xref: /openbmc/qemu/target/i386/kvm/kvm.c (revision 65e728a2)
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             {.func = HV_CPUID_ENLIGHTMENT_INFO, .reg = R_EAX,
826              .bits = HV_APIC_ACCESS_RECOMMENDED}
827         }
828     },
829     [HYPERV_FEAT_TIME] = {
830         .desc = "clocksources (hv-time)",
831         .flags = {
832             {.func = HV_CPUID_FEATURES, .reg = R_EAX,
833              .bits = HV_TIME_REF_COUNT_AVAILABLE | HV_REFERENCE_TSC_AVAILABLE}
834         }
835     },
836     [HYPERV_FEAT_CRASH] = {
837         .desc = "crash MSRs (hv-crash)",
838         .flags = {
839             {.func = HV_CPUID_FEATURES, .reg = R_EDX,
840              .bits = HV_GUEST_CRASH_MSR_AVAILABLE}
841         }
842     },
843     [HYPERV_FEAT_RESET] = {
844         .desc = "reset MSR (hv-reset)",
845         .flags = {
846             {.func = HV_CPUID_FEATURES, .reg = R_EAX,
847              .bits = HV_RESET_AVAILABLE}
848         }
849     },
850     [HYPERV_FEAT_VPINDEX] = {
851         .desc = "VP_INDEX MSR (hv-vpindex)",
852         .flags = {
853             {.func = HV_CPUID_FEATURES, .reg = R_EAX,
854              .bits = HV_VP_INDEX_AVAILABLE}
855         }
856     },
857     [HYPERV_FEAT_RUNTIME] = {
858         .desc = "VP_RUNTIME MSR (hv-runtime)",
859         .flags = {
860             {.func = HV_CPUID_FEATURES, .reg = R_EAX,
861              .bits = HV_VP_RUNTIME_AVAILABLE}
862         }
863     },
864     [HYPERV_FEAT_SYNIC] = {
865         .desc = "synthetic interrupt controller (hv-synic)",
866         .flags = {
867             {.func = HV_CPUID_FEATURES, .reg = R_EAX,
868              .bits = HV_SYNIC_AVAILABLE}
869         }
870     },
871     [HYPERV_FEAT_STIMER] = {
872         .desc = "synthetic timers (hv-stimer)",
873         .flags = {
874             {.func = HV_CPUID_FEATURES, .reg = R_EAX,
875              .bits = HV_SYNTIMERS_AVAILABLE}
876         },
877         .dependencies = BIT(HYPERV_FEAT_SYNIC) | BIT(HYPERV_FEAT_TIME)
878     },
879     [HYPERV_FEAT_FREQUENCIES] = {
880         .desc = "frequency MSRs (hv-frequencies)",
881         .flags = {
882             {.func = HV_CPUID_FEATURES, .reg = R_EAX,
883              .bits = HV_ACCESS_FREQUENCY_MSRS},
884             {.func = HV_CPUID_FEATURES, .reg = R_EDX,
885              .bits = HV_FREQUENCY_MSRS_AVAILABLE}
886         }
887     },
888     [HYPERV_FEAT_REENLIGHTENMENT] = {
889         .desc = "reenlightenment MSRs (hv-reenlightenment)",
890         .flags = {
891             {.func = HV_CPUID_FEATURES, .reg = R_EAX,
892              .bits = HV_ACCESS_REENLIGHTENMENTS_CONTROL}
893         }
894     },
895     [HYPERV_FEAT_TLBFLUSH] = {
896         .desc = "paravirtualized TLB flush (hv-tlbflush)",
897         .flags = {
898             {.func = HV_CPUID_ENLIGHTMENT_INFO, .reg = R_EAX,
899              .bits = HV_REMOTE_TLB_FLUSH_RECOMMENDED |
900              HV_EX_PROCESSOR_MASKS_RECOMMENDED}
901         },
902         .dependencies = BIT(HYPERV_FEAT_VPINDEX)
903     },
904     [HYPERV_FEAT_EVMCS] = {
905         .desc = "enlightened VMCS (hv-evmcs)",
906         .flags = {
907             {.func = HV_CPUID_ENLIGHTMENT_INFO, .reg = R_EAX,
908              .bits = HV_ENLIGHTENED_VMCS_RECOMMENDED}
909         },
910         .dependencies = BIT(HYPERV_FEAT_VAPIC)
911     },
912     [HYPERV_FEAT_IPI] = {
913         .desc = "paravirtualized IPI (hv-ipi)",
914         .flags = {
915             {.func = HV_CPUID_ENLIGHTMENT_INFO, .reg = R_EAX,
916              .bits = HV_CLUSTER_IPI_RECOMMENDED |
917              HV_EX_PROCESSOR_MASKS_RECOMMENDED}
918         },
919         .dependencies = BIT(HYPERV_FEAT_VPINDEX)
920     },
921     [HYPERV_FEAT_STIMER_DIRECT] = {
922         .desc = "direct mode synthetic timers (hv-stimer-direct)",
923         .flags = {
924             {.func = HV_CPUID_FEATURES, .reg = R_EDX,
925              .bits = HV_STIMER_DIRECT_MODE_AVAILABLE}
926         },
927         .dependencies = BIT(HYPERV_FEAT_STIMER)
928     },
929 };
930 
931 static struct kvm_cpuid2 *try_get_hv_cpuid(CPUState *cs, int max,
932                                            bool do_sys_ioctl)
933 {
934     struct kvm_cpuid2 *cpuid;
935     int r, size;
936 
937     size = sizeof(*cpuid) + max * sizeof(*cpuid->entries);
938     cpuid = g_malloc0(size);
939     cpuid->nent = max;
940 
941     if (do_sys_ioctl) {
942         r = kvm_ioctl(kvm_state, KVM_GET_SUPPORTED_HV_CPUID, cpuid);
943     } else {
944         r = kvm_vcpu_ioctl(cs, KVM_GET_SUPPORTED_HV_CPUID, cpuid);
945     }
946     if (r == 0 && cpuid->nent >= max) {
947         r = -E2BIG;
948     }
949     if (r < 0) {
950         if (r == -E2BIG) {
951             g_free(cpuid);
952             return NULL;
953         } else {
954             fprintf(stderr, "KVM_GET_SUPPORTED_HV_CPUID failed: %s\n",
955                     strerror(-r));
956             exit(1);
957         }
958     }
959     return cpuid;
960 }
961 
962 /*
963  * Run KVM_GET_SUPPORTED_HV_CPUID ioctl(), allocating a buffer large enough
964  * for all entries.
965  */
966 static struct kvm_cpuid2 *get_supported_hv_cpuid(CPUState *cs)
967 {
968     struct kvm_cpuid2 *cpuid;
969     /* 0x40000000..0x40000005, 0x4000000A, 0x40000080..0x40000080 leaves */
970     int max = 10;
971     int i;
972     bool do_sys_ioctl;
973 
974     do_sys_ioctl =
975         kvm_check_extension(kvm_state, KVM_CAP_SYS_HYPERV_CPUID) > 0;
976 
977     /*
978      * Non-empty KVM context is needed when KVM_CAP_SYS_HYPERV_CPUID is
979      * unsupported, kvm_hyperv_expand_features() checks for that.
980      */
981     assert(do_sys_ioctl || cs->kvm_state);
982 
983     /*
984      * When the buffer is too small, KVM_GET_SUPPORTED_HV_CPUID fails with
985      * -E2BIG, however, it doesn't report back the right size. Keep increasing
986      * it and re-trying until we succeed.
987      */
988     while ((cpuid = try_get_hv_cpuid(cs, max, do_sys_ioctl)) == NULL) {
989         max++;
990     }
991 
992     /*
993      * KVM_GET_SUPPORTED_HV_CPUID does not set EVMCS CPUID bit before
994      * KVM_CAP_HYPERV_ENLIGHTENED_VMCS is enabled but we want to get the
995      * information early, just check for the capability and set the bit
996      * manually.
997      */
998     if (!do_sys_ioctl && kvm_check_extension(cs->kvm_state,
999                             KVM_CAP_HYPERV_ENLIGHTENED_VMCS) > 0) {
1000         for (i = 0; i < cpuid->nent; i++) {
1001             if (cpuid->entries[i].function == HV_CPUID_ENLIGHTMENT_INFO) {
1002                 cpuid->entries[i].eax |= HV_ENLIGHTENED_VMCS_RECOMMENDED;
1003             }
1004         }
1005     }
1006 
1007     return cpuid;
1008 }
1009 
1010 /*
1011  * When KVM_GET_SUPPORTED_HV_CPUID is not supported we fill CPUID feature
1012  * leaves from KVM_CAP_HYPERV* and present MSRs data.
1013  */
1014 static struct kvm_cpuid2 *get_supported_hv_cpuid_legacy(CPUState *cs)
1015 {
1016     X86CPU *cpu = X86_CPU(cs);
1017     struct kvm_cpuid2 *cpuid;
1018     struct kvm_cpuid_entry2 *entry_feat, *entry_recomm;
1019 
1020     /* HV_CPUID_FEATURES, HV_CPUID_ENLIGHTMENT_INFO */
1021     cpuid = g_malloc0(sizeof(*cpuid) + 2 * sizeof(*cpuid->entries));
1022     cpuid->nent = 2;
1023 
1024     /* HV_CPUID_VENDOR_AND_MAX_FUNCTIONS */
1025     entry_feat = &cpuid->entries[0];
1026     entry_feat->function = HV_CPUID_FEATURES;
1027 
1028     entry_recomm = &cpuid->entries[1];
1029     entry_recomm->function = HV_CPUID_ENLIGHTMENT_INFO;
1030     entry_recomm->ebx = cpu->hyperv_spinlock_attempts;
1031 
1032     if (kvm_check_extension(cs->kvm_state, KVM_CAP_HYPERV) > 0) {
1033         entry_feat->eax |= HV_HYPERCALL_AVAILABLE;
1034         entry_feat->eax |= HV_APIC_ACCESS_AVAILABLE;
1035         entry_feat->edx |= HV_CPU_DYNAMIC_PARTITIONING_AVAILABLE;
1036         entry_recomm->eax |= HV_RELAXED_TIMING_RECOMMENDED;
1037         entry_recomm->eax |= HV_APIC_ACCESS_RECOMMENDED;
1038     }
1039 
1040     if (kvm_check_extension(cs->kvm_state, KVM_CAP_HYPERV_TIME) > 0) {
1041         entry_feat->eax |= HV_TIME_REF_COUNT_AVAILABLE;
1042         entry_feat->eax |= HV_REFERENCE_TSC_AVAILABLE;
1043     }
1044 
1045     if (has_msr_hv_frequencies) {
1046         entry_feat->eax |= HV_ACCESS_FREQUENCY_MSRS;
1047         entry_feat->edx |= HV_FREQUENCY_MSRS_AVAILABLE;
1048     }
1049 
1050     if (has_msr_hv_crash) {
1051         entry_feat->edx |= HV_GUEST_CRASH_MSR_AVAILABLE;
1052     }
1053 
1054     if (has_msr_hv_reenlightenment) {
1055         entry_feat->eax |= HV_ACCESS_REENLIGHTENMENTS_CONTROL;
1056     }
1057 
1058     if (has_msr_hv_reset) {
1059         entry_feat->eax |= HV_RESET_AVAILABLE;
1060     }
1061 
1062     if (has_msr_hv_vpindex) {
1063         entry_feat->eax |= HV_VP_INDEX_AVAILABLE;
1064     }
1065 
1066     if (has_msr_hv_runtime) {
1067         entry_feat->eax |= HV_VP_RUNTIME_AVAILABLE;
1068     }
1069 
1070     if (has_msr_hv_synic) {
1071         unsigned int cap = cpu->hyperv_synic_kvm_only ?
1072             KVM_CAP_HYPERV_SYNIC : KVM_CAP_HYPERV_SYNIC2;
1073 
1074         if (kvm_check_extension(cs->kvm_state, cap) > 0) {
1075             entry_feat->eax |= HV_SYNIC_AVAILABLE;
1076         }
1077     }
1078 
1079     if (has_msr_hv_stimer) {
1080         entry_feat->eax |= HV_SYNTIMERS_AVAILABLE;
1081     }
1082 
1083     if (kvm_check_extension(cs->kvm_state,
1084                             KVM_CAP_HYPERV_TLBFLUSH) > 0) {
1085         entry_recomm->eax |= HV_REMOTE_TLB_FLUSH_RECOMMENDED;
1086         entry_recomm->eax |= HV_EX_PROCESSOR_MASKS_RECOMMENDED;
1087     }
1088 
1089     if (kvm_check_extension(cs->kvm_state,
1090                             KVM_CAP_HYPERV_ENLIGHTENED_VMCS) > 0) {
1091         entry_recomm->eax |= HV_ENLIGHTENED_VMCS_RECOMMENDED;
1092     }
1093 
1094     if (kvm_check_extension(cs->kvm_state,
1095                             KVM_CAP_HYPERV_SEND_IPI) > 0) {
1096         entry_recomm->eax |= HV_CLUSTER_IPI_RECOMMENDED;
1097         entry_recomm->eax |= HV_EX_PROCESSOR_MASKS_RECOMMENDED;
1098     }
1099 
1100     return cpuid;
1101 }
1102 
1103 static uint32_t hv_cpuid_get_host(CPUState *cs, uint32_t func, int reg)
1104 {
1105     struct kvm_cpuid_entry2 *entry;
1106     struct kvm_cpuid2 *cpuid;
1107 
1108     if (hv_cpuid_cache) {
1109         cpuid = hv_cpuid_cache;
1110     } else {
1111         if (kvm_check_extension(kvm_state, KVM_CAP_HYPERV_CPUID) > 0) {
1112             cpuid = get_supported_hv_cpuid(cs);
1113         } else {
1114             /*
1115              * 'cs->kvm_state' may be NULL when Hyper-V features are expanded
1116              * before KVM context is created but this is only done when
1117              * KVM_CAP_SYS_HYPERV_CPUID is supported and it implies
1118              * KVM_CAP_HYPERV_CPUID.
1119              */
1120             assert(cs->kvm_state);
1121 
1122             cpuid = get_supported_hv_cpuid_legacy(cs);
1123         }
1124         hv_cpuid_cache = cpuid;
1125     }
1126 
1127     if (!cpuid) {
1128         return 0;
1129     }
1130 
1131     entry = cpuid_find_entry(cpuid, func, 0);
1132     if (!entry) {
1133         return 0;
1134     }
1135 
1136     return cpuid_entry_get_reg(entry, reg);
1137 }
1138 
1139 static bool hyperv_feature_supported(CPUState *cs, int feature)
1140 {
1141     uint32_t func, bits;
1142     int i, reg;
1143 
1144     for (i = 0; i < ARRAY_SIZE(kvm_hyperv_properties[feature].flags); i++) {
1145 
1146         func = kvm_hyperv_properties[feature].flags[i].func;
1147         reg = kvm_hyperv_properties[feature].flags[i].reg;
1148         bits = kvm_hyperv_properties[feature].flags[i].bits;
1149 
1150         if (!func) {
1151             continue;
1152         }
1153 
1154         if ((hv_cpuid_get_host(cs, func, reg) & bits) != bits) {
1155             return false;
1156         }
1157     }
1158 
1159     return true;
1160 }
1161 
1162 /* Checks that all feature dependencies are enabled */
1163 static bool hv_feature_check_deps(X86CPU *cpu, int feature, Error **errp)
1164 {
1165     uint64_t deps;
1166     int dep_feat;
1167 
1168     deps = kvm_hyperv_properties[feature].dependencies;
1169     while (deps) {
1170         dep_feat = ctz64(deps);
1171         if (!(hyperv_feat_enabled(cpu, dep_feat))) {
1172             error_setg(errp, "Hyper-V %s requires Hyper-V %s",
1173                        kvm_hyperv_properties[feature].desc,
1174                        kvm_hyperv_properties[dep_feat].desc);
1175             return false;
1176         }
1177         deps &= ~(1ull << dep_feat);
1178     }
1179 
1180     return true;
1181 }
1182 
1183 static uint32_t hv_build_cpuid_leaf(CPUState *cs, uint32_t func, int reg)
1184 {
1185     X86CPU *cpu = X86_CPU(cs);
1186     uint32_t r = 0;
1187     int i, j;
1188 
1189     for (i = 0; i < ARRAY_SIZE(kvm_hyperv_properties); i++) {
1190         if (!hyperv_feat_enabled(cpu, i)) {
1191             continue;
1192         }
1193 
1194         for (j = 0; j < ARRAY_SIZE(kvm_hyperv_properties[i].flags); j++) {
1195             if (kvm_hyperv_properties[i].flags[j].func != func) {
1196                 continue;
1197             }
1198             if (kvm_hyperv_properties[i].flags[j].reg != reg) {
1199                 continue;
1200             }
1201 
1202             r |= kvm_hyperv_properties[i].flags[j].bits;
1203         }
1204     }
1205 
1206     return r;
1207 }
1208 
1209 /*
1210  * Expand Hyper-V CPU features. In partucular, check that all the requested
1211  * features are supported by the host and the sanity of the configuration
1212  * (that all the required dependencies are included). Also, this takes care
1213  * of 'hv_passthrough' mode and fills the environment with all supported
1214  * Hyper-V features.
1215  */
1216 bool kvm_hyperv_expand_features(X86CPU *cpu, Error **errp)
1217 {
1218     CPUState *cs = CPU(cpu);
1219     Error *local_err = NULL;
1220     int feat;
1221 
1222     if (!hyperv_enabled(cpu))
1223         return true;
1224 
1225     /*
1226      * When kvm_hyperv_expand_features is called at CPU feature expansion
1227      * time per-CPU kvm_state is not available yet so we can only proceed
1228      * when KVM_CAP_SYS_HYPERV_CPUID is supported.
1229      */
1230     if (!cs->kvm_state &&
1231         !kvm_check_extension(kvm_state, KVM_CAP_SYS_HYPERV_CPUID))
1232         return true;
1233 
1234     if (cpu->hyperv_passthrough) {
1235         cpu->hyperv_vendor_id[0] =
1236             hv_cpuid_get_host(cs, HV_CPUID_VENDOR_AND_MAX_FUNCTIONS, R_EBX);
1237         cpu->hyperv_vendor_id[1] =
1238             hv_cpuid_get_host(cs, HV_CPUID_VENDOR_AND_MAX_FUNCTIONS, R_ECX);
1239         cpu->hyperv_vendor_id[2] =
1240             hv_cpuid_get_host(cs, HV_CPUID_VENDOR_AND_MAX_FUNCTIONS, R_EDX);
1241         cpu->hyperv_vendor = g_realloc(cpu->hyperv_vendor,
1242                                        sizeof(cpu->hyperv_vendor_id) + 1);
1243         memcpy(cpu->hyperv_vendor, cpu->hyperv_vendor_id,
1244                sizeof(cpu->hyperv_vendor_id));
1245         cpu->hyperv_vendor[sizeof(cpu->hyperv_vendor_id)] = 0;
1246 
1247         cpu->hyperv_interface_id[0] =
1248             hv_cpuid_get_host(cs, HV_CPUID_INTERFACE, R_EAX);
1249         cpu->hyperv_interface_id[1] =
1250             hv_cpuid_get_host(cs, HV_CPUID_INTERFACE, R_EBX);
1251         cpu->hyperv_interface_id[2] =
1252             hv_cpuid_get_host(cs, HV_CPUID_INTERFACE, R_ECX);
1253         cpu->hyperv_interface_id[3] =
1254             hv_cpuid_get_host(cs, HV_CPUID_INTERFACE, R_EDX);
1255 
1256         cpu->hyperv_version_id[0] =
1257             hv_cpuid_get_host(cs, HV_CPUID_VERSION, R_EAX);
1258         cpu->hyperv_version_id[1] =
1259             hv_cpuid_get_host(cs, HV_CPUID_VERSION, R_EBX);
1260         cpu->hyperv_version_id[2] =
1261             hv_cpuid_get_host(cs, HV_CPUID_VERSION, R_ECX);
1262         cpu->hyperv_version_id[3] =
1263             hv_cpuid_get_host(cs, HV_CPUID_VERSION, R_EDX);
1264 
1265         cpu->hv_max_vps = hv_cpuid_get_host(cs, HV_CPUID_IMPLEMENT_LIMITS,
1266                                             R_EAX);
1267         cpu->hyperv_limits[0] =
1268             hv_cpuid_get_host(cs, HV_CPUID_IMPLEMENT_LIMITS, R_EBX);
1269         cpu->hyperv_limits[1] =
1270             hv_cpuid_get_host(cs, HV_CPUID_IMPLEMENT_LIMITS, R_ECX);
1271         cpu->hyperv_limits[2] =
1272             hv_cpuid_get_host(cs, HV_CPUID_IMPLEMENT_LIMITS, R_EDX);
1273 
1274         cpu->hyperv_spinlock_attempts =
1275             hv_cpuid_get_host(cs, HV_CPUID_ENLIGHTMENT_INFO, R_EBX);
1276 
1277         /*
1278          * Mark feature as enabled in 'cpu->hyperv_features' as
1279          * hv_build_cpuid_leaf() uses this info to build guest CPUIDs.
1280          */
1281         for (feat = 0; feat < ARRAY_SIZE(kvm_hyperv_properties); feat++) {
1282             if (hyperv_feature_supported(cs, feat)) {
1283                 cpu->hyperv_features |= BIT(feat);
1284             }
1285         }
1286     } else {
1287         /* Check features availability and dependencies */
1288         for (feat = 0; feat < ARRAY_SIZE(kvm_hyperv_properties); feat++) {
1289             /* If the feature was not requested skip it. */
1290             if (!hyperv_feat_enabled(cpu, feat)) {
1291                 continue;
1292             }
1293 
1294             /* Check if the feature is supported by KVM */
1295             if (!hyperv_feature_supported(cs, feat)) {
1296                 error_setg(errp, "Hyper-V %s is not supported by kernel",
1297                            kvm_hyperv_properties[feat].desc);
1298                 return false;
1299             }
1300 
1301             /* Check dependencies */
1302             if (!hv_feature_check_deps(cpu, feat, &local_err)) {
1303                 error_propagate(errp, local_err);
1304                 return false;
1305             }
1306         }
1307     }
1308 
1309     /* Additional dependencies not covered by kvm_hyperv_properties[] */
1310     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_SYNIC) &&
1311         !cpu->hyperv_synic_kvm_only &&
1312         !hyperv_feat_enabled(cpu, HYPERV_FEAT_VPINDEX)) {
1313         error_setg(errp, "Hyper-V %s requires Hyper-V %s",
1314                    kvm_hyperv_properties[HYPERV_FEAT_SYNIC].desc,
1315                    kvm_hyperv_properties[HYPERV_FEAT_VPINDEX].desc);
1316         return false;
1317     }
1318 
1319     return true;
1320 }
1321 
1322 /*
1323  * Fill in Hyper-V CPUIDs. Returns the number of entries filled in cpuid_ent.
1324  */
1325 static int hyperv_fill_cpuids(CPUState *cs,
1326                               struct kvm_cpuid_entry2 *cpuid_ent)
1327 {
1328     X86CPU *cpu = X86_CPU(cs);
1329     struct kvm_cpuid_entry2 *c;
1330     uint32_t cpuid_i = 0;
1331 
1332     c = &cpuid_ent[cpuid_i++];
1333     c->function = HV_CPUID_VENDOR_AND_MAX_FUNCTIONS;
1334     c->eax = hyperv_feat_enabled(cpu, HYPERV_FEAT_EVMCS) ?
1335         HV_CPUID_NESTED_FEATURES : HV_CPUID_IMPLEMENT_LIMITS;
1336     c->ebx = cpu->hyperv_vendor_id[0];
1337     c->ecx = cpu->hyperv_vendor_id[1];
1338     c->edx = cpu->hyperv_vendor_id[2];
1339 
1340     c = &cpuid_ent[cpuid_i++];
1341     c->function = HV_CPUID_INTERFACE;
1342     c->eax = cpu->hyperv_interface_id[0];
1343     c->ebx = cpu->hyperv_interface_id[1];
1344     c->ecx = cpu->hyperv_interface_id[2];
1345     c->edx = cpu->hyperv_interface_id[3];
1346 
1347     c = &cpuid_ent[cpuid_i++];
1348     c->function = HV_CPUID_VERSION;
1349     c->eax = cpu->hyperv_version_id[0];
1350     c->ebx = cpu->hyperv_version_id[1];
1351     c->ecx = cpu->hyperv_version_id[2];
1352     c->edx = cpu->hyperv_version_id[3];
1353 
1354     c = &cpuid_ent[cpuid_i++];
1355     c->function = HV_CPUID_FEATURES;
1356     c->eax = hv_build_cpuid_leaf(cs, HV_CPUID_FEATURES, R_EAX);
1357     c->ebx = hv_build_cpuid_leaf(cs, HV_CPUID_FEATURES, R_EBX);
1358     c->edx = hv_build_cpuid_leaf(cs, HV_CPUID_FEATURES, R_EDX);
1359 
1360     /* Unconditionally required with any Hyper-V enlightenment */
1361     c->eax |= HV_HYPERCALL_AVAILABLE;
1362 
1363     /* SynIC and Vmbus devices require messages/signals hypercalls */
1364     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_SYNIC) &&
1365         !cpu->hyperv_synic_kvm_only) {
1366         c->ebx |= HV_POST_MESSAGES | HV_SIGNAL_EVENTS;
1367     }
1368 
1369     /* Not exposed by KVM but needed to make CPU hotplug in Windows work */
1370     c->edx |= HV_CPU_DYNAMIC_PARTITIONING_AVAILABLE;
1371 
1372     c = &cpuid_ent[cpuid_i++];
1373     c->function = HV_CPUID_ENLIGHTMENT_INFO;
1374     c->eax = hv_build_cpuid_leaf(cs, HV_CPUID_ENLIGHTMENT_INFO, R_EAX);
1375     c->ebx = cpu->hyperv_spinlock_attempts;
1376 
1377     if (cpu->hyperv_no_nonarch_cs == ON_OFF_AUTO_ON) {
1378         c->eax |= HV_NO_NONARCH_CORESHARING;
1379     } else if (cpu->hyperv_no_nonarch_cs == ON_OFF_AUTO_AUTO) {
1380         c->eax |= hv_cpuid_get_host(cs, HV_CPUID_ENLIGHTMENT_INFO, R_EAX) &
1381             HV_NO_NONARCH_CORESHARING;
1382     }
1383 
1384     c = &cpuid_ent[cpuid_i++];
1385     c->function = HV_CPUID_IMPLEMENT_LIMITS;
1386     c->eax = cpu->hv_max_vps;
1387     c->ebx = cpu->hyperv_limits[0];
1388     c->ecx = cpu->hyperv_limits[1];
1389     c->edx = cpu->hyperv_limits[2];
1390 
1391     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_EVMCS)) {
1392         __u32 function;
1393 
1394         /* Create zeroed 0x40000006..0x40000009 leaves */
1395         for (function = HV_CPUID_IMPLEMENT_LIMITS + 1;
1396              function < HV_CPUID_NESTED_FEATURES; function++) {
1397             c = &cpuid_ent[cpuid_i++];
1398             c->function = function;
1399         }
1400 
1401         c = &cpuid_ent[cpuid_i++];
1402         c->function = HV_CPUID_NESTED_FEATURES;
1403         c->eax = cpu->hyperv_nested[0];
1404     }
1405 
1406     return cpuid_i;
1407 }
1408 
1409 static Error *hv_passthrough_mig_blocker;
1410 static Error *hv_no_nonarch_cs_mig_blocker;
1411 
1412 /* Checks that the exposed eVMCS version range is supported by KVM */
1413 static bool evmcs_version_supported(uint16_t evmcs_version,
1414                                     uint16_t supported_evmcs_version)
1415 {
1416     uint8_t min_version = evmcs_version & 0xff;
1417     uint8_t max_version = evmcs_version >> 8;
1418     uint8_t min_supported_version = supported_evmcs_version & 0xff;
1419     uint8_t max_supported_version = supported_evmcs_version >> 8;
1420 
1421     return (min_version >= min_supported_version) &&
1422         (max_version <= max_supported_version);
1423 }
1424 
1425 #define DEFAULT_EVMCS_VERSION ((1 << 8) | 1)
1426 
1427 static int hyperv_init_vcpu(X86CPU *cpu)
1428 {
1429     CPUState *cs = CPU(cpu);
1430     Error *local_err = NULL;
1431     int ret;
1432 
1433     if (cpu->hyperv_passthrough && hv_passthrough_mig_blocker == NULL) {
1434         error_setg(&hv_passthrough_mig_blocker,
1435                    "'hv-passthrough' CPU flag prevents migration, use explicit"
1436                    " set of hv-* flags instead");
1437         ret = migrate_add_blocker(hv_passthrough_mig_blocker, &local_err);
1438         if (ret < 0) {
1439             error_report_err(local_err);
1440             return ret;
1441         }
1442     }
1443 
1444     if (cpu->hyperv_no_nonarch_cs == ON_OFF_AUTO_AUTO &&
1445         hv_no_nonarch_cs_mig_blocker == NULL) {
1446         error_setg(&hv_no_nonarch_cs_mig_blocker,
1447                    "'hv-no-nonarch-coresharing=auto' CPU flag prevents migration"
1448                    " use explicit 'hv-no-nonarch-coresharing=on' instead (but"
1449                    " make sure SMT is disabled and/or that vCPUs are properly"
1450                    " pinned)");
1451         ret = migrate_add_blocker(hv_no_nonarch_cs_mig_blocker, &local_err);
1452         if (ret < 0) {
1453             error_report_err(local_err);
1454             return ret;
1455         }
1456     }
1457 
1458     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_VPINDEX) && !hv_vpindex_settable) {
1459         /*
1460          * the kernel doesn't support setting vp_index; assert that its value
1461          * is in sync
1462          */
1463         struct {
1464             struct kvm_msrs info;
1465             struct kvm_msr_entry entries[1];
1466         } msr_data = {
1467             .info.nmsrs = 1,
1468             .entries[0].index = HV_X64_MSR_VP_INDEX,
1469         };
1470 
1471         ret = kvm_vcpu_ioctl(cs, KVM_GET_MSRS, &msr_data);
1472         if (ret < 0) {
1473             return ret;
1474         }
1475         assert(ret == 1);
1476 
1477         if (msr_data.entries[0].data != hyperv_vp_index(CPU(cpu))) {
1478             error_report("kernel's vp_index != QEMU's vp_index");
1479             return -ENXIO;
1480         }
1481     }
1482 
1483     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_SYNIC)) {
1484         uint32_t synic_cap = cpu->hyperv_synic_kvm_only ?
1485             KVM_CAP_HYPERV_SYNIC : KVM_CAP_HYPERV_SYNIC2;
1486         ret = kvm_vcpu_enable_cap(cs, synic_cap, 0);
1487         if (ret < 0) {
1488             error_report("failed to turn on HyperV SynIC in KVM: %s",
1489                          strerror(-ret));
1490             return ret;
1491         }
1492 
1493         if (!cpu->hyperv_synic_kvm_only) {
1494             ret = hyperv_x86_synic_add(cpu);
1495             if (ret < 0) {
1496                 error_report("failed to create HyperV SynIC: %s",
1497                              strerror(-ret));
1498                 return ret;
1499             }
1500         }
1501     }
1502 
1503     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_EVMCS)) {
1504         uint16_t evmcs_version = DEFAULT_EVMCS_VERSION;
1505         uint16_t supported_evmcs_version;
1506 
1507         ret = kvm_vcpu_enable_cap(cs, KVM_CAP_HYPERV_ENLIGHTENED_VMCS, 0,
1508                                   (uintptr_t)&supported_evmcs_version);
1509 
1510         /*
1511          * KVM is required to support EVMCS ver.1. as that's what 'hv-evmcs'
1512          * option sets. Note: we hardcode the maximum supported eVMCS version
1513          * to '1' as well so 'hv-evmcs' feature is migratable even when (and if)
1514          * ver.2 is implemented. A new option (e.g. 'hv-evmcs=2') will then have
1515          * to be added.
1516          */
1517         if (ret < 0) {
1518             error_report("Hyper-V %s is not supported by kernel",
1519                          kvm_hyperv_properties[HYPERV_FEAT_EVMCS].desc);
1520             return ret;
1521         }
1522 
1523         if (!evmcs_version_supported(evmcs_version, supported_evmcs_version)) {
1524             error_report("eVMCS version range [%d..%d] is not supported by "
1525                          "kernel (supported: [%d..%d])", evmcs_version & 0xff,
1526                          evmcs_version >> 8, supported_evmcs_version & 0xff,
1527                          supported_evmcs_version >> 8);
1528             return -ENOTSUP;
1529         }
1530 
1531         cpu->hyperv_nested[0] = evmcs_version;
1532     }
1533 
1534     return 0;
1535 }
1536 
1537 static Error *invtsc_mig_blocker;
1538 
1539 #define KVM_MAX_CPUID_ENTRIES  100
1540 
1541 int kvm_arch_init_vcpu(CPUState *cs)
1542 {
1543     struct {
1544         struct kvm_cpuid2 cpuid;
1545         struct kvm_cpuid_entry2 entries[KVM_MAX_CPUID_ENTRIES];
1546     } cpuid_data;
1547     /*
1548      * The kernel defines these structs with padding fields so there
1549      * should be no extra padding in our cpuid_data struct.
1550      */
1551     QEMU_BUILD_BUG_ON(sizeof(cpuid_data) !=
1552                       sizeof(struct kvm_cpuid2) +
1553                       sizeof(struct kvm_cpuid_entry2) * KVM_MAX_CPUID_ENTRIES);
1554 
1555     X86CPU *cpu = X86_CPU(cs);
1556     CPUX86State *env = &cpu->env;
1557     uint32_t limit, i, j, cpuid_i;
1558     uint32_t unused;
1559     struct kvm_cpuid_entry2 *c;
1560     uint32_t signature[3];
1561     int kvm_base = KVM_CPUID_SIGNATURE;
1562     int max_nested_state_len;
1563     int r;
1564     Error *local_err = NULL;
1565 
1566     memset(&cpuid_data, 0, sizeof(cpuid_data));
1567 
1568     cpuid_i = 0;
1569 
1570     r = kvm_arch_set_tsc_khz(cs);
1571     if (r < 0) {
1572         return r;
1573     }
1574 
1575     /* vcpu's TSC frequency is either specified by user, or following
1576      * the value used by KVM if the former is not present. In the
1577      * latter case, we query it from KVM and record in env->tsc_khz,
1578      * so that vcpu's TSC frequency can be migrated later via this field.
1579      */
1580     if (!env->tsc_khz) {
1581         r = kvm_check_extension(cs->kvm_state, KVM_CAP_GET_TSC_KHZ) ?
1582             kvm_vcpu_ioctl(cs, KVM_GET_TSC_KHZ) :
1583             -ENOTSUP;
1584         if (r > 0) {
1585             env->tsc_khz = r;
1586         }
1587     }
1588 
1589     env->apic_bus_freq = KVM_APIC_BUS_FREQUENCY;
1590 
1591     /*
1592      * kvm_hyperv_expand_features() is called here for the second time in case
1593      * KVM_CAP_SYS_HYPERV_CPUID is not supported. While we can't possibly handle
1594      * 'query-cpu-model-expansion' in this case as we don't have a KVM vCPU to
1595      * check which Hyper-V enlightenments are supported and which are not, we
1596      * can still proceed and check/expand Hyper-V enlightenments here so legacy
1597      * behavior is preserved.
1598      */
1599     if (!kvm_hyperv_expand_features(cpu, &local_err)) {
1600         error_report_err(local_err);
1601         return -ENOSYS;
1602     }
1603 
1604     if (hyperv_enabled(cpu)) {
1605         r = hyperv_init_vcpu(cpu);
1606         if (r) {
1607             return r;
1608         }
1609 
1610         cpuid_i = hyperv_fill_cpuids(cs, cpuid_data.entries);
1611         kvm_base = KVM_CPUID_SIGNATURE_NEXT;
1612         has_msr_hv_hypercall = true;
1613     }
1614 
1615     if (cpu->expose_kvm) {
1616         memcpy(signature, "KVMKVMKVM\0\0\0", 12);
1617         c = &cpuid_data.entries[cpuid_i++];
1618         c->function = KVM_CPUID_SIGNATURE | kvm_base;
1619         c->eax = KVM_CPUID_FEATURES | kvm_base;
1620         c->ebx = signature[0];
1621         c->ecx = signature[1];
1622         c->edx = signature[2];
1623 
1624         c = &cpuid_data.entries[cpuid_i++];
1625         c->function = KVM_CPUID_FEATURES | kvm_base;
1626         c->eax = env->features[FEAT_KVM];
1627         c->edx = env->features[FEAT_KVM_HINTS];
1628     }
1629 
1630     cpu_x86_cpuid(env, 0, 0, &limit, &unused, &unused, &unused);
1631 
1632     for (i = 0; i <= limit; i++) {
1633         if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1634             fprintf(stderr, "unsupported level value: 0x%x\n", limit);
1635             abort();
1636         }
1637         c = &cpuid_data.entries[cpuid_i++];
1638 
1639         switch (i) {
1640         case 2: {
1641             /* Keep reading function 2 till all the input is received */
1642             int times;
1643 
1644             c->function = i;
1645             c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC |
1646                        KVM_CPUID_FLAG_STATE_READ_NEXT;
1647             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1648             times = c->eax & 0xff;
1649 
1650             for (j = 1; j < times; ++j) {
1651                 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1652                     fprintf(stderr, "cpuid_data is full, no space for "
1653                             "cpuid(eax:2):eax & 0xf = 0x%x\n", times);
1654                     abort();
1655                 }
1656                 c = &cpuid_data.entries[cpuid_i++];
1657                 c->function = i;
1658                 c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC;
1659                 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1660             }
1661             break;
1662         }
1663         case 0x1f:
1664             if (env->nr_dies < 2) {
1665                 break;
1666             }
1667             /* fallthrough */
1668         case 4:
1669         case 0xb:
1670         case 0xd:
1671             for (j = 0; ; j++) {
1672                 if (i == 0xd && j == 64) {
1673                     break;
1674                 }
1675 
1676                 if (i == 0x1f && j == 64) {
1677                     break;
1678                 }
1679 
1680                 c->function = i;
1681                 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1682                 c->index = j;
1683                 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
1684 
1685                 if (i == 4 && c->eax == 0) {
1686                     break;
1687                 }
1688                 if (i == 0xb && !(c->ecx & 0xff00)) {
1689                     break;
1690                 }
1691                 if (i == 0x1f && !(c->ecx & 0xff00)) {
1692                     break;
1693                 }
1694                 if (i == 0xd && c->eax == 0) {
1695                     continue;
1696                 }
1697                 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1698                     fprintf(stderr, "cpuid_data is full, no space for "
1699                             "cpuid(eax:0x%x,ecx:0x%x)\n", i, j);
1700                     abort();
1701                 }
1702                 c = &cpuid_data.entries[cpuid_i++];
1703             }
1704             break;
1705         case 0x7:
1706         case 0x14: {
1707             uint32_t times;
1708 
1709             c->function = i;
1710             c->index = 0;
1711             c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1712             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1713             times = c->eax;
1714 
1715             for (j = 1; j <= times; ++j) {
1716                 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1717                     fprintf(stderr, "cpuid_data is full, no space for "
1718                                 "cpuid(eax:0x%x,ecx:0x%x)\n", i, j);
1719                     abort();
1720                 }
1721                 c = &cpuid_data.entries[cpuid_i++];
1722                 c->function = i;
1723                 c->index = j;
1724                 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1725                 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
1726             }
1727             break;
1728         }
1729         default:
1730             c->function = i;
1731             c->flags = 0;
1732             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1733             if (!c->eax && !c->ebx && !c->ecx && !c->edx) {
1734                 /*
1735                  * KVM already returns all zeroes if a CPUID entry is missing,
1736                  * so we can omit it and avoid hitting KVM's 80-entry limit.
1737                  */
1738                 cpuid_i--;
1739             }
1740             break;
1741         }
1742     }
1743 
1744     if (limit >= 0x0a) {
1745         uint32_t eax, edx;
1746 
1747         cpu_x86_cpuid(env, 0x0a, 0, &eax, &unused, &unused, &edx);
1748 
1749         has_architectural_pmu_version = eax & 0xff;
1750         if (has_architectural_pmu_version > 0) {
1751             num_architectural_pmu_gp_counters = (eax & 0xff00) >> 8;
1752 
1753             /* Shouldn't be more than 32, since that's the number of bits
1754              * available in EBX to tell us _which_ counters are available.
1755              * Play it safe.
1756              */
1757             if (num_architectural_pmu_gp_counters > MAX_GP_COUNTERS) {
1758                 num_architectural_pmu_gp_counters = MAX_GP_COUNTERS;
1759             }
1760 
1761             if (has_architectural_pmu_version > 1) {
1762                 num_architectural_pmu_fixed_counters = edx & 0x1f;
1763 
1764                 if (num_architectural_pmu_fixed_counters > MAX_FIXED_COUNTERS) {
1765                     num_architectural_pmu_fixed_counters = MAX_FIXED_COUNTERS;
1766                 }
1767             }
1768         }
1769     }
1770 
1771     cpu_x86_cpuid(env, 0x80000000, 0, &limit, &unused, &unused, &unused);
1772 
1773     for (i = 0x80000000; i <= limit; i++) {
1774         if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1775             fprintf(stderr, "unsupported xlevel value: 0x%x\n", limit);
1776             abort();
1777         }
1778         c = &cpuid_data.entries[cpuid_i++];
1779 
1780         switch (i) {
1781         case 0x8000001d:
1782             /* Query for all AMD cache information leaves */
1783             for (j = 0; ; j++) {
1784                 c->function = i;
1785                 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1786                 c->index = j;
1787                 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
1788 
1789                 if (c->eax == 0) {
1790                     break;
1791                 }
1792                 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1793                     fprintf(stderr, "cpuid_data is full, no space for "
1794                             "cpuid(eax:0x%x,ecx:0x%x)\n", i, j);
1795                     abort();
1796                 }
1797                 c = &cpuid_data.entries[cpuid_i++];
1798             }
1799             break;
1800         default:
1801             c->function = i;
1802             c->flags = 0;
1803             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1804             if (!c->eax && !c->ebx && !c->ecx && !c->edx) {
1805                 /*
1806                  * KVM already returns all zeroes if a CPUID entry is missing,
1807                  * so we can omit it and avoid hitting KVM's 80-entry limit.
1808                  */
1809                 cpuid_i--;
1810             }
1811             break;
1812         }
1813     }
1814 
1815     /* Call Centaur's CPUID instructions they are supported. */
1816     if (env->cpuid_xlevel2 > 0) {
1817         cpu_x86_cpuid(env, 0xC0000000, 0, &limit, &unused, &unused, &unused);
1818 
1819         for (i = 0xC0000000; i <= limit; i++) {
1820             if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1821                 fprintf(stderr, "unsupported xlevel2 value: 0x%x\n", limit);
1822                 abort();
1823             }
1824             c = &cpuid_data.entries[cpuid_i++];
1825 
1826             c->function = i;
1827             c->flags = 0;
1828             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1829         }
1830     }
1831 
1832     cpuid_data.cpuid.nent = cpuid_i;
1833 
1834     if (((env->cpuid_version >> 8)&0xF) >= 6
1835         && (env->features[FEAT_1_EDX] & (CPUID_MCE | CPUID_MCA)) ==
1836            (CPUID_MCE | CPUID_MCA)
1837         && kvm_check_extension(cs->kvm_state, KVM_CAP_MCE) > 0) {
1838         uint64_t mcg_cap, unsupported_caps;
1839         int banks;
1840         int ret;
1841 
1842         ret = kvm_get_mce_cap_supported(cs->kvm_state, &mcg_cap, &banks);
1843         if (ret < 0) {
1844             fprintf(stderr, "kvm_get_mce_cap_supported: %s", strerror(-ret));
1845             return ret;
1846         }
1847 
1848         if (banks < (env->mcg_cap & MCG_CAP_BANKS_MASK)) {
1849             error_report("kvm: Unsupported MCE bank count (QEMU = %d, KVM = %d)",
1850                          (int)(env->mcg_cap & MCG_CAP_BANKS_MASK), banks);
1851             return -ENOTSUP;
1852         }
1853 
1854         unsupported_caps = env->mcg_cap & ~(mcg_cap | MCG_CAP_BANKS_MASK);
1855         if (unsupported_caps) {
1856             if (unsupported_caps & MCG_LMCE_P) {
1857                 error_report("kvm: LMCE not supported");
1858                 return -ENOTSUP;
1859             }
1860             warn_report("Unsupported MCG_CAP bits: 0x%" PRIx64,
1861                         unsupported_caps);
1862         }
1863 
1864         env->mcg_cap &= mcg_cap | MCG_CAP_BANKS_MASK;
1865         ret = kvm_vcpu_ioctl(cs, KVM_X86_SETUP_MCE, &env->mcg_cap);
1866         if (ret < 0) {
1867             fprintf(stderr, "KVM_X86_SETUP_MCE: %s", strerror(-ret));
1868             return ret;
1869         }
1870     }
1871 
1872     cpu->vmsentry = qemu_add_vm_change_state_handler(cpu_update_state, env);
1873 
1874     c = cpuid_find_entry(&cpuid_data.cpuid, 1, 0);
1875     if (c) {
1876         has_msr_feature_control = !!(c->ecx & CPUID_EXT_VMX) ||
1877                                   !!(c->ecx & CPUID_EXT_SMX);
1878     }
1879 
1880     if (env->mcg_cap & MCG_LMCE_P) {
1881         has_msr_mcg_ext_ctl = has_msr_feature_control = true;
1882     }
1883 
1884     if (!env->user_tsc_khz) {
1885         if ((env->features[FEAT_8000_0007_EDX] & CPUID_APM_INVTSC) &&
1886             invtsc_mig_blocker == NULL) {
1887             error_setg(&invtsc_mig_blocker,
1888                        "State blocked by non-migratable CPU device"
1889                        " (invtsc flag)");
1890             r = migrate_add_blocker(invtsc_mig_blocker, &local_err);
1891             if (r < 0) {
1892                 error_report_err(local_err);
1893                 return r;
1894             }
1895         }
1896     }
1897 
1898     if (cpu->vmware_cpuid_freq
1899         /* Guests depend on 0x40000000 to detect this feature, so only expose
1900          * it if KVM exposes leaf 0x40000000. (Conflicts with Hyper-V) */
1901         && cpu->expose_kvm
1902         && kvm_base == KVM_CPUID_SIGNATURE
1903         /* TSC clock must be stable and known for this feature. */
1904         && tsc_is_stable_and_known(env)) {
1905 
1906         c = &cpuid_data.entries[cpuid_i++];
1907         c->function = KVM_CPUID_SIGNATURE | 0x10;
1908         c->eax = env->tsc_khz;
1909         c->ebx = env->apic_bus_freq / 1000; /* Hz to KHz */
1910         c->ecx = c->edx = 0;
1911 
1912         c = cpuid_find_entry(&cpuid_data.cpuid, kvm_base, 0);
1913         c->eax = MAX(c->eax, KVM_CPUID_SIGNATURE | 0x10);
1914     }
1915 
1916     cpuid_data.cpuid.nent = cpuid_i;
1917 
1918     cpuid_data.cpuid.padding = 0;
1919     r = kvm_vcpu_ioctl(cs, KVM_SET_CPUID2, &cpuid_data);
1920     if (r) {
1921         goto fail;
1922     }
1923 
1924     if (has_xsave) {
1925         env->xsave_buf_len = sizeof(struct kvm_xsave);
1926         env->xsave_buf = qemu_memalign(4096, env->xsave_buf_len);
1927         memset(env->xsave_buf, 0, env->xsave_buf_len);
1928 
1929         /*
1930          * The allocated storage must be large enough for all of the
1931          * possible XSAVE state components.
1932          */
1933         assert(kvm_arch_get_supported_cpuid(kvm_state, 0xd, 0, R_ECX)
1934                <= env->xsave_buf_len);
1935     }
1936 
1937     max_nested_state_len = kvm_max_nested_state_length();
1938     if (max_nested_state_len > 0) {
1939         assert(max_nested_state_len >= offsetof(struct kvm_nested_state, data));
1940 
1941         if (cpu_has_vmx(env) || cpu_has_svm(env)) {
1942             struct kvm_vmx_nested_state_hdr *vmx_hdr;
1943 
1944             env->nested_state = g_malloc0(max_nested_state_len);
1945             env->nested_state->size = max_nested_state_len;
1946 
1947             if (cpu_has_vmx(env)) {
1948                 env->nested_state->format = KVM_STATE_NESTED_FORMAT_VMX;
1949                 vmx_hdr = &env->nested_state->hdr.vmx;
1950                 vmx_hdr->vmxon_pa = -1ull;
1951                 vmx_hdr->vmcs12_pa = -1ull;
1952             } else {
1953                 env->nested_state->format = KVM_STATE_NESTED_FORMAT_SVM;
1954             }
1955         }
1956     }
1957 
1958     cpu->kvm_msr_buf = g_malloc0(MSR_BUF_SIZE);
1959 
1960     if (!(env->features[FEAT_8000_0001_EDX] & CPUID_EXT2_RDTSCP)) {
1961         has_msr_tsc_aux = false;
1962     }
1963 
1964     kvm_init_msrs(cpu);
1965 
1966     return 0;
1967 
1968  fail:
1969     migrate_del_blocker(invtsc_mig_blocker);
1970 
1971     return r;
1972 }
1973 
1974 int kvm_arch_destroy_vcpu(CPUState *cs)
1975 {
1976     X86CPU *cpu = X86_CPU(cs);
1977     CPUX86State *env = &cpu->env;
1978 
1979     if (cpu->kvm_msr_buf) {
1980         g_free(cpu->kvm_msr_buf);
1981         cpu->kvm_msr_buf = NULL;
1982     }
1983 
1984     if (env->nested_state) {
1985         g_free(env->nested_state);
1986         env->nested_state = NULL;
1987     }
1988 
1989     qemu_del_vm_change_state_handler(cpu->vmsentry);
1990 
1991     return 0;
1992 }
1993 
1994 void kvm_arch_reset_vcpu(X86CPU *cpu)
1995 {
1996     CPUX86State *env = &cpu->env;
1997 
1998     env->xcr0 = 1;
1999     if (kvm_irqchip_in_kernel()) {
2000         env->mp_state = cpu_is_bsp(cpu) ? KVM_MP_STATE_RUNNABLE :
2001                                           KVM_MP_STATE_UNINITIALIZED;
2002     } else {
2003         env->mp_state = KVM_MP_STATE_RUNNABLE;
2004     }
2005 
2006     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_SYNIC)) {
2007         int i;
2008         for (i = 0; i < ARRAY_SIZE(env->msr_hv_synic_sint); i++) {
2009             env->msr_hv_synic_sint[i] = HV_SINT_MASKED;
2010         }
2011 
2012         hyperv_x86_synic_reset(cpu);
2013     }
2014     /* enabled by default */
2015     env->poll_control_msr = 1;
2016 
2017     sev_es_set_reset_vector(CPU(cpu));
2018 }
2019 
2020 void kvm_arch_do_init_vcpu(X86CPU *cpu)
2021 {
2022     CPUX86State *env = &cpu->env;
2023 
2024     /* APs get directly into wait-for-SIPI state.  */
2025     if (env->mp_state == KVM_MP_STATE_UNINITIALIZED) {
2026         env->mp_state = KVM_MP_STATE_INIT_RECEIVED;
2027     }
2028 }
2029 
2030 static int kvm_get_supported_feature_msrs(KVMState *s)
2031 {
2032     int ret = 0;
2033 
2034     if (kvm_feature_msrs != NULL) {
2035         return 0;
2036     }
2037 
2038     if (!kvm_check_extension(s, KVM_CAP_GET_MSR_FEATURES)) {
2039         return 0;
2040     }
2041 
2042     struct kvm_msr_list msr_list;
2043 
2044     msr_list.nmsrs = 0;
2045     ret = kvm_ioctl(s, KVM_GET_MSR_FEATURE_INDEX_LIST, &msr_list);
2046     if (ret < 0 && ret != -E2BIG) {
2047         error_report("Fetch KVM feature MSR list failed: %s",
2048             strerror(-ret));
2049         return ret;
2050     }
2051 
2052     assert(msr_list.nmsrs > 0);
2053     kvm_feature_msrs = (struct kvm_msr_list *) \
2054         g_malloc0(sizeof(msr_list) +
2055                  msr_list.nmsrs * sizeof(msr_list.indices[0]));
2056 
2057     kvm_feature_msrs->nmsrs = msr_list.nmsrs;
2058     ret = kvm_ioctl(s, KVM_GET_MSR_FEATURE_INDEX_LIST, kvm_feature_msrs);
2059 
2060     if (ret < 0) {
2061         error_report("Fetch KVM feature MSR list failed: %s",
2062             strerror(-ret));
2063         g_free(kvm_feature_msrs);
2064         kvm_feature_msrs = NULL;
2065         return ret;
2066     }
2067 
2068     return 0;
2069 }
2070 
2071 static int kvm_get_supported_msrs(KVMState *s)
2072 {
2073     int ret = 0;
2074     struct kvm_msr_list msr_list, *kvm_msr_list;
2075 
2076     /*
2077      *  Obtain MSR list from KVM.  These are the MSRs that we must
2078      *  save/restore.
2079      */
2080     msr_list.nmsrs = 0;
2081     ret = kvm_ioctl(s, KVM_GET_MSR_INDEX_LIST, &msr_list);
2082     if (ret < 0 && ret != -E2BIG) {
2083         return ret;
2084     }
2085     /*
2086      * Old kernel modules had a bug and could write beyond the provided
2087      * memory. Allocate at least a safe amount of 1K.
2088      */
2089     kvm_msr_list = g_malloc0(MAX(1024, sizeof(msr_list) +
2090                                           msr_list.nmsrs *
2091                                           sizeof(msr_list.indices[0])));
2092 
2093     kvm_msr_list->nmsrs = msr_list.nmsrs;
2094     ret = kvm_ioctl(s, KVM_GET_MSR_INDEX_LIST, kvm_msr_list);
2095     if (ret >= 0) {
2096         int i;
2097 
2098         for (i = 0; i < kvm_msr_list->nmsrs; i++) {
2099             switch (kvm_msr_list->indices[i]) {
2100             case MSR_STAR:
2101                 has_msr_star = true;
2102                 break;
2103             case MSR_VM_HSAVE_PA:
2104                 has_msr_hsave_pa = true;
2105                 break;
2106             case MSR_TSC_AUX:
2107                 has_msr_tsc_aux = true;
2108                 break;
2109             case MSR_TSC_ADJUST:
2110                 has_msr_tsc_adjust = true;
2111                 break;
2112             case MSR_IA32_TSCDEADLINE:
2113                 has_msr_tsc_deadline = true;
2114                 break;
2115             case MSR_IA32_SMBASE:
2116                 has_msr_smbase = true;
2117                 break;
2118             case MSR_SMI_COUNT:
2119                 has_msr_smi_count = true;
2120                 break;
2121             case MSR_IA32_MISC_ENABLE:
2122                 has_msr_misc_enable = true;
2123                 break;
2124             case MSR_IA32_BNDCFGS:
2125                 has_msr_bndcfgs = true;
2126                 break;
2127             case MSR_IA32_XSS:
2128                 has_msr_xss = true;
2129                 break;
2130             case MSR_IA32_UMWAIT_CONTROL:
2131                 has_msr_umwait = true;
2132                 break;
2133             case HV_X64_MSR_CRASH_CTL:
2134                 has_msr_hv_crash = true;
2135                 break;
2136             case HV_X64_MSR_RESET:
2137                 has_msr_hv_reset = true;
2138                 break;
2139             case HV_X64_MSR_VP_INDEX:
2140                 has_msr_hv_vpindex = true;
2141                 break;
2142             case HV_X64_MSR_VP_RUNTIME:
2143                 has_msr_hv_runtime = true;
2144                 break;
2145             case HV_X64_MSR_SCONTROL:
2146                 has_msr_hv_synic = true;
2147                 break;
2148             case HV_X64_MSR_STIMER0_CONFIG:
2149                 has_msr_hv_stimer = true;
2150                 break;
2151             case HV_X64_MSR_TSC_FREQUENCY:
2152                 has_msr_hv_frequencies = true;
2153                 break;
2154             case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2155                 has_msr_hv_reenlightenment = true;
2156                 break;
2157             case MSR_IA32_SPEC_CTRL:
2158                 has_msr_spec_ctrl = true;
2159                 break;
2160             case MSR_IA32_TSX_CTRL:
2161                 has_msr_tsx_ctrl = true;
2162                 break;
2163             case MSR_VIRT_SSBD:
2164                 has_msr_virt_ssbd = true;
2165                 break;
2166             case MSR_IA32_ARCH_CAPABILITIES:
2167                 has_msr_arch_capabs = true;
2168                 break;
2169             case MSR_IA32_CORE_CAPABILITY:
2170                 has_msr_core_capabs = true;
2171                 break;
2172             case MSR_IA32_PERF_CAPABILITIES:
2173                 has_msr_perf_capabs = true;
2174                 break;
2175             case MSR_IA32_VMX_VMFUNC:
2176                 has_msr_vmx_vmfunc = true;
2177                 break;
2178             case MSR_IA32_UCODE_REV:
2179                 has_msr_ucode_rev = true;
2180                 break;
2181             case MSR_IA32_VMX_PROCBASED_CTLS2:
2182                 has_msr_vmx_procbased_ctls2 = true;
2183                 break;
2184             case MSR_IA32_PKRS:
2185                 has_msr_pkrs = true;
2186                 break;
2187             }
2188         }
2189     }
2190 
2191     g_free(kvm_msr_list);
2192 
2193     return ret;
2194 }
2195 
2196 static Notifier smram_machine_done;
2197 static KVMMemoryListener smram_listener;
2198 static AddressSpace smram_address_space;
2199 static MemoryRegion smram_as_root;
2200 static MemoryRegion smram_as_mem;
2201 
2202 static void register_smram_listener(Notifier *n, void *unused)
2203 {
2204     MemoryRegion *smram =
2205         (MemoryRegion *) object_resolve_path("/machine/smram", NULL);
2206 
2207     /* Outer container... */
2208     memory_region_init(&smram_as_root, OBJECT(kvm_state), "mem-container-smram", ~0ull);
2209     memory_region_set_enabled(&smram_as_root, true);
2210 
2211     /* ... with two regions inside: normal system memory with low
2212      * priority, and...
2213      */
2214     memory_region_init_alias(&smram_as_mem, OBJECT(kvm_state), "mem-smram",
2215                              get_system_memory(), 0, ~0ull);
2216     memory_region_add_subregion_overlap(&smram_as_root, 0, &smram_as_mem, 0);
2217     memory_region_set_enabled(&smram_as_mem, true);
2218 
2219     if (smram) {
2220         /* ... SMRAM with higher priority */
2221         memory_region_add_subregion_overlap(&smram_as_root, 0, smram, 10);
2222         memory_region_set_enabled(smram, true);
2223     }
2224 
2225     address_space_init(&smram_address_space, &smram_as_root, "KVM-SMRAM");
2226     kvm_memory_listener_register(kvm_state, &smram_listener,
2227                                  &smram_address_space, 1);
2228 }
2229 
2230 int kvm_arch_init(MachineState *ms, KVMState *s)
2231 {
2232     uint64_t identity_base = 0xfffbc000;
2233     uint64_t shadow_mem;
2234     int ret;
2235     struct utsname utsname;
2236     Error *local_err = NULL;
2237 
2238     /*
2239      * Initialize SEV context, if required
2240      *
2241      * If no memory encryption is requested (ms->cgs == NULL) this is
2242      * a no-op.
2243      *
2244      * It's also a no-op if a non-SEV confidential guest support
2245      * mechanism is selected.  SEV is the only mechanism available to
2246      * select on x86 at present, so this doesn't arise, but if new
2247      * mechanisms are supported in future (e.g. TDX), they'll need
2248      * their own initialization either here or elsewhere.
2249      */
2250     ret = sev_kvm_init(ms->cgs, &local_err);
2251     if (ret < 0) {
2252         error_report_err(local_err);
2253         return ret;
2254     }
2255 
2256     if (!kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
2257         error_report("kvm: KVM_CAP_IRQ_ROUTING not supported by KVM");
2258         return -ENOTSUP;
2259     }
2260 
2261     has_xsave = kvm_check_extension(s, KVM_CAP_XSAVE);
2262     has_xcrs = kvm_check_extension(s, KVM_CAP_XCRS);
2263     has_pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2);
2264 
2265     hv_vpindex_settable = kvm_check_extension(s, KVM_CAP_HYPERV_VP_INDEX);
2266 
2267     has_exception_payload = kvm_check_extension(s, KVM_CAP_EXCEPTION_PAYLOAD);
2268     if (has_exception_payload) {
2269         ret = kvm_vm_enable_cap(s, KVM_CAP_EXCEPTION_PAYLOAD, 0, true);
2270         if (ret < 0) {
2271             error_report("kvm: Failed to enable exception payload cap: %s",
2272                          strerror(-ret));
2273             return ret;
2274         }
2275     }
2276 
2277     ret = kvm_get_supported_msrs(s);
2278     if (ret < 0) {
2279         return ret;
2280     }
2281 
2282     kvm_get_supported_feature_msrs(s);
2283 
2284     uname(&utsname);
2285     lm_capable_kernel = strcmp(utsname.machine, "x86_64") == 0;
2286 
2287     /*
2288      * On older Intel CPUs, KVM uses vm86 mode to emulate 16-bit code directly.
2289      * In order to use vm86 mode, an EPT identity map and a TSS  are needed.
2290      * Since these must be part of guest physical memory, we need to allocate
2291      * them, both by setting their start addresses in the kernel and by
2292      * creating a corresponding e820 entry. We need 4 pages before the BIOS.
2293      *
2294      * Older KVM versions may not support setting the identity map base. In
2295      * that case we need to stick with the default, i.e. a 256K maximum BIOS
2296      * size.
2297      */
2298     if (kvm_check_extension(s, KVM_CAP_SET_IDENTITY_MAP_ADDR)) {
2299         /* Allows up to 16M BIOSes. */
2300         identity_base = 0xfeffc000;
2301 
2302         ret = kvm_vm_ioctl(s, KVM_SET_IDENTITY_MAP_ADDR, &identity_base);
2303         if (ret < 0) {
2304             return ret;
2305         }
2306     }
2307 
2308     /* Set TSS base one page after EPT identity map. */
2309     ret = kvm_vm_ioctl(s, KVM_SET_TSS_ADDR, identity_base + 0x1000);
2310     if (ret < 0) {
2311         return ret;
2312     }
2313 
2314     /* Tell fw_cfg to notify the BIOS to reserve the range. */
2315     ret = e820_add_entry(identity_base, 0x4000, E820_RESERVED);
2316     if (ret < 0) {
2317         fprintf(stderr, "e820_add_entry() table is full\n");
2318         return ret;
2319     }
2320 
2321     shadow_mem = object_property_get_int(OBJECT(s), "kvm-shadow-mem", &error_abort);
2322     if (shadow_mem != -1) {
2323         shadow_mem /= 4096;
2324         ret = kvm_vm_ioctl(s, KVM_SET_NR_MMU_PAGES, shadow_mem);
2325         if (ret < 0) {
2326             return ret;
2327         }
2328     }
2329 
2330     if (kvm_check_extension(s, KVM_CAP_X86_SMM) &&
2331         object_dynamic_cast(OBJECT(ms), TYPE_X86_MACHINE) &&
2332         x86_machine_is_smm_enabled(X86_MACHINE(ms))) {
2333         smram_machine_done.notify = register_smram_listener;
2334         qemu_add_machine_init_done_notifier(&smram_machine_done);
2335     }
2336 
2337     if (enable_cpu_pm) {
2338         int disable_exits = kvm_check_extension(s, KVM_CAP_X86_DISABLE_EXITS);
2339         int ret;
2340 
2341 /* Work around for kernel header with a typo. TODO: fix header and drop. */
2342 #if defined(KVM_X86_DISABLE_EXITS_HTL) && !defined(KVM_X86_DISABLE_EXITS_HLT)
2343 #define KVM_X86_DISABLE_EXITS_HLT KVM_X86_DISABLE_EXITS_HTL
2344 #endif
2345         if (disable_exits) {
2346             disable_exits &= (KVM_X86_DISABLE_EXITS_MWAIT |
2347                               KVM_X86_DISABLE_EXITS_HLT |
2348                               KVM_X86_DISABLE_EXITS_PAUSE |
2349                               KVM_X86_DISABLE_EXITS_CSTATE);
2350         }
2351 
2352         ret = kvm_vm_enable_cap(s, KVM_CAP_X86_DISABLE_EXITS, 0,
2353                                 disable_exits);
2354         if (ret < 0) {
2355             error_report("kvm: guest stopping CPU not supported: %s",
2356                          strerror(-ret));
2357         }
2358     }
2359 
2360     if (object_dynamic_cast(OBJECT(ms), TYPE_X86_MACHINE)) {
2361         X86MachineState *x86ms = X86_MACHINE(ms);
2362 
2363         if (x86ms->bus_lock_ratelimit > 0) {
2364             ret = kvm_check_extension(s, KVM_CAP_X86_BUS_LOCK_EXIT);
2365             if (!(ret & KVM_BUS_LOCK_DETECTION_EXIT)) {
2366                 error_report("kvm: bus lock detection unsupported");
2367                 return -ENOTSUP;
2368             }
2369             ret = kvm_vm_enable_cap(s, KVM_CAP_X86_BUS_LOCK_EXIT, 0,
2370                                     KVM_BUS_LOCK_DETECTION_EXIT);
2371             if (ret < 0) {
2372                 error_report("kvm: Failed to enable bus lock detection cap: %s",
2373                              strerror(-ret));
2374                 return ret;
2375             }
2376             ratelimit_init(&bus_lock_ratelimit_ctrl);
2377             ratelimit_set_speed(&bus_lock_ratelimit_ctrl,
2378                                 x86ms->bus_lock_ratelimit, BUS_LOCK_SLICE_TIME);
2379         }
2380     }
2381 
2382     return 0;
2383 }
2384 
2385 static void set_v8086_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
2386 {
2387     lhs->selector = rhs->selector;
2388     lhs->base = rhs->base;
2389     lhs->limit = rhs->limit;
2390     lhs->type = 3;
2391     lhs->present = 1;
2392     lhs->dpl = 3;
2393     lhs->db = 0;
2394     lhs->s = 1;
2395     lhs->l = 0;
2396     lhs->g = 0;
2397     lhs->avl = 0;
2398     lhs->unusable = 0;
2399 }
2400 
2401 static void set_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
2402 {
2403     unsigned flags = rhs->flags;
2404     lhs->selector = rhs->selector;
2405     lhs->base = rhs->base;
2406     lhs->limit = rhs->limit;
2407     lhs->type = (flags >> DESC_TYPE_SHIFT) & 15;
2408     lhs->present = (flags & DESC_P_MASK) != 0;
2409     lhs->dpl = (flags >> DESC_DPL_SHIFT) & 3;
2410     lhs->db = (flags >> DESC_B_SHIFT) & 1;
2411     lhs->s = (flags & DESC_S_MASK) != 0;
2412     lhs->l = (flags >> DESC_L_SHIFT) & 1;
2413     lhs->g = (flags & DESC_G_MASK) != 0;
2414     lhs->avl = (flags & DESC_AVL_MASK) != 0;
2415     lhs->unusable = !lhs->present;
2416     lhs->padding = 0;
2417 }
2418 
2419 static void get_seg(SegmentCache *lhs, const struct kvm_segment *rhs)
2420 {
2421     lhs->selector = rhs->selector;
2422     lhs->base = rhs->base;
2423     lhs->limit = rhs->limit;
2424     lhs->flags = (rhs->type << DESC_TYPE_SHIFT) |
2425                  ((rhs->present && !rhs->unusable) * DESC_P_MASK) |
2426                  (rhs->dpl << DESC_DPL_SHIFT) |
2427                  (rhs->db << DESC_B_SHIFT) |
2428                  (rhs->s * DESC_S_MASK) |
2429                  (rhs->l << DESC_L_SHIFT) |
2430                  (rhs->g * DESC_G_MASK) |
2431                  (rhs->avl * DESC_AVL_MASK);
2432 }
2433 
2434 static void kvm_getput_reg(__u64 *kvm_reg, target_ulong *qemu_reg, int set)
2435 {
2436     if (set) {
2437         *kvm_reg = *qemu_reg;
2438     } else {
2439         *qemu_reg = *kvm_reg;
2440     }
2441 }
2442 
2443 static int kvm_getput_regs(X86CPU *cpu, int set)
2444 {
2445     CPUX86State *env = &cpu->env;
2446     struct kvm_regs regs;
2447     int ret = 0;
2448 
2449     if (!set) {
2450         ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_REGS, &regs);
2451         if (ret < 0) {
2452             return ret;
2453         }
2454     }
2455 
2456     kvm_getput_reg(&regs.rax, &env->regs[R_EAX], set);
2457     kvm_getput_reg(&regs.rbx, &env->regs[R_EBX], set);
2458     kvm_getput_reg(&regs.rcx, &env->regs[R_ECX], set);
2459     kvm_getput_reg(&regs.rdx, &env->regs[R_EDX], set);
2460     kvm_getput_reg(&regs.rsi, &env->regs[R_ESI], set);
2461     kvm_getput_reg(&regs.rdi, &env->regs[R_EDI], set);
2462     kvm_getput_reg(&regs.rsp, &env->regs[R_ESP], set);
2463     kvm_getput_reg(&regs.rbp, &env->regs[R_EBP], set);
2464 #ifdef TARGET_X86_64
2465     kvm_getput_reg(&regs.r8, &env->regs[8], set);
2466     kvm_getput_reg(&regs.r9, &env->regs[9], set);
2467     kvm_getput_reg(&regs.r10, &env->regs[10], set);
2468     kvm_getput_reg(&regs.r11, &env->regs[11], set);
2469     kvm_getput_reg(&regs.r12, &env->regs[12], set);
2470     kvm_getput_reg(&regs.r13, &env->regs[13], set);
2471     kvm_getput_reg(&regs.r14, &env->regs[14], set);
2472     kvm_getput_reg(&regs.r15, &env->regs[15], set);
2473 #endif
2474 
2475     kvm_getput_reg(&regs.rflags, &env->eflags, set);
2476     kvm_getput_reg(&regs.rip, &env->eip, set);
2477 
2478     if (set) {
2479         ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_REGS, &regs);
2480     }
2481 
2482     return ret;
2483 }
2484 
2485 static int kvm_put_fpu(X86CPU *cpu)
2486 {
2487     CPUX86State *env = &cpu->env;
2488     struct kvm_fpu fpu;
2489     int i;
2490 
2491     memset(&fpu, 0, sizeof fpu);
2492     fpu.fsw = env->fpus & ~(7 << 11);
2493     fpu.fsw |= (env->fpstt & 7) << 11;
2494     fpu.fcw = env->fpuc;
2495     fpu.last_opcode = env->fpop;
2496     fpu.last_ip = env->fpip;
2497     fpu.last_dp = env->fpdp;
2498     for (i = 0; i < 8; ++i) {
2499         fpu.ftwx |= (!env->fptags[i]) << i;
2500     }
2501     memcpy(fpu.fpr, env->fpregs, sizeof env->fpregs);
2502     for (i = 0; i < CPU_NB_REGS; i++) {
2503         stq_p(&fpu.xmm[i][0], env->xmm_regs[i].ZMM_Q(0));
2504         stq_p(&fpu.xmm[i][8], env->xmm_regs[i].ZMM_Q(1));
2505     }
2506     fpu.mxcsr = env->mxcsr;
2507 
2508     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_FPU, &fpu);
2509 }
2510 
2511 static int kvm_put_xsave(X86CPU *cpu)
2512 {
2513     CPUX86State *env = &cpu->env;
2514     void *xsave = env->xsave_buf;
2515 
2516     if (!has_xsave) {
2517         return kvm_put_fpu(cpu);
2518     }
2519     x86_cpu_xsave_all_areas(cpu, xsave, env->xsave_buf_len);
2520 
2521     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_XSAVE, xsave);
2522 }
2523 
2524 static int kvm_put_xcrs(X86CPU *cpu)
2525 {
2526     CPUX86State *env = &cpu->env;
2527     struct kvm_xcrs xcrs = {};
2528 
2529     if (!has_xcrs) {
2530         return 0;
2531     }
2532 
2533     xcrs.nr_xcrs = 1;
2534     xcrs.flags = 0;
2535     xcrs.xcrs[0].xcr = 0;
2536     xcrs.xcrs[0].value = env->xcr0;
2537     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_XCRS, &xcrs);
2538 }
2539 
2540 static int kvm_put_sregs(X86CPU *cpu)
2541 {
2542     CPUX86State *env = &cpu->env;
2543     struct kvm_sregs sregs;
2544 
2545     memset(sregs.interrupt_bitmap, 0, sizeof(sregs.interrupt_bitmap));
2546     if (env->interrupt_injected >= 0) {
2547         sregs.interrupt_bitmap[env->interrupt_injected / 64] |=
2548                 (uint64_t)1 << (env->interrupt_injected % 64);
2549     }
2550 
2551     if ((env->eflags & VM_MASK)) {
2552         set_v8086_seg(&sregs.cs, &env->segs[R_CS]);
2553         set_v8086_seg(&sregs.ds, &env->segs[R_DS]);
2554         set_v8086_seg(&sregs.es, &env->segs[R_ES]);
2555         set_v8086_seg(&sregs.fs, &env->segs[R_FS]);
2556         set_v8086_seg(&sregs.gs, &env->segs[R_GS]);
2557         set_v8086_seg(&sregs.ss, &env->segs[R_SS]);
2558     } else {
2559         set_seg(&sregs.cs, &env->segs[R_CS]);
2560         set_seg(&sregs.ds, &env->segs[R_DS]);
2561         set_seg(&sregs.es, &env->segs[R_ES]);
2562         set_seg(&sregs.fs, &env->segs[R_FS]);
2563         set_seg(&sregs.gs, &env->segs[R_GS]);
2564         set_seg(&sregs.ss, &env->segs[R_SS]);
2565     }
2566 
2567     set_seg(&sregs.tr, &env->tr);
2568     set_seg(&sregs.ldt, &env->ldt);
2569 
2570     sregs.idt.limit = env->idt.limit;
2571     sregs.idt.base = env->idt.base;
2572     memset(sregs.idt.padding, 0, sizeof sregs.idt.padding);
2573     sregs.gdt.limit = env->gdt.limit;
2574     sregs.gdt.base = env->gdt.base;
2575     memset(sregs.gdt.padding, 0, sizeof sregs.gdt.padding);
2576 
2577     sregs.cr0 = env->cr[0];
2578     sregs.cr2 = env->cr[2];
2579     sregs.cr3 = env->cr[3];
2580     sregs.cr4 = env->cr[4];
2581 
2582     sregs.cr8 = cpu_get_apic_tpr(cpu->apic_state);
2583     sregs.apic_base = cpu_get_apic_base(cpu->apic_state);
2584 
2585     sregs.efer = env->efer;
2586 
2587     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_SREGS, &sregs);
2588 }
2589 
2590 static void kvm_msr_buf_reset(X86CPU *cpu)
2591 {
2592     memset(cpu->kvm_msr_buf, 0, MSR_BUF_SIZE);
2593 }
2594 
2595 static void kvm_msr_entry_add(X86CPU *cpu, uint32_t index, uint64_t value)
2596 {
2597     struct kvm_msrs *msrs = cpu->kvm_msr_buf;
2598     void *limit = ((void *)msrs) + MSR_BUF_SIZE;
2599     struct kvm_msr_entry *entry = &msrs->entries[msrs->nmsrs];
2600 
2601     assert((void *)(entry + 1) <= limit);
2602 
2603     entry->index = index;
2604     entry->reserved = 0;
2605     entry->data = value;
2606     msrs->nmsrs++;
2607 }
2608 
2609 static int kvm_put_one_msr(X86CPU *cpu, int index, uint64_t value)
2610 {
2611     kvm_msr_buf_reset(cpu);
2612     kvm_msr_entry_add(cpu, index, value);
2613 
2614     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MSRS, cpu->kvm_msr_buf);
2615 }
2616 
2617 void kvm_put_apicbase(X86CPU *cpu, uint64_t value)
2618 {
2619     int ret;
2620 
2621     ret = kvm_put_one_msr(cpu, MSR_IA32_APICBASE, value);
2622     assert(ret == 1);
2623 }
2624 
2625 static int kvm_put_tscdeadline_msr(X86CPU *cpu)
2626 {
2627     CPUX86State *env = &cpu->env;
2628     int ret;
2629 
2630     if (!has_msr_tsc_deadline) {
2631         return 0;
2632     }
2633 
2634     ret = kvm_put_one_msr(cpu, MSR_IA32_TSCDEADLINE, env->tsc_deadline);
2635     if (ret < 0) {
2636         return ret;
2637     }
2638 
2639     assert(ret == 1);
2640     return 0;
2641 }
2642 
2643 /*
2644  * Provide a separate write service for the feature control MSR in order to
2645  * kick the VCPU out of VMXON or even guest mode on reset. This has to be done
2646  * before writing any other state because forcibly leaving nested mode
2647  * invalidates the VCPU state.
2648  */
2649 static int kvm_put_msr_feature_control(X86CPU *cpu)
2650 {
2651     int ret;
2652 
2653     if (!has_msr_feature_control) {
2654         return 0;
2655     }
2656 
2657     ret = kvm_put_one_msr(cpu, MSR_IA32_FEATURE_CONTROL,
2658                           cpu->env.msr_ia32_feature_control);
2659     if (ret < 0) {
2660         return ret;
2661     }
2662 
2663     assert(ret == 1);
2664     return 0;
2665 }
2666 
2667 static uint64_t make_vmx_msr_value(uint32_t index, uint32_t features)
2668 {
2669     uint32_t default1, can_be_one, can_be_zero;
2670     uint32_t must_be_one;
2671 
2672     switch (index) {
2673     case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2674         default1 = 0x00000016;
2675         break;
2676     case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2677         default1 = 0x0401e172;
2678         break;
2679     case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2680         default1 = 0x000011ff;
2681         break;
2682     case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2683         default1 = 0x00036dff;
2684         break;
2685     case MSR_IA32_VMX_PROCBASED_CTLS2:
2686         default1 = 0;
2687         break;
2688     default:
2689         abort();
2690     }
2691 
2692     /* If a feature bit is set, the control can be either set or clear.
2693      * Otherwise the value is limited to either 0 or 1 by default1.
2694      */
2695     can_be_one = features | default1;
2696     can_be_zero = features | ~default1;
2697     must_be_one = ~can_be_zero;
2698 
2699     /*
2700      * Bit 0:31 -> 0 if the control bit can be zero (i.e. 1 if it must be one).
2701      * Bit 32:63 -> 1 if the control bit can be one.
2702      */
2703     return must_be_one | (((uint64_t)can_be_one) << 32);
2704 }
2705 
2706 static void kvm_msr_entry_add_vmx(X86CPU *cpu, FeatureWordArray f)
2707 {
2708     uint64_t kvm_vmx_basic =
2709         kvm_arch_get_supported_msr_feature(kvm_state,
2710                                            MSR_IA32_VMX_BASIC);
2711 
2712     if (!kvm_vmx_basic) {
2713         /* If the kernel doesn't support VMX feature (kvm_intel.nested=0),
2714          * then kvm_vmx_basic will be 0 and KVM_SET_MSR will fail.
2715          */
2716         return;
2717     }
2718 
2719     uint64_t kvm_vmx_misc =
2720         kvm_arch_get_supported_msr_feature(kvm_state,
2721                                            MSR_IA32_VMX_MISC);
2722     uint64_t kvm_vmx_ept_vpid =
2723         kvm_arch_get_supported_msr_feature(kvm_state,
2724                                            MSR_IA32_VMX_EPT_VPID_CAP);
2725 
2726     /*
2727      * If the guest is 64-bit, a value of 1 is allowed for the host address
2728      * space size vmexit control.
2729      */
2730     uint64_t fixed_vmx_exit = f[FEAT_8000_0001_EDX] & CPUID_EXT2_LM
2731         ? (uint64_t)VMX_VM_EXIT_HOST_ADDR_SPACE_SIZE << 32 : 0;
2732 
2733     /*
2734      * Bits 0-30, 32-44 and 50-53 come from the host.  KVM should
2735      * not change them for backwards compatibility.
2736      */
2737     uint64_t fixed_vmx_basic = kvm_vmx_basic &
2738         (MSR_VMX_BASIC_VMCS_REVISION_MASK |
2739          MSR_VMX_BASIC_VMXON_REGION_SIZE_MASK |
2740          MSR_VMX_BASIC_VMCS_MEM_TYPE_MASK);
2741 
2742     /*
2743      * Same for bits 0-4 and 25-27.  Bits 16-24 (CR3 target count) can
2744      * change in the future but are always zero for now, clear them to be
2745      * future proof.  Bits 32-63 in theory could change, though KVM does
2746      * not support dual-monitor treatment and probably never will; mask
2747      * them out as well.
2748      */
2749     uint64_t fixed_vmx_misc = kvm_vmx_misc &
2750         (MSR_VMX_MISC_PREEMPTION_TIMER_SHIFT_MASK |
2751          MSR_VMX_MISC_MAX_MSR_LIST_SIZE_MASK);
2752 
2753     /*
2754      * EPT memory types should not change either, so we do not bother
2755      * adding features for them.
2756      */
2757     uint64_t fixed_vmx_ept_mask =
2758             (f[FEAT_VMX_SECONDARY_CTLS] & VMX_SECONDARY_EXEC_ENABLE_EPT ?
2759              MSR_VMX_EPT_UC | MSR_VMX_EPT_WB : 0);
2760     uint64_t fixed_vmx_ept_vpid = kvm_vmx_ept_vpid & fixed_vmx_ept_mask;
2761 
2762     kvm_msr_entry_add(cpu, MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
2763                       make_vmx_msr_value(MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
2764                                          f[FEAT_VMX_PROCBASED_CTLS]));
2765     kvm_msr_entry_add(cpu, MSR_IA32_VMX_TRUE_PINBASED_CTLS,
2766                       make_vmx_msr_value(MSR_IA32_VMX_TRUE_PINBASED_CTLS,
2767                                          f[FEAT_VMX_PINBASED_CTLS]));
2768     kvm_msr_entry_add(cpu, MSR_IA32_VMX_TRUE_EXIT_CTLS,
2769                       make_vmx_msr_value(MSR_IA32_VMX_TRUE_EXIT_CTLS,
2770                                          f[FEAT_VMX_EXIT_CTLS]) | fixed_vmx_exit);
2771     kvm_msr_entry_add(cpu, MSR_IA32_VMX_TRUE_ENTRY_CTLS,
2772                       make_vmx_msr_value(MSR_IA32_VMX_TRUE_ENTRY_CTLS,
2773                                          f[FEAT_VMX_ENTRY_CTLS]));
2774     kvm_msr_entry_add(cpu, MSR_IA32_VMX_PROCBASED_CTLS2,
2775                       make_vmx_msr_value(MSR_IA32_VMX_PROCBASED_CTLS2,
2776                                          f[FEAT_VMX_SECONDARY_CTLS]));
2777     kvm_msr_entry_add(cpu, MSR_IA32_VMX_EPT_VPID_CAP,
2778                       f[FEAT_VMX_EPT_VPID_CAPS] | fixed_vmx_ept_vpid);
2779     kvm_msr_entry_add(cpu, MSR_IA32_VMX_BASIC,
2780                       f[FEAT_VMX_BASIC] | fixed_vmx_basic);
2781     kvm_msr_entry_add(cpu, MSR_IA32_VMX_MISC,
2782                       f[FEAT_VMX_MISC] | fixed_vmx_misc);
2783     if (has_msr_vmx_vmfunc) {
2784         kvm_msr_entry_add(cpu, MSR_IA32_VMX_VMFUNC, f[FEAT_VMX_VMFUNC]);
2785     }
2786 
2787     /*
2788      * Just to be safe, write these with constant values.  The CRn_FIXED1
2789      * MSRs are generated by KVM based on the vCPU's CPUID.
2790      */
2791     kvm_msr_entry_add(cpu, MSR_IA32_VMX_CR0_FIXED0,
2792                       CR0_PE_MASK | CR0_PG_MASK | CR0_NE_MASK);
2793     kvm_msr_entry_add(cpu, MSR_IA32_VMX_CR4_FIXED0,
2794                       CR4_VMXE_MASK);
2795 
2796     if (f[FEAT_VMX_SECONDARY_CTLS] & VMX_SECONDARY_EXEC_TSC_SCALING) {
2797         /* TSC multiplier (0x2032).  */
2798         kvm_msr_entry_add(cpu, MSR_IA32_VMX_VMCS_ENUM, 0x32);
2799     } else {
2800         /* Preemption timer (0x482E).  */
2801         kvm_msr_entry_add(cpu, MSR_IA32_VMX_VMCS_ENUM, 0x2E);
2802     }
2803 }
2804 
2805 static void kvm_msr_entry_add_perf(X86CPU *cpu, FeatureWordArray f)
2806 {
2807     uint64_t kvm_perf_cap =
2808         kvm_arch_get_supported_msr_feature(kvm_state,
2809                                            MSR_IA32_PERF_CAPABILITIES);
2810 
2811     if (kvm_perf_cap) {
2812         kvm_msr_entry_add(cpu, MSR_IA32_PERF_CAPABILITIES,
2813                         kvm_perf_cap & f[FEAT_PERF_CAPABILITIES]);
2814     }
2815 }
2816 
2817 static int kvm_buf_set_msrs(X86CPU *cpu)
2818 {
2819     int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MSRS, cpu->kvm_msr_buf);
2820     if (ret < 0) {
2821         return ret;
2822     }
2823 
2824     if (ret < cpu->kvm_msr_buf->nmsrs) {
2825         struct kvm_msr_entry *e = &cpu->kvm_msr_buf->entries[ret];
2826         error_report("error: failed to set MSR 0x%" PRIx32 " to 0x%" PRIx64,
2827                      (uint32_t)e->index, (uint64_t)e->data);
2828     }
2829 
2830     assert(ret == cpu->kvm_msr_buf->nmsrs);
2831     return 0;
2832 }
2833 
2834 static void kvm_init_msrs(X86CPU *cpu)
2835 {
2836     CPUX86State *env = &cpu->env;
2837 
2838     kvm_msr_buf_reset(cpu);
2839     if (has_msr_arch_capabs) {
2840         kvm_msr_entry_add(cpu, MSR_IA32_ARCH_CAPABILITIES,
2841                           env->features[FEAT_ARCH_CAPABILITIES]);
2842     }
2843 
2844     if (has_msr_core_capabs) {
2845         kvm_msr_entry_add(cpu, MSR_IA32_CORE_CAPABILITY,
2846                           env->features[FEAT_CORE_CAPABILITY]);
2847     }
2848 
2849     if (has_msr_perf_capabs && cpu->enable_pmu) {
2850         kvm_msr_entry_add_perf(cpu, env->features);
2851     }
2852 
2853     if (has_msr_ucode_rev) {
2854         kvm_msr_entry_add(cpu, MSR_IA32_UCODE_REV, cpu->ucode_rev);
2855     }
2856 
2857     /*
2858      * Older kernels do not include VMX MSRs in KVM_GET_MSR_INDEX_LIST, but
2859      * all kernels with MSR features should have them.
2860      */
2861     if (kvm_feature_msrs && cpu_has_vmx(env)) {
2862         kvm_msr_entry_add_vmx(cpu, env->features);
2863     }
2864 
2865     assert(kvm_buf_set_msrs(cpu) == 0);
2866 }
2867 
2868 static int kvm_put_msrs(X86CPU *cpu, int level)
2869 {
2870     CPUX86State *env = &cpu->env;
2871     int i;
2872 
2873     kvm_msr_buf_reset(cpu);
2874 
2875     kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_CS, env->sysenter_cs);
2876     kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_ESP, env->sysenter_esp);
2877     kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_EIP, env->sysenter_eip);
2878     kvm_msr_entry_add(cpu, MSR_PAT, env->pat);
2879     if (has_msr_star) {
2880         kvm_msr_entry_add(cpu, MSR_STAR, env->star);
2881     }
2882     if (has_msr_hsave_pa) {
2883         kvm_msr_entry_add(cpu, MSR_VM_HSAVE_PA, env->vm_hsave);
2884     }
2885     if (has_msr_tsc_aux) {
2886         kvm_msr_entry_add(cpu, MSR_TSC_AUX, env->tsc_aux);
2887     }
2888     if (has_msr_tsc_adjust) {
2889         kvm_msr_entry_add(cpu, MSR_TSC_ADJUST, env->tsc_adjust);
2890     }
2891     if (has_msr_misc_enable) {
2892         kvm_msr_entry_add(cpu, MSR_IA32_MISC_ENABLE,
2893                           env->msr_ia32_misc_enable);
2894     }
2895     if (has_msr_smbase) {
2896         kvm_msr_entry_add(cpu, MSR_IA32_SMBASE, env->smbase);
2897     }
2898     if (has_msr_smi_count) {
2899         kvm_msr_entry_add(cpu, MSR_SMI_COUNT, env->msr_smi_count);
2900     }
2901     if (has_msr_pkrs) {
2902         kvm_msr_entry_add(cpu, MSR_IA32_PKRS, env->pkrs);
2903     }
2904     if (has_msr_bndcfgs) {
2905         kvm_msr_entry_add(cpu, MSR_IA32_BNDCFGS, env->msr_bndcfgs);
2906     }
2907     if (has_msr_xss) {
2908         kvm_msr_entry_add(cpu, MSR_IA32_XSS, env->xss);
2909     }
2910     if (has_msr_umwait) {
2911         kvm_msr_entry_add(cpu, MSR_IA32_UMWAIT_CONTROL, env->umwait);
2912     }
2913     if (has_msr_spec_ctrl) {
2914         kvm_msr_entry_add(cpu, MSR_IA32_SPEC_CTRL, env->spec_ctrl);
2915     }
2916     if (has_msr_tsx_ctrl) {
2917         kvm_msr_entry_add(cpu, MSR_IA32_TSX_CTRL, env->tsx_ctrl);
2918     }
2919     if (has_msr_virt_ssbd) {
2920         kvm_msr_entry_add(cpu, MSR_VIRT_SSBD, env->virt_ssbd);
2921     }
2922 
2923 #ifdef TARGET_X86_64
2924     if (lm_capable_kernel) {
2925         kvm_msr_entry_add(cpu, MSR_CSTAR, env->cstar);
2926         kvm_msr_entry_add(cpu, MSR_KERNELGSBASE, env->kernelgsbase);
2927         kvm_msr_entry_add(cpu, MSR_FMASK, env->fmask);
2928         kvm_msr_entry_add(cpu, MSR_LSTAR, env->lstar);
2929     }
2930 #endif
2931 
2932     /*
2933      * The following MSRs have side effects on the guest or are too heavy
2934      * for normal writeback. Limit them to reset or full state updates.
2935      */
2936     if (level >= KVM_PUT_RESET_STATE) {
2937         kvm_msr_entry_add(cpu, MSR_IA32_TSC, env->tsc);
2938         kvm_msr_entry_add(cpu, MSR_KVM_SYSTEM_TIME, env->system_time_msr);
2939         kvm_msr_entry_add(cpu, MSR_KVM_WALL_CLOCK, env->wall_clock_msr);
2940         if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_ASYNC_PF_INT)) {
2941             kvm_msr_entry_add(cpu, MSR_KVM_ASYNC_PF_INT, env->async_pf_int_msr);
2942         }
2943         if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_ASYNC_PF)) {
2944             kvm_msr_entry_add(cpu, MSR_KVM_ASYNC_PF_EN, env->async_pf_en_msr);
2945         }
2946         if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_PV_EOI)) {
2947             kvm_msr_entry_add(cpu, MSR_KVM_PV_EOI_EN, env->pv_eoi_en_msr);
2948         }
2949         if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_STEAL_TIME)) {
2950             kvm_msr_entry_add(cpu, MSR_KVM_STEAL_TIME, env->steal_time_msr);
2951         }
2952 
2953         if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_POLL_CONTROL)) {
2954             kvm_msr_entry_add(cpu, MSR_KVM_POLL_CONTROL, env->poll_control_msr);
2955         }
2956 
2957         if (has_architectural_pmu_version > 0) {
2958             if (has_architectural_pmu_version > 1) {
2959                 /* Stop the counter.  */
2960                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR_CTRL, 0);
2961                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_CTRL, 0);
2962             }
2963 
2964             /* Set the counter values.  */
2965             for (i = 0; i < num_architectural_pmu_fixed_counters; i++) {
2966                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR0 + i,
2967                                   env->msr_fixed_counters[i]);
2968             }
2969             for (i = 0; i < num_architectural_pmu_gp_counters; i++) {
2970                 kvm_msr_entry_add(cpu, MSR_P6_PERFCTR0 + i,
2971                                   env->msr_gp_counters[i]);
2972                 kvm_msr_entry_add(cpu, MSR_P6_EVNTSEL0 + i,
2973                                   env->msr_gp_evtsel[i]);
2974             }
2975             if (has_architectural_pmu_version > 1) {
2976                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_STATUS,
2977                                   env->msr_global_status);
2978                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
2979                                   env->msr_global_ovf_ctrl);
2980 
2981                 /* Now start the PMU.  */
2982                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR_CTRL,
2983                                   env->msr_fixed_ctr_ctrl);
2984                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_CTRL,
2985                                   env->msr_global_ctrl);
2986             }
2987         }
2988         /*
2989          * Hyper-V partition-wide MSRs: to avoid clearing them on cpu hot-add,
2990          * only sync them to KVM on the first cpu
2991          */
2992         if (current_cpu == first_cpu) {
2993             if (has_msr_hv_hypercall) {
2994                 kvm_msr_entry_add(cpu, HV_X64_MSR_GUEST_OS_ID,
2995                                   env->msr_hv_guest_os_id);
2996                 kvm_msr_entry_add(cpu, HV_X64_MSR_HYPERCALL,
2997                                   env->msr_hv_hypercall);
2998             }
2999             if (hyperv_feat_enabled(cpu, HYPERV_FEAT_TIME)) {
3000                 kvm_msr_entry_add(cpu, HV_X64_MSR_REFERENCE_TSC,
3001                                   env->msr_hv_tsc);
3002             }
3003             if (hyperv_feat_enabled(cpu, HYPERV_FEAT_REENLIGHTENMENT)) {
3004                 kvm_msr_entry_add(cpu, HV_X64_MSR_REENLIGHTENMENT_CONTROL,
3005                                   env->msr_hv_reenlightenment_control);
3006                 kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_CONTROL,
3007                                   env->msr_hv_tsc_emulation_control);
3008                 kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_STATUS,
3009                                   env->msr_hv_tsc_emulation_status);
3010             }
3011         }
3012         if (hyperv_feat_enabled(cpu, HYPERV_FEAT_VAPIC)) {
3013             kvm_msr_entry_add(cpu, HV_X64_MSR_APIC_ASSIST_PAGE,
3014                               env->msr_hv_vapic);
3015         }
3016         if (has_msr_hv_crash) {
3017             int j;
3018 
3019             for (j = 0; j < HV_CRASH_PARAMS; j++)
3020                 kvm_msr_entry_add(cpu, HV_X64_MSR_CRASH_P0 + j,
3021                                   env->msr_hv_crash_params[j]);
3022 
3023             kvm_msr_entry_add(cpu, HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_NOTIFY);
3024         }
3025         if (has_msr_hv_runtime) {
3026             kvm_msr_entry_add(cpu, HV_X64_MSR_VP_RUNTIME, env->msr_hv_runtime);
3027         }
3028         if (hyperv_feat_enabled(cpu, HYPERV_FEAT_VPINDEX)
3029             && hv_vpindex_settable) {
3030             kvm_msr_entry_add(cpu, HV_X64_MSR_VP_INDEX,
3031                               hyperv_vp_index(CPU(cpu)));
3032         }
3033         if (hyperv_feat_enabled(cpu, HYPERV_FEAT_SYNIC)) {
3034             int j;
3035 
3036             kvm_msr_entry_add(cpu, HV_X64_MSR_SVERSION, HV_SYNIC_VERSION);
3037 
3038             kvm_msr_entry_add(cpu, HV_X64_MSR_SCONTROL,
3039                               env->msr_hv_synic_control);
3040             kvm_msr_entry_add(cpu, HV_X64_MSR_SIEFP,
3041                               env->msr_hv_synic_evt_page);
3042             kvm_msr_entry_add(cpu, HV_X64_MSR_SIMP,
3043                               env->msr_hv_synic_msg_page);
3044 
3045             for (j = 0; j < ARRAY_SIZE(env->msr_hv_synic_sint); j++) {
3046                 kvm_msr_entry_add(cpu, HV_X64_MSR_SINT0 + j,
3047                                   env->msr_hv_synic_sint[j]);
3048             }
3049         }
3050         if (has_msr_hv_stimer) {
3051             int j;
3052 
3053             for (j = 0; j < ARRAY_SIZE(env->msr_hv_stimer_config); j++) {
3054                 kvm_msr_entry_add(cpu, HV_X64_MSR_STIMER0_CONFIG + j * 2,
3055                                 env->msr_hv_stimer_config[j]);
3056             }
3057 
3058             for (j = 0; j < ARRAY_SIZE(env->msr_hv_stimer_count); j++) {
3059                 kvm_msr_entry_add(cpu, HV_X64_MSR_STIMER0_COUNT + j * 2,
3060                                 env->msr_hv_stimer_count[j]);
3061             }
3062         }
3063         if (env->features[FEAT_1_EDX] & CPUID_MTRR) {
3064             uint64_t phys_mask = MAKE_64BIT_MASK(0, cpu->phys_bits);
3065 
3066             kvm_msr_entry_add(cpu, MSR_MTRRdefType, env->mtrr_deftype);
3067             kvm_msr_entry_add(cpu, MSR_MTRRfix64K_00000, env->mtrr_fixed[0]);
3068             kvm_msr_entry_add(cpu, MSR_MTRRfix16K_80000, env->mtrr_fixed[1]);
3069             kvm_msr_entry_add(cpu, MSR_MTRRfix16K_A0000, env->mtrr_fixed[2]);
3070             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C0000, env->mtrr_fixed[3]);
3071             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C8000, env->mtrr_fixed[4]);
3072             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D0000, env->mtrr_fixed[5]);
3073             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D8000, env->mtrr_fixed[6]);
3074             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E0000, env->mtrr_fixed[7]);
3075             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E8000, env->mtrr_fixed[8]);
3076             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F0000, env->mtrr_fixed[9]);
3077             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F8000, env->mtrr_fixed[10]);
3078             for (i = 0; i < MSR_MTRRcap_VCNT; i++) {
3079                 /* The CPU GPs if we write to a bit above the physical limit of
3080                  * the host CPU (and KVM emulates that)
3081                  */
3082                 uint64_t mask = env->mtrr_var[i].mask;
3083                 mask &= phys_mask;
3084 
3085                 kvm_msr_entry_add(cpu, MSR_MTRRphysBase(i),
3086                                   env->mtrr_var[i].base);
3087                 kvm_msr_entry_add(cpu, MSR_MTRRphysMask(i), mask);
3088             }
3089         }
3090         if (env->features[FEAT_7_0_EBX] & CPUID_7_0_EBX_INTEL_PT) {
3091             int addr_num = kvm_arch_get_supported_cpuid(kvm_state,
3092                                                     0x14, 1, R_EAX) & 0x7;
3093 
3094             kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CTL,
3095                             env->msr_rtit_ctrl);
3096             kvm_msr_entry_add(cpu, MSR_IA32_RTIT_STATUS,
3097                             env->msr_rtit_status);
3098             kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_BASE,
3099                             env->msr_rtit_output_base);
3100             kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_MASK,
3101                             env->msr_rtit_output_mask);
3102             kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CR3_MATCH,
3103                             env->msr_rtit_cr3_match);
3104             for (i = 0; i < addr_num; i++) {
3105                 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_ADDR0_A + i,
3106                             env->msr_rtit_addrs[i]);
3107             }
3108         }
3109 
3110         /* Note: MSR_IA32_FEATURE_CONTROL is written separately, see
3111          *       kvm_put_msr_feature_control. */
3112     }
3113 
3114     if (env->mcg_cap) {
3115         int i;
3116 
3117         kvm_msr_entry_add(cpu, MSR_MCG_STATUS, env->mcg_status);
3118         kvm_msr_entry_add(cpu, MSR_MCG_CTL, env->mcg_ctl);
3119         if (has_msr_mcg_ext_ctl) {
3120             kvm_msr_entry_add(cpu, MSR_MCG_EXT_CTL, env->mcg_ext_ctl);
3121         }
3122         for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++) {
3123             kvm_msr_entry_add(cpu, MSR_MC0_CTL + i, env->mce_banks[i]);
3124         }
3125     }
3126 
3127     return kvm_buf_set_msrs(cpu);
3128 }
3129 
3130 
3131 static int kvm_get_fpu(X86CPU *cpu)
3132 {
3133     CPUX86State *env = &cpu->env;
3134     struct kvm_fpu fpu;
3135     int i, ret;
3136 
3137     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_FPU, &fpu);
3138     if (ret < 0) {
3139         return ret;
3140     }
3141 
3142     env->fpstt = (fpu.fsw >> 11) & 7;
3143     env->fpus = fpu.fsw;
3144     env->fpuc = fpu.fcw;
3145     env->fpop = fpu.last_opcode;
3146     env->fpip = fpu.last_ip;
3147     env->fpdp = fpu.last_dp;
3148     for (i = 0; i < 8; ++i) {
3149         env->fptags[i] = !((fpu.ftwx >> i) & 1);
3150     }
3151     memcpy(env->fpregs, fpu.fpr, sizeof env->fpregs);
3152     for (i = 0; i < CPU_NB_REGS; i++) {
3153         env->xmm_regs[i].ZMM_Q(0) = ldq_p(&fpu.xmm[i][0]);
3154         env->xmm_regs[i].ZMM_Q(1) = ldq_p(&fpu.xmm[i][8]);
3155     }
3156     env->mxcsr = fpu.mxcsr;
3157 
3158     return 0;
3159 }
3160 
3161 static int kvm_get_xsave(X86CPU *cpu)
3162 {
3163     CPUX86State *env = &cpu->env;
3164     void *xsave = env->xsave_buf;
3165     int ret;
3166 
3167     if (!has_xsave) {
3168         return kvm_get_fpu(cpu);
3169     }
3170 
3171     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_XSAVE, xsave);
3172     if (ret < 0) {
3173         return ret;
3174     }
3175     x86_cpu_xrstor_all_areas(cpu, xsave, env->xsave_buf_len);
3176 
3177     return 0;
3178 }
3179 
3180 static int kvm_get_xcrs(X86CPU *cpu)
3181 {
3182     CPUX86State *env = &cpu->env;
3183     int i, ret;
3184     struct kvm_xcrs xcrs;
3185 
3186     if (!has_xcrs) {
3187         return 0;
3188     }
3189 
3190     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_XCRS, &xcrs);
3191     if (ret < 0) {
3192         return ret;
3193     }
3194 
3195     for (i = 0; i < xcrs.nr_xcrs; i++) {
3196         /* Only support xcr0 now */
3197         if (xcrs.xcrs[i].xcr == 0) {
3198             env->xcr0 = xcrs.xcrs[i].value;
3199             break;
3200         }
3201     }
3202     return 0;
3203 }
3204 
3205 static int kvm_get_sregs(X86CPU *cpu)
3206 {
3207     CPUX86State *env = &cpu->env;
3208     struct kvm_sregs sregs;
3209     int bit, i, ret;
3210 
3211     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
3212     if (ret < 0) {
3213         return ret;
3214     }
3215 
3216     /* There can only be one pending IRQ set in the bitmap at a time, so try
3217        to find it and save its number instead (-1 for none). */
3218     env->interrupt_injected = -1;
3219     for (i = 0; i < ARRAY_SIZE(sregs.interrupt_bitmap); i++) {
3220         if (sregs.interrupt_bitmap[i]) {
3221             bit = ctz64(sregs.interrupt_bitmap[i]);
3222             env->interrupt_injected = i * 64 + bit;
3223             break;
3224         }
3225     }
3226 
3227     get_seg(&env->segs[R_CS], &sregs.cs);
3228     get_seg(&env->segs[R_DS], &sregs.ds);
3229     get_seg(&env->segs[R_ES], &sregs.es);
3230     get_seg(&env->segs[R_FS], &sregs.fs);
3231     get_seg(&env->segs[R_GS], &sregs.gs);
3232     get_seg(&env->segs[R_SS], &sregs.ss);
3233 
3234     get_seg(&env->tr, &sregs.tr);
3235     get_seg(&env->ldt, &sregs.ldt);
3236 
3237     env->idt.limit = sregs.idt.limit;
3238     env->idt.base = sregs.idt.base;
3239     env->gdt.limit = sregs.gdt.limit;
3240     env->gdt.base = sregs.gdt.base;
3241 
3242     env->cr[0] = sregs.cr0;
3243     env->cr[2] = sregs.cr2;
3244     env->cr[3] = sregs.cr3;
3245     env->cr[4] = sregs.cr4;
3246 
3247     env->efer = sregs.efer;
3248 
3249     /* changes to apic base and cr8/tpr are read back via kvm_arch_post_run */
3250     x86_update_hflags(env);
3251 
3252     return 0;
3253 }
3254 
3255 static int kvm_get_msrs(X86CPU *cpu)
3256 {
3257     CPUX86State *env = &cpu->env;
3258     struct kvm_msr_entry *msrs = cpu->kvm_msr_buf->entries;
3259     int ret, i;
3260     uint64_t mtrr_top_bits;
3261 
3262     kvm_msr_buf_reset(cpu);
3263 
3264     kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_CS, 0);
3265     kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_ESP, 0);
3266     kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_EIP, 0);
3267     kvm_msr_entry_add(cpu, MSR_PAT, 0);
3268     if (has_msr_star) {
3269         kvm_msr_entry_add(cpu, MSR_STAR, 0);
3270     }
3271     if (has_msr_hsave_pa) {
3272         kvm_msr_entry_add(cpu, MSR_VM_HSAVE_PA, 0);
3273     }
3274     if (has_msr_tsc_aux) {
3275         kvm_msr_entry_add(cpu, MSR_TSC_AUX, 0);
3276     }
3277     if (has_msr_tsc_adjust) {
3278         kvm_msr_entry_add(cpu, MSR_TSC_ADJUST, 0);
3279     }
3280     if (has_msr_tsc_deadline) {
3281         kvm_msr_entry_add(cpu, MSR_IA32_TSCDEADLINE, 0);
3282     }
3283     if (has_msr_misc_enable) {
3284         kvm_msr_entry_add(cpu, MSR_IA32_MISC_ENABLE, 0);
3285     }
3286     if (has_msr_smbase) {
3287         kvm_msr_entry_add(cpu, MSR_IA32_SMBASE, 0);
3288     }
3289     if (has_msr_smi_count) {
3290         kvm_msr_entry_add(cpu, MSR_SMI_COUNT, 0);
3291     }
3292     if (has_msr_feature_control) {
3293         kvm_msr_entry_add(cpu, MSR_IA32_FEATURE_CONTROL, 0);
3294     }
3295     if (has_msr_pkrs) {
3296         kvm_msr_entry_add(cpu, MSR_IA32_PKRS, 0);
3297     }
3298     if (has_msr_bndcfgs) {
3299         kvm_msr_entry_add(cpu, MSR_IA32_BNDCFGS, 0);
3300     }
3301     if (has_msr_xss) {
3302         kvm_msr_entry_add(cpu, MSR_IA32_XSS, 0);
3303     }
3304     if (has_msr_umwait) {
3305         kvm_msr_entry_add(cpu, MSR_IA32_UMWAIT_CONTROL, 0);
3306     }
3307     if (has_msr_spec_ctrl) {
3308         kvm_msr_entry_add(cpu, MSR_IA32_SPEC_CTRL, 0);
3309     }
3310     if (has_msr_tsx_ctrl) {
3311         kvm_msr_entry_add(cpu, MSR_IA32_TSX_CTRL, 0);
3312     }
3313     if (has_msr_virt_ssbd) {
3314         kvm_msr_entry_add(cpu, MSR_VIRT_SSBD, 0);
3315     }
3316     if (!env->tsc_valid) {
3317         kvm_msr_entry_add(cpu, MSR_IA32_TSC, 0);
3318         env->tsc_valid = !runstate_is_running();
3319     }
3320 
3321 #ifdef TARGET_X86_64
3322     if (lm_capable_kernel) {
3323         kvm_msr_entry_add(cpu, MSR_CSTAR, 0);
3324         kvm_msr_entry_add(cpu, MSR_KERNELGSBASE, 0);
3325         kvm_msr_entry_add(cpu, MSR_FMASK, 0);
3326         kvm_msr_entry_add(cpu, MSR_LSTAR, 0);
3327     }
3328 #endif
3329     kvm_msr_entry_add(cpu, MSR_KVM_SYSTEM_TIME, 0);
3330     kvm_msr_entry_add(cpu, MSR_KVM_WALL_CLOCK, 0);
3331     if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_ASYNC_PF_INT)) {
3332         kvm_msr_entry_add(cpu, MSR_KVM_ASYNC_PF_INT, 0);
3333     }
3334     if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_ASYNC_PF)) {
3335         kvm_msr_entry_add(cpu, MSR_KVM_ASYNC_PF_EN, 0);
3336     }
3337     if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_PV_EOI)) {
3338         kvm_msr_entry_add(cpu, MSR_KVM_PV_EOI_EN, 0);
3339     }
3340     if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_STEAL_TIME)) {
3341         kvm_msr_entry_add(cpu, MSR_KVM_STEAL_TIME, 0);
3342     }
3343     if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_POLL_CONTROL)) {
3344         kvm_msr_entry_add(cpu, MSR_KVM_POLL_CONTROL, 1);
3345     }
3346     if (has_architectural_pmu_version > 0) {
3347         if (has_architectural_pmu_version > 1) {
3348             kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR_CTRL, 0);
3349             kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_CTRL, 0);
3350             kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_STATUS, 0);
3351             kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_OVF_CTRL, 0);
3352         }
3353         for (i = 0; i < num_architectural_pmu_fixed_counters; i++) {
3354             kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR0 + i, 0);
3355         }
3356         for (i = 0; i < num_architectural_pmu_gp_counters; i++) {
3357             kvm_msr_entry_add(cpu, MSR_P6_PERFCTR0 + i, 0);
3358             kvm_msr_entry_add(cpu, MSR_P6_EVNTSEL0 + i, 0);
3359         }
3360     }
3361 
3362     if (env->mcg_cap) {
3363         kvm_msr_entry_add(cpu, MSR_MCG_STATUS, 0);
3364         kvm_msr_entry_add(cpu, MSR_MCG_CTL, 0);
3365         if (has_msr_mcg_ext_ctl) {
3366             kvm_msr_entry_add(cpu, MSR_MCG_EXT_CTL, 0);
3367         }
3368         for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++) {
3369             kvm_msr_entry_add(cpu, MSR_MC0_CTL + i, 0);
3370         }
3371     }
3372 
3373     if (has_msr_hv_hypercall) {
3374         kvm_msr_entry_add(cpu, HV_X64_MSR_HYPERCALL, 0);
3375         kvm_msr_entry_add(cpu, HV_X64_MSR_GUEST_OS_ID, 0);
3376     }
3377     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_VAPIC)) {
3378         kvm_msr_entry_add(cpu, HV_X64_MSR_APIC_ASSIST_PAGE, 0);
3379     }
3380     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_TIME)) {
3381         kvm_msr_entry_add(cpu, HV_X64_MSR_REFERENCE_TSC, 0);
3382     }
3383     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_REENLIGHTENMENT)) {
3384         kvm_msr_entry_add(cpu, HV_X64_MSR_REENLIGHTENMENT_CONTROL, 0);
3385         kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_CONTROL, 0);
3386         kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_STATUS, 0);
3387     }
3388     if (has_msr_hv_crash) {
3389         int j;
3390 
3391         for (j = 0; j < HV_CRASH_PARAMS; j++) {
3392             kvm_msr_entry_add(cpu, HV_X64_MSR_CRASH_P0 + j, 0);
3393         }
3394     }
3395     if (has_msr_hv_runtime) {
3396         kvm_msr_entry_add(cpu, HV_X64_MSR_VP_RUNTIME, 0);
3397     }
3398     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_SYNIC)) {
3399         uint32_t msr;
3400 
3401         kvm_msr_entry_add(cpu, HV_X64_MSR_SCONTROL, 0);
3402         kvm_msr_entry_add(cpu, HV_X64_MSR_SIEFP, 0);
3403         kvm_msr_entry_add(cpu, HV_X64_MSR_SIMP, 0);
3404         for (msr = HV_X64_MSR_SINT0; msr <= HV_X64_MSR_SINT15; msr++) {
3405             kvm_msr_entry_add(cpu, msr, 0);
3406         }
3407     }
3408     if (has_msr_hv_stimer) {
3409         uint32_t msr;
3410 
3411         for (msr = HV_X64_MSR_STIMER0_CONFIG; msr <= HV_X64_MSR_STIMER3_COUNT;
3412              msr++) {
3413             kvm_msr_entry_add(cpu, msr, 0);
3414         }
3415     }
3416     if (env->features[FEAT_1_EDX] & CPUID_MTRR) {
3417         kvm_msr_entry_add(cpu, MSR_MTRRdefType, 0);
3418         kvm_msr_entry_add(cpu, MSR_MTRRfix64K_00000, 0);
3419         kvm_msr_entry_add(cpu, MSR_MTRRfix16K_80000, 0);
3420         kvm_msr_entry_add(cpu, MSR_MTRRfix16K_A0000, 0);
3421         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C0000, 0);
3422         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C8000, 0);
3423         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D0000, 0);
3424         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D8000, 0);
3425         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E0000, 0);
3426         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E8000, 0);
3427         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F0000, 0);
3428         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F8000, 0);
3429         for (i = 0; i < MSR_MTRRcap_VCNT; i++) {
3430             kvm_msr_entry_add(cpu, MSR_MTRRphysBase(i), 0);
3431             kvm_msr_entry_add(cpu, MSR_MTRRphysMask(i), 0);
3432         }
3433     }
3434 
3435     if (env->features[FEAT_7_0_EBX] & CPUID_7_0_EBX_INTEL_PT) {
3436         int addr_num =
3437             kvm_arch_get_supported_cpuid(kvm_state, 0x14, 1, R_EAX) & 0x7;
3438 
3439         kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CTL, 0);
3440         kvm_msr_entry_add(cpu, MSR_IA32_RTIT_STATUS, 0);
3441         kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_BASE, 0);
3442         kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_MASK, 0);
3443         kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CR3_MATCH, 0);
3444         for (i = 0; i < addr_num; i++) {
3445             kvm_msr_entry_add(cpu, MSR_IA32_RTIT_ADDR0_A + i, 0);
3446         }
3447     }
3448 
3449     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MSRS, cpu->kvm_msr_buf);
3450     if (ret < 0) {
3451         return ret;
3452     }
3453 
3454     if (ret < cpu->kvm_msr_buf->nmsrs) {
3455         struct kvm_msr_entry *e = &cpu->kvm_msr_buf->entries[ret];
3456         error_report("error: failed to get MSR 0x%" PRIx32,
3457                      (uint32_t)e->index);
3458     }
3459 
3460     assert(ret == cpu->kvm_msr_buf->nmsrs);
3461     /*
3462      * MTRR masks: Each mask consists of 5 parts
3463      * a  10..0: must be zero
3464      * b  11   : valid bit
3465      * c n-1.12: actual mask bits
3466      * d  51..n: reserved must be zero
3467      * e  63.52: reserved must be zero
3468      *
3469      * 'n' is the number of physical bits supported by the CPU and is
3470      * apparently always <= 52.   We know our 'n' but don't know what
3471      * the destinations 'n' is; it might be smaller, in which case
3472      * it masks (c) on loading. It might be larger, in which case
3473      * we fill 'd' so that d..c is consistent irrespetive of the 'n'
3474      * we're migrating to.
3475      */
3476 
3477     if (cpu->fill_mtrr_mask) {
3478         QEMU_BUILD_BUG_ON(TARGET_PHYS_ADDR_SPACE_BITS > 52);
3479         assert(cpu->phys_bits <= TARGET_PHYS_ADDR_SPACE_BITS);
3480         mtrr_top_bits = MAKE_64BIT_MASK(cpu->phys_bits, 52 - cpu->phys_bits);
3481     } else {
3482         mtrr_top_bits = 0;
3483     }
3484 
3485     for (i = 0; i < ret; i++) {
3486         uint32_t index = msrs[i].index;
3487         switch (index) {
3488         case MSR_IA32_SYSENTER_CS:
3489             env->sysenter_cs = msrs[i].data;
3490             break;
3491         case MSR_IA32_SYSENTER_ESP:
3492             env->sysenter_esp = msrs[i].data;
3493             break;
3494         case MSR_IA32_SYSENTER_EIP:
3495             env->sysenter_eip = msrs[i].data;
3496             break;
3497         case MSR_PAT:
3498             env->pat = msrs[i].data;
3499             break;
3500         case MSR_STAR:
3501             env->star = msrs[i].data;
3502             break;
3503 #ifdef TARGET_X86_64
3504         case MSR_CSTAR:
3505             env->cstar = msrs[i].data;
3506             break;
3507         case MSR_KERNELGSBASE:
3508             env->kernelgsbase = msrs[i].data;
3509             break;
3510         case MSR_FMASK:
3511             env->fmask = msrs[i].data;
3512             break;
3513         case MSR_LSTAR:
3514             env->lstar = msrs[i].data;
3515             break;
3516 #endif
3517         case MSR_IA32_TSC:
3518             env->tsc = msrs[i].data;
3519             break;
3520         case MSR_TSC_AUX:
3521             env->tsc_aux = msrs[i].data;
3522             break;
3523         case MSR_TSC_ADJUST:
3524             env->tsc_adjust = msrs[i].data;
3525             break;
3526         case MSR_IA32_TSCDEADLINE:
3527             env->tsc_deadline = msrs[i].data;
3528             break;
3529         case MSR_VM_HSAVE_PA:
3530             env->vm_hsave = msrs[i].data;
3531             break;
3532         case MSR_KVM_SYSTEM_TIME:
3533             env->system_time_msr = msrs[i].data;
3534             break;
3535         case MSR_KVM_WALL_CLOCK:
3536             env->wall_clock_msr = msrs[i].data;
3537             break;
3538         case MSR_MCG_STATUS:
3539             env->mcg_status = msrs[i].data;
3540             break;
3541         case MSR_MCG_CTL:
3542             env->mcg_ctl = msrs[i].data;
3543             break;
3544         case MSR_MCG_EXT_CTL:
3545             env->mcg_ext_ctl = msrs[i].data;
3546             break;
3547         case MSR_IA32_MISC_ENABLE:
3548             env->msr_ia32_misc_enable = msrs[i].data;
3549             break;
3550         case MSR_IA32_SMBASE:
3551             env->smbase = msrs[i].data;
3552             break;
3553         case MSR_SMI_COUNT:
3554             env->msr_smi_count = msrs[i].data;
3555             break;
3556         case MSR_IA32_FEATURE_CONTROL:
3557             env->msr_ia32_feature_control = msrs[i].data;
3558             break;
3559         case MSR_IA32_BNDCFGS:
3560             env->msr_bndcfgs = msrs[i].data;
3561             break;
3562         case MSR_IA32_XSS:
3563             env->xss = msrs[i].data;
3564             break;
3565         case MSR_IA32_UMWAIT_CONTROL:
3566             env->umwait = msrs[i].data;
3567             break;
3568         case MSR_IA32_PKRS:
3569             env->pkrs = msrs[i].data;
3570             break;
3571         default:
3572             if (msrs[i].index >= MSR_MC0_CTL &&
3573                 msrs[i].index < MSR_MC0_CTL + (env->mcg_cap & 0xff) * 4) {
3574                 env->mce_banks[msrs[i].index - MSR_MC0_CTL] = msrs[i].data;
3575             }
3576             break;
3577         case MSR_KVM_ASYNC_PF_EN:
3578             env->async_pf_en_msr = msrs[i].data;
3579             break;
3580         case MSR_KVM_ASYNC_PF_INT:
3581             env->async_pf_int_msr = msrs[i].data;
3582             break;
3583         case MSR_KVM_PV_EOI_EN:
3584             env->pv_eoi_en_msr = msrs[i].data;
3585             break;
3586         case MSR_KVM_STEAL_TIME:
3587             env->steal_time_msr = msrs[i].data;
3588             break;
3589         case MSR_KVM_POLL_CONTROL: {
3590             env->poll_control_msr = msrs[i].data;
3591             break;
3592         }
3593         case MSR_CORE_PERF_FIXED_CTR_CTRL:
3594             env->msr_fixed_ctr_ctrl = msrs[i].data;
3595             break;
3596         case MSR_CORE_PERF_GLOBAL_CTRL:
3597             env->msr_global_ctrl = msrs[i].data;
3598             break;
3599         case MSR_CORE_PERF_GLOBAL_STATUS:
3600             env->msr_global_status = msrs[i].data;
3601             break;
3602         case MSR_CORE_PERF_GLOBAL_OVF_CTRL:
3603             env->msr_global_ovf_ctrl = msrs[i].data;
3604             break;
3605         case MSR_CORE_PERF_FIXED_CTR0 ... MSR_CORE_PERF_FIXED_CTR0 + MAX_FIXED_COUNTERS - 1:
3606             env->msr_fixed_counters[index - MSR_CORE_PERF_FIXED_CTR0] = msrs[i].data;
3607             break;
3608         case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR0 + MAX_GP_COUNTERS - 1:
3609             env->msr_gp_counters[index - MSR_P6_PERFCTR0] = msrs[i].data;
3610             break;
3611         case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL0 + MAX_GP_COUNTERS - 1:
3612             env->msr_gp_evtsel[index - MSR_P6_EVNTSEL0] = msrs[i].data;
3613             break;
3614         case HV_X64_MSR_HYPERCALL:
3615             env->msr_hv_hypercall = msrs[i].data;
3616             break;
3617         case HV_X64_MSR_GUEST_OS_ID:
3618             env->msr_hv_guest_os_id = msrs[i].data;
3619             break;
3620         case HV_X64_MSR_APIC_ASSIST_PAGE:
3621             env->msr_hv_vapic = msrs[i].data;
3622             break;
3623         case HV_X64_MSR_REFERENCE_TSC:
3624             env->msr_hv_tsc = msrs[i].data;
3625             break;
3626         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3627             env->msr_hv_crash_params[index - HV_X64_MSR_CRASH_P0] = msrs[i].data;
3628             break;
3629         case HV_X64_MSR_VP_RUNTIME:
3630             env->msr_hv_runtime = msrs[i].data;
3631             break;
3632         case HV_X64_MSR_SCONTROL:
3633             env->msr_hv_synic_control = msrs[i].data;
3634             break;
3635         case HV_X64_MSR_SIEFP:
3636             env->msr_hv_synic_evt_page = msrs[i].data;
3637             break;
3638         case HV_X64_MSR_SIMP:
3639             env->msr_hv_synic_msg_page = msrs[i].data;
3640             break;
3641         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
3642             env->msr_hv_synic_sint[index - HV_X64_MSR_SINT0] = msrs[i].data;
3643             break;
3644         case HV_X64_MSR_STIMER0_CONFIG:
3645         case HV_X64_MSR_STIMER1_CONFIG:
3646         case HV_X64_MSR_STIMER2_CONFIG:
3647         case HV_X64_MSR_STIMER3_CONFIG:
3648             env->msr_hv_stimer_config[(index - HV_X64_MSR_STIMER0_CONFIG)/2] =
3649                                 msrs[i].data;
3650             break;
3651         case HV_X64_MSR_STIMER0_COUNT:
3652         case HV_X64_MSR_STIMER1_COUNT:
3653         case HV_X64_MSR_STIMER2_COUNT:
3654         case HV_X64_MSR_STIMER3_COUNT:
3655             env->msr_hv_stimer_count[(index - HV_X64_MSR_STIMER0_COUNT)/2] =
3656                                 msrs[i].data;
3657             break;
3658         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3659             env->msr_hv_reenlightenment_control = msrs[i].data;
3660             break;
3661         case HV_X64_MSR_TSC_EMULATION_CONTROL:
3662             env->msr_hv_tsc_emulation_control = msrs[i].data;
3663             break;
3664         case HV_X64_MSR_TSC_EMULATION_STATUS:
3665             env->msr_hv_tsc_emulation_status = msrs[i].data;
3666             break;
3667         case MSR_MTRRdefType:
3668             env->mtrr_deftype = msrs[i].data;
3669             break;
3670         case MSR_MTRRfix64K_00000:
3671             env->mtrr_fixed[0] = msrs[i].data;
3672             break;
3673         case MSR_MTRRfix16K_80000:
3674             env->mtrr_fixed[1] = msrs[i].data;
3675             break;
3676         case MSR_MTRRfix16K_A0000:
3677             env->mtrr_fixed[2] = msrs[i].data;
3678             break;
3679         case MSR_MTRRfix4K_C0000:
3680             env->mtrr_fixed[3] = msrs[i].data;
3681             break;
3682         case MSR_MTRRfix4K_C8000:
3683             env->mtrr_fixed[4] = msrs[i].data;
3684             break;
3685         case MSR_MTRRfix4K_D0000:
3686             env->mtrr_fixed[5] = msrs[i].data;
3687             break;
3688         case MSR_MTRRfix4K_D8000:
3689             env->mtrr_fixed[6] = msrs[i].data;
3690             break;
3691         case MSR_MTRRfix4K_E0000:
3692             env->mtrr_fixed[7] = msrs[i].data;
3693             break;
3694         case MSR_MTRRfix4K_E8000:
3695             env->mtrr_fixed[8] = msrs[i].data;
3696             break;
3697         case MSR_MTRRfix4K_F0000:
3698             env->mtrr_fixed[9] = msrs[i].data;
3699             break;
3700         case MSR_MTRRfix4K_F8000:
3701             env->mtrr_fixed[10] = msrs[i].data;
3702             break;
3703         case MSR_MTRRphysBase(0) ... MSR_MTRRphysMask(MSR_MTRRcap_VCNT - 1):
3704             if (index & 1) {
3705                 env->mtrr_var[MSR_MTRRphysIndex(index)].mask = msrs[i].data |
3706                                                                mtrr_top_bits;
3707             } else {
3708                 env->mtrr_var[MSR_MTRRphysIndex(index)].base = msrs[i].data;
3709             }
3710             break;
3711         case MSR_IA32_SPEC_CTRL:
3712             env->spec_ctrl = msrs[i].data;
3713             break;
3714         case MSR_IA32_TSX_CTRL:
3715             env->tsx_ctrl = msrs[i].data;
3716             break;
3717         case MSR_VIRT_SSBD:
3718             env->virt_ssbd = msrs[i].data;
3719             break;
3720         case MSR_IA32_RTIT_CTL:
3721             env->msr_rtit_ctrl = msrs[i].data;
3722             break;
3723         case MSR_IA32_RTIT_STATUS:
3724             env->msr_rtit_status = msrs[i].data;
3725             break;
3726         case MSR_IA32_RTIT_OUTPUT_BASE:
3727             env->msr_rtit_output_base = msrs[i].data;
3728             break;
3729         case MSR_IA32_RTIT_OUTPUT_MASK:
3730             env->msr_rtit_output_mask = msrs[i].data;
3731             break;
3732         case MSR_IA32_RTIT_CR3_MATCH:
3733             env->msr_rtit_cr3_match = msrs[i].data;
3734             break;
3735         case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
3736             env->msr_rtit_addrs[index - MSR_IA32_RTIT_ADDR0_A] = msrs[i].data;
3737             break;
3738         }
3739     }
3740 
3741     return 0;
3742 }
3743 
3744 static int kvm_put_mp_state(X86CPU *cpu)
3745 {
3746     struct kvm_mp_state mp_state = { .mp_state = cpu->env.mp_state };
3747 
3748     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
3749 }
3750 
3751 static int kvm_get_mp_state(X86CPU *cpu)
3752 {
3753     CPUState *cs = CPU(cpu);
3754     CPUX86State *env = &cpu->env;
3755     struct kvm_mp_state mp_state;
3756     int ret;
3757 
3758     ret = kvm_vcpu_ioctl(cs, KVM_GET_MP_STATE, &mp_state);
3759     if (ret < 0) {
3760         return ret;
3761     }
3762     env->mp_state = mp_state.mp_state;
3763     if (kvm_irqchip_in_kernel()) {
3764         cs->halted = (mp_state.mp_state == KVM_MP_STATE_HALTED);
3765     }
3766     return 0;
3767 }
3768 
3769 static int kvm_get_apic(X86CPU *cpu)
3770 {
3771     DeviceState *apic = cpu->apic_state;
3772     struct kvm_lapic_state kapic;
3773     int ret;
3774 
3775     if (apic && kvm_irqchip_in_kernel()) {
3776         ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_LAPIC, &kapic);
3777         if (ret < 0) {
3778             return ret;
3779         }
3780 
3781         kvm_get_apic_state(apic, &kapic);
3782     }
3783     return 0;
3784 }
3785 
3786 static int kvm_put_vcpu_events(X86CPU *cpu, int level)
3787 {
3788     CPUState *cs = CPU(cpu);
3789     CPUX86State *env = &cpu->env;
3790     struct kvm_vcpu_events events = {};
3791 
3792     if (!kvm_has_vcpu_events()) {
3793         return 0;
3794     }
3795 
3796     events.flags = 0;
3797 
3798     if (has_exception_payload) {
3799         events.flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3800         events.exception.pending = env->exception_pending;
3801         events.exception_has_payload = env->exception_has_payload;
3802         events.exception_payload = env->exception_payload;
3803     }
3804     events.exception.nr = env->exception_nr;
3805     events.exception.injected = env->exception_injected;
3806     events.exception.has_error_code = env->has_error_code;
3807     events.exception.error_code = env->error_code;
3808 
3809     events.interrupt.injected = (env->interrupt_injected >= 0);
3810     events.interrupt.nr = env->interrupt_injected;
3811     events.interrupt.soft = env->soft_interrupt;
3812 
3813     events.nmi.injected = env->nmi_injected;
3814     events.nmi.pending = env->nmi_pending;
3815     events.nmi.masked = !!(env->hflags2 & HF2_NMI_MASK);
3816 
3817     events.sipi_vector = env->sipi_vector;
3818 
3819     if (has_msr_smbase) {
3820         events.smi.smm = !!(env->hflags & HF_SMM_MASK);
3821         events.smi.smm_inside_nmi = !!(env->hflags2 & HF2_SMM_INSIDE_NMI_MASK);
3822         if (kvm_irqchip_in_kernel()) {
3823             /* As soon as these are moved to the kernel, remove them
3824              * from cs->interrupt_request.
3825              */
3826             events.smi.pending = cs->interrupt_request & CPU_INTERRUPT_SMI;
3827             events.smi.latched_init = cs->interrupt_request & CPU_INTERRUPT_INIT;
3828             cs->interrupt_request &= ~(CPU_INTERRUPT_INIT | CPU_INTERRUPT_SMI);
3829         } else {
3830             /* Keep these in cs->interrupt_request.  */
3831             events.smi.pending = 0;
3832             events.smi.latched_init = 0;
3833         }
3834         /* Stop SMI delivery on old machine types to avoid a reboot
3835          * on an inward migration of an old VM.
3836          */
3837         if (!cpu->kvm_no_smi_migration) {
3838             events.flags |= KVM_VCPUEVENT_VALID_SMM;
3839         }
3840     }
3841 
3842     if (level >= KVM_PUT_RESET_STATE) {
3843         events.flags |= KVM_VCPUEVENT_VALID_NMI_PENDING;
3844         if (env->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
3845             events.flags |= KVM_VCPUEVENT_VALID_SIPI_VECTOR;
3846         }
3847     }
3848 
3849     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_VCPU_EVENTS, &events);
3850 }
3851 
3852 static int kvm_get_vcpu_events(X86CPU *cpu)
3853 {
3854     CPUX86State *env = &cpu->env;
3855     struct kvm_vcpu_events events;
3856     int ret;
3857 
3858     if (!kvm_has_vcpu_events()) {
3859         return 0;
3860     }
3861 
3862     memset(&events, 0, sizeof(events));
3863     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_VCPU_EVENTS, &events);
3864     if (ret < 0) {
3865        return ret;
3866     }
3867 
3868     if (events.flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3869         env->exception_pending = events.exception.pending;
3870         env->exception_has_payload = events.exception_has_payload;
3871         env->exception_payload = events.exception_payload;
3872     } else {
3873         env->exception_pending = 0;
3874         env->exception_has_payload = false;
3875     }
3876     env->exception_injected = events.exception.injected;
3877     env->exception_nr =
3878         (env->exception_pending || env->exception_injected) ?
3879         events.exception.nr : -1;
3880     env->has_error_code = events.exception.has_error_code;
3881     env->error_code = events.exception.error_code;
3882 
3883     env->interrupt_injected =
3884         events.interrupt.injected ? events.interrupt.nr : -1;
3885     env->soft_interrupt = events.interrupt.soft;
3886 
3887     env->nmi_injected = events.nmi.injected;
3888     env->nmi_pending = events.nmi.pending;
3889     if (events.nmi.masked) {
3890         env->hflags2 |= HF2_NMI_MASK;
3891     } else {
3892         env->hflags2 &= ~HF2_NMI_MASK;
3893     }
3894 
3895     if (events.flags & KVM_VCPUEVENT_VALID_SMM) {
3896         if (events.smi.smm) {
3897             env->hflags |= HF_SMM_MASK;
3898         } else {
3899             env->hflags &= ~HF_SMM_MASK;
3900         }
3901         if (events.smi.pending) {
3902             cpu_interrupt(CPU(cpu), CPU_INTERRUPT_SMI);
3903         } else {
3904             cpu_reset_interrupt(CPU(cpu), CPU_INTERRUPT_SMI);
3905         }
3906         if (events.smi.smm_inside_nmi) {
3907             env->hflags2 |= HF2_SMM_INSIDE_NMI_MASK;
3908         } else {
3909             env->hflags2 &= ~HF2_SMM_INSIDE_NMI_MASK;
3910         }
3911         if (events.smi.latched_init) {
3912             cpu_interrupt(CPU(cpu), CPU_INTERRUPT_INIT);
3913         } else {
3914             cpu_reset_interrupt(CPU(cpu), CPU_INTERRUPT_INIT);
3915         }
3916     }
3917 
3918     env->sipi_vector = events.sipi_vector;
3919 
3920     return 0;
3921 }
3922 
3923 static int kvm_guest_debug_workarounds(X86CPU *cpu)
3924 {
3925     CPUState *cs = CPU(cpu);
3926     CPUX86State *env = &cpu->env;
3927     int ret = 0;
3928     unsigned long reinject_trap = 0;
3929 
3930     if (!kvm_has_vcpu_events()) {
3931         if (env->exception_nr == EXCP01_DB) {
3932             reinject_trap = KVM_GUESTDBG_INJECT_DB;
3933         } else if (env->exception_injected == EXCP03_INT3) {
3934             reinject_trap = KVM_GUESTDBG_INJECT_BP;
3935         }
3936         kvm_reset_exception(env);
3937     }
3938 
3939     /*
3940      * Kernels before KVM_CAP_X86_ROBUST_SINGLESTEP overwrote flags.TF
3941      * injected via SET_GUEST_DEBUG while updating GP regs. Work around this
3942      * by updating the debug state once again if single-stepping is on.
3943      * Another reason to call kvm_update_guest_debug here is a pending debug
3944      * trap raise by the guest. On kernels without SET_VCPU_EVENTS we have to
3945      * reinject them via SET_GUEST_DEBUG.
3946      */
3947     if (reinject_trap ||
3948         (!kvm_has_robust_singlestep() && cs->singlestep_enabled)) {
3949         ret = kvm_update_guest_debug(cs, reinject_trap);
3950     }
3951     return ret;
3952 }
3953 
3954 static int kvm_put_debugregs(X86CPU *cpu)
3955 {
3956     CPUX86State *env = &cpu->env;
3957     struct kvm_debugregs dbgregs;
3958     int i;
3959 
3960     if (!kvm_has_debugregs()) {
3961         return 0;
3962     }
3963 
3964     memset(&dbgregs, 0, sizeof(dbgregs));
3965     for (i = 0; i < 4; i++) {
3966         dbgregs.db[i] = env->dr[i];
3967     }
3968     dbgregs.dr6 = env->dr[6];
3969     dbgregs.dr7 = env->dr[7];
3970     dbgregs.flags = 0;
3971 
3972     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_DEBUGREGS, &dbgregs);
3973 }
3974 
3975 static int kvm_get_debugregs(X86CPU *cpu)
3976 {
3977     CPUX86State *env = &cpu->env;
3978     struct kvm_debugregs dbgregs;
3979     int i, ret;
3980 
3981     if (!kvm_has_debugregs()) {
3982         return 0;
3983     }
3984 
3985     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_DEBUGREGS, &dbgregs);
3986     if (ret < 0) {
3987         return ret;
3988     }
3989     for (i = 0; i < 4; i++) {
3990         env->dr[i] = dbgregs.db[i];
3991     }
3992     env->dr[4] = env->dr[6] = dbgregs.dr6;
3993     env->dr[5] = env->dr[7] = dbgregs.dr7;
3994 
3995     return 0;
3996 }
3997 
3998 static int kvm_put_nested_state(X86CPU *cpu)
3999 {
4000     CPUX86State *env = &cpu->env;
4001     int max_nested_state_len = kvm_max_nested_state_length();
4002 
4003     if (!env->nested_state) {
4004         return 0;
4005     }
4006 
4007     /*
4008      * Copy flags that are affected by reset from env->hflags and env->hflags2.
4009      */
4010     if (env->hflags & HF_GUEST_MASK) {
4011         env->nested_state->flags |= KVM_STATE_NESTED_GUEST_MODE;
4012     } else {
4013         env->nested_state->flags &= ~KVM_STATE_NESTED_GUEST_MODE;
4014     }
4015 
4016     /* Don't set KVM_STATE_NESTED_GIF_SET on VMX as it is illegal */
4017     if (cpu_has_svm(env) && (env->hflags2 & HF2_GIF_MASK)) {
4018         env->nested_state->flags |= KVM_STATE_NESTED_GIF_SET;
4019     } else {
4020         env->nested_state->flags &= ~KVM_STATE_NESTED_GIF_SET;
4021     }
4022 
4023     assert(env->nested_state->size <= max_nested_state_len);
4024     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_NESTED_STATE, env->nested_state);
4025 }
4026 
4027 static int kvm_get_nested_state(X86CPU *cpu)
4028 {
4029     CPUX86State *env = &cpu->env;
4030     int max_nested_state_len = kvm_max_nested_state_length();
4031     int ret;
4032 
4033     if (!env->nested_state) {
4034         return 0;
4035     }
4036 
4037     /*
4038      * It is possible that migration restored a smaller size into
4039      * nested_state->hdr.size than what our kernel support.
4040      * We preserve migration origin nested_state->hdr.size for
4041      * call to KVM_SET_NESTED_STATE but wish that our next call
4042      * to KVM_GET_NESTED_STATE will use max size our kernel support.
4043      */
4044     env->nested_state->size = max_nested_state_len;
4045 
4046     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_NESTED_STATE, env->nested_state);
4047     if (ret < 0) {
4048         return ret;
4049     }
4050 
4051     /*
4052      * Copy flags that are affected by reset to env->hflags and env->hflags2.
4053      */
4054     if (env->nested_state->flags & KVM_STATE_NESTED_GUEST_MODE) {
4055         env->hflags |= HF_GUEST_MASK;
4056     } else {
4057         env->hflags &= ~HF_GUEST_MASK;
4058     }
4059 
4060     /* Keep HF2_GIF_MASK set on !SVM as x86_cpu_pending_interrupt() needs it */
4061     if (cpu_has_svm(env)) {
4062         if (env->nested_state->flags & KVM_STATE_NESTED_GIF_SET) {
4063             env->hflags2 |= HF2_GIF_MASK;
4064         } else {
4065             env->hflags2 &= ~HF2_GIF_MASK;
4066         }
4067     }
4068 
4069     return ret;
4070 }
4071 
4072 int kvm_arch_put_registers(CPUState *cpu, int level)
4073 {
4074     X86CPU *x86_cpu = X86_CPU(cpu);
4075     int ret;
4076 
4077     assert(cpu_is_stopped(cpu) || qemu_cpu_is_self(cpu));
4078 
4079     /* must be before kvm_put_nested_state so that EFER.SVME is set */
4080     ret = kvm_put_sregs(x86_cpu);
4081     if (ret < 0) {
4082         return ret;
4083     }
4084 
4085     if (level >= KVM_PUT_RESET_STATE) {
4086         ret = kvm_put_nested_state(x86_cpu);
4087         if (ret < 0) {
4088             return ret;
4089         }
4090 
4091         ret = kvm_put_msr_feature_control(x86_cpu);
4092         if (ret < 0) {
4093             return ret;
4094         }
4095     }
4096 
4097     if (level == KVM_PUT_FULL_STATE) {
4098         /* We don't check for kvm_arch_set_tsc_khz() errors here,
4099          * because TSC frequency mismatch shouldn't abort migration,
4100          * unless the user explicitly asked for a more strict TSC
4101          * setting (e.g. using an explicit "tsc-freq" option).
4102          */
4103         kvm_arch_set_tsc_khz(cpu);
4104     }
4105 
4106     ret = kvm_getput_regs(x86_cpu, 1);
4107     if (ret < 0) {
4108         return ret;
4109     }
4110     ret = kvm_put_xsave(x86_cpu);
4111     if (ret < 0) {
4112         return ret;
4113     }
4114     ret = kvm_put_xcrs(x86_cpu);
4115     if (ret < 0) {
4116         return ret;
4117     }
4118     /* must be before kvm_put_msrs */
4119     ret = kvm_inject_mce_oldstyle(x86_cpu);
4120     if (ret < 0) {
4121         return ret;
4122     }
4123     ret = kvm_put_msrs(x86_cpu, level);
4124     if (ret < 0) {
4125         return ret;
4126     }
4127     ret = kvm_put_vcpu_events(x86_cpu, level);
4128     if (ret < 0) {
4129         return ret;
4130     }
4131     if (level >= KVM_PUT_RESET_STATE) {
4132         ret = kvm_put_mp_state(x86_cpu);
4133         if (ret < 0) {
4134             return ret;
4135         }
4136     }
4137 
4138     ret = kvm_put_tscdeadline_msr(x86_cpu);
4139     if (ret < 0) {
4140         return ret;
4141     }
4142     ret = kvm_put_debugregs(x86_cpu);
4143     if (ret < 0) {
4144         return ret;
4145     }
4146     /* must be last */
4147     ret = kvm_guest_debug_workarounds(x86_cpu);
4148     if (ret < 0) {
4149         return ret;
4150     }
4151     return 0;
4152 }
4153 
4154 int kvm_arch_get_registers(CPUState *cs)
4155 {
4156     X86CPU *cpu = X86_CPU(cs);
4157     int ret;
4158 
4159     assert(cpu_is_stopped(cs) || qemu_cpu_is_self(cs));
4160 
4161     ret = kvm_get_vcpu_events(cpu);
4162     if (ret < 0) {
4163         goto out;
4164     }
4165     /*
4166      * KVM_GET_MPSTATE can modify CS and RIP, call it before
4167      * KVM_GET_REGS and KVM_GET_SREGS.
4168      */
4169     ret = kvm_get_mp_state(cpu);
4170     if (ret < 0) {
4171         goto out;
4172     }
4173     ret = kvm_getput_regs(cpu, 0);
4174     if (ret < 0) {
4175         goto out;
4176     }
4177     ret = kvm_get_xsave(cpu);
4178     if (ret < 0) {
4179         goto out;
4180     }
4181     ret = kvm_get_xcrs(cpu);
4182     if (ret < 0) {
4183         goto out;
4184     }
4185     ret = kvm_get_sregs(cpu);
4186     if (ret < 0) {
4187         goto out;
4188     }
4189     ret = kvm_get_msrs(cpu);
4190     if (ret < 0) {
4191         goto out;
4192     }
4193     ret = kvm_get_apic(cpu);
4194     if (ret < 0) {
4195         goto out;
4196     }
4197     ret = kvm_get_debugregs(cpu);
4198     if (ret < 0) {
4199         goto out;
4200     }
4201     ret = kvm_get_nested_state(cpu);
4202     if (ret < 0) {
4203         goto out;
4204     }
4205     ret = 0;
4206  out:
4207     cpu_sync_bndcs_hflags(&cpu->env);
4208     return ret;
4209 }
4210 
4211 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
4212 {
4213     X86CPU *x86_cpu = X86_CPU(cpu);
4214     CPUX86State *env = &x86_cpu->env;
4215     int ret;
4216 
4217     /* Inject NMI */
4218     if (cpu->interrupt_request & (CPU_INTERRUPT_NMI | CPU_INTERRUPT_SMI)) {
4219         if (cpu->interrupt_request & CPU_INTERRUPT_NMI) {
4220             qemu_mutex_lock_iothread();
4221             cpu->interrupt_request &= ~CPU_INTERRUPT_NMI;
4222             qemu_mutex_unlock_iothread();
4223             DPRINTF("injected NMI\n");
4224             ret = kvm_vcpu_ioctl(cpu, KVM_NMI);
4225             if (ret < 0) {
4226                 fprintf(stderr, "KVM: injection failed, NMI lost (%s)\n",
4227                         strerror(-ret));
4228             }
4229         }
4230         if (cpu->interrupt_request & CPU_INTERRUPT_SMI) {
4231             qemu_mutex_lock_iothread();
4232             cpu->interrupt_request &= ~CPU_INTERRUPT_SMI;
4233             qemu_mutex_unlock_iothread();
4234             DPRINTF("injected SMI\n");
4235             ret = kvm_vcpu_ioctl(cpu, KVM_SMI);
4236             if (ret < 0) {
4237                 fprintf(stderr, "KVM: injection failed, SMI lost (%s)\n",
4238                         strerror(-ret));
4239             }
4240         }
4241     }
4242 
4243     if (!kvm_pic_in_kernel()) {
4244         qemu_mutex_lock_iothread();
4245     }
4246 
4247     /* Force the VCPU out of its inner loop to process any INIT requests
4248      * or (for userspace APIC, but it is cheap to combine the checks here)
4249      * pending TPR access reports.
4250      */
4251     if (cpu->interrupt_request & (CPU_INTERRUPT_INIT | CPU_INTERRUPT_TPR)) {
4252         if ((cpu->interrupt_request & CPU_INTERRUPT_INIT) &&
4253             !(env->hflags & HF_SMM_MASK)) {
4254             cpu->exit_request = 1;
4255         }
4256         if (cpu->interrupt_request & CPU_INTERRUPT_TPR) {
4257             cpu->exit_request = 1;
4258         }
4259     }
4260 
4261     if (!kvm_pic_in_kernel()) {
4262         /* Try to inject an interrupt if the guest can accept it */
4263         if (run->ready_for_interrupt_injection &&
4264             (cpu->interrupt_request & CPU_INTERRUPT_HARD) &&
4265             (env->eflags & IF_MASK)) {
4266             int irq;
4267 
4268             cpu->interrupt_request &= ~CPU_INTERRUPT_HARD;
4269             irq = cpu_get_pic_interrupt(env);
4270             if (irq >= 0) {
4271                 struct kvm_interrupt intr;
4272 
4273                 intr.irq = irq;
4274                 DPRINTF("injected interrupt %d\n", irq);
4275                 ret = kvm_vcpu_ioctl(cpu, KVM_INTERRUPT, &intr);
4276                 if (ret < 0) {
4277                     fprintf(stderr,
4278                             "KVM: injection failed, interrupt lost (%s)\n",
4279                             strerror(-ret));
4280                 }
4281             }
4282         }
4283 
4284         /* If we have an interrupt but the guest is not ready to receive an
4285          * interrupt, request an interrupt window exit.  This will
4286          * cause a return to userspace as soon as the guest is ready to
4287          * receive interrupts. */
4288         if ((cpu->interrupt_request & CPU_INTERRUPT_HARD)) {
4289             run->request_interrupt_window = 1;
4290         } else {
4291             run->request_interrupt_window = 0;
4292         }
4293 
4294         DPRINTF("setting tpr\n");
4295         run->cr8 = cpu_get_apic_tpr(x86_cpu->apic_state);
4296 
4297         qemu_mutex_unlock_iothread();
4298     }
4299 }
4300 
4301 static void kvm_rate_limit_on_bus_lock(void)
4302 {
4303     uint64_t delay_ns = ratelimit_calculate_delay(&bus_lock_ratelimit_ctrl, 1);
4304 
4305     if (delay_ns) {
4306         g_usleep(delay_ns / SCALE_US);
4307     }
4308 }
4309 
4310 MemTxAttrs kvm_arch_post_run(CPUState *cpu, struct kvm_run *run)
4311 {
4312     X86CPU *x86_cpu = X86_CPU(cpu);
4313     CPUX86State *env = &x86_cpu->env;
4314 
4315     if (run->flags & KVM_RUN_X86_SMM) {
4316         env->hflags |= HF_SMM_MASK;
4317     } else {
4318         env->hflags &= ~HF_SMM_MASK;
4319     }
4320     if (run->if_flag) {
4321         env->eflags |= IF_MASK;
4322     } else {
4323         env->eflags &= ~IF_MASK;
4324     }
4325     if (run->flags & KVM_RUN_X86_BUS_LOCK) {
4326         kvm_rate_limit_on_bus_lock();
4327     }
4328 
4329     /* We need to protect the apic state against concurrent accesses from
4330      * different threads in case the userspace irqchip is used. */
4331     if (!kvm_irqchip_in_kernel()) {
4332         qemu_mutex_lock_iothread();
4333     }
4334     cpu_set_apic_tpr(x86_cpu->apic_state, run->cr8);
4335     cpu_set_apic_base(x86_cpu->apic_state, run->apic_base);
4336     if (!kvm_irqchip_in_kernel()) {
4337         qemu_mutex_unlock_iothread();
4338     }
4339     return cpu_get_mem_attrs(env);
4340 }
4341 
4342 int kvm_arch_process_async_events(CPUState *cs)
4343 {
4344     X86CPU *cpu = X86_CPU(cs);
4345     CPUX86State *env = &cpu->env;
4346 
4347     if (cs->interrupt_request & CPU_INTERRUPT_MCE) {
4348         /* We must not raise CPU_INTERRUPT_MCE if it's not supported. */
4349         assert(env->mcg_cap);
4350 
4351         cs->interrupt_request &= ~CPU_INTERRUPT_MCE;
4352 
4353         kvm_cpu_synchronize_state(cs);
4354 
4355         if (env->exception_nr == EXCP08_DBLE) {
4356             /* this means triple fault */
4357             qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET);
4358             cs->exit_request = 1;
4359             return 0;
4360         }
4361         kvm_queue_exception(env, EXCP12_MCHK, 0, 0);
4362         env->has_error_code = 0;
4363 
4364         cs->halted = 0;
4365         if (kvm_irqchip_in_kernel() && env->mp_state == KVM_MP_STATE_HALTED) {
4366             env->mp_state = KVM_MP_STATE_RUNNABLE;
4367         }
4368     }
4369 
4370     if ((cs->interrupt_request & CPU_INTERRUPT_INIT) &&
4371         !(env->hflags & HF_SMM_MASK)) {
4372         kvm_cpu_synchronize_state(cs);
4373         do_cpu_init(cpu);
4374     }
4375 
4376     if (kvm_irqchip_in_kernel()) {
4377         return 0;
4378     }
4379 
4380     if (cs->interrupt_request & CPU_INTERRUPT_POLL) {
4381         cs->interrupt_request &= ~CPU_INTERRUPT_POLL;
4382         apic_poll_irq(cpu->apic_state);
4383     }
4384     if (((cs->interrupt_request & CPU_INTERRUPT_HARD) &&
4385          (env->eflags & IF_MASK)) ||
4386         (cs->interrupt_request & CPU_INTERRUPT_NMI)) {
4387         cs->halted = 0;
4388     }
4389     if (cs->interrupt_request & CPU_INTERRUPT_SIPI) {
4390         kvm_cpu_synchronize_state(cs);
4391         do_cpu_sipi(cpu);
4392     }
4393     if (cs->interrupt_request & CPU_INTERRUPT_TPR) {
4394         cs->interrupt_request &= ~CPU_INTERRUPT_TPR;
4395         kvm_cpu_synchronize_state(cs);
4396         apic_handle_tpr_access_report(cpu->apic_state, env->eip,
4397                                       env->tpr_access_type);
4398     }
4399 
4400     return cs->halted;
4401 }
4402 
4403 static int kvm_handle_halt(X86CPU *cpu)
4404 {
4405     CPUState *cs = CPU(cpu);
4406     CPUX86State *env = &cpu->env;
4407 
4408     if (!((cs->interrupt_request & CPU_INTERRUPT_HARD) &&
4409           (env->eflags & IF_MASK)) &&
4410         !(cs->interrupt_request & CPU_INTERRUPT_NMI)) {
4411         cs->halted = 1;
4412         return EXCP_HLT;
4413     }
4414 
4415     return 0;
4416 }
4417 
4418 static int kvm_handle_tpr_access(X86CPU *cpu)
4419 {
4420     CPUState *cs = CPU(cpu);
4421     struct kvm_run *run = cs->kvm_run;
4422 
4423     apic_handle_tpr_access_report(cpu->apic_state, run->tpr_access.rip,
4424                                   run->tpr_access.is_write ? TPR_ACCESS_WRITE
4425                                                            : TPR_ACCESS_READ);
4426     return 1;
4427 }
4428 
4429 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
4430 {
4431     static const uint8_t int3 = 0xcc;
4432 
4433     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 1, 0) ||
4434         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&int3, 1, 1)) {
4435         return -EINVAL;
4436     }
4437     return 0;
4438 }
4439 
4440 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
4441 {
4442     uint8_t int3;
4443 
4444     if (cpu_memory_rw_debug(cs, bp->pc, &int3, 1, 0)) {
4445         return -EINVAL;
4446     }
4447     if (int3 != 0xcc) {
4448         return 0;
4449     }
4450     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 1, 1)) {
4451         return -EINVAL;
4452     }
4453     return 0;
4454 }
4455 
4456 static struct {
4457     target_ulong addr;
4458     int len;
4459     int type;
4460 } hw_breakpoint[4];
4461 
4462 static int nb_hw_breakpoint;
4463 
4464 static int find_hw_breakpoint(target_ulong addr, int len, int type)
4465 {
4466     int n;
4467 
4468     for (n = 0; n < nb_hw_breakpoint; n++) {
4469         if (hw_breakpoint[n].addr == addr && hw_breakpoint[n].type == type &&
4470             (hw_breakpoint[n].len == len || len == -1)) {
4471             return n;
4472         }
4473     }
4474     return -1;
4475 }
4476 
4477 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
4478                                   target_ulong len, int type)
4479 {
4480     switch (type) {
4481     case GDB_BREAKPOINT_HW:
4482         len = 1;
4483         break;
4484     case GDB_WATCHPOINT_WRITE:
4485     case GDB_WATCHPOINT_ACCESS:
4486         switch (len) {
4487         case 1:
4488             break;
4489         case 2:
4490         case 4:
4491         case 8:
4492             if (addr & (len - 1)) {
4493                 return -EINVAL;
4494             }
4495             break;
4496         default:
4497             return -EINVAL;
4498         }
4499         break;
4500     default:
4501         return -ENOSYS;
4502     }
4503 
4504     if (nb_hw_breakpoint == 4) {
4505         return -ENOBUFS;
4506     }
4507     if (find_hw_breakpoint(addr, len, type) >= 0) {
4508         return -EEXIST;
4509     }
4510     hw_breakpoint[nb_hw_breakpoint].addr = addr;
4511     hw_breakpoint[nb_hw_breakpoint].len = len;
4512     hw_breakpoint[nb_hw_breakpoint].type = type;
4513     nb_hw_breakpoint++;
4514 
4515     return 0;
4516 }
4517 
4518 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
4519                                   target_ulong len, int type)
4520 {
4521     int n;
4522 
4523     n = find_hw_breakpoint(addr, (type == GDB_BREAKPOINT_HW) ? 1 : len, type);
4524     if (n < 0) {
4525         return -ENOENT;
4526     }
4527     nb_hw_breakpoint--;
4528     hw_breakpoint[n] = hw_breakpoint[nb_hw_breakpoint];
4529 
4530     return 0;
4531 }
4532 
4533 void kvm_arch_remove_all_hw_breakpoints(void)
4534 {
4535     nb_hw_breakpoint = 0;
4536 }
4537 
4538 static CPUWatchpoint hw_watchpoint;
4539 
4540 static int kvm_handle_debug(X86CPU *cpu,
4541                             struct kvm_debug_exit_arch *arch_info)
4542 {
4543     CPUState *cs = CPU(cpu);
4544     CPUX86State *env = &cpu->env;
4545     int ret = 0;
4546     int n;
4547 
4548     if (arch_info->exception == EXCP01_DB) {
4549         if (arch_info->dr6 & DR6_BS) {
4550             if (cs->singlestep_enabled) {
4551                 ret = EXCP_DEBUG;
4552             }
4553         } else {
4554             for (n = 0; n < 4; n++) {
4555                 if (arch_info->dr6 & (1 << n)) {
4556                     switch ((arch_info->dr7 >> (16 + n*4)) & 0x3) {
4557                     case 0x0:
4558                         ret = EXCP_DEBUG;
4559                         break;
4560                     case 0x1:
4561                         ret = EXCP_DEBUG;
4562                         cs->watchpoint_hit = &hw_watchpoint;
4563                         hw_watchpoint.vaddr = hw_breakpoint[n].addr;
4564                         hw_watchpoint.flags = BP_MEM_WRITE;
4565                         break;
4566                     case 0x3:
4567                         ret = EXCP_DEBUG;
4568                         cs->watchpoint_hit = &hw_watchpoint;
4569                         hw_watchpoint.vaddr = hw_breakpoint[n].addr;
4570                         hw_watchpoint.flags = BP_MEM_ACCESS;
4571                         break;
4572                     }
4573                 }
4574             }
4575         }
4576     } else if (kvm_find_sw_breakpoint(cs, arch_info->pc)) {
4577         ret = EXCP_DEBUG;
4578     }
4579     if (ret == 0) {
4580         cpu_synchronize_state(cs);
4581         assert(env->exception_nr == -1);
4582 
4583         /* pass to guest */
4584         kvm_queue_exception(env, arch_info->exception,
4585                             arch_info->exception == EXCP01_DB,
4586                             arch_info->dr6);
4587         env->has_error_code = 0;
4588     }
4589 
4590     return ret;
4591 }
4592 
4593 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
4594 {
4595     const uint8_t type_code[] = {
4596         [GDB_BREAKPOINT_HW] = 0x0,
4597         [GDB_WATCHPOINT_WRITE] = 0x1,
4598         [GDB_WATCHPOINT_ACCESS] = 0x3
4599     };
4600     const uint8_t len_code[] = {
4601         [1] = 0x0, [2] = 0x1, [4] = 0x3, [8] = 0x2
4602     };
4603     int n;
4604 
4605     if (kvm_sw_breakpoints_active(cpu)) {
4606         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
4607     }
4608     if (nb_hw_breakpoint > 0) {
4609         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
4610         dbg->arch.debugreg[7] = 0x0600;
4611         for (n = 0; n < nb_hw_breakpoint; n++) {
4612             dbg->arch.debugreg[n] = hw_breakpoint[n].addr;
4613             dbg->arch.debugreg[7] |= (2 << (n * 2)) |
4614                 (type_code[hw_breakpoint[n].type] << (16 + n*4)) |
4615                 ((uint32_t)len_code[hw_breakpoint[n].len] << (18 + n*4));
4616         }
4617     }
4618 }
4619 
4620 static bool host_supports_vmx(void)
4621 {
4622     uint32_t ecx, unused;
4623 
4624     host_cpuid(1, 0, &unused, &unused, &ecx, &unused);
4625     return ecx & CPUID_EXT_VMX;
4626 }
4627 
4628 #define VMX_INVALID_GUEST_STATE 0x80000021
4629 
4630 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
4631 {
4632     X86CPU *cpu = X86_CPU(cs);
4633     uint64_t code;
4634     int ret;
4635 
4636     switch (run->exit_reason) {
4637     case KVM_EXIT_HLT:
4638         DPRINTF("handle_hlt\n");
4639         qemu_mutex_lock_iothread();
4640         ret = kvm_handle_halt(cpu);
4641         qemu_mutex_unlock_iothread();
4642         break;
4643     case KVM_EXIT_SET_TPR:
4644         ret = 0;
4645         break;
4646     case KVM_EXIT_TPR_ACCESS:
4647         qemu_mutex_lock_iothread();
4648         ret = kvm_handle_tpr_access(cpu);
4649         qemu_mutex_unlock_iothread();
4650         break;
4651     case KVM_EXIT_FAIL_ENTRY:
4652         code = run->fail_entry.hardware_entry_failure_reason;
4653         fprintf(stderr, "KVM: entry failed, hardware error 0x%" PRIx64 "\n",
4654                 code);
4655         if (host_supports_vmx() && code == VMX_INVALID_GUEST_STATE) {
4656             fprintf(stderr,
4657                     "\nIf you're running a guest on an Intel machine without "
4658                         "unrestricted mode\n"
4659                     "support, the failure can be most likely due to the guest "
4660                         "entering an invalid\n"
4661                     "state for Intel VT. For example, the guest maybe running "
4662                         "in big real mode\n"
4663                     "which is not supported on less recent Intel processors."
4664                         "\n\n");
4665         }
4666         ret = -1;
4667         break;
4668     case KVM_EXIT_EXCEPTION:
4669         fprintf(stderr, "KVM: exception %d exit (error code 0x%x)\n",
4670                 run->ex.exception, run->ex.error_code);
4671         ret = -1;
4672         break;
4673     case KVM_EXIT_DEBUG:
4674         DPRINTF("kvm_exit_debug\n");
4675         qemu_mutex_lock_iothread();
4676         ret = kvm_handle_debug(cpu, &run->debug.arch);
4677         qemu_mutex_unlock_iothread();
4678         break;
4679     case KVM_EXIT_HYPERV:
4680         ret = kvm_hv_handle_exit(cpu, &run->hyperv);
4681         break;
4682     case KVM_EXIT_IOAPIC_EOI:
4683         ioapic_eoi_broadcast(run->eoi.vector);
4684         ret = 0;
4685         break;
4686     case KVM_EXIT_X86_BUS_LOCK:
4687         /* already handled in kvm_arch_post_run */
4688         ret = 0;
4689         break;
4690     default:
4691         fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
4692         ret = -1;
4693         break;
4694     }
4695 
4696     return ret;
4697 }
4698 
4699 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
4700 {
4701     X86CPU *cpu = X86_CPU(cs);
4702     CPUX86State *env = &cpu->env;
4703 
4704     kvm_cpu_synchronize_state(cs);
4705     return !(env->cr[0] & CR0_PE_MASK) ||
4706            ((env->segs[R_CS].selector  & 3) != 3);
4707 }
4708 
4709 void kvm_arch_init_irq_routing(KVMState *s)
4710 {
4711     /* We know at this point that we're using the in-kernel
4712      * irqchip, so we can use irqfds, and on x86 we know
4713      * we can use msi via irqfd and GSI routing.
4714      */
4715     kvm_msi_via_irqfd_allowed = true;
4716     kvm_gsi_routing_allowed = true;
4717 
4718     if (kvm_irqchip_is_split()) {
4719         int i;
4720 
4721         /* If the ioapic is in QEMU and the lapics are in KVM, reserve
4722            MSI routes for signaling interrupts to the local apics. */
4723         for (i = 0; i < IOAPIC_NUM_PINS; i++) {
4724             if (kvm_irqchip_add_msi_route(s, 0, NULL) < 0) {
4725                 error_report("Could not enable split IRQ mode.");
4726                 exit(1);
4727             }
4728         }
4729     }
4730 }
4731 
4732 int kvm_arch_irqchip_create(KVMState *s)
4733 {
4734     int ret;
4735     if (kvm_kernel_irqchip_split()) {
4736         ret = kvm_vm_enable_cap(s, KVM_CAP_SPLIT_IRQCHIP, 0, 24);
4737         if (ret) {
4738             error_report("Could not enable split irqchip mode: %s",
4739                          strerror(-ret));
4740             exit(1);
4741         } else {
4742             DPRINTF("Enabled KVM_CAP_SPLIT_IRQCHIP\n");
4743             kvm_split_irqchip = true;
4744             return 1;
4745         }
4746     } else {
4747         return 0;
4748     }
4749 }
4750 
4751 uint64_t kvm_swizzle_msi_ext_dest_id(uint64_t address)
4752 {
4753     CPUX86State *env;
4754     uint64_t ext_id;
4755 
4756     if (!first_cpu) {
4757         return address;
4758     }
4759     env = &X86_CPU(first_cpu)->env;
4760     if (!(env->features[FEAT_KVM] & (1 << KVM_FEATURE_MSI_EXT_DEST_ID))) {
4761         return address;
4762     }
4763 
4764     /*
4765      * If the remappable format bit is set, or the upper bits are
4766      * already set in address_hi, or the low extended bits aren't
4767      * there anyway, do nothing.
4768      */
4769     ext_id = address & (0xff << MSI_ADDR_DEST_IDX_SHIFT);
4770     if (!ext_id || (ext_id & (1 << MSI_ADDR_DEST_IDX_SHIFT)) || (address >> 32)) {
4771         return address;
4772     }
4773 
4774     address &= ~ext_id;
4775     address |= ext_id << 35;
4776     return address;
4777 }
4778 
4779 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
4780                              uint64_t address, uint32_t data, PCIDevice *dev)
4781 {
4782     X86IOMMUState *iommu = x86_iommu_get_default();
4783 
4784     if (iommu) {
4785         X86IOMMUClass *class = X86_IOMMU_DEVICE_GET_CLASS(iommu);
4786 
4787         if (class->int_remap) {
4788             int ret;
4789             MSIMessage src, dst;
4790 
4791             src.address = route->u.msi.address_hi;
4792             src.address <<= VTD_MSI_ADDR_HI_SHIFT;
4793             src.address |= route->u.msi.address_lo;
4794             src.data = route->u.msi.data;
4795 
4796             ret = class->int_remap(iommu, &src, &dst, dev ?     \
4797                                    pci_requester_id(dev) :      \
4798                                    X86_IOMMU_SID_INVALID);
4799             if (ret) {
4800                 trace_kvm_x86_fixup_msi_error(route->gsi);
4801                 return 1;
4802             }
4803 
4804             /*
4805              * Handled untranslated compatibilty format interrupt with
4806              * extended destination ID in the low bits 11-5. */
4807             dst.address = kvm_swizzle_msi_ext_dest_id(dst.address);
4808 
4809             route->u.msi.address_hi = dst.address >> VTD_MSI_ADDR_HI_SHIFT;
4810             route->u.msi.address_lo = dst.address & VTD_MSI_ADDR_LO_MASK;
4811             route->u.msi.data = dst.data;
4812             return 0;
4813         }
4814     }
4815 
4816     address = kvm_swizzle_msi_ext_dest_id(address);
4817     route->u.msi.address_hi = address >> VTD_MSI_ADDR_HI_SHIFT;
4818     route->u.msi.address_lo = address & VTD_MSI_ADDR_LO_MASK;
4819     return 0;
4820 }
4821 
4822 typedef struct MSIRouteEntry MSIRouteEntry;
4823 
4824 struct MSIRouteEntry {
4825     PCIDevice *dev;             /* Device pointer */
4826     int vector;                 /* MSI/MSIX vector index */
4827     int virq;                   /* Virtual IRQ index */
4828     QLIST_ENTRY(MSIRouteEntry) list;
4829 };
4830 
4831 /* List of used GSI routes */
4832 static QLIST_HEAD(, MSIRouteEntry) msi_route_list = \
4833     QLIST_HEAD_INITIALIZER(msi_route_list);
4834 
4835 static void kvm_update_msi_routes_all(void *private, bool global,
4836                                       uint32_t index, uint32_t mask)
4837 {
4838     int cnt = 0, vector;
4839     MSIRouteEntry *entry;
4840     MSIMessage msg;
4841     PCIDevice *dev;
4842 
4843     /* TODO: explicit route update */
4844     QLIST_FOREACH(entry, &msi_route_list, list) {
4845         cnt++;
4846         vector = entry->vector;
4847         dev = entry->dev;
4848         if (msix_enabled(dev) && !msix_is_masked(dev, vector)) {
4849             msg = msix_get_message(dev, vector);
4850         } else if (msi_enabled(dev) && !msi_is_masked(dev, vector)) {
4851             msg = msi_get_message(dev, vector);
4852         } else {
4853             /*
4854              * Either MSI/MSIX is disabled for the device, or the
4855              * specific message was masked out.  Skip this one.
4856              */
4857             continue;
4858         }
4859         kvm_irqchip_update_msi_route(kvm_state, entry->virq, msg, dev);
4860     }
4861     kvm_irqchip_commit_routes(kvm_state);
4862     trace_kvm_x86_update_msi_routes(cnt);
4863 }
4864 
4865 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
4866                                 int vector, PCIDevice *dev)
4867 {
4868     static bool notify_list_inited = false;
4869     MSIRouteEntry *entry;
4870 
4871     if (!dev) {
4872         /* These are (possibly) IOAPIC routes only used for split
4873          * kernel irqchip mode, while what we are housekeeping are
4874          * PCI devices only. */
4875         return 0;
4876     }
4877 
4878     entry = g_new0(MSIRouteEntry, 1);
4879     entry->dev = dev;
4880     entry->vector = vector;
4881     entry->virq = route->gsi;
4882     QLIST_INSERT_HEAD(&msi_route_list, entry, list);
4883 
4884     trace_kvm_x86_add_msi_route(route->gsi);
4885 
4886     if (!notify_list_inited) {
4887         /* For the first time we do add route, add ourselves into
4888          * IOMMU's IEC notify list if needed. */
4889         X86IOMMUState *iommu = x86_iommu_get_default();
4890         if (iommu) {
4891             x86_iommu_iec_register_notifier(iommu,
4892                                             kvm_update_msi_routes_all,
4893                                             NULL);
4894         }
4895         notify_list_inited = true;
4896     }
4897     return 0;
4898 }
4899 
4900 int kvm_arch_release_virq_post(int virq)
4901 {
4902     MSIRouteEntry *entry, *next;
4903     QLIST_FOREACH_SAFE(entry, &msi_route_list, list, next) {
4904         if (entry->virq == virq) {
4905             trace_kvm_x86_remove_msi_route(virq);
4906             QLIST_REMOVE(entry, list);
4907             g_free(entry);
4908             break;
4909         }
4910     }
4911     return 0;
4912 }
4913 
4914 int kvm_arch_msi_data_to_gsi(uint32_t data)
4915 {
4916     abort();
4917 }
4918 
4919 bool kvm_has_waitpkg(void)
4920 {
4921     return has_msr_umwait;
4922 }
4923 
4924 bool kvm_arch_cpu_check_are_resettable(void)
4925 {
4926     return !sev_es_enabled();
4927 }
4928