSR_MCG_STATUS:
3617             env->mcg_status = msrs[i].data;
3618             break;
3619         case MSR_MCG_CTL:
3620             env->mcg_ctl = msrs[i].data;
3621             break;
3622         case MSR_MCG_EXT_CTL:
3623             env->mcg_ext_ctl = msrs[i].data;
3624             break;
3625         case MSR_IA32_MISC_ENABLE:
3626             env->msr_ia32_misc_enable = msrs[i].data;
3627             break;
3628         case MSR_IA32_SMBASE:
3629             env->smbase = msrs[i].data;
3630             break;
3631         case MSR_SMI_COUNT:
3632             env->msr_smi_count = msrs[i].data;
3633             break;
3634         case MSR_IA32_FEATURE_CONTROL:
3635             env->msr_ia32_feature_control = msrs[i].data;
3636             break;
3637         case MSR_IA32_BNDCFGS:
3638             env->msr_bndcfgs = msrs[i].data;
3639             break;
3640         case MSR_IA32_XSS:
3641             env->xss = msrs[i].data;
3642             break;
3643         case MSR_IA32_UMWAIT_CONTROL:
3644             env->umwait = msrs[i].data;
3645             break;
3646         case MSR_IA32_PKRS:
3647             env->pkrs = msrs[i].data;
3648             break;
3649         default:
3650             if (msrs[i].index >= MSR_MC0_CTL &&
3651                 msrs[i].index < MSR_MC0_CTL + (env->mcg_cap & 0xff) * 4) {
3652                 env->mce_banks[msrs[i].index - MSR_MC0_CTL] = msrs[i].data;
3653             }
3654             break;
3655         case MSR_KVM_ASYNC_PF_EN:
3656             env->async_pf_en_msr = msrs[i].data;
3657             break;
3658         case MSR_KVM_ASYNC_PF_INT:
3659             env->async_pf_int_msr = msrs[i].data;
3660             break;
3661         case MSR_KVM_PV_EOI_EN:
3662             env->pv_eoi_en_msr = msrs[i].data;
3663             break;
3664         case MSR_KVM_STEAL_TIME:
3665             env->steal_time_msr = msrs[i].data;
3666             break;
3667         case MSR_KVM_POLL_CONTROL: {
3668             env->poll_control_msr = msrs[i].data;
3669             break;
3670         }
3671         case MSR_CORE_PERF_FIXED_CTR_CTRL:
3672             env->msr_fixed_ctr_ctrl = msrs[i].data;
3673             break;
3674         case MSR_CORE_PERF_GLOBAL_CTRL:
3675             env->msr_global_ctrl = msrs[i].data;
3676             break;
3677         case MSR_CORE_PERF_GLOBAL_STATUS:
3678             env->msr_global_status = msrs[i].data;
3679             break;
3680         case MSR_CORE_PERF_GLOBAL_OVF_CTRL:
3681             env->msr_global_ovf_ctrl = msrs[i].data;
3682             break;
3683         case MSR_CORE_PERF_FIXED_CTR0 ... MSR_CORE_PERF_FIXED_CTR0 + MAX_FIXED_COUNTERS - 1:
3684             env->msr_fixed_counters[index - MSR_CORE_PERF_FIXED_CTR0] = msrs[i].data;
3685             break;
3686         case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR0 + MAX_GP_COUNTERS - 1:
3687             env->msr_gp_counters[index - MSR_P6_PERFCTR0] = msrs[i].data;
3688             break;
3689         case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL0 + MAX_GP_COUNTERS - 1:
3690             env->msr_gp_evtsel[index - MSR_P6_EVNTSEL0] = msrs[i].data;
3691             break;
3692         case HV_X64_MSR_HYPERCALL:
3693             env->msr_hv_hypercall = msrs[i].data;
3694             break;
3695         case HV_X64_MSR_GUEST_OS_ID:
3696             env->msr_hv_guest_os_id = msrs[i].data;
3697             break;
3698         case HV_X64_MSR_APIC_ASSIST_PAGE:
3699             env->msr_hv_vapic = msrs[i].data;
3700             break;
3701         case HV_X64_MSR_REFERENCE_TSC:
3702             env->msr_hv_tsc = msrs[i].data;
3703             break;
3704         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3705             env->msr_hv_crash_params[index - HV_X64_MSR_CRASH_P0] = msrs[i].data;
3706             break;
3707         case HV_X64_MSR_VP_RUNTIME:
3708             env->msr_hv_runtime = msrs[i].data;
3709             break;
3710         case HV_X64_MSR_SCONTROL:
3711             env->msr_hv_synic_control = msrs[i].data;
3712             break;
3713         case HV_X64_MSR_SIEFP:
3714             env->msr_hv_synic_evt_page = msrs[i].data;
3715             break;
3716         case HV_X64_MSR_SIMP:
3717             env->msr_hv_synic_msg_page = msrs[i].data;
3718             break;
3719         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
3720             env->msr_hv_synic_sint[index - HV_X64_MSR_SINT0] = msrs[i].data;
3721             break;
3722         case HV_X64_MSR_STIMER0_CONFIG:
3723         case HV_X64_MSR_STIMER1_CONFIG:
3724         case HV_X64_MSR_STIMER2_CONFIG:
3725         case HV_X64_MSR_STIMER3_CONFIG:
3726             env->msr_hv_stimer_config[(index - HV_X64_MSR_STIMER0_CONFIG)/2] =
3727                                 msrs[i].data;
3728             break;
3729         case HV_X64_MSR_STIMER0_COUNT:
3730         case HV_X64_MSR_STIMER1_COUNT:
3731         case HV_X64_MSR_STIMER2_COUNT:
3732         case HV_X64_MSR_STIMER3_COUNT:
3733             env->msr_hv_stimer_count[(index - HV_X64_MSR_STIMER0_COUNT)/2] =
3734                                 msrs[i].data;
3735             break;
3736         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3737             env->msr_hv_reenlightenment_control = msrs[i].data;
3738             break;
3739         case HV_X64_MSR_TSC_EMULATION_CONTROL:
3740             env->msr_hv_tsc_emulation_control = msrs[i].data;
3741             break;
3742         case HV_X64_MSR_TSC_EMULATION_STATUS:
3743             env->msr_hv_tsc_emulation_status = msrs[i].data;
3744             break;
3745         case MSR_MTRRdefType:
3746             env->mtrr_deftype = msrs[i].data;
3747             break;
3748         case MSR_MTRRfix64K_00000:
3749             env->mtrr_fixed[0] = msrs[i].data;
3750             break;
3751         case MSR_MTRRfix16K_80000:
3752             env->mtrr_fixed[1] = msrs[i].data;
3753             break;
3754         case MSR_MTRRfix16K_A0000:
3755             env->mtrr_fixed[2] = msrs[i].data;
3756             break;
3757         case MSR_MTRRfix4K_C0000:
3758             env->mtrr_fixed[3] = msrs[i].data;
3759             break;
3760         case MSR_MTRRfix4K_C8000:
3761             env->mtrr_fixed[4] = msrs[i].data;
3762             break;
3763         case MSR_MTRRfix4K_D0000:
3764             env->mtrr_fixed[5] = msrs[i].data;
3765             break;
3766         case MSR_MTRRfix4K_D8000:
3767             env->mtrr_fixed[6] = msrs[i].data;
3768             break;
3769         case MSR_MTRRfix4K_E0000:
3770             env->mtrr_fixed[7] = msrs[i].data;
3771             break;
3772         case MSR_MTRRfix4K_E8000:
3773             env->mtrr_fixed[8] = msrs[i].data;
3774             break;
3775         case MSR_MTRRfix4K_F0000:
3776             env->mtrr_fixed[9] = msrs[i].data;
3777             break;
3778         case MSR_MTRRfix4K_F8000:
3779             env->mtrr_fixed[10] = msrs[i].data;
3780             break;
3781         case MSR_MTRRphysBase(0) ... MSR_MTRRphysMask(MSR_MTRRcap_VCNT - 1):
3782             if (index & 1) {
3783                 env->mtrr_var[MSR_MTRRphysIndex(index)].mask = msrs[i].data |
3784                                                                mtrr_top_bits;
3785             } else {
3786                 env->mtrr_var[MSR_MTRRphysIndex(index)].base = msrs[i].data;
3787             }
3788             break;
3789         case MSR_IA32_SPEC_CTRL:
3790             env->spec_ctrl = msrs[i].data;
3791             break;
3792         case MSR_IA32_TSX_CTRL:
3793             env->tsx_ctrl = msrs[i].data;
3794             break;
3795         case MSR_VIRT_SSBD:
3796             env->virt_ssbd = msrs[i].data;
3797             break;
3798         case MSR_IA32_RTIT_CTL:
3799             env->msr_rtit_ctrl = msrs[i].data;
3800             break;
3801         case MSR_IA32_RTIT_STATUS:
3802             env->msr_rtit_status = msrs[i].data;
3803             break;
3804         case MSR_IA32_RTIT_OUTPUT_BASE:
3805             env->msr_rtit_output_base = msrs[i].data;
3806             break;
3807         case MSR_IA32_RTIT_OUTPUT_MASK:
3808             env->msr_rtit_output_mask = msrs[i].data;
3809             break;
3810         case MSR_IA32_RTIT_CR3_MATCH:
3811             env->msr_rtit_cr3_match = msrs[i].data;
3812             break;
3813         case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
3814             env->msr_rtit_addrs[index - MSR_IA32_RTIT_ADDR0_A] = msrs[i].data;
3815             break;
3816         case MSR_IA32_SGXLEPUBKEYHASH0 ... MSR_IA32_SGXLEPUBKEYHASH3:
3817             env->msr_ia32_sgxlepubkeyhash[index - MSR_IA32_SGXLEPUBKEYHASH0] =
3818                            msrs[i].data;
3819             break;
3820         }
3821     }
3822 
3823     return 0;
3824 }
3825 
3826 static int kvm_put_mp_state(X86CPU *cpu)
3827 {
3828     struct kvm_mp_state mp_state = { .mp_state = cpu->env.mp_state };
3829 
3830     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
3831 }
3832 
3833 static int kvm_get_mp_state(X86CPU *cpu)
3834 {
3835     CPUState *cs = CPU(cpu);
3836     CPUX86State *env = &cpu->env;
3837     struct kvm_mp_state mp_state;
3838     int ret;
3839 
3840     ret = kvm_vcpu_ioctl(cs, KVM_GET_MP_STATE, &mp_state);
3841     if (ret < 0) {
3842         return ret;
3843     }
3844     env->mp_state = mp_state.mp_state;
3845     if (kvm_irqchip_in_kernel()) {
3846         cs->halted = (mp_state.mp_state == KVM_MP_STATE_HALTED);
3847     }
3848     return 0;
3849 }
3850 
3851 static int kvm_get_apic(X86CPU *cpu)
3852 {
3853     DeviceState *apic = cpu->apic_state;
3854     struct kvm_lapic_state kapic;
3855     int ret;
3856 
3857     if (apic && kvm_irqchip_in_kernel()) {
3858         ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_LAPIC, &kapic);
3859         if (ret < 0) {
3860             return ret;
3861         }
3862 
3863         kvm_get_apic_state(apic, &kapic);
3864     }
3865     return 0;
3866 }
3867 
3868 static int kvm_put_vcpu_events(X86CPU *cpu, int level)
3869 {
3870     CPUState *cs = CPU(cpu);
3871     CPUX86State *env = &cpu->env;
3872     struct kvm_vcpu_events events = {};
3873 
3874     if (!kvm_has_vcpu_events()) {
3875         return 0;
3876     }
3877 
3878     events.flags = 0;
3879 
3880     if (has_exception_payload) {
3881         events.flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3882         events.exception.pending = env->exception_pending;
3883         events.exception_has_payload = env->exception_has_payload;
3884         events.exception_payload = env->exception_payload;
3885     }
3886     events.exception.nr = env->exception_nr;
3887     events.exception.injected = env->exception_injected;
3888     events.exception.has_error_code = env->has_error_code;
3889     events.exception.error_code = env->error_code;
3890 
3891     events.interrupt.injected = (env->interrupt_injected >= 0);
3892     events.interrupt.nr = env->interrupt_injected;
3893     events.interrupt.soft = env->soft_interrupt;
3894 
3895     events.nmi.injected = env->nmi_injected;
3896     events.nmi.pending = env->nmi_pending;
3897     events.nmi.masked = !!(env->hflags2 & HF2_NMI_MASK);
3898 
3899     events.sipi_vector = env->sipi_vector;
3900 
3901     if (has_msr_smbase) {
3902         events.smi.smm = !!(env->hflags & HF_SMM_MASK);
3903         events.smi.smm_inside_nmi = !!(env->hflags2 & HF2_SMM_INSIDE_NMI_MASK);
3904         if (kvm_irqchip_in_kernel()) {
3905             /* As soon as these are moved to the kernel, remove them
3906              * from cs->interrupt_request.
3907              */
3908             events.smi.pending = cs->interrupt_request & CPU_INTERRUPT_SMI;
3909             events.smi.latched_init = cs->interrupt_request & CPU_INTERRUPT_INIT;
3910             cs->interrupt_request &= ~(CPU_INTERRUPT_INIT | CPU_INTERRUPT_SMI);
3911         } else {
3912             /* Keep these in cs->interrupt_request.  */
3913             events.smi.pending = 0;
3914             events.smi.latched_init = 0;
3915         }
3916         /* Stop SMI delivery on old machine types to avoid a reboot
3917          * on an inward migration of an old VM.
3918          */
3919         if (!cpu->kvm_no_smi_migration) {
3920             events.flags |= KVM_VCPUEVENT_VALID_SMM;
3921         }
3922     }
3923 
3924     if (level >= KVM_PUT_RESET_STATE) {
3925         events.flags |= KVM_VCPUEVENT_VALID_NMI_PENDING;
3926         if (env->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
3927             events.flags |= KVM_VCPUEVENT_VALID_SIPI_VECTOR;
3928         }
3929     }
3930 
3931     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_VCPU_EVENTS, &events);
3932 }
3933 
3934 static int kvm_get_vcpu_events(X86CPU *cpu)
3935 {
3936     CPUX86State *env = &cpu->env;
3937     struct kvm_vcpu_events events;
3938     int ret;
3939 
3940     if (!kvm_has_vcpu_events()) {
3941         return 0;
3942     }
3943 
3944     memset(&events, 0, sizeof(events));
3945     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_VCPU_EVENTS, &events);
3946     if (ret < 0) {
3947        return ret;
3948     }
3949 
3950     if (events.flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3951         env->exception_pending = events.exception.pending;
3952         env->exception_has_payload = events.exception_has_payload;
3953         env->exception_payload = events.exception_payload;
3954     } else {
3955         env->exception_pending = 0;
3956         env->exception_has_payload = false;
3957     }
3958     env->exception_injected = events.exception.injected;
3959     env->exception_nr =
3960         (env->exception_pending || env->exception_injected) ?
3961         events.exception.nr : -1;
3962     env->has_error_code = events.exception.has_error_code;
3963     env->error_code = events.exception.error_code;
3964 
3965     env->interrupt_injected =
3966         events.interrupt.injected ? events.interrupt.nr : -1;
3967     env->soft_interrupt = events.interrupt.soft;
3968 
3969     env->nmi_injected = events.nmi.injected;
3970     env->nmi_pending = events.nmi.pending;
3971     if (events.nmi.masked) {
3972         env->hflags2 |= HF2_NMI_MASK;
3973     } else {
3974         env->hflags2 &= ~HF2_NMI_MASK;
3975     }
3976 
3977     if (events.flags & KVM_VCPUEVENT_VALID_SMM) {
3978         if (events.smi.smm) {
3979             env->hflags |= HF_SMM_MASK;
3980         } else {
3981             env->hflags &= ~HF_SMM_MASK;
3982         }
3983         if (events.smi.pending) {
3984             cpu_interrupt(CPU(cpu), CPU_INTERRUPT_SMI);
3985         } else {
3986             cpu_reset_interrupt(CPU(cpu), CPU_INTERRUPT_SMI);
3987         }
3988         if (events.smi.smm_inside_nmi) {
3989             env->hflags2 |= HF2_SMM_INSIDE_NMI_MASK;
3990         } else {
3991             env->hflags2 &= ~HF2_SMM_INSIDE_NMI_MASK;
3992         }
3993         if (events.smi.latched_init) {
3994             cpu_interrupt(CPU(cpu), CPU_INTERRUPT_INIT);
3995         } else {
3996             cpu_reset_interrupt(CPU(cpu), CPU_INTERRUPT_INIT);
3997         }
3998     }
3999 
4000     env->sipi_vector = events.sipi_vector;
4001 
4002     return 0;
4003 }
4004 
4005 static int kvm_guest_debug_workarounds(X86CPU *cpu)
4006 {
4007     CPUState *cs = CPU(cpu);
4008     CPUX86State *env = &cpu->env;
4009     int ret = 0;
4010     unsigned long reinject_trap = 0;
4011 
4012     if (!kvm_has_vcpu_events()) {
4013         if (env->exception_nr == EXCP01_DB) {
4014             reinject_trap = KVM_GUESTDBG_INJECT_DB;
4015         } else if (env->exception_injected == EXCP03_INT3) {
4016             reinject_trap = KVM_GUESTDBG_INJECT_BP;
4017         }
4018         kvm_reset_exception(env);
4019     }
4020 
4021     /*
4022      * Kernels before KVM_CAP_X86_ROBUST_SINGLESTEP overwrote flags.TF
4023      * injected via SET_GUEST_DEBUG while updating GP regs. Work around this
4024      * by updating the debug state once again if single-stepping is on.
4025      * Another reason to call kvm_update_guest_debug here is a pending debug
4026      * trap raise by the guest. On kernels without SET_VCPU_EVENTS we have to
4027      * reinject them via SET_GUEST_DEBUG.
4028      */
4029     if (reinject_trap ||
4030         (!kvm_has_robust_singlestep() && cs->singlestep_enabled)) {
4031         ret = kvm_update_guest_debug(cs, reinject_trap);
4032     }
4033     return ret;
4034 }
4035 
4036 static int kvm_put_debugregs(X86CPU *cpu)
4037 {
4038     CPUX86State *env = &cpu->env;
4039     struct kvm_debugregs dbgregs;
4040     int i;
4041 
4042     if (!kvm_has_debugregs()) {
4043         return 0;
4044     }
4045 
4046     memset(&dbgregs, 0, sizeof(dbgregs));
4047     for (i = 0; i < 4; i++) {
4048         dbgregs.db[i] = env->dr[i];
4049     }
4050     dbgregs.dr6 = env->dr[6];
4051     dbgregs.dr7 = env->dr[7];
4052     dbgregs.flags = 0;
4053 
4054     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_DEBUGREGS, &dbgregs);
4055 }
4056 
4057 static int kvm_get_debugregs(X86CPU *cpu)
4058 {
4059     CPUX86State *env = &cpu->env;
4060     struct kvm_debugregs dbgregs;
4061     int i, ret;
4062 
4063     if (!kvm_has_debugregs()) {
4064         return 0;
4065     }
4066 
4067     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_DEBUGREGS, &dbgregs);
4068     if (ret < 0) {
4069         return ret;
4070     }
4071     for (i = 0; i < 4; i++) {
4072         env->dr[i] = dbgregs.db[i];
4073     }
4074     env->dr[4] = env->dr[6] = dbgregs.dr6;
4075     env->dr[5] = env->dr[7] = dbgregs.dr7;
4076 
4077     return 0;
4078 }
4079 
4080 static int kvm_put_nested_state(X86CPU *cpu)
4081 {
4082     CPUX86State *env = &cpu->env;
4083     int max_nested_state_len = kvm_max_nested_state_length();
4084 
4085     if (!env->nested_state) {
4086         return 0;
4087     }
4088 
4089     /*
4090      * Copy flags that are affected by reset from env->hflags and env->hflags2.
4091      */
4092     if (env->hflags & HF_GUEST_MASK) {
4093         env->nested_state->flags |= KVM_STATE_NESTED_GUEST_MODE;
4094     } else {
4095         env->nested_state->flags &= ~KVM_STATE_NESTED_GUEST_MODE;
4096     }
4097 
4098     /* Don't set KVM_STATE_NESTED_GIF_SET on VMX as it is illegal */
4099     if (cpu_has_svm(env) && (env->hflags2 & HF2_GIF_MASK)) {
4100         env->nested_state->flags |= KVM_STATE_NESTED_GIF_SET;
4101     } else {
4102         env->nested_state->flags &= ~KVM_STATE_NESTED_GIF_SET;
4103     }
4104 
4105     assert(env->nested_state->size <= max_nested_state_len);
4106     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_NESTED_STATE, env->nested_state);
4107 }
4108 
4109 static int kvm_get_nested_state(X86CPU *cpu)
4110 {
4111     CPUX86State *env = &cpu->env;
4112     int max_nested_state_len = kvm_max_nested_state_length();
4113     int ret;
4114 
4115     if (!env->nested_state) {
4116         return 0;
4117     }
4118 
4119     /*
4120      * It is possible that migration restored a smaller size into
4121      * nested_state->hdr.size than what our kernel support.
4122      * We preserve migration origin nested_state->hdr.size for
4123      * call to KVM_SET_NESTED_STATE but wish that our next call
4124      * to KVM_GET_NESTED_STATE will use max size our kernel support.
4125      */
4126     env->nested_state->size = max_nested_state_len;
4127 
4128     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_NESTED_STATE, env->nested_state);
4129     if (ret < 0) {
4130         return ret;
4131     }
4132 
4133     /*
4134      * Copy flags that are affected by reset to env->hflags and env->hflags2.
4135      */
4136     if (env->nested_state->flags & KVM_STATE_NESTED_GUEST_MODE) {
4137         env->hflags |= HF_GUEST_MASK;
4138     } else {
4139         env->hflags &= ~HF_GUEST_MASK;
4140     }
4141 
4142     /* Keep HF2_GIF_MASK set on !SVM as x86_cpu_pending_interrupt() needs it */
4143     if (cpu_has_svm(env)) {
4144         if (env->nested_state->flags & KVM_STATE_NESTED_GIF_SET) {
4145             env->hflags2 |= HF2_GIF_MASK;
4146         } else {
4147             env->hflags2 &= ~HF2_GIF_MASK;
4148         }
4149     }
4150 
4151     return ret;
4152 }
4153 
4154 int kvm_arch_put_registers(CPUState *cpu, int level)
4155 {
4156     X86CPU *x86_cpu = X86_CPU(cpu);
4157     int ret;
4158 
4159     assert(cpu_is_stopped(cpu) || qemu_cpu_is_self(cpu));
4160 
4161     /* must be before kvm_put_nested_state so that EFER.SVME is set */
4162     ret = kvm_put_sregs(x86_cpu);
4163     if (ret < 0) {
4164         return ret;
4165     }
4166 
4167     if (level >= KVM_PUT_RESET_STATE) {
4168         ret = kvm_put_nested_state(x86_cpu);
4169         if (ret < 0) {
4170             return ret;
4171         }
4172 
4173         ret = kvm_put_msr_feature_control(x86_cpu);
4174         if (ret < 0) {
4175             return ret;
4176         }
4177     }
4178 
4179     if (level == KVM_PUT_FULL_STATE) {
4180         /* We don't check for kvm_arch_set_tsc_khz() errors here,
4181          * because TSC frequency mismatch shouldn't abort migration,
4182          * unless the user explicitly asked for a more strict TSC
4183          * setting (e.g. using an explicit "tsc-freq" option).
4184          */
4185         kvm_arch_set_tsc_khz(cpu);
4186     }
4187 
4188     ret = kvm_getput_regs(x86_cpu, 1);
4189     if (ret < 0) {
4190         return ret;
4191     }
4192     ret = kvm_put_xsave(x86_cpu);
4193     if (ret < 0) {
4194         return ret;
4195     }
4196     ret = kvm_put_xcrs(x86_cpu);
4197     if (ret < 0) {
4198         return ret;
4199     }
4200     /* must be before kvm_put_msrs */
4201     ret = kvm_inject_mce_oldstyle(x86_cpu);
4202     if (ret < 0) {
4203         return ret;
4204     }
4205     ret = kvm_put_msrs(x86_cpu, level);
4206     if (ret < 0) {
4207         return ret;
4208     }
4209     ret = kvm_put_vcpu_events(x86_cpu, level);
4210     if (ret < 0) {
4211         return ret;
4212     }
4213     if (level >= KVM_PUT_RESET_STATE) {
4214         ret = kvm_put_mp_state(x86_cpu);
4215         if (ret < 0) {
4216             return ret;
4217         }
4218     }
4219 
4220     ret = kvm_put_tscdeadline_msr(x86_cpu);
4221     if (ret < 0) {
4222         return ret;
4223     }
4224     ret = kvm_put_debugregs(x86_cpu);
4225     if (ret < 0) {
4226         return ret;
4227     }
4228     /* must be last */
4229     ret = kvm_guest_debug_workarounds(x86_cpu);
4230     if (ret < 0) {
4231         return ret;
4232     }
4233     return 0;
4234 }
4235 
4236 int kvm_arch_get_registers(CPUState *cs)
4237 {
4238     X86CPU *cpu = X86_CPU(cs);
4239     int ret;
4240 
4241     assert(cpu_is_stopped(cs) || qemu_cpu_is_self(cs));
4242 
4243     ret = kvm_get_vcpu_events(cpu);
4244     if (ret < 0) {
4245         goto out;
4246     }
4247     /*
4248      * KVM_GET_MPSTATE can modify CS and RIP, call it before
4249      * KVM_GET_REGS and KVM_GET_SREGS.
4250      */
4251     ret = kvm_get_mp_state(cpu);
4252     if (ret < 0) {
4253         goto out;
4254     }
4255     ret = kvm_getput_regs(cpu, 0);
4256     if (ret < 0) {
4257         goto out;
4258     }
4259     ret = kvm_get_xsave(cpu);
4260     if (ret < 0) {
4261         goto out;
4262     }
4263     ret = kvm_get_xcrs(cpu);
4264     if (ret < 0) {
4265         goto out;
4266     }
4267     ret = kvm_get_sregs(cpu);
4268     if (ret < 0) {
4269         goto out;
4270     }
4271     ret = kvm_get_msrs(cpu);
4272     if (ret < 0) {
4273         goto out;
4274     }
4275     ret = kvm_get_apic(cpu);
4276     if (ret < 0) {
4277         goto out;
4278     }
4279     ret = kvm_get_debugregs(cpu);
4280     if (ret < 0) {
4281         goto out;
4282     }
4283     ret = kvm_get_nested_state(cpu);
4284     if (ret < 0) {
4285         goto out;
4286     }
4287     ret = 0;
4288  out:
4289     cpu_sync_bndcs_hflags(&cpu->env);
4290     return ret;
4291 }
4292 
4293 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
4294 {
4295     X86CPU *x86_cpu = X86_CPU(cpu);
4296     CPUX86State *env = &x86_cpu->env;
4297     int ret;
4298 
4299     /* Inject NMI */
4300     if (cpu->interrupt_request & (CPU_INTERRUPT_NMI | CPU_INTERRUPT_SMI)) {
4301         if (cpu->interrupt_request & CPU_INTERRUPT_NMI) {
4302             qemu_mutex_lock_iothread();
4303             cpu->interrupt_request &= ~CPU_INTERRUPT_NMI;
4304             qemu_mutex_unlock_iothread();
4305             DPRINTF("injected NMI\n");
4306             ret = kvm_vcpu_ioctl(cpu, KVM_NMI);
4307             if (ret < 0) {
4308                 fprintf(stderr, "KVM: injection failed, NMI lost (%s)\n",
4309                         strerror(-ret));
4310             }
4311         }
4312         if (cpu->interrupt_request & CPU_INTERRUPT_SMI) {
4313             qemu_mutex_lock_iothread();
4314             cpu->interrupt_request &= ~CPU_INTERRUPT_SMI;
4315             qemu_mutex_unlock_iothread();
4316             DPRINTF("injected SMI\n");
4317             ret = kvm_vcpu_ioctl(cpu, KVM_SMI);
4318             if (ret < 0) {
4319                 fprintf(stderr, "KVM: injection failed, SMI lost (%s)\n",
4320                         strerror(-ret));
4321             }
4322         }
4323     }
4324 
4325     if (!kvm_pic_in_kernel()) {
4326         qemu_mutex_lock_iothread();
4327     }
4328 
4329     /* Force the VCPU out of its inner loop to process any INIT requests
4330      * or (for userspace APIC, but it is cheap to combine the checks here)
4331      * pending TPR access reports.
4332      */
4333     if (cpu->interrupt_request & (CPU_INTERRUPT_INIT | CPU_INTERRUPT_TPR)) {
4334         if ((cpu->interrupt_request & CPU_INTERRUPT_INIT) &&
4335             !(env->hflags & HF_SMM_MASK)) {
4336             cpu->exit_request = 1;
4337         }
4338         if (cpu->interrupt_request & CPU_INTERRUPT_TPR) {
4339             cpu->exit_request = 1;
4340         }
4341     }
4342 
4343     if (!kvm_pic_in_kernel()) {
4344         /* Try to inject an interrupt if the guest can accept it */
4345         if (run->ready_for_interrupt_injection &&
4346             (cpu->interrupt_request & CPU_INTERRUPT_HARD) &&
4347             (env->eflags & IF_MASK)) {
4348             int irq;
4349 
4350             cpu->interrupt_request &= ~CPU_INTERRUPT_HARD;
4351             irq = cpu_get_pic_interrupt(env);
4352             if (irq >= 0) {
4353                 struct kvm_interrupt intr;
4354 
4355                 intr.irq = irq;
4356                 DPRINTF("injected interrupt %d\n", irq);
4357                 ret = kvm_vcpu_ioctl(cpu, KVM_INTERRUPT, &intr);
4358                 if (ret < 0) {
4359                     fprintf(stderr,
4360                             "KVM: injection failed, interrupt lost (%s)\n",
4361                             strerror(-ret));
4362                 }
4363             }
4364         }
4365 
4366         /* If we have an interrupt but the guest is not ready to receive an
4367          * interrupt, request an interrupt window exit.  This will
4368          * cause a return to userspace as soon as the guest is ready to
4369          * receive interrupts. */
4370         if ((cpu->interrupt_request & CPU_INTERRUPT_HARD)) {
4371             run->request_interrupt_window = 1;
4372         } else {
4373             run->request_interrupt_window = 0;
4374         }
4375 
4376         DPRINTF("setting tpr\n");
4377         run->cr8 = cpu_get_apic_tpr(x86_cpu->apic_state);
4378 
4379         qemu_mutex_unlock_iothread();
4380     }
4381 }
4382 
4383 static void kvm_rate_limit_on_bus_lock(void)
4384 {
4385     uint64_t delay_ns = ratelimit_calculate_delay(&bus_lock_ratelimit_ctrl, 1);
4386 
4387     if (delay_ns) {
4388         g_usleep(delay_ns / SCALE_US);
4389     }
4390 }
4391 
4392 MemTxAttrs kvm_arch_post_run(CPUState *cpu, struct kvm_run *run)
4393 {
4394     X86CPU *x86_cpu = X86_CPU(cpu);
4395     CPUX86State *env = &x86_cpu->env;
4396 
4397     if (run->flags & KVM_RUN_X86_SMM) {
4398         env->hflags |= HF_SMM_MASK;
4399     } else {
4400         env->hflags &= ~HF_SMM_MASK;
4401     }
4402     if (run->if_flag) {
4403         env->eflags |= IF_MASK;
4404     } else {
4405         env->eflags &= ~IF_MASK;
4406     }
4407     if (run->flags & KVM_RUN_X86_BUS_LOCK) {
4408         kvm_rate_limit_on_bus_lock();
4409     }
4410 
4411     /* We need to protect the apic state against concurrent accesses from
4412      * different threads in case the userspace irqchip is used. */
4413     if (!kvm_irqchip_in_kernel()) {
4414         qemu_mutex_lock_iothread();
4415     }
4416     cpu_set_apic_tpr(x86_cpu->apic_state, run->cr8);
4417     cpu_set_apic_base(x86_cpu->apic_state, run->apic_base);
4418     if (!kvm_irqchip_in_kernel()) {
4419         qemu_mutex_unlock_iothread();
4420     }
4421     return cpu_get_mem_attrs(env);
4422 }
4423 
4424 int kvm_arch_process_async_events(CPUState *cs)
4425 {
4426     X86CPU *cpu = X86_CPU(cs);
4427     CPUX86State *env = &cpu->env;
4428 
4429     if (cs->interrupt_request & CPU_INTERRUPT_MCE) {
4430         /* We must not raise CPU_INTERRUPT_MCE if it's not supported. */
4431         assert(env->mcg_cap);
4432 
4433         cs->interrupt_request &= ~CPU_INTERRUPT_MCE;
4434 
4435         kvm_cpu_synchronize_state(cs);
4436 
4437         if (env->exception_nr == EXCP08_DBLE) {
4438             /* this means triple fault */
4439             qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET);
4440             cs->exit_request = 1;
4441             return 0;
4442         }
4443         kvm_queue_exception(env, EXCP12_MCHK, 0, 0);
4444         env->has_error_code = 0;
4445 
4446         cs->halted = 0;
4447         if (kvm_irqchip_in_kernel() && env->mp_state == KVM_MP_STATE_HALTED) {
4448             env->mp_state = KVM_MP_STATE_RUNNABLE;
4449         }
4450     }
4451 
4452     if ((cs->interrupt_request & CPU_INTERRUPT_INIT) &&
4453         !(env->hflags & HF_SMM_MASK)) {
4454         kvm_cpu_synchronize_state(cs);
4455         do_cpu_init(cpu);
4456     }
4457 
4458     if (kvm_irqchip_in_kernel()) {
4459         return 0;
4460     }
4461 
4462     if (cs->interrupt_request & CPU_INTERRUPT_POLL) {
4463         cs->interrupt_request &= ~CPU_INTERRUPT_POLL;
4464         apic_poll_irq(cpu->apic_state);
4465     }
4466     if (((cs->interrupt_request & CPU_INTERRUPT_HARD) &&
4467          (env->eflags & IF_MASK)) ||
4468         (cs->interrupt_request & CPU_INTERRUPT_NMI)) {
4469         cs->halted = 0;
4470     }
4471     if (cs->interrupt_request & CPU_INTERRUPT_SIPI) {
4472         kvm_cpu_synchronize_state(cs);
4473         do_cpu_sipi(cpu);
4474     }
4475     if (cs->interrupt_request & CPU_INTERRUPT_TPR) {
4476         cs->interrupt_request &= ~CPU_INTERRUPT_TPR;
4477         kvm_cpu_synchronize_state(cs);
4478         apic_handle_tpr_access_report(cpu->apic_state, env->eip,
4479                                       env->tpr_access_type);
4480     }
4481 
4482     return cs->halted;
4483 }
4484 
4485 static int kvm_handle_halt(X86CPU *cpu)
4486 {
4487     CPUState *cs = CPU(cpu);
4488     CPUX86State *env = &cpu->env;
4489 
4490     if (!((cs->interrupt_request & CPU_INTERRUPT_HARD) &&
4491           (env->eflags & IF_MASK)) &&
4492         !(cs->interrupt_request & CPU_INTERRUPT_NMI)) {
4493         cs->halted = 1;
4494         return EXCP_HLT;
4495     }
4496 
4497     return 0;
4498 }
4499 
4500 static int kvm_handle_tpr_access(X86CPU *cpu)
4501 {
4502     CPUState *cs = CPU(cpu);
4503     struct kvm_run *run = cs->kvm_run;
4504 
4505     apic_handle_tpr_access_report(cpu->apic_state, run->tpr_access.rip,
4506                                   run->tpr_access.is_write ? TPR_ACCESS_WRITE
4507                                                            : TPR_ACCESS_READ);
4508     return 1;
4509 }
4510 
4511 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
4512 {
4513     static const uint8_t int3 = 0xcc;
4514 
4515     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 1, 0) ||
4516         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&int3, 1, 1)) {
4517         return -EINVAL;
4518     }
4519     return 0;
4520 }
4521 
4522 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
4523 {
4524     uint8_t int3;
4525 
4526     if (cpu_memory_rw_debug(cs, bp->pc, &int3, 1, 0)) {
4527         return -EINVAL;
4528     }
4529     if (int3 != 0xcc) {
4530         return 0;
4531     }
4532     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 1, 1)) {
4533         return -EINVAL;
4534     }
4535     return 0;
4536 }
4537 
4538 static struct {
4539     target_ulong addr;
4540     int len;
4541     int type;
4542 } hw_breakpoint[4];
4543 
4544 static int nb_hw_breakpoint;
4545 
4546 static int find_hw_breakpoint(target_ulong addr, int len, int type)
4547 {
4548     int n;
4549 
4550     for (n = 0; n < nb_hw_breakpoint; n++) {
4551         if (hw_breakpoint[n].addr == addr && hw_breakpoint[n].type == type &&
4552             (hw_breakpoint[n].len == len || len == -1)) {
4553             return n;
4554         }
4555     }
4556     return -1;
4557 }
4558 
4559 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
4560                                   target_ulong len, int type)
4561 {
4562     switch (type) {
4563     case GDB_BREAKPOINT_HW:
4564         len = 1;
4565         break;
4566     case GDB_WATCHPOINT_WRITE:
4567     case GDB_WATCHPOINT_ACCESS:
4568         switch (len) {
4569         case 1:
4570             break;
4571         case 2:
4572         case 4:
4573         case 8:
4574             if (addr & (len - 1)) {
4575                 return -EINVAL;
4576             }
4577             break;
4578         default:
4579             return -EINVAL;
4580         }
4581         break;
4582     default:
4583         return -ENOSYS;
4584     }
4585 
4586     if (nb_hw_breakpoint == 4) {
4587         return -ENOBUFS;
4588     }
4589     if (find_hw_breakpoint(addr, len, type) >= 0) {
4590         return -EEXIST;
4591     }
4592     hw_breakpoint[nb_hw_breakpoint].addr = addr;
4593     hw_breakpoint[nb_hw_breakpoint].len = len;
4594     hw_breakpoint[nb_hw_breakpoint].type = type;
4595     nb_hw_breakpoint++;
4596 
4597     return 0;
4598 }
4599 
4600 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
4601                                   target_ulong len, int type)
4602 {
4603     int n;
4604 
4605     n = find_hw_breakpoint(addr, (type == GDB_BREAKPOINT_HW) ? 1 : len, type);
4606     if (n < 0) {
4607         return -ENOENT;
4608     }
4609     nb_hw_breakpoint--;
4610     hw_breakpoint[n] = hw_breakpoint[nb_hw_breakpoint];
4611 
4612     return 0;
4613 }
4614 
4615 void kvm_arch_remove_all_hw_breakpoints(void)
4616 {
4617     nb_hw_breakpoint = 0;
4618 }
4619 
4620 static CPUWatchpoint hw_watchpoint;
4621 
4622 static int kvm_handle_debug(X86CPU *cpu,
4623                             struct kvm_debug_exit_arch *arch_info)
4624 {
4625     CPUState *cs = CPU(cpu);
4626     CPUX86State *env = &cpu->env;
4627     int ret = 0;
4628     int n;
4629 
4630     if (arch_info->exception == EXCP01_DB) {
4631         if (arch_info->dr6 & DR6_BS) {
4632             if (cs->singlestep_enabled) {
4633                 ret = EXCP_DEBUG;
4634             }
4635         } else {
4636             for (n = 0; n < 4; n++) {
4637                 if (arch_info->dr6 & (1 << n)) {
4638                     switch ((arch_info->dr7 >> (16 + n*4)) & 0x3) {
4639                     case 0x0:
4640                         ret = EXCP_DEBUG;
4641                         break;
4642                     case 0x1:
4643                         ret = EXCP_DEBUG;
4644                         cs->watchpoint_hit = &hw_watchpoint;
4645                         hw_watchpoint.vaddr = hw_breakpoint[n].addr;
4646                         hw_watchpoint.flags = BP_MEM_WRITE;
4647                         break;
4648                     case 0x3:
4649                         ret = EXCP_DEBUG;
4650                         cs->watchpoint_hit = &hw_watchpoint;
4651                         hw_watchpoint.vaddr = hw_breakpoint[n].addr;
4652                         hw_watchpoint.flags = BP_MEM_ACCESS;
4653                         break;
4654                     }
4655                 }
4656             }
4657         }
4658     } else if (kvm_find_sw_breakpoint(cs, arch_info->pc)) {
4659         ret = EXCP_DEBUG;
4660     }
4661     if (ret == 0) {
4662         cpu_synchronize_state(cs);
4663         assert(env->exception_nr == -1);
4664 
4665         /* pass to guest */
4666         kvm_queue_exception(env, arch_info->exception,
4667                             arch_info->exception == EXCP01_DB,
4668                             arch_info->dr6);
4669         env->has_error_code = 0;
4670     }
4671 
4672     return ret;
4673 }
4674 
4675 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
4676 {
4677     const uint8_t type_code[] = {
4678         [GDB_BREAKPOINT_HW] = 0x0,
4679         [GDB_WATCHPOINT_WRITE] = 0x1,
4680         [GDB_WATCHPOINT_ACCESS] = 0x3
4681     };
4682     const uint8_t len_code[] = {
4683         [1] = 0x0, [2] = 0x1, [4] = 0x3, [8] = 0x2
4684     };
4685     int n;
4686 
4687     if (kvm_sw_breakpoints_active(cpu)) {
4688         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
4689     }
4690     if (nb_hw_breakpoint > 0) {
4691         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
4692         dbg->arch.debugreg[7] = 0x0600;
4693         for (n = 0; n < nb_hw_breakpoint; n++) {
4694             dbg->arch.debugreg[n] = hw_breakpoint[n].addr;
4695             dbg->arch.debugreg[7] |= (2 << (n * 2)) |
4696                 (type_code[hw_breakpoint[n].type] << (16 + n*4)) |
4697                 ((uint32_t)len_code[hw_breakpoint[n].len] << (18 + n*4));
4698         }
4699     }
4700 }
4701 
4702 static bool has_sgx_provisioning;
4703 
4704 static bool __kvm_enable_sgx_provisioning(KVMState *s)
4705 {
4706     int fd, ret;
4707 
4708     if (!kvm_vm_check_extension(s, KVM_CAP_SGX_ATTRIBUTE)) {
4709         return false;
4710     }
4711 
4712     fd = qemu_open_old("/dev/sgx_provision", O_RDONLY);
4713     if (fd < 0) {
4714         return false;
4715     }
4716 
4717     ret = kvm_vm_enable_cap(s, KVM_CAP_SGX_ATTRIBUTE, 0, fd);
4718     if (ret) {
4719         error_report("Could not enable SGX PROVISIONKEY: %s", strerror(-ret));
4720         exit(1);
4721     }
4722     close(fd);
4723     return true;
4724 }
4725 
4726 bool kvm_enable_sgx_provisioning(KVMState *s)
4727 {
4728     return MEMORIZE(__kvm_enable_sgx_provisioning(s), has_sgx_provisioning);
4729 }
4730 
4731 static bool host_supports_vmx(void)
4732 {
4733     uint32_t ecx, unused;
4734 
4735     host_cpuid(1, 0, &unused, &unused, &ecx, &unused);
4736     return ecx & CPUID_EXT_VMX;
4737 }
4738 
4739 #define VMX_INVALID_GUEST_STATE 0x80000021
4740 
4741 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
4742 {
4743     X86CPU *cpu = X86_CPU(cs);
4744     uint64_t code;
4745     int ret;
4746 
4747     switch (run->exit_reason) {
4748     case KVM_EXIT_HLT:
4749         DPRINTF("handle_hlt\n");
4750         qemu_mutex_lock_iothread();
4751         ret = kvm_handle_halt(cpu);
4752         qemu_mutex_unlock_iothread();
4753         break;
4754     case KVM_EXIT_SET_TPR:
4755         ret = 0;
4756         break;
4757     case KVM_EXIT_TPR_ACCESS:
4758         qemu_mutex_lock_iothread();
4759         ret = kvm_handle_tpr_access(cpu);
4760         qemu_mutex_unlock_iothread();
4761         break;
4762     case KVM_EXIT_FAIL_ENTRY:
4763         code = run->fail_entry.hardware_entry_failure_reason;
4764         fprintf(stderr, "KVM: entry failed, hardware error 0x%" PRIx64 "\n",
4765                 code);
4766         if (host_supports_vmx() && code == VMX_INVALID_GUEST_STATE) {
4767             fprintf(stderr,
4768                     "\nIf you're running a guest on an Intel machine without "
4769                         "unrestricted mode\n"
4770                     "support, the failure can be most likely due to the guest "
4771                         "entering an invalid\n"
4772                     "state for Intel VT. For example, the guest maybe running "
4773                         "in big real mode\n"
4774                     "which is not supported on less recent Intel processors."
4775                         "\n\n");
4776         }
4777         ret = -1;
4778         break;
4779     case KVM_EXIT_EXCEPTION:
4780         fprintf(stderr, "KVM: exception %d exit (error code 0x%x)\n",
4781                 run->ex.exception, run->ex.error_code);
4782         ret = -1;
4783         break;
4784     case KVM_EXIT_DEBUG:
4785         DPRINTF("kvm_exit_debug\n");
4786         qemu_mutex_lock_iothread();
4787         ret = kvm_handle_debug(cpu, &run->debug.arch);
4788         qemu_mutex_unlock_iothread();
4789         break;
4790     case KVM_EXIT_HYPERV:
4791         ret = kvm_hv_handle_exit(cpu, &run->hyperv);
4792         break;
4793     case KVM_EXIT_IOAPIC_EOI:
4794         ioapic_eoi_broadcast(run->eoi.vector);
4795         ret = 0;
4796         break;
4797     case KVM_EXIT_X86_BUS_LOCK:
4798         /* already handled in kvm_arch_post_run */
4799         ret = 0;
4800         break;
4801     default:
4802         fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
4803         ret = -1;
4804         break;
4805     }
4806 
4807     return ret;
4808 }
4809 
4810 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
4811 {
4812     X86CPU *cpu = X86_CPU(cs);
4813     CPUX86State *env = &cpu->env;
4814 
4815     kvm_cpu_synchronize_state(cs);
4816     return !(env->cr[0] & CR0_PE_MASK) ||
4817            ((env->segs[R_CS].selector  & 3) != 3);
4818 }
4819 
4820 void kvm_arch_init_irq_routing(KVMState *s)
4821 {
4822     /* We know at this point that we're using the in-kernel
4823      * irqchip, so we can use irqfds, and on x86 we know
4824      * we can use msi via irqfd and GSI routing.
4825      */
4826     kvm_msi_via_irqfd_allowed = true;
4827     kvm_gsi_routing_allowed = true;
4828 
4829     if (kvm_irqchip_is_split()) {
4830         int i;
4831 
4832         /* If the ioapic is in QEMU and the lapics are in KVM, reserve
4833            MSI routes for signaling interrupts to the local apics. */
4834         for (i = 0; i < IOAPIC_NUM_PINS; i++) {
4835             if (kvm_irqchip_add_msi_route(s, 0, NULL) < 0) {
4836                 error_report("Could not enable split IRQ mode.");
4837                 exit(1);
4838             }
4839         }
4840     }
4841 }
4842 
4843 int kvm_arch_irqchip_create(KVMState *s)
4844 {
4845     int ret;
4846     if (kvm_kernel_irqchip_split()) {
4847         ret = kvm_vm_enable_cap(s, KVM_CAP_SPLIT_IRQCHIP, 0, 24);
4848         if (ret) {
4849             error_report("Could not enable split irqchip mode: %s",
4850                          strerror(-ret));
4851             exit(1);
4852         } else {
4853             DPRINTF("Enabled KVM_CAP_SPLIT_IRQCHIP\n");
4854             kvm_split_irqchip = true;
4855             return 1;
4856         }
4857     } else {
4858         return 0;
4859     }
4860 }
4861 
4862 uint64_t kvm_swizzle_msi_ext_dest_id(uint64_t address)
4863 {
4864     CPUX86State *env;
4865     uint64_t ext_id;
4866 
4867     if (!first_cpu) {
4868         return address;
4869     }
4870     env = &X86_CPU(first_cpu)->env;
4871     if (!(env->features[FEAT_KVM] & (1 << KVM_FEATURE_MSI_EXT_DEST_ID))) {
4872         return address;
4873     }
4874 
4875     /*
4876      * If the remappable format bit is set, or the upper bits are
4877      * already set in address_hi, or the low extended bits aren't
4878      * there anyway, do nothing.
4879      */
4880     ext_id = address & (0xff << MSI_ADDR_DEST_IDX_SHIFT);
4881     if (!ext_id || (ext_id & (1 << MSI_ADDR_DEST_IDX_SHIFT)) || (address >> 32)) {
4882         return address;
4883     }
4884 
4885     address &= ~ext_id;
4886     address |= ext_id << 35;
4887     return address;
4888 }
4889 
4890 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
4891                              uint64_t address, uint32_t data, PCIDevice *dev)
4892 {
4893     X86IOMMUState *iommu = x86_iommu_get_default();
4894 
4895     if (iommu) {
4896         X86IOMMUClass *class = X86_IOMMU_DEVICE_GET_CLASS(iommu);
4897 
4898         if (class->int_remap) {
4899             int ret;
4900             MSIMessage src, dst;
4901 
4902             src.address = route->u.msi.address_hi;
4903             src.address <<= VTD_MSI_ADDR_HI_SHIFT;
4904             src.address |= route->u.msi.address_lo;
4905             src.data = route->u.msi.data;
4906 
4907             ret = class->int_remap(iommu, &src, &dst, dev ?     \
4908                                    pci_requester_id(dev) :      \
4909                                    X86_IOMMU_SID_INVALID);
4910             if (ret) {
4911                 trace_kvm_x86_fixup_msi_error(route->gsi);
4912                 return 1;
4913             }
4914 
4915             /*
4916              * Handled untranslated compatibilty format interrupt with
4917              * extended destination ID in the low bits 11-5. */
4918             dst.address = kvm_swizzle_msi_ext_dest_id(dst.address);
4919 
4920             route->u.msi.address_hi = dst.address >> VTD_MSI_ADDR_HI_SHIFT;
4921             route->u.msi.address_lo = dst.address & VTD_MSI_ADDR_LO_MASK;
4922             route->u.msi.data = dst.data;
4923             return 0;
4924         }
4925     }
4926 
4927     address = kvm_swizzle_msi_ext_dest_id(address);
4928     route->u.msi.address_hi = address >> VTD_MSI_ADDR_HI_SHIFT;
4929     route->u.msi.address_lo = address & VTD_MSI_ADDR_LO_MASK;
4930     return 0;
4931 }
4932 
4933 typedef struct MSIRouteEntry MSIRouteEntry;
4934 
4935 struct MSIRouteEntry {
4936     PCIDevice *dev;             /* Device pointer */
4937     int vector;                 /* MSI/MSIX vector index */
4938     int virq;                   /* Virtual IRQ index */
4939     QLIST_ENTRY(MSIRouteEntry) list;
4940 };
4941 
4942 /* List of used GSI routes */
4943 static QLIST_HEAD(, MSIRouteEntry) msi_route_list = \
4944     QLIST_HEAD_INITIALIZER(msi_route_list);
4945 
4946 static void kvm_update_msi_routes_all(void *private, bool global,
4947                                       uint32_t index, uint32_t mask)
4948 {
4949     int cnt = 0, vector;
4950     MSIRouteEntry *entry;
4951     MSIMessage msg;
4952     PCIDevice *dev;
4953 
4954     /* TODO: explicit route update */
4955     QLIST_FOREACH(entry, &msi_route_list, list) {
4956         cnt++;
4957         vector = entry->vector;
4958         dev = entry->dev;
4959         if (msix_enabled(dev) && !msix_is_masked(dev, vector)) {
4960             msg = msix_get_message(dev, vector);
4961         } else if (msi_enabled(dev) && !msi_is_masked(dev, vector)) {
4962             msg = msi_get_message(dev, vector);
4963         } else {
4964             /*
4965              * Either MSI/MSIX is disabled for the device, or the
4966              * specific message was masked out.  Skip this one.
4967              */
4968             continue;
4969         }
4970         kvm_irqchip_update_msi_route(kvm_state, entry->virq, msg, dev);
4971     }
4972     kvm_irqchip_commit_routes(kvm_state);
4973     trace_kvm_x86_update_msi_routes(cnt);
4974 }
4975 
4976 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
4977                                 int vector, PCIDevice *dev)
4978 {
4979     static bool notify_list_inited = false;
4980     MSIRouteEntry *entry;
4981 
4982     if (!dev) {
4983         /* These are (possibly) IOAPIC routes only used for split
4984          * kernel irqchip mode, while what we are housekeeping are
4985          * PCI devices only. */
4986         return 0;
4987     }
4988 
4989     entry = g_new0(MSIRouteEntry, 1);
4990     entry->dev = dev;
4991     entry->vector = vector;
4992     entry->virq = route->gsi;
4993     QLIST_INSERT_HEAD(&msi_route_list, entry, list);
4994 
4995     trace_kvm_x86_add_msi_route(route->gsi);
4996 
4997     if (!notify_list_inited) {
4998         /* For the first time we do add route, add ourselves into
4999          * IOMMU's IEC notify list if needed. */
5000         X86IOMMUState *iommu = x86_iommu_get_default();
5001         if (iommu) {
5002             x86_iommu_iec_register_notifier(iommu,
5003                                             kvm_update_msi_routes_all,
5004                                             NULL);
5005         }
5006         notify_list_inited = true;
5007     }
5008     return 0;
5009 }
5010 
5011 int kvm_arch_release_virq_post(int virq)
5012 {
5013     MSIRouteEntry *entry, *next;
5014     QLIST_FOREACH_SAFE(entry, &msi_route_list, list, next) {
5015         if (entry->virq == virq) {
5016             trace_kvm_x86_remove_msi_route(virq);
5017             QLIST_REMOVE(entry, list);
5018             g_free(entry);
5019             break;
5020         }
5021     }
5022     return 0;
5023 }
5024 
5025 int kvm_arch_msi_data_to_gsi(uint32_t data)
5026 {
5027     abort();
5028 }
5029 
5030 bool kvm_has_waitpkg(void)
5031 {
5032     return has_msr_umwait;
5033 }
5034 
5035 bool kvm_arch_cpu_check_are_resettable(void)
5036 {
5037     return !sev_es_enabled();
5038 }
5039