?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
pci: Simplify pci_bus_is_root()
[qemu.git] / target / i386 / kvm.c
blob3b29ce5c0d087b80c71950428442ef651b261a3c
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/error.h"
17 #include <sys/ioctl.h>
18 #include <sys/utsname.h>
19
20 #include <linux/kvm.h>
21 #include "standard-headers/asm-x86/kvm_para.h"
22
23 #include "qemu-common.h"
24 #include "cpu.h"
25 #include "sysemu/sysemu.h"
26 #include "sysemu/hw_accel.h"
27 #include "sysemu/kvm_int.h"
28 #include "kvm_i386.h"
29 #include "hyperv.h"
30 #include "hyperv-proto.h"
31
32 #include "exec/gdbstub.h"
33 #include "qemu/host-utils.h"
34 #include "qemu/config-file.h"
35 #include "qemu/error-report.h"
36 #include "hw/i386/pc.h"
37 #include "hw/i386/apic.h"
38 #include "hw/i386/apic_internal.h"
39 #include "hw/i386/apic-msidef.h"
40 #include "hw/i386/intel_iommu.h"
41 #include "hw/i386/x86-iommu.h"
42
43 #include "hw/pci/pci.h"
44 #include "hw/pci/msi.h"
45 #include "hw/pci/msix.h"
46 #include "migration/blocker.h"
47 #include "exec/memattrs.h"
48 #include "trace.h"
49
50 //#define DEBUG_KVM
51
52 #ifdef DEBUG_KVM
53 #define DPRINTF(fmt, ...) \
54 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
55 #else
56 #define DPRINTF(fmt, ...) \
57 do { } while (0)
58 #endif
59
60 #define MSR_KVM_WALL_CLOCK 0x11
61 #define MSR_KVM_SYSTEM_TIME 0x12
62
63 /* A 4096-byte buffer can hold the 8-byte kvm_msrs header, plus
64 * 255 kvm_msr_entry structs */
65 #define MSR_BUF_SIZE 4096
66
67 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
68 KVM_CAP_INFO(SET_TSS_ADDR),
69 KVM_CAP_INFO(EXT_CPUID),
70 KVM_CAP_INFO(MP_STATE),
71 KVM_CAP_LAST_INFO
72 };
73
74 static bool has_msr_star;
75 static bool has_msr_hsave_pa;
76 static bool has_msr_tsc_aux;
77 static bool has_msr_tsc_adjust;
78 static bool has_msr_tsc_deadline;
79 static bool has_msr_feature_control;
80 static bool has_msr_misc_enable;
81 static bool has_msr_smbase;
82 static bool has_msr_bndcfgs;
83 static int lm_capable_kernel;
84 static bool has_msr_hv_hypercall;
85 static bool has_msr_hv_crash;
86 static bool has_msr_hv_reset;
87 static bool has_msr_hv_vpindex;
88 static bool hv_vpindex_settable;
89 static bool has_msr_hv_runtime;
90 static bool has_msr_hv_synic;
91 static bool has_msr_hv_stimer;
92 static bool has_msr_hv_frequencies;
93 static bool has_msr_hv_reenlightenment;
94 static bool has_msr_xss;
95 static bool has_msr_spec_ctrl;
96 static bool has_msr_virt_ssbd;
97 static bool has_msr_smi_count;
98 static bool has_msr_arch_capabs;
99
100 static uint32_t has_architectural_pmu_version;
101 static uint32_t num_architectural_pmu_gp_counters;
102 static uint32_t num_architectural_pmu_fixed_counters;
103
104 static int has_xsave;
105 static int has_xcrs;
106 static int has_pit_state2;
107
108 static bool has_msr_mcg_ext_ctl;
109
110 static struct kvm_cpuid2 *cpuid_cache;
111 static struct kvm_msr_list *kvm_feature_msrs;
112
113 int kvm_has_pit_state2(void)
114 {
115 return has_pit_state2;
116 }
117
118 bool kvm_has_smm(void)
119 {
120 return kvm_check_extension(kvm_state, KVM_CAP_X86_SMM);
121 }
122
123 bool kvm_has_adjust_clock_stable(void)
124 {
125 int ret = kvm_check_extension(kvm_state, KVM_CAP_ADJUST_CLOCK);
126
127 return (ret == KVM_CLOCK_TSC_STABLE);
128 }
129
130 bool kvm_allows_irq0_override(void)
131 {
132 return !kvm_irqchip_in_kernel() || kvm_has_gsi_routing();
133 }
134
135 static bool kvm_x2apic_api_set_flags(uint64_t flags)
136 {
137 KVMState *s = KVM_STATE(current_machine->accelerator);
138
139 return !kvm_vm_enable_cap(s, KVM_CAP_X2APIC_API, 0, flags);
140 }
141
142 #define MEMORIZE(fn, _result) \
143 ({ \
144 static bool _memorized; \
145 \
146 if (_memorized) { \
147 return _result; \
148 } \
149 _memorized = true; \
150 _result = fn; \
151 })
152
153 static bool has_x2apic_api;
154
155 bool kvm_has_x2apic_api(void)
156 {
157 return has_x2apic_api;
158 }
159
160 bool kvm_enable_x2apic(void)
161 {
162 return MEMORIZE(
163 kvm_x2apic_api_set_flags(KVM_X2APIC_API_USE_32BIT_IDS |
164 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK),
165 has_x2apic_api);
166 }
167
168 bool kvm_hv_vpindex_settable(void)
169 {
170 return hv_vpindex_settable;
171 }
172
173 static int kvm_get_tsc(CPUState *cs)
174 {
175 X86CPU *cpu = X86_CPU(cs);
176 CPUX86State *env = &cpu->env;
177 struct {
178 struct kvm_msrs info;
179 struct kvm_msr_entry entries[1];
180 } msr_data;
181 int ret;
182
183 if (env->tsc_valid) {
184 return 0;
185 }
186
187 msr_data.info.nmsrs = 1;
188 msr_data.entries[0].index = MSR_IA32_TSC;
189 env->tsc_valid = !runstate_is_running();
190
191 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MSRS, &msr_data);
192 if (ret < 0) {
193 return ret;
194 }
195
196 assert(ret == 1);
197 env->tsc = msr_data.entries[0].data;
198 return 0;
199 }
200
201 static inline void do_kvm_synchronize_tsc(CPUState *cpu, run_on_cpu_data arg)
202 {
203 kvm_get_tsc(cpu);
204 }
205
206 void kvm_synchronize_all_tsc(void)
207 {
208 CPUState *cpu;
209
210 if (kvm_enabled()) {
211 CPU_FOREACH(cpu) {
212 run_on_cpu(cpu, do_kvm_synchronize_tsc, RUN_ON_CPU_NULL);
213 }
214 }
215 }
216
217 static struct kvm_cpuid2 *try_get_cpuid(KVMState *s, int max)
218 {
219 struct kvm_cpuid2 *cpuid;
220 int r, size;
221
222 size = sizeof(*cpuid) + max * sizeof(*cpuid->entries);
223 cpuid = g_malloc0(size);
224 cpuid->nent = max;
225 r = kvm_ioctl(s, KVM_GET_SUPPORTED_CPUID, cpuid);
226 if (r == 0 && cpuid->nent >= max) {
227 r = -E2BIG;
228 }
229 if (r < 0) {
230 if (r == -E2BIG) {
231 g_free(cpuid);
232 return NULL;
233 } else {
234 fprintf(stderr, "KVM_GET_SUPPORTED_CPUID failed: %s\n",
235 strerror(-r));
236 exit(1);
237 }
238 }
239 return cpuid;
240 }
241
242 /* Run KVM_GET_SUPPORTED_CPUID ioctl(), allocating a buffer large enough
243 * for all entries.
244 */
245 static struct kvm_cpuid2 *get_supported_cpuid(KVMState *s)
246 {
247 struct kvm_cpuid2 *cpuid;
248 int max = 1;
249
250 if (cpuid_cache != NULL) {
251 return cpuid_cache;
252 }
253 while ((cpuid = try_get_cpuid(s, max)) == NULL) {
254 max *= 2;
255 }
256 cpuid_cache = cpuid;
257 return cpuid;
258 }
259
260 static const struct kvm_para_features {
261 int cap;
262 int feature;
263 } para_features[] = {
264 { KVM_CAP_CLOCKSOURCE, KVM_FEATURE_CLOCKSOURCE },
265 { KVM_CAP_NOP_IO_DELAY, KVM_FEATURE_NOP_IO_DELAY },
266 { KVM_CAP_PV_MMU, KVM_FEATURE_MMU_OP },
267 { KVM_CAP_ASYNC_PF, KVM_FEATURE_ASYNC_PF },
268 };
269
270 static int get_para_features(KVMState *s)
271 {
272 int i, features = 0;
273
274 for (i = 0; i < ARRAY_SIZE(para_features); i++) {
275 if (kvm_check_extension(s, para_features[i].cap)) {
276 features |= (1 << para_features[i].feature);
277 }
278 }
279
280 return features;
281 }
282
283 static bool host_tsx_blacklisted(void)
284 {
285 int family, model, stepping;\
286 char vendor[CPUID_VENDOR_SZ + 1];
287
288 host_vendor_fms(vendor, &family, &model, &stepping);
289
290 /* Check if we are running on a Haswell host known to have broken TSX */
291 return !strcmp(vendor, CPUID_VENDOR_INTEL) &&
292 (family == 6) &&
293 ((model == 63 && stepping < 4) ||
294 model == 60 || model == 69 || model == 70);
295 }
296
297 /* Returns the value for a specific register on the cpuid entry
298 */
299 static uint32_t cpuid_entry_get_reg(struct kvm_cpuid_entry2 *entry, int reg)
300 {
301 uint32_t ret = 0;
302 switch (reg) {
303 case R_EAX:
304 ret = entry->eax;
305 break;
306 case R_EBX:
307 ret = entry->ebx;
308 break;
309 case R_ECX:
310 ret = entry->ecx;
311 break;
312 case R_EDX:
313 ret = entry->edx;
314 break;
315 }
316 return ret;
317 }
318
319 /* Find matching entry for function/index on kvm_cpuid2 struct
320 */
321 static struct kvm_cpuid_entry2 *cpuid_find_entry(struct kvm_cpuid2 *cpuid,
322 uint32_t function,
323 uint32_t index)
324 {
325 int i;
326 for (i = 0; i < cpuid->nent; ++i) {
327 if (cpuid->entries[i].function == function &&
328 cpuid->entries[i].index == index) {
329 return &cpuid->entries[i];
330 }
331 }
332 /* not found: */
333 return NULL;
334 }
335
336 uint32_t kvm_arch_get_supported_cpuid(KVMState *s, uint32_t function,
337 uint32_t index, int reg)
338 {
339 struct kvm_cpuid2 *cpuid;
340 uint32_t ret = 0;
341 uint32_t cpuid_1_edx;
342 bool found = false;
343
344 cpuid = get_supported_cpuid(s);
345
346 struct kvm_cpuid_entry2 *entry = cpuid_find_entry(cpuid, function, index);
347 if (entry) {
348 found = true;
349 ret = cpuid_entry_get_reg(entry, reg);
350 }
351
352 /* Fixups for the data returned by KVM, below */
353
354 if (function == 1 && reg == R_EDX) {
355 /* KVM before 2.6.30 misreports the following features */
356 ret |= CPUID_MTRR | CPUID_PAT | CPUID_MCE | CPUID_MCA;
357 } else if (function == 1 && reg == R_ECX) {
358 /* We can set the hypervisor flag, even if KVM does not return it on
359 * GET_SUPPORTED_CPUID
360 */
361 ret |= CPUID_EXT_HYPERVISOR;
362 /* tsc-deadline flag is not returned by GET_SUPPORTED_CPUID, but it
363 * can be enabled if the kernel has KVM_CAP_TSC_DEADLINE_TIMER,
364 * and the irqchip is in the kernel.
365 */
366 if (kvm_irqchip_in_kernel() &&
367 kvm_check_extension(s, KVM_CAP_TSC_DEADLINE_TIMER)) {
368 ret |= CPUID_EXT_TSC_DEADLINE_TIMER;
369 }
370
371 /* x2apic is reported by GET_SUPPORTED_CPUID, but it can't be enabled
372 * without the in-kernel irqchip
373 */
374 if (!kvm_irqchip_in_kernel()) {
375 ret &= ~CPUID_EXT_X2APIC;
376 }
377
378 if (enable_cpu_pm) {
379 int disable_exits = kvm_check_extension(s,
380 KVM_CAP_X86_DISABLE_EXITS);
381
382 if (disable_exits & KVM_X86_DISABLE_EXITS_MWAIT) {
383 ret |= CPUID_EXT_MONITOR;
384 }
385 }
386 } else if (function == 6 && reg == R_EAX) {
387 ret |= CPUID_6_EAX_ARAT; /* safe to allow because of emulated APIC */
388 } else if (function == 7 && index == 0 && reg == R_EBX) {
389 if (host_tsx_blacklisted()) {
390 ret &= ~(CPUID_7_0_EBX_RTM | CPUID_7_0_EBX_HLE);
391 }
392 } else if (function == 7 && index == 0 && reg == R_EDX) {
393 /*
394 * Linux v4.17-v4.20 incorrectly return ARCH_CAPABILITIES on SVM hosts.
395 * We can detect the bug by checking if MSR_IA32_ARCH_CAPABILITIES is
396 * returned by KVM_GET_MSR_INDEX_LIST.
397 */
398 if (!has_msr_arch_capabs) {
399 ret &= ~CPUID_7_0_EDX_ARCH_CAPABILITIES;
400 }
401 } else if (function == 0x80000001 && reg == R_ECX) {
402 /*
403 * It's safe to enable TOPOEXT even if it's not returned by
404 * GET_SUPPORTED_CPUID. Unconditionally enabling TOPOEXT here allows
405 * us to keep CPU models including TOPOEXT runnable on older kernels.
406 */
407 ret |= CPUID_EXT3_TOPOEXT;
408 } else if (function == 0x80000001 && reg == R_EDX) {
409 /* On Intel, kvm returns cpuid according to the Intel spec,
410 * so add missing bits according to the AMD spec:
411 */
412 cpuid_1_edx = kvm_arch_get_supported_cpuid(s, 1, 0, R_EDX);
413 ret |= cpuid_1_edx & CPUID_EXT2_AMD_ALIASES;
414 } else if (function == KVM_CPUID_FEATURES && reg == R_EAX) {
415 /* kvm_pv_unhalt is reported by GET_SUPPORTED_CPUID, but it can't
416 * be enabled without the in-kernel irqchip
417 */
418 if (!kvm_irqchip_in_kernel()) {
419 ret &= ~(1U << KVM_FEATURE_PV_UNHALT);
420 }
421 } else if (function == KVM_CPUID_FEATURES && reg == R_EDX) {
422 ret |= 1U << KVM_HINTS_REALTIME;
423 found = 1;
424 }
425
426 /* fallback for older kernels */
427 if ((function == KVM_CPUID_FEATURES) && !found) {
428 ret = get_para_features(s);
429 }
430
431 return ret;
432 }
433
434 uint32_t kvm_arch_get_supported_msr_feature(KVMState *s, uint32_t index)
435 {
436 struct {
437 struct kvm_msrs info;
438 struct kvm_msr_entry entries[1];
439 } msr_data;
440 uint32_t ret;
441
442 if (kvm_feature_msrs == NULL) { /* Host doesn't support feature MSRs */
443 return 0;
444 }
445
446 /* Check if requested MSR is supported feature MSR */
447 int i;
448 for (i = 0; i < kvm_feature_msrs->nmsrs; i++)
449 if (kvm_feature_msrs->indices[i] == index) {
450 break;
451 }
452 if (i == kvm_feature_msrs->nmsrs) {
453 return 0; /* if the feature MSR is not supported, simply return 0 */
454 }
455
456 msr_data.info.nmsrs = 1;
457 msr_data.entries[0].index = index;
458
459 ret = kvm_ioctl(s, KVM_GET_MSRS, &msr_data);
460 if (ret != 1) {
461 error_report("KVM get MSR (index=0x%x) feature failed, %s",
462 index, strerror(-ret));
463 exit(1);
464 }
465
466 return msr_data.entries[0].data;
467 }
468
469
470 typedef struct HWPoisonPage {
471 ram_addr_t ram_addr;
472 QLIST_ENTRY(HWPoisonPage) list;
473 } HWPoisonPage;
474
475 static QLIST_HEAD(, HWPoisonPage) hwpoison_page_list =
476 QLIST_HEAD_INITIALIZER(hwpoison_page_list);
477
478 static void kvm_unpoison_all(void *param)
479 {
480 HWPoisonPage *page, *next_page;
481
482 QLIST_FOREACH_SAFE(page, &hwpoison_page_list, list, next_page) {
483 QLIST_REMOVE(page, list);
484 qemu_ram_remap(page->ram_addr, TARGET_PAGE_SIZE);
485 g_free(page);
486 }
487 }
488
489 static void kvm_hwpoison_page_add(ram_addr_t ram_addr)
490 {
491 HWPoisonPage *page;
492
493 QLIST_FOREACH(page, &hwpoison_page_list, list) {
494 if (page->ram_addr == ram_addr) {
495 return;
496 }
497 }
498 page = g_new(HWPoisonPage, 1);
499 page->ram_addr = ram_addr;
500 QLIST_INSERT_HEAD(&hwpoison_page_list, page, list);
501 }
502
503 static int kvm_get_mce_cap_supported(KVMState *s, uint64_t *mce_cap,
504 int *max_banks)
505 {
506 int r;
507
508 r = kvm_check_extension(s, KVM_CAP_MCE);
509 if (r > 0) {
510 *max_banks = r;
511 return kvm_ioctl(s, KVM_X86_GET_MCE_CAP_SUPPORTED, mce_cap);
512 }
513 return -ENOSYS;
514 }
515
516 static void kvm_mce_inject(X86CPU *cpu, hwaddr paddr, int code)
517 {
518 CPUState *cs = CPU(cpu);
519 CPUX86State *env = &cpu->env;
520 uint64_t status = MCI_STATUS_VAL | MCI_STATUS_UC | MCI_STATUS_EN |
521 MCI_STATUS_MISCV | MCI_STATUS_ADDRV | MCI_STATUS_S;
522 uint64_t mcg_status = MCG_STATUS_MCIP;
523 int flags = 0;
524
525 if (code == BUS_MCEERR_AR) {
526 status |= MCI_STATUS_AR | 0x134;
527 mcg_status |= MCG_STATUS_EIPV;
528 } else {
529 status |= 0xc0;
530 mcg_status |= MCG_STATUS_RIPV;
531 }
532
533 flags = cpu_x86_support_mca_broadcast(env) ? MCE_INJECT_BROADCAST : 0;
534 /* We need to read back the value of MSR_EXT_MCG_CTL that was set by the
535 * guest kernel back into env->mcg_ext_ctl.
536 */
537 cpu_synchronize_state(cs);
538 if (env->mcg_ext_ctl & MCG_EXT_CTL_LMCE_EN) {
539 mcg_status |= MCG_STATUS_LMCE;
540 flags = 0;
541 }
542
543 cpu_x86_inject_mce(NULL, cpu, 9, status, mcg_status, paddr,
544 (MCM_ADDR_PHYS << 6) | 0xc, flags);
545 }
546
547 static void hardware_memory_error(void)
548 {
549 fprintf(stderr, "Hardware memory error!\n");
550 exit(1);
551 }
552
553 void kvm_arch_on_sigbus_vcpu(CPUState *c, int code, void *addr)
554 {
555 X86CPU *cpu = X86_CPU(c);
556 CPUX86State *env = &cpu->env;
557 ram_addr_t ram_addr;
558 hwaddr paddr;
559
560 /* If we get an action required MCE, it has been injected by KVM
561 * while the VM was running. An action optional MCE instead should
562 * be coming from the main thread, which qemu_init_sigbus identifies
563 * as the "early kill" thread.
564 */
565 assert(code == BUS_MCEERR_AR || code == BUS_MCEERR_AO);
566
567 if ((env->mcg_cap & MCG_SER_P) && addr) {
568 ram_addr = qemu_ram_addr_from_host(addr);
569 if (ram_addr != RAM_ADDR_INVALID &&
570 kvm_physical_memory_addr_from_host(c->kvm_state, addr, &paddr)) {
571 kvm_hwpoison_page_add(ram_addr);
572 kvm_mce_inject(cpu, paddr, code);
573 return;
574 }
575
576 fprintf(stderr, "Hardware memory error for memory used by "
577 "QEMU itself instead of guest system!\n");
578 }
579
580 if (code == BUS_MCEERR_AR) {
581 hardware_memory_error();
582 }
583
584 /* Hope we are lucky for AO MCE */
585 }
586
587 static int kvm_inject_mce_oldstyle(X86CPU *cpu)
588 {
589 CPUX86State *env = &cpu->env;
590
591 if (!kvm_has_vcpu_events() && env->exception_injected == EXCP12_MCHK) {
592 unsigned int bank, bank_num = env->mcg_cap & 0xff;
593 struct kvm_x86_mce mce;
594
595 env->exception_injected = -1;
596
597 /*
598 * There must be at least one bank in use if an MCE is pending.
599 * Find it and use its values for the event injection.
600 */
601 for (bank = 0; bank < bank_num; bank++) {
602 if (env->mce_banks[bank * 4 + 1] & MCI_STATUS_VAL) {
603 break;
604 }
605 }
606 assert(bank < bank_num);
607
608 mce.bank = bank;
609 mce.status = env->mce_banks[bank * 4 + 1];
610 mce.mcg_status = env->mcg_status;
611 mce.addr = env->mce_banks[bank * 4 + 2];
612 mce.misc = env->mce_banks[bank * 4 + 3];
613
614 return kvm_vcpu_ioctl(CPU(cpu), KVM_X86_SET_MCE, &mce);
615 }
616 return 0;
617 }
618
619 static void cpu_update_state(void *opaque, int running, RunState state)
620 {
621 CPUX86State *env = opaque;
622
623 if (running) {
624 env->tsc_valid = false;
625 }
626 }
627
628 unsigned long kvm_arch_vcpu_id(CPUState *cs)
629 {
630 X86CPU *cpu = X86_CPU(cs);
631 return cpu->apic_id;
632 }
633
634 #ifndef KVM_CPUID_SIGNATURE_NEXT
635 #define KVM_CPUID_SIGNATURE_NEXT 0x40000100
636 #endif
637
638 static bool hyperv_hypercall_available(X86CPU *cpu)
639 {
640 return cpu->hyperv_vapic ||
641 (cpu->hyperv_spinlock_attempts != HYPERV_SPINLOCK_NEVER_RETRY);
642 }
643
644 static bool hyperv_enabled(X86CPU *cpu)
645 {
646 CPUState *cs = CPU(cpu);
647 return kvm_check_extension(cs->kvm_state, KVM_CAP_HYPERV) > 0 &&
648 (hyperv_hypercall_available(cpu) ||
649 cpu->hyperv_time ||
650 cpu->hyperv_relaxed_timing ||
651 cpu->hyperv_crash ||
652 cpu->hyperv_reset ||
653 cpu->hyperv_vpindex ||
654 cpu->hyperv_runtime ||
655 cpu->hyperv_synic ||
656 cpu->hyperv_stimer ||
657 cpu->hyperv_reenlightenment ||
658 cpu->hyperv_tlbflush ||
659 cpu->hyperv_ipi);
660 }
661
662 static int kvm_arch_set_tsc_khz(CPUState *cs)
663 {
664 X86CPU *cpu = X86_CPU(cs);
665 CPUX86State *env = &cpu->env;
666 int r;
667
668 if (!env->tsc_khz) {
669 return 0;
670 }
671
672 r = kvm_check_extension(cs->kvm_state, KVM_CAP_TSC_CONTROL) ?
673 kvm_vcpu_ioctl(cs, KVM_SET_TSC_KHZ, env->tsc_khz) :
674 -ENOTSUP;
675 if (r < 0) {
676 /* When KVM_SET_TSC_KHZ fails, it's an error only if the current
677 * TSC frequency doesn't match the one we want.
678 */
679 int cur_freq = kvm_check_extension(cs->kvm_state, KVM_CAP_GET_TSC_KHZ) ?
680 kvm_vcpu_ioctl(cs, KVM_GET_TSC_KHZ) :
681 -ENOTSUP;
682 if (cur_freq <= 0 || cur_freq != env->tsc_khz) {
683 warn_report("TSC frequency mismatch between "
684 "VM (%" PRId64 " kHz) and host (%d kHz), "
685 "and TSC scaling unavailable",
686 env->tsc_khz, cur_freq);
687 return r;
688 }
689 }
690
691 return 0;
692 }
693
694 static bool tsc_is_stable_and_known(CPUX86State *env)
695 {
696 if (!env->tsc_khz) {
697 return false;
698 }
699 return (env->features[FEAT_8000_0007_EDX] & CPUID_APM_INVTSC)
700 || env->user_tsc_khz;
701 }
702
703 static int hyperv_handle_properties(CPUState *cs)
704 {
705 X86CPU *cpu = X86_CPU(cs);
706 CPUX86State *env = &cpu->env;
707
708 if (cpu->hyperv_relaxed_timing) {
709 env->features[FEAT_HYPERV_EAX] |= HV_HYPERCALL_AVAILABLE;
710 }
711 if (cpu->hyperv_vapic) {
712 env->features[FEAT_HYPERV_EAX] |= HV_HYPERCALL_AVAILABLE;
713 env->features[FEAT_HYPERV_EAX] |= HV_APIC_ACCESS_AVAILABLE;
714 }
715 if (cpu->hyperv_time) {
716 if (kvm_check_extension(cs->kvm_state, KVM_CAP_HYPERV_TIME) <= 0) {
717 fprintf(stderr, "Hyper-V clocksources "
718 "(requested by 'hv-time' cpu flag) "
719 "are not supported by kernel\n");
720 return -ENOSYS;
721 }
722 env->features[FEAT_HYPERV_EAX] |= HV_HYPERCALL_AVAILABLE;
723 env->features[FEAT_HYPERV_EAX] |= HV_TIME_REF_COUNT_AVAILABLE;
724 env->features[FEAT_HYPERV_EAX] |= HV_REFERENCE_TSC_AVAILABLE;
725 }
726 if (cpu->hyperv_frequencies) {
727 if (!has_msr_hv_frequencies) {
728 fprintf(stderr, "Hyper-V frequency MSRs "
729 "(requested by 'hv-frequencies' cpu flag) "
730 "are not supported by kernel\n");
731 return -ENOSYS;
732 }
733 env->features[FEAT_HYPERV_EAX] |= HV_ACCESS_FREQUENCY_MSRS;
734 env->features[FEAT_HYPERV_EDX] |= HV_FREQUENCY_MSRS_AVAILABLE;
735 }
736 if (cpu->hyperv_crash) {
737 if (!has_msr_hv_crash) {
738 fprintf(stderr, "Hyper-V crash MSRs "
739 "(requested by 'hv-crash' cpu flag) "
740 "are not supported by kernel\n");
741 return -ENOSYS;
742 }
743 env->features[FEAT_HYPERV_EDX] |= HV_GUEST_CRASH_MSR_AVAILABLE;
744 }
745 if (cpu->hyperv_reenlightenment) {
746 if (!has_msr_hv_reenlightenment) {
747 fprintf(stderr,
748 "Hyper-V Reenlightenment MSRs "
749 "(requested by 'hv-reenlightenment' cpu flag) "
750 "are not supported by kernel\n");
751 return -ENOSYS;
752 }
753 env->features[FEAT_HYPERV_EAX] |= HV_ACCESS_REENLIGHTENMENTS_CONTROL;
754 }
755 env->features[FEAT_HYPERV_EDX] |= HV_CPU_DYNAMIC_PARTITIONING_AVAILABLE;
756 if (cpu->hyperv_reset) {
757 if (!has_msr_hv_reset) {
758 fprintf(stderr, "Hyper-V reset MSR "
759 "(requested by 'hv-reset' cpu flag) "
760 "is not supported by kernel\n");
761 return -ENOSYS;
762 }
763 env->features[FEAT_HYPERV_EAX] |= HV_RESET_AVAILABLE;
764 }
765 if (cpu->hyperv_vpindex) {
766 if (!has_msr_hv_vpindex) {
767 fprintf(stderr, "Hyper-V VP_INDEX MSR "
768 "(requested by 'hv-vpindex' cpu flag) "
769 "is not supported by kernel\n");
770 return -ENOSYS;
771 }
772 env->features[FEAT_HYPERV_EAX] |= HV_VP_INDEX_AVAILABLE;
773 }
774 if (cpu->hyperv_runtime) {
775 if (!has_msr_hv_runtime) {
776 fprintf(stderr, "Hyper-V VP_RUNTIME MSR "
777 "(requested by 'hv-runtime' cpu flag) "
778 "is not supported by kernel\n");
779 return -ENOSYS;
780 }
781 env->features[FEAT_HYPERV_EAX] |= HV_VP_RUNTIME_AVAILABLE;
782 }
783 if (cpu->hyperv_synic) {
784 unsigned int cap = KVM_CAP_HYPERV_SYNIC;
785 if (!cpu->hyperv_synic_kvm_only) {
786 if (!cpu->hyperv_vpindex) {
787 fprintf(stderr, "Hyper-V SynIC "
788 "(requested by 'hv-synic' cpu flag) "
789 "requires Hyper-V VP_INDEX ('hv-vpindex')\n");
790 return -ENOSYS;
791 }
792 cap = KVM_CAP_HYPERV_SYNIC2;
793 }
794
795 if (!has_msr_hv_synic || !kvm_check_extension(cs->kvm_state, cap)) {
796 fprintf(stderr, "Hyper-V SynIC (requested by 'hv-synic' cpu flag) "
797 "is not supported by kernel\n");
798 return -ENOSYS;
799 }
800
801 env->features[FEAT_HYPERV_EAX] |= HV_SYNIC_AVAILABLE;
802 }
803 if (cpu->hyperv_stimer) {
804 if (!has_msr_hv_stimer) {
805 fprintf(stderr, "Hyper-V timers aren't supported by kernel\n");
806 return -ENOSYS;
807 }
808 env->features[FEAT_HYPERV_EAX] |= HV_SYNTIMERS_AVAILABLE;
809 }
810 if (cpu->hyperv_relaxed_timing) {
811 env->features[FEAT_HV_RECOMM_EAX] |= HV_RELAXED_TIMING_RECOMMENDED;
812 }
813 if (cpu->hyperv_vapic) {
814 env->features[FEAT_HV_RECOMM_EAX] |= HV_APIC_ACCESS_RECOMMENDED;
815 }
816 if (cpu->hyperv_tlbflush) {
817 if (kvm_check_extension(cs->kvm_state,
818 KVM_CAP_HYPERV_TLBFLUSH) <= 0) {
819 fprintf(stderr, "Hyper-V TLB flush support "
820 "(requested by 'hv-tlbflush' cpu flag) "
821 " is not supported by kernel\n");
822 return -ENOSYS;
823 }
824 env->features[FEAT_HV_RECOMM_EAX] |= HV_REMOTE_TLB_FLUSH_RECOMMENDED;
825 env->features[FEAT_HV_RECOMM_EAX] |= HV_EX_PROCESSOR_MASKS_RECOMMENDED;
826 }
827 if (cpu->hyperv_ipi) {
828 if (kvm_check_extension(cs->kvm_state,
829 KVM_CAP_HYPERV_SEND_IPI) <= 0) {
830 fprintf(stderr, "Hyper-V IPI send support "
831 "(requested by 'hv-ipi' cpu flag) "
832 " is not supported by kernel\n");
833 return -ENOSYS;
834 }
835 env->features[FEAT_HV_RECOMM_EAX] |= HV_CLUSTER_IPI_RECOMMENDED;
836 env->features[FEAT_HV_RECOMM_EAX] |= HV_EX_PROCESSOR_MASKS_RECOMMENDED;
837 }
838 if (cpu->hyperv_evmcs) {
839 uint16_t evmcs_version;
840
841 if (kvm_vcpu_enable_cap(cs, KVM_CAP_HYPERV_ENLIGHTENED_VMCS, 0,
842 (uintptr_t)&evmcs_version)) {
843 fprintf(stderr, "Hyper-V Enlightened VMCS "
844 "(requested by 'hv-evmcs' cpu flag) "
845 "is not supported by kernel\n");
846 return -ENOSYS;
847 }
848 env->features[FEAT_HV_RECOMM_EAX] |= HV_ENLIGHTENED_VMCS_RECOMMENDED;
849 env->features[FEAT_HV_NESTED_EAX] = evmcs_version;
850 }
851
852 return 0;
853 }
854
855 static int hyperv_init_vcpu(X86CPU *cpu)
856 {
857 CPUState *cs = CPU(cpu);
858 int ret;
859
860 if (cpu->hyperv_vpindex && !hv_vpindex_settable) {
861 /*
862 * the kernel doesn't support setting vp_index; assert that its value
863 * is in sync
864 */
865 struct {
866 struct kvm_msrs info;
867 struct kvm_msr_entry entries[1];
868 } msr_data = {
869 .info.nmsrs = 1,
870 .entries[0].index = HV_X64_MSR_VP_INDEX,
871 };
872
873 ret = kvm_vcpu_ioctl(cs, KVM_GET_MSRS, &msr_data);
874 if (ret < 0) {
875 return ret;
876 }
877 assert(ret == 1);
878
879 if (msr_data.entries[0].data != hyperv_vp_index(CPU(cpu))) {
880 error_report("kernel's vp_index != QEMU's vp_index");
881 return -ENXIO;
882 }
883 }
884
885 if (cpu->hyperv_synic) {
886 uint32_t synic_cap = cpu->hyperv_synic_kvm_only ?
887 KVM_CAP_HYPERV_SYNIC : KVM_CAP_HYPERV_SYNIC2;
888 ret = kvm_vcpu_enable_cap(cs, synic_cap, 0);
889 if (ret < 0) {
890 error_report("failed to turn on HyperV SynIC in KVM: %s",
891 strerror(-ret));
892 return ret;
893 }
894
895 if (!cpu->hyperv_synic_kvm_only) {
896 ret = hyperv_x86_synic_add(cpu);
897 if (ret < 0) {
898 error_report("failed to create HyperV SynIC: %s",
899 strerror(-ret));
900 return ret;
901 }
902 }
903 }
904
905 return 0;
906 }
907
908 static Error *invtsc_mig_blocker;
909 static Error *vmx_mig_blocker;
910
911 #define KVM_MAX_CPUID_ENTRIES 100
912
913 int kvm_arch_init_vcpu(CPUState *cs)
914 {
915 struct {
916 struct kvm_cpuid2 cpuid;
917 struct kvm_cpuid_entry2 entries[KVM_MAX_CPUID_ENTRIES];
918 } cpuid_data;
919 /*
920 * The kernel defines these structs with padding fields so there
921 * should be no extra padding in our cpuid_data struct.
922 */
923 QEMU_BUILD_BUG_ON(sizeof(cpuid_data) !=
924 sizeof(struct kvm_cpuid2) +
925 sizeof(struct kvm_cpuid_entry2) * KVM_MAX_CPUID_ENTRIES);
926
927 X86CPU *cpu = X86_CPU(cs);
928 CPUX86State *env = &cpu->env;
929 uint32_t limit, i, j, cpuid_i;
930 uint32_t unused;
931 struct kvm_cpuid_entry2 *c;
932 uint32_t signature[3];
933 int kvm_base = KVM_CPUID_SIGNATURE;
934 int r;
935 Error *local_err = NULL;
936
937 memset(&cpuid_data, 0, sizeof(cpuid_data));
938
939 cpuid_i = 0;
940
941 r = kvm_arch_set_tsc_khz(cs);
942 if (r < 0) {
943 goto fail;
944 }
945
946 /* vcpu's TSC frequency is either specified by user, or following
947 * the value used by KVM if the former is not present. In the
948 * latter case, we query it from KVM and record in env->tsc_khz,
949 * so that vcpu's TSC frequency can be migrated later via this field.
950 */
951 if (!env->tsc_khz) {
952 r = kvm_check_extension(cs->kvm_state, KVM_CAP_GET_TSC_KHZ) ?
953 kvm_vcpu_ioctl(cs, KVM_GET_TSC_KHZ) :
954 -ENOTSUP;
955 if (r > 0) {
956 env->tsc_khz = r;
957 }
958 }
959
960 /* Paravirtualization CPUIDs */
961 if (hyperv_enabled(cpu)) {
962 c = &cpuid_data.entries[cpuid_i++];
963 c->function = HV_CPUID_VENDOR_AND_MAX_FUNCTIONS;
964 if (!cpu->hyperv_vendor_id) {
965 memcpy(signature, "Microsoft Hv", 12);
966 } else {
967 size_t len = strlen(cpu->hyperv_vendor_id);
968
969 if (len > 12) {
970 error_report("hv-vendor-id truncated to 12 characters");
971 len = 12;
972 }
973 memset(signature, 0, 12);
974 memcpy(signature, cpu->hyperv_vendor_id, len);
975 }
976 c->eax = cpu->hyperv_evmcs ?
977 HV_CPUID_NESTED_FEATURES : HV_CPUID_IMPLEMENT_LIMITS;
978 c->ebx = signature[0];
979 c->ecx = signature[1];
980 c->edx = signature[2];
981
982 c = &cpuid_data.entries[cpuid_i++];
983 c->function = HV_CPUID_INTERFACE;
984 memcpy(signature, "Hv#1\0\0\0\0\0\0\0\0", 12);
985 c->eax = signature[0];
986 c->ebx = 0;
987 c->ecx = 0;
988 c->edx = 0;
989
990 c = &cpuid_data.entries[cpuid_i++];
991 c->function = HV_CPUID_VERSION;
992 c->eax = 0x00001bbc;
993 c->ebx = 0x00060001;
994
995 c = &cpuid_data.entries[cpuid_i++];
996 c->function = HV_CPUID_FEATURES;
997 r = hyperv_handle_properties(cs);
998 if (r) {
999 return r;
1000 }
1001 c->eax = env->features[FEAT_HYPERV_EAX];
1002 c->ebx = env->features[FEAT_HYPERV_EBX];
1003 c->edx = env->features[FEAT_HYPERV_EDX];
1004
1005 c = &cpuid_data.entries[cpuid_i++];
1006 c->function = HV_CPUID_ENLIGHTMENT_INFO;
1007
1008 c->eax = env->features[FEAT_HV_RECOMM_EAX];
1009 c->ebx = cpu->hyperv_spinlock_attempts;
1010
1011 c = &cpuid_data.entries[cpuid_i++];
1012 c->function = HV_CPUID_IMPLEMENT_LIMITS;
1013
1014 c->eax = cpu->hv_max_vps;
1015 c->ebx = 0x40;
1016
1017 kvm_base = KVM_CPUID_SIGNATURE_NEXT;
1018 has_msr_hv_hypercall = true;
1019
1020 if (cpu->hyperv_evmcs) {
1021 __u32 function;
1022
1023 /* Create zeroed 0x40000006..0x40000009 leaves */
1024 for (function = HV_CPUID_IMPLEMENT_LIMITS + 1;
1025 function < HV_CPUID_NESTED_FEATURES; function++) {
1026 c = &cpuid_data.entries[cpuid_i++];
1027 c->function = function;
1028 }
1029
1030 c = &cpuid_data.entries[cpuid_i++];
1031 c->function = HV_CPUID_NESTED_FEATURES;
1032 c->eax = env->features[FEAT_HV_NESTED_EAX];
1033 }
1034 }
1035
1036 if (cpu->expose_kvm) {
1037 memcpy(signature, "KVMKVMKVM\0\0\0", 12);
1038 c = &cpuid_data.entries[cpuid_i++];
1039 c->function = KVM_CPUID_SIGNATURE | kvm_base;
1040 c->eax = KVM_CPUID_FEATURES | kvm_base;
1041 c->ebx = signature[0];
1042 c->ecx = signature[1];
1043 c->edx = signature[2];
1044
1045 c = &cpuid_data.entries[cpuid_i++];
1046 c->function = KVM_CPUID_FEATURES | kvm_base;
1047 c->eax = env->features[FEAT_KVM];
1048 c->edx = env->features[FEAT_KVM_HINTS];
1049 }
1050
1051 cpu_x86_cpuid(env, 0, 0, &limit, &unused, &unused, &unused);
1052
1053 for (i = 0; i <= limit; i++) {
1054 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1055 fprintf(stderr, "unsupported level value: 0x%x\n", limit);
1056 abort();
1057 }
1058 c = &cpuid_data.entries[cpuid_i++];
1059
1060 switch (i) {
1061 case 2: {
1062 /* Keep reading function 2 till all the input is received */
1063 int times;
1064
1065 c->function = i;
1066 c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC |
1067 KVM_CPUID_FLAG_STATE_READ_NEXT;
1068 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1069 times = c->eax & 0xff;
1070
1071 for (j = 1; j < times; ++j) {
1072 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1073 fprintf(stderr, "cpuid_data is full, no space for "
1074 "cpuid(eax:2):eax & 0xf = 0x%x\n", times);
1075 abort();
1076 }
1077 c = &cpuid_data.entries[cpuid_i++];
1078 c->function = i;
1079 c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC;
1080 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1081 }
1082 break;
1083 }
1084 case 4:
1085 case 0xb:
1086 case 0xd:
1087 for (j = 0; ; j++) {
1088 if (i == 0xd && j == 64) {
1089 break;
1090 }
1091 c->function = i;
1092 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1093 c->index = j;
1094 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
1095
1096 if (i == 4 && c->eax == 0) {
1097 break;
1098 }
1099 if (i == 0xb && !(c->ecx & 0xff00)) {
1100 break;
1101 }
1102 if (i == 0xd && c->eax == 0) {
1103 continue;
1104 }
1105 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1106 fprintf(stderr, "cpuid_data is full, no space for "
1107 "cpuid(eax:0x%x,ecx:0x%x)\n", i, j);
1108 abort();
1109 }
1110 c = &cpuid_data.entries[cpuid_i++];
1111 }
1112 break;
1113 case 0x14: {
1114 uint32_t times;
1115
1116 c->function = i;
1117 c->index = 0;
1118 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1119 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1120 times = c->eax;
1121
1122 for (j = 1; j <= times; ++j) {
1123 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1124 fprintf(stderr, "cpuid_data is full, no space for "
1125 "cpuid(eax:0x14,ecx:0x%x)\n", j);
1126 abort();
1127 }
1128 c = &cpuid_data.entries[cpuid_i++];
1129 c->function = i;
1130 c->index = j;
1131 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1132 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
1133 }
1134 break;
1135 }
1136 default:
1137 c->function = i;
1138 c->flags = 0;
1139 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1140 break;
1141 }
1142 }
1143
1144 if (limit >= 0x0a) {
1145 uint32_t eax, edx;
1146
1147 cpu_x86_cpuid(env, 0x0a, 0, &eax, &unused, &unused, &edx);
1148
1149 has_architectural_pmu_version = eax & 0xff;
1150 if (has_architectural_pmu_version > 0) {
1151 num_architectural_pmu_gp_counters = (eax & 0xff00) >> 8;
1152
1153 /* Shouldn't be more than 32, since that's the number of bits
1154 * available in EBX to tell us _which_ counters are available.
1155 * Play it safe.
1156 */
1157 if (num_architectural_pmu_gp_counters > MAX_GP_COUNTERS) {
1158 num_architectural_pmu_gp_counters = MAX_GP_COUNTERS;
1159 }
1160
1161 if (has_architectural_pmu_version > 1) {
1162 num_architectural_pmu_fixed_counters = edx & 0x1f;
1163
1164 if (num_architectural_pmu_fixed_counters > MAX_FIXED_COUNTERS) {
1165 num_architectural_pmu_fixed_counters = MAX_FIXED_COUNTERS;
1166 }
1167 }
1168 }
1169 }
1170
1171 cpu_x86_cpuid(env, 0x80000000, 0, &limit, &unused, &unused, &unused);
1172
1173 for (i = 0x80000000; i <= limit; i++) {
1174 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1175 fprintf(stderr, "unsupported xlevel value: 0x%x\n", limit);
1176 abort();
1177 }
1178 c = &cpuid_data.entries[cpuid_i++];
1179
1180 switch (i) {
1181 case 0x8000001d:
1182 /* Query for all AMD cache information leaves */
1183 for (j = 0; ; j++) {
1184 c->function = i;
1185 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1186 c->index = j;
1187 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
1188
1189 if (c->eax == 0) {
1190 break;
1191 }
1192 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1193 fprintf(stderr, "cpuid_data is full, no space for "
1194 "cpuid(eax:0x%x,ecx:0x%x)\n", i, j);
1195 abort();
1196 }
1197 c = &cpuid_data.entries[cpuid_i++];
1198 }
1199 break;
1200 default:
1201 c->function = i;
1202 c->flags = 0;
1203 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1204 break;
1205 }
1206 }
1207
1208 /* Call Centaur's CPUID instructions they are supported. */
1209 if (env->cpuid_xlevel2 > 0) {
1210 cpu_x86_cpuid(env, 0xC0000000, 0, &limit, &unused, &unused, &unused);
1211
1212 for (i = 0xC0000000; i <= limit; i++) {
1213 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1214 fprintf(stderr, "unsupported xlevel2 value: 0x%x\n", limit);
1215 abort();
1216 }
1217 c = &cpuid_data.entries[cpuid_i++];
1218
1219 c->function = i;
1220 c->flags = 0;
1221 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1222 }
1223 }
1224
1225 cpuid_data.cpuid.nent = cpuid_i;
1226
1227 if (((env->cpuid_version >> 8)&0xF) >= 6
1228 && (env->features[FEAT_1_EDX] & (CPUID_MCE | CPUID_MCA)) ==
1229 (CPUID_MCE | CPUID_MCA)
1230 && kvm_check_extension(cs->kvm_state, KVM_CAP_MCE) > 0) {
1231 uint64_t mcg_cap, unsupported_caps;
1232 int banks;
1233 int ret;
1234
1235 ret = kvm_get_mce_cap_supported(cs->kvm_state, &mcg_cap, &banks);
1236 if (ret < 0) {
1237 fprintf(stderr, "kvm_get_mce_cap_supported: %s", strerror(-ret));
1238 return ret;
1239 }
1240
1241 if (banks < (env->mcg_cap & MCG_CAP_BANKS_MASK)) {
1242 error_report("kvm: Unsupported MCE bank count (QEMU = %d, KVM = %d)",
1243 (int)(env->mcg_cap & MCG_CAP_BANKS_MASK), banks);
1244 return -ENOTSUP;
1245 }
1246
1247 unsupported_caps = env->mcg_cap & ~(mcg_cap | MCG_CAP_BANKS_MASK);
1248 if (unsupported_caps) {
1249 if (unsupported_caps & MCG_LMCE_P) {
1250 error_report("kvm: LMCE not supported");
1251 return -ENOTSUP;
1252 }
1253 warn_report("Unsupported MCG_CAP bits: 0x%" PRIx64,
1254 unsupported_caps);
1255 }
1256
1257 env->mcg_cap &= mcg_cap | MCG_CAP_BANKS_MASK;
1258 ret = kvm_vcpu_ioctl(cs, KVM_X86_SETUP_MCE, &env->mcg_cap);
1259 if (ret < 0) {
1260 fprintf(stderr, "KVM_X86_SETUP_MCE: %s", strerror(-ret));
1261 return ret;
1262 }
1263 }
1264
1265 qemu_add_vm_change_state_handler(cpu_update_state, env);
1266
1267 c = cpuid_find_entry(&cpuid_data.cpuid, 1, 0);
1268 if (c) {
1269 has_msr_feature_control = !!(c->ecx & CPUID_EXT_VMX) ||
1270 !!(c->ecx & CPUID_EXT_SMX);
1271 }
1272
1273 if ((env->features[FEAT_1_ECX] & CPUID_EXT_VMX) && !vmx_mig_blocker) {
1274 error_setg(&vmx_mig_blocker,
1275 "Nested VMX virtualization does not support live migration yet");
1276 r = migrate_add_blocker(vmx_mig_blocker, &local_err);
1277 if (local_err) {
1278 error_report_err(local_err);
1279 error_free(vmx_mig_blocker);
1280 return r;
1281 }
1282 }
1283
1284 if (env->mcg_cap & MCG_LMCE_P) {
1285 has_msr_mcg_ext_ctl = has_msr_feature_control = true;
1286 }
1287
1288 if (!env->user_tsc_khz) {
1289 if ((env->features[FEAT_8000_0007_EDX] & CPUID_APM_INVTSC) &&
1290 invtsc_mig_blocker == NULL) {
1291 error_setg(&invtsc_mig_blocker,
1292 "State blocked by non-migratable CPU device"
1293 " (invtsc flag)");
1294 r = migrate_add_blocker(invtsc_mig_blocker, &local_err);
1295 if (local_err) {
1296 error_report_err(local_err);
1297 error_free(invtsc_mig_blocker);
1298 return r;
1299 }
1300 }
1301 }
1302
1303 if (cpu->vmware_cpuid_freq
1304 /* Guests depend on 0x40000000 to detect this feature, so only expose
1305 * it if KVM exposes leaf 0x40000000. (Conflicts with Hyper-V) */
1306 && cpu->expose_kvm
1307 && kvm_base == KVM_CPUID_SIGNATURE
1308 /* TSC clock must be stable and known for this feature. */
1309 && tsc_is_stable_and_known(env)) {
1310
1311 c = &cpuid_data.entries[cpuid_i++];
1312 c->function = KVM_CPUID_SIGNATURE | 0x10;
1313 c->eax = env->tsc_khz;
1314 /* LAPIC resolution of 1ns (freq: 1GHz) is hardcoded in KVM's
1315 * APIC_BUS_CYCLE_NS */
1316 c->ebx = 1000000;
1317 c->ecx = c->edx = 0;
1318
1319 c = cpuid_find_entry(&cpuid_data.cpuid, kvm_base, 0);
1320 c->eax = MAX(c->eax, KVM_CPUID_SIGNATURE | 0x10);
1321 }
1322
1323 cpuid_data.cpuid.nent = cpuid_i;
1324
1325 cpuid_data.cpuid.padding = 0;
1326 r = kvm_vcpu_ioctl(cs, KVM_SET_CPUID2, &cpuid_data);
1327 if (r) {
1328 goto fail;
1329 }
1330
1331 if (has_xsave) {
1332 env->xsave_buf = qemu_memalign(4096, sizeof(struct kvm_xsave));
1333 }
1334 cpu->kvm_msr_buf = g_malloc0(MSR_BUF_SIZE);
1335
1336 if (!(env->features[FEAT_8000_0001_EDX] & CPUID_EXT2_RDTSCP)) {
1337 has_msr_tsc_aux = false;
1338 }
1339
1340 r = hyperv_init_vcpu(cpu);
1341 if (r) {
1342 goto fail;
1343 }
1344
1345 return 0;
1346
1347 fail:
1348 migrate_del_blocker(invtsc_mig_blocker);
1349 return r;
1350 }
1351
1352 void kvm_arch_reset_vcpu(X86CPU *cpu)
1353 {
1354 CPUX86State *env = &cpu->env;
1355
1356 env->xcr0 = 1;
1357 if (kvm_irqchip_in_kernel()) {
1358 env->mp_state = cpu_is_bsp(cpu) ? KVM_MP_STATE_RUNNABLE :
1359 KVM_MP_STATE_UNINITIALIZED;
1360 } else {
1361 env->mp_state = KVM_MP_STATE_RUNNABLE;
1362 }
1363
1364 if (cpu->hyperv_synic) {
1365 int i;
1366 for (i = 0; i < ARRAY_SIZE(env->msr_hv_synic_sint); i++) {
1367 env->msr_hv_synic_sint[i] = HV_SINT_MASKED;
1368 }
1369
1370 hyperv_x86_synic_reset(cpu);
1371 }
1372 }
1373
1374 void kvm_arch_do_init_vcpu(X86CPU *cpu)
1375 {
1376 CPUX86State *env = &cpu->env;
1377
1378 /* APs get directly into wait-for-SIPI state. */
1379 if (env->mp_state == KVM_MP_STATE_UNINITIALIZED) {
1380 env->mp_state = KVM_MP_STATE_INIT_RECEIVED;
1381 }
1382 }
1383
1384 static int kvm_get_supported_feature_msrs(KVMState *s)
1385 {
1386 int ret = 0;
1387
1388 if (kvm_feature_msrs != NULL) {
1389 return 0;
1390 }
1391
1392 if (!kvm_check_extension(s, KVM_CAP_GET_MSR_FEATURES)) {
1393 return 0;
1394 }
1395
1396 struct kvm_msr_list msr_list;
1397
1398 msr_list.nmsrs = 0;
1399 ret = kvm_ioctl(s, KVM_GET_MSR_FEATURE_INDEX_LIST, &msr_list);
1400 if (ret < 0 && ret != -E2BIG) {
1401 error_report("Fetch KVM feature MSR list failed: %s",
1402 strerror(-ret));
1403 return ret;
1404 }
1405
1406 assert(msr_list.nmsrs > 0);
1407 kvm_feature_msrs = (struct kvm_msr_list *) \
1408 g_malloc0(sizeof(msr_list) +
1409 msr_list.nmsrs * sizeof(msr_list.indices[0]));
1410
1411 kvm_feature_msrs->nmsrs = msr_list.nmsrs;
1412 ret = kvm_ioctl(s, KVM_GET_MSR_FEATURE_INDEX_LIST, kvm_feature_msrs);
1413
1414 if (ret < 0) {
1415 error_report("Fetch KVM feature MSR list failed: %s",
1416 strerror(-ret));
1417 g_free(kvm_feature_msrs);
1418 kvm_feature_msrs = NULL;
1419 return ret;
1420 }
1421
1422 return 0;
1423 }
1424
1425 static int kvm_get_supported_msrs(KVMState *s)
1426 {
1427 static int kvm_supported_msrs;
1428 int ret = 0;
1429
1430 /* first time */
1431 if (kvm_supported_msrs == 0) {
1432 struct kvm_msr_list msr_list, *kvm_msr_list;
1433
1434 kvm_supported_msrs = -1;
1435
1436 /* Obtain MSR list from KVM. These are the MSRs that we must
1437 * save/restore */
1438 msr_list.nmsrs = 0;
1439 ret = kvm_ioctl(s, KVM_GET_MSR_INDEX_LIST, &msr_list);
1440 if (ret < 0 && ret != -E2BIG) {
1441 return ret;
1442 }
1443 /* Old kernel modules had a bug and could write beyond the provided
1444 memory. Allocate at least a safe amount of 1K. */
1445 kvm_msr_list = g_malloc0(MAX(1024, sizeof(msr_list) +
1446 msr_list.nmsrs *
1447 sizeof(msr_list.indices[0])));
1448
1449 kvm_msr_list->nmsrs = msr_list.nmsrs;
1450 ret = kvm_ioctl(s, KVM_GET_MSR_INDEX_LIST, kvm_msr_list);
1451 if (ret >= 0) {
1452 int i;
1453
1454 for (i = 0; i < kvm_msr_list->nmsrs; i++) {
1455 switch (kvm_msr_list->indices[i]) {
1456 case MSR_STAR:
1457 has_msr_star = true;
1458 break;
1459 case MSR_VM_HSAVE_PA:
1460 has_msr_hsave_pa = true;
1461 break;
1462 case MSR_TSC_AUX:
1463 has_msr_tsc_aux = true;
1464 break;
1465 case MSR_TSC_ADJUST:
1466 has_msr_tsc_adjust = true;
1467 break;
1468 case MSR_IA32_TSCDEADLINE:
1469 has_msr_tsc_deadline = true;
1470 break;
1471 case MSR_IA32_SMBASE:
1472 has_msr_smbase = true;
1473 break;
1474 case MSR_SMI_COUNT:
1475 has_msr_smi_count = true;
1476 break;
1477 case MSR_IA32_MISC_ENABLE:
1478 has_msr_misc_enable = true;
1479 break;
1480 case MSR_IA32_BNDCFGS:
1481 has_msr_bndcfgs = true;
1482 break;
1483 case MSR_IA32_XSS:
1484 has_msr_xss = true;
1485 break;
1486 case HV_X64_MSR_CRASH_CTL:
1487 has_msr_hv_crash = true;
1488 break;
1489 case HV_X64_MSR_RESET:
1490 has_msr_hv_reset = true;
1491 break;
1492 case HV_X64_MSR_VP_INDEX:
1493 has_msr_hv_vpindex = true;
1494 break;
1495 case HV_X64_MSR_VP_RUNTIME:
1496 has_msr_hv_runtime = true;
1497 break;
1498 case HV_X64_MSR_SCONTROL:
1499 has_msr_hv_synic = true;
1500 break;
1501 case HV_X64_MSR_STIMER0_CONFIG:
1502 has_msr_hv_stimer = true;
1503 break;
1504 case HV_X64_MSR_TSC_FREQUENCY:
1505 has_msr_hv_frequencies = true;
1506 break;
1507 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1508 has_msr_hv_reenlightenment = true;
1509 break;
1510 case MSR_IA32_SPEC_CTRL:
1511 has_msr_spec_ctrl = true;
1512 break;
1513 case MSR_VIRT_SSBD:
1514 has_msr_virt_ssbd = true;
1515 break;
1516 case MSR_IA32_ARCH_CAPABILITIES:
1517 has_msr_arch_capabs = true;
1518 break;
1519 }
1520 }
1521 }
1522
1523 g_free(kvm_msr_list);
1524 }
1525
1526 return ret;
1527 }
1528
1529 static Notifier smram_machine_done;
1530 static KVMMemoryListener smram_listener;
1531 static AddressSpace smram_address_space;
1532 static MemoryRegion smram_as_root;
1533 static MemoryRegion smram_as_mem;
1534
1535 static void register_smram_listener(Notifier *n, void *unused)
1536 {
1537 MemoryRegion *smram =
1538 (MemoryRegion *) object_resolve_path("/machine/smram", NULL);
1539
1540 /* Outer container... */
1541 memory_region_init(&smram_as_root, OBJECT(kvm_state), "mem-container-smram", ~0ull);
1542 memory_region_set_enabled(&smram_as_root, true);
1543
1544 /* ... with two regions inside: normal system memory with low
1545 * priority, and...
1546 */
1547 memory_region_init_alias(&smram_as_mem, OBJECT(kvm_state), "mem-smram",
1548 get_system_memory(), 0, ~0ull);
1549 memory_region_add_subregion_overlap(&smram_as_root, 0, &smram_as_mem, 0);
1550 memory_region_set_enabled(&smram_as_mem, true);
1551
1552 if (smram) {
1553 /* ... SMRAM with higher priority */
1554 memory_region_add_subregion_overlap(&smram_as_root, 0, smram, 10);
1555 memory_region_set_enabled(smram, true);
1556 }
1557
1558 address_space_init(&smram_address_space, &smram_as_root, "KVM-SMRAM");
1559 kvm_memory_listener_register(kvm_state, &smram_listener,
1560 &smram_address_space, 1);
1561 }
1562
1563 int kvm_arch_init(MachineState *ms, KVMState *s)
1564 {
1565 uint64_t identity_base = 0xfffbc000;
1566 uint64_t shadow_mem;
1567 int ret;
1568 struct utsname utsname;
1569
1570 has_xsave = kvm_check_extension(s, KVM_CAP_XSAVE);
1571 has_xcrs = kvm_check_extension(s, KVM_CAP_XCRS);
1572 has_pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2);
1573
1574 hv_vpindex_settable = kvm_check_extension(s, KVM_CAP_HYPERV_VP_INDEX);
1575
1576 ret = kvm_get_supported_msrs(s);
1577 if (ret < 0) {
1578 return ret;
1579 }
1580
1581 kvm_get_supported_feature_msrs(s);
1582
1583 uname(&utsname);
1584 lm_capable_kernel = strcmp(utsname.machine, "x86_64") == 0;
1585
1586 /*
1587 * On older Intel CPUs, KVM uses vm86 mode to emulate 16-bit code directly.
1588 * In order to use vm86 mode, an EPT identity map and a TSS are needed.
1589 * Since these must be part of guest physical memory, we need to allocate
1590 * them, both by setting their start addresses in the kernel and by
1591 * creating a corresponding e820 entry. We need 4 pages before the BIOS.
1592 *
1593 * Older KVM versions may not support setting the identity map base. In
1594 * that case we need to stick with the default, i.e. a 256K maximum BIOS
1595 * size.
1596 */
1597 if (kvm_check_extension(s, KVM_CAP_SET_IDENTITY_MAP_ADDR)) {
1598 /* Allows up to 16M BIOSes. */
1599 identity_base = 0xfeffc000;
1600
1601 ret = kvm_vm_ioctl(s, KVM_SET_IDENTITY_MAP_ADDR, &identity_base);
1602 if (ret < 0) {
1603 return ret;
1604 }
1605 }
1606
1607 /* Set TSS base one page after EPT identity map. */
1608 ret = kvm_vm_ioctl(s, KVM_SET_TSS_ADDR, identity_base + 0x1000);
1609 if (ret < 0) {
1610 return ret;
1611 }
1612
1613 /* Tell fw_cfg to notify the BIOS to reserve the range. */
1614 ret = e820_add_entry(identity_base, 0x4000, E820_RESERVED);
1615 if (ret < 0) {
1616 fprintf(stderr, "e820_add_entry() table is full\n");
1617 return ret;
1618 }
1619 qemu_register_reset(kvm_unpoison_all, NULL);
1620
1621 shadow_mem = machine_kvm_shadow_mem(ms);
1622 if (shadow_mem != -1) {
1623 shadow_mem /= 4096;
1624 ret = kvm_vm_ioctl(s, KVM_SET_NR_MMU_PAGES, shadow_mem);
1625 if (ret < 0) {
1626 return ret;
1627 }
1628 }
1629
1630 if (kvm_check_extension(s, KVM_CAP_X86_SMM) &&
1631 object_dynamic_cast(OBJECT(ms), TYPE_PC_MACHINE) &&
1632 pc_machine_is_smm_enabled(PC_MACHINE(ms))) {
1633 smram_machine_done.notify = register_smram_listener;
1634 qemu_add_machine_init_done_notifier(&smram_machine_done);
1635 }
1636
1637 if (enable_cpu_pm) {
1638 int disable_exits = kvm_check_extension(s, KVM_CAP_X86_DISABLE_EXITS);
1639 int ret;
1640
1641 /* Work around for kernel header with a typo. TODO: fix header and drop. */
1642 #if defined(KVM_X86_DISABLE_EXITS_HTL) && !defined(KVM_X86_DISABLE_EXITS_HLT)
1643 #define KVM_X86_DISABLE_EXITS_HLT KVM_X86_DISABLE_EXITS_HTL
1644 #endif
1645 if (disable_exits) {
1646 disable_exits &= (KVM_X86_DISABLE_EXITS_MWAIT |
1647 KVM_X86_DISABLE_EXITS_HLT |
1648 KVM_X86_DISABLE_EXITS_PAUSE);
1649 }
1650
1651 ret = kvm_vm_enable_cap(s, KVM_CAP_X86_DISABLE_EXITS, 0,
1652 disable_exits);
1653 if (ret < 0) {
1654 error_report("kvm: guest stopping CPU not supported: %s",
1655 strerror(-ret));
1656 }
1657 }
1658
1659 return 0;
1660 }
1661
1662 static void set_v8086_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
1663 {
1664 lhs->selector = rhs->selector;
1665 lhs->base = rhs->base;
1666 lhs->limit = rhs->limit;
1667 lhs->type = 3;
1668 lhs->present = 1;
1669 lhs->dpl = 3;
1670 lhs->db = 0;
1671 lhs->s = 1;
1672 lhs->l = 0;
1673 lhs->g = 0;
1674 lhs->avl = 0;
1675 lhs->unusable = 0;
1676 }
1677
1678 static void set_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
1679 {
1680 unsigned flags = rhs->flags;
1681 lhs->selector = rhs->selector;
1682 lhs->base = rhs->base;
1683 lhs->limit = rhs->limit;
1684 lhs->type = (flags >> DESC_TYPE_SHIFT) & 15;
1685 lhs->present = (flags & DESC_P_MASK) != 0;
1686 lhs->dpl = (flags >> DESC_DPL_SHIFT) & 3;
1687 lhs->db = (flags >> DESC_B_SHIFT) & 1;
1688 lhs->s = (flags & DESC_S_MASK) != 0;
1689 lhs->l = (flags >> DESC_L_SHIFT) & 1;
1690 lhs->g = (flags & DESC_G_MASK) != 0;
1691 lhs->avl = (flags & DESC_AVL_MASK) != 0;
1692 lhs->unusable = !lhs->present;
1693 lhs->padding = 0;
1694 }
1695
1696 static void get_seg(SegmentCache *lhs, const struct kvm_segment *rhs)
1697 {
1698 lhs->selector = rhs->selector;
1699 lhs->base = rhs->base;
1700 lhs->limit = rhs->limit;
1701 lhs->flags = (rhs->type << DESC_TYPE_SHIFT) |
1702 ((rhs->present && !rhs->unusable) * DESC_P_MASK) |
1703 (rhs->dpl << DESC_DPL_SHIFT) |
1704 (rhs->db << DESC_B_SHIFT) |
1705 (rhs->s * DESC_S_MASK) |
1706 (rhs->l << DESC_L_SHIFT) |
1707 (rhs->g * DESC_G_MASK) |
1708 (rhs->avl * DESC_AVL_MASK);
1709 }
1710
1711 static void kvm_getput_reg(__u64 *kvm_reg, target_ulong *qemu_reg, int set)
1712 {
1713 if (set) {
1714 *kvm_reg = *qemu_reg;
1715 } else {
1716 *qemu_reg = *kvm_reg;
1717 }
1718 }
1719
1720 static int kvm_getput_regs(X86CPU *cpu, int set)
1721 {
1722 CPUX86State *env = &cpu->env;
1723 struct kvm_regs regs;
1724 int ret = 0;
1725
1726 if (!set) {
1727 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_REGS, &regs);
1728 if (ret < 0) {
1729 return ret;
1730 }
1731 }
1732
1733 kvm_getput_reg(&regs.rax, &env->regs[R_EAX], set);
1734 kvm_getput_reg(&regs.rbx, &env->regs[R_EBX], set);
1735 kvm_getput_reg(&regs.rcx, &env->regs[R_ECX], set);
1736 kvm_getput_reg(&regs.rdx, &env->regs[R_EDX], set);
1737 kvm_getput_reg(&regs.rsi, &env->regs[R_ESI], set);
1738 kvm_getput_reg(&regs.rdi, &env->regs[R_EDI], set);
1739 kvm_getput_reg(&regs.rsp, &env->regs[R_ESP], set);
1740 kvm_getput_reg(&regs.rbp, &env->regs[R_EBP], set);
1741 #ifdef TARGET_X86_64
1742 kvm_getput_reg(&regs.r8, &env->regs[8], set);
1743 kvm_getput_reg(&regs.r9, &env->regs[9], set);
1744 kvm_getput_reg(&regs.r10, &env->regs[10], set);
1745 kvm_getput_reg(&regs.r11, &env->regs[11], set);
1746 kvm_getput_reg(&regs.r12, &env->regs[12], set);
1747 kvm_getput_reg(&regs.r13, &env->regs[13], set);
1748 kvm_getput_reg(&regs.r14, &env->regs[14], set);
1749 kvm_getput_reg(&regs.r15, &env->regs[15], set);
1750 #endif
1751
1752 kvm_getput_reg(&regs.rflags, &env->eflags, set);
1753 kvm_getput_reg(&regs.rip, &env->eip, set);
1754
1755 if (set) {
1756 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_REGS, &regs);
1757 }
1758
1759 return ret;
1760 }
1761
1762 static int kvm_put_fpu(X86CPU *cpu)
1763 {
1764 CPUX86State *env = &cpu->env;
1765 struct kvm_fpu fpu;
1766 int i;
1767
1768 memset(&fpu, 0, sizeof fpu);
1769 fpu.fsw = env->fpus & ~(7 << 11);
1770 fpu.fsw |= (env->fpstt & 7) << 11;
1771 fpu.fcw = env->fpuc;
1772 fpu.last_opcode = env->fpop;
1773 fpu.last_ip = env->fpip;
1774 fpu.last_dp = env->fpdp;
1775 for (i = 0; i < 8; ++i) {
1776 fpu.ftwx |= (!env->fptags[i]) << i;
1777 }
1778 memcpy(fpu.fpr, env->fpregs, sizeof env->fpregs);
1779 for (i = 0; i < CPU_NB_REGS; i++) {
1780 stq_p(&fpu.xmm[i][0], env->xmm_regs[i].ZMM_Q(0));
1781 stq_p(&fpu.xmm[i][8], env->xmm_regs[i].ZMM_Q(1));
1782 }
1783 fpu.mxcsr = env->mxcsr;
1784
1785 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_FPU, &fpu);
1786 }
1787
1788 #define XSAVE_FCW_FSW 0
1789 #define XSAVE_FTW_FOP 1
1790 #define XSAVE_CWD_RIP 2
1791 #define XSAVE_CWD_RDP 4
1792 #define XSAVE_MXCSR 6
1793 #define XSAVE_ST_SPACE 8
1794 #define XSAVE_XMM_SPACE 40
1795 #define XSAVE_XSTATE_BV 128
1796 #define XSAVE_YMMH_SPACE 144
1797 #define XSAVE_BNDREGS 240
1798 #define XSAVE_BNDCSR 256
1799 #define XSAVE_OPMASK 272
1800 #define XSAVE_ZMM_Hi256 288
1801 #define XSAVE_Hi16_ZMM 416
1802 #define XSAVE_PKRU 672
1803
1804 #define XSAVE_BYTE_OFFSET(word_offset) \
1805 ((word_offset) * sizeof_field(struct kvm_xsave, region[0]))
1806
1807 #define ASSERT_OFFSET(word_offset, field) \
1808 QEMU_BUILD_BUG_ON(XSAVE_BYTE_OFFSET(word_offset) != \
1809 offsetof(X86XSaveArea, field))
1810
1811 ASSERT_OFFSET(XSAVE_FCW_FSW, legacy.fcw);
1812 ASSERT_OFFSET(XSAVE_FTW_FOP, legacy.ftw);
1813 ASSERT_OFFSET(XSAVE_CWD_RIP, legacy.fpip);
1814 ASSERT_OFFSET(XSAVE_CWD_RDP, legacy.fpdp);
1815 ASSERT_OFFSET(XSAVE_MXCSR, legacy.mxcsr);
1816 ASSERT_OFFSET(XSAVE_ST_SPACE, legacy.fpregs);
1817 ASSERT_OFFSET(XSAVE_XMM_SPACE, legacy.xmm_regs);
1818 ASSERT_OFFSET(XSAVE_XSTATE_BV, header.xstate_bv);
1819 ASSERT_OFFSET(XSAVE_YMMH_SPACE, avx_state);
1820 ASSERT_OFFSET(XSAVE_BNDREGS, bndreg_state);
1821 ASSERT_OFFSET(XSAVE_BNDCSR, bndcsr_state);
1822 ASSERT_OFFSET(XSAVE_OPMASK, opmask_state);
1823 ASSERT_OFFSET(XSAVE_ZMM_Hi256, zmm_hi256_state);
1824 ASSERT_OFFSET(XSAVE_Hi16_ZMM, hi16_zmm_state);
1825 ASSERT_OFFSET(XSAVE_PKRU, pkru_state);
1826
1827 static int kvm_put_xsave(X86CPU *cpu)
1828 {
1829 CPUX86State *env = &cpu->env;
1830 X86XSaveArea *xsave = env->xsave_buf;
1831
1832 if (!has_xsave) {
1833 return kvm_put_fpu(cpu);
1834 }
1835 x86_cpu_xsave_all_areas(cpu, xsave);
1836
1837 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_XSAVE, xsave);
1838 }
1839
1840 static int kvm_put_xcrs(X86CPU *cpu)
1841 {
1842 CPUX86State *env = &cpu->env;
1843 struct kvm_xcrs xcrs = {};
1844
1845 if (!has_xcrs) {
1846 return 0;
1847 }
1848
1849 xcrs.nr_xcrs = 1;
1850 xcrs.flags = 0;
1851 xcrs.xcrs[0].xcr = 0;
1852 xcrs.xcrs[0].value = env->xcr0;
1853 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_XCRS, &xcrs);
1854 }
1855
1856 static int kvm_put_sregs(X86CPU *cpu)
1857 {
1858 CPUX86State *env = &cpu->env;
1859 struct kvm_sregs sregs;
1860
1861 memset(sregs.interrupt_bitmap, 0, sizeof(sregs.interrupt_bitmap));
1862 if (env->interrupt_injected >= 0) {
1863 sregs.interrupt_bitmap[env->interrupt_injected / 64] |=
1864 (uint64_t)1 << (env->interrupt_injected % 64);
1865 }
1866
1867 if ((env->eflags & VM_MASK)) {
1868 set_v8086_seg(&sregs.cs, &env->segs[R_CS]);
1869 set_v8086_seg(&sregs.ds, &env->segs[R_DS]);
1870 set_v8086_seg(&sregs.es, &env->segs[R_ES]);
1871 set_v8086_seg(&sregs.fs, &env->segs[R_FS]);
1872 set_v8086_seg(&sregs.gs, &env->segs[R_GS]);
1873 set_v8086_seg(&sregs.ss, &env->segs[R_SS]);
1874 } else {
1875 set_seg(&sregs.cs, &env->segs[R_CS]);
1876 set_seg(&sregs.ds, &env->segs[R_DS]);
1877 set_seg(&sregs.es, &env->segs[R_ES]);
1878 set_seg(&sregs.fs, &env->segs[R_FS]);
1879 set_seg(&sregs.gs, &env->segs[R_GS]);
1880 set_seg(&sregs.ss, &env->segs[R_SS]);
1881 }
1882
1883 set_seg(&sregs.tr, &env->tr);
1884 set_seg(&sregs.ldt, &env->ldt);
1885
1886 sregs.idt.limit = env->idt.limit;
1887 sregs.idt.base = env->idt.base;
1888 memset(sregs.idt.padding, 0, sizeof sregs.idt.padding);
1889 sregs.gdt.limit = env->gdt.limit;
1890 sregs.gdt.base = env->gdt.base;
1891 memset(sregs.gdt.padding, 0, sizeof sregs.gdt.padding);
1892
1893 sregs.cr0 = env->cr[0];
1894 sregs.cr2 = env->cr[2];
1895 sregs.cr3 = env->cr[3];
1896 sregs.cr4 = env->cr[4];
1897
1898 sregs.cr8 = cpu_get_apic_tpr(cpu->apic_state);
1899 sregs.apic_base = cpu_get_apic_base(cpu->apic_state);
1900
1901 sregs.efer = env->efer;
1902
1903 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_SREGS, &sregs);
1904 }
1905
1906 static void kvm_msr_buf_reset(X86CPU *cpu)
1907 {
1908 memset(cpu->kvm_msr_buf, 0, MSR_BUF_SIZE);
1909 }
1910
1911 static void kvm_msr_entry_add(X86CPU *cpu, uint32_t index, uint64_t value)
1912 {
1913 struct kvm_msrs *msrs = cpu->kvm_msr_buf;
1914 void *limit = ((void *)msrs) + MSR_BUF_SIZE;
1915 struct kvm_msr_entry *entry = &msrs->entries[msrs->nmsrs];
1916
1917 assert((void *)(entry + 1) <= limit);
1918
1919 entry->index = index;
1920 entry->reserved = 0;
1921 entry->data = value;
1922 msrs->nmsrs++;
1923 }
1924
1925 static int kvm_put_one_msr(X86CPU *cpu, int index, uint64_t value)
1926 {
1927 kvm_msr_buf_reset(cpu);
1928 kvm_msr_entry_add(cpu, index, value);
1929
1930 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MSRS, cpu->kvm_msr_buf);
1931 }
1932
1933 void kvm_put_apicbase(X86CPU *cpu, uint64_t value)
1934 {
1935 int ret;
1936
1937 ret = kvm_put_one_msr(cpu, MSR_IA32_APICBASE, value);
1938 assert(ret == 1);
1939 }
1940
1941 static int kvm_put_tscdeadline_msr(X86CPU *cpu)
1942 {
1943 CPUX86State *env = &cpu->env;
1944 int ret;
1945
1946 if (!has_msr_tsc_deadline) {
1947 return 0;
1948 }
1949
1950 ret = kvm_put_one_msr(cpu, MSR_IA32_TSCDEADLINE, env->tsc_deadline);
1951 if (ret < 0) {
1952 return ret;
1953 }
1954
1955 assert(ret == 1);
1956 return 0;
1957 }
1958
1959 /*
1960 * Provide a separate write service for the feature control MSR in order to
1961 * kick the VCPU out of VMXON or even guest mode on reset. This has to be done
1962 * before writing any other state because forcibly leaving nested mode
1963 * invalidates the VCPU state.
1964 */
1965 static int kvm_put_msr_feature_control(X86CPU *cpu)
1966 {
1967 int ret;
1968
1969 if (!has_msr_feature_control) {
1970 return 0;
1971 }
1972
1973 ret = kvm_put_one_msr(cpu, MSR_IA32_FEATURE_CONTROL,
1974 cpu->env.msr_ia32_feature_control);
1975 if (ret < 0) {
1976 return ret;
1977 }
1978
1979 assert(ret == 1);
1980 return 0;
1981 }
1982
1983 static int kvm_put_msrs(X86CPU *cpu, int level)
1984 {
1985 CPUX86State *env = &cpu->env;
1986 int i;
1987 int ret;
1988
1989 kvm_msr_buf_reset(cpu);
1990
1991 kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_CS, env->sysenter_cs);
1992 kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_ESP, env->sysenter_esp);
1993 kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_EIP, env->sysenter_eip);
1994 kvm_msr_entry_add(cpu, MSR_PAT, env->pat);
1995 if (has_msr_star) {
1996 kvm_msr_entry_add(cpu, MSR_STAR, env->star);
1997 }
1998 if (has_msr_hsave_pa) {
1999 kvm_msr_entry_add(cpu, MSR_VM_HSAVE_PA, env->vm_hsave);
2000 }
2001 if (has_msr_tsc_aux) {
2002 kvm_msr_entry_add(cpu, MSR_TSC_AUX, env->tsc_aux);
2003 }
2004 if (has_msr_tsc_adjust) {
2005 kvm_msr_entry_add(cpu, MSR_TSC_ADJUST, env->tsc_adjust);
2006 }
2007 if (has_msr_misc_enable) {
2008 kvm_msr_entry_add(cpu, MSR_IA32_MISC_ENABLE,
2009 env->msr_ia32_misc_enable);
2010 }
2011 if (has_msr_smbase) {
2012 kvm_msr_entry_add(cpu, MSR_IA32_SMBASE, env->smbase);
2013 }
2014 if (has_msr_smi_count) {
2015 kvm_msr_entry_add(cpu, MSR_SMI_COUNT, env->msr_smi_count);
2016 }
2017 if (has_msr_bndcfgs) {
2018 kvm_msr_entry_add(cpu, MSR_IA32_BNDCFGS, env->msr_bndcfgs);
2019 }
2020 if (has_msr_xss) {
2021 kvm_msr_entry_add(cpu, MSR_IA32_XSS, env->xss);
2022 }
2023 if (has_msr_spec_ctrl) {
2024 kvm_msr_entry_add(cpu, MSR_IA32_SPEC_CTRL, env->spec_ctrl);
2025 }
2026 if (has_msr_virt_ssbd) {
2027 kvm_msr_entry_add(cpu, MSR_VIRT_SSBD, env->virt_ssbd);
2028 }
2029
2030 #ifdef TARGET_X86_64
2031 if (lm_capable_kernel) {
2032 kvm_msr_entry_add(cpu, MSR_CSTAR, env->cstar);
2033 kvm_msr_entry_add(cpu, MSR_KERNELGSBASE, env->kernelgsbase);
2034 kvm_msr_entry_add(cpu, MSR_FMASK, env->fmask);
2035 kvm_msr_entry_add(cpu, MSR_LSTAR, env->lstar);
2036 }
2037 #endif
2038
2039 /* If host supports feature MSR, write down. */
2040 if (has_msr_arch_capabs) {
2041 kvm_msr_entry_add(cpu, MSR_IA32_ARCH_CAPABILITIES,
2042 env->features[FEAT_ARCH_CAPABILITIES]);
2043 }
2044
2045 /*
2046 * The following MSRs have side effects on the guest or are too heavy
2047 * for normal writeback. Limit them to reset or full state updates.
2048 */
2049 if (level >= KVM_PUT_RESET_STATE) {
2050 kvm_msr_entry_add(cpu, MSR_IA32_TSC, env->tsc);
2051 kvm_msr_entry_add(cpu, MSR_KVM_SYSTEM_TIME, env->system_time_msr);
2052 kvm_msr_entry_add(cpu, MSR_KVM_WALL_CLOCK, env->wall_clock_msr);
2053 if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_ASYNC_PF)) {
2054 kvm_msr_entry_add(cpu, MSR_KVM_ASYNC_PF_EN, env->async_pf_en_msr);
2055 }
2056 if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_PV_EOI)) {
2057 kvm_msr_entry_add(cpu, MSR_KVM_PV_EOI_EN, env->pv_eoi_en_msr);
2058 }
2059 if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_STEAL_TIME)) {
2060 kvm_msr_entry_add(cpu, MSR_KVM_STEAL_TIME, env->steal_time_msr);
2061 }
2062 if (has_architectural_pmu_version > 0) {
2063 if (has_architectural_pmu_version > 1) {
2064 /* Stop the counter. */
2065 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR_CTRL, 0);
2066 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_CTRL, 0);
2067 }
2068
2069 /* Set the counter values. */
2070 for (i = 0; i < num_architectural_pmu_fixed_counters; i++) {
2071 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR0 + i,
2072 env->msr_fixed_counters[i]);
2073 }
2074 for (i = 0; i < num_architectural_pmu_gp_counters; i++) {
2075 kvm_msr_entry_add(cpu, MSR_P6_PERFCTR0 + i,
2076 env->msr_gp_counters[i]);
2077 kvm_msr_entry_add(cpu, MSR_P6_EVNTSEL0 + i,
2078 env->msr_gp_evtsel[i]);
2079 }
2080 if (has_architectural_pmu_version > 1) {
2081 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_STATUS,
2082 env->msr_global_status);
2083 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
2084 env->msr_global_ovf_ctrl);
2085
2086 /* Now start the PMU. */
2087 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR_CTRL,
2088 env->msr_fixed_ctr_ctrl);
2089 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_CTRL,
2090 env->msr_global_ctrl);
2091 }
2092 }
2093 /*
2094 * Hyper-V partition-wide MSRs: to avoid clearing them on cpu hot-add,
2095 * only sync them to KVM on the first cpu
2096 */
2097 if (current_cpu == first_cpu) {
2098 if (has_msr_hv_hypercall) {
2099 kvm_msr_entry_add(cpu, HV_X64_MSR_GUEST_OS_ID,
2100 env->msr_hv_guest_os_id);
2101 kvm_msr_entry_add(cpu, HV_X64_MSR_HYPERCALL,
2102 env->msr_hv_hypercall);
2103 }
2104 if (cpu->hyperv_time) {
2105 kvm_msr_entry_add(cpu, HV_X64_MSR_REFERENCE_TSC,
2106 env->msr_hv_tsc);
2107 }
2108 if (cpu->hyperv_reenlightenment) {
2109 kvm_msr_entry_add(cpu, HV_X64_MSR_REENLIGHTENMENT_CONTROL,
2110 env->msr_hv_reenlightenment_control);
2111 kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_CONTROL,
2112 env->msr_hv_tsc_emulation_control);
2113 kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_STATUS,
2114 env->msr_hv_tsc_emulation_status);
2115 }
2116 }
2117 if (cpu->hyperv_vapic) {
2118 kvm_msr_entry_add(cpu, HV_X64_MSR_APIC_ASSIST_PAGE,
2119 env->msr_hv_vapic);
2120 }
2121 if (has_msr_hv_crash) {
2122 int j;
2123
2124 for (j = 0; j < HV_CRASH_PARAMS; j++)
2125 kvm_msr_entry_add(cpu, HV_X64_MSR_CRASH_P0 + j,
2126 env->msr_hv_crash_params[j]);
2127
2128 kvm_msr_entry_add(cpu, HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_NOTIFY);
2129 }
2130 if (has_msr_hv_runtime) {
2131 kvm_msr_entry_add(cpu, HV_X64_MSR_VP_RUNTIME, env->msr_hv_runtime);
2132 }
2133 if (cpu->hyperv_vpindex && hv_vpindex_settable) {
2134 kvm_msr_entry_add(cpu, HV_X64_MSR_VP_INDEX,
2135 hyperv_vp_index(CPU(cpu)));
2136 }
2137 if (cpu->hyperv_synic) {
2138 int j;
2139
2140 kvm_msr_entry_add(cpu, HV_X64_MSR_SVERSION, HV_SYNIC_VERSION);
2141
2142 kvm_msr_entry_add(cpu, HV_X64_MSR_SCONTROL,
2143 env->msr_hv_synic_control);
2144 kvm_msr_entry_add(cpu, HV_X64_MSR_SIEFP,
2145 env->msr_hv_synic_evt_page);
2146 kvm_msr_entry_add(cpu, HV_X64_MSR_SIMP,
2147 env->msr_hv_synic_msg_page);
2148
2149 for (j = 0; j < ARRAY_SIZE(env->msr_hv_synic_sint); j++) {
2150 kvm_msr_entry_add(cpu, HV_X64_MSR_SINT0 + j,
2151 env->msr_hv_synic_sint[j]);
2152 }
2153 }
2154 if (has_msr_hv_stimer) {
2155 int j;
2156
2157 for (j = 0; j < ARRAY_SIZE(env->msr_hv_stimer_config); j++) {
2158 kvm_msr_entry_add(cpu, HV_X64_MSR_STIMER0_CONFIG + j * 2,
2159 env->msr_hv_stimer_config[j]);
2160 }
2161
2162 for (j = 0; j < ARRAY_SIZE(env->msr_hv_stimer_count); j++) {
2163 kvm_msr_entry_add(cpu, HV_X64_MSR_STIMER0_COUNT + j * 2,
2164 env->msr_hv_stimer_count[j]);
2165 }
2166 }
2167 if (env->features[FEAT_1_EDX] & CPUID_MTRR) {
2168 uint64_t phys_mask = MAKE_64BIT_MASK(0, cpu->phys_bits);
2169
2170 kvm_msr_entry_add(cpu, MSR_MTRRdefType, env->mtrr_deftype);
2171 kvm_msr_entry_add(cpu, MSR_MTRRfix64K_00000, env->mtrr_fixed[0]);
2172 kvm_msr_entry_add(cpu, MSR_MTRRfix16K_80000, env->mtrr_fixed[1]);
2173 kvm_msr_entry_add(cpu, MSR_MTRRfix16K_A0000, env->mtrr_fixed[2]);
2174 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C0000, env->mtrr_fixed[3]);
2175 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C8000, env->mtrr_fixed[4]);
2176 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D0000, env->mtrr_fixed[5]);
2177 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D8000, env->mtrr_fixed[6]);
2178 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E0000, env->mtrr_fixed[7]);
2179 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E8000, env->mtrr_fixed[8]);
2180 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F0000, env->mtrr_fixed[9]);
2181 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F8000, env->mtrr_fixed[10]);
2182 for (i = 0; i < MSR_MTRRcap_VCNT; i++) {
2183 /* The CPU GPs if we write to a bit above the physical limit of
2184 * the host CPU (and KVM emulates that)
2185 */
2186 uint64_t mask = env->mtrr_var[i].mask;
2187 mask &= phys_mask;
2188
2189 kvm_msr_entry_add(cpu, MSR_MTRRphysBase(i),
2190 env->mtrr_var[i].base);
2191 kvm_msr_entry_add(cpu, MSR_MTRRphysMask(i), mask);
2192 }
2193 }
2194 if (env->features[FEAT_7_0_EBX] & CPUID_7_0_EBX_INTEL_PT) {
2195 int addr_num = kvm_arch_get_supported_cpuid(kvm_state,
2196 0x14, 1, R_EAX) & 0x7;
2197
2198 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CTL,
2199 env->msr_rtit_ctrl);
2200 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_STATUS,
2201 env->msr_rtit_status);
2202 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_BASE,
2203 env->msr_rtit_output_base);
2204 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_MASK,
2205 env->msr_rtit_output_mask);
2206 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CR3_MATCH,
2207 env->msr_rtit_cr3_match);
2208 for (i = 0; i < addr_num; i++) {
2209 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_ADDR0_A + i,
2210 env->msr_rtit_addrs[i]);
2211 }
2212 }
2213
2214 /* Note: MSR_IA32_FEATURE_CONTROL is written separately, see
2215 * kvm_put_msr_feature_control. */
2216 }
2217 if (env->mcg_cap) {
2218 int i;
2219
2220 kvm_msr_entry_add(cpu, MSR_MCG_STATUS, env->mcg_status);
2221 kvm_msr_entry_add(cpu, MSR_MCG_CTL, env->mcg_ctl);
2222 if (has_msr_mcg_ext_ctl) {
2223 kvm_msr_entry_add(cpu, MSR_MCG_EXT_CTL, env->mcg_ext_ctl);
2224 }
2225 for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++) {
2226 kvm_msr_entry_add(cpu, MSR_MC0_CTL + i, env->mce_banks[i]);
2227 }
2228 }
2229
2230 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MSRS, cpu->kvm_msr_buf);
2231 if (ret < 0) {
2232 return ret;
2233 }
2234
2235 if (ret < cpu->kvm_msr_buf->nmsrs) {
2236 struct kvm_msr_entry *e = &cpu->kvm_msr_buf->entries[ret];
2237 error_report("error: failed to set MSR 0x%" PRIx32 " to 0x%" PRIx64,
2238 (uint32_t)e->index, (uint64_t)e->data);
2239 }
2240
2241 assert(ret == cpu->kvm_msr_buf->nmsrs);
2242 return 0;
2243 }
2244
2245
2246 static int kvm_get_fpu(X86CPU *cpu)
2247 {
2248 CPUX86State *env = &cpu->env;
2249 struct kvm_fpu fpu;
2250 int i, ret;
2251
2252 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_FPU, &fpu);
2253 if (ret < 0) {
2254 return ret;
2255 }
2256
2257 env->fpstt = (fpu.fsw >> 11) & 7;
2258 env->fpus = fpu.fsw;
2259 env->fpuc = fpu.fcw;
2260 env->fpop = fpu.last_opcode;
2261 env->fpip = fpu.last_ip;
2262 env->fpdp = fpu.last_dp;
2263 for (i = 0; i < 8; ++i) {
2264 env->fptags[i] = !((fpu.ftwx >> i) & 1);
2265 }
2266 memcpy(env->fpregs, fpu.fpr, sizeof env->fpregs);
2267 for (i = 0; i < CPU_NB_REGS; i++) {
2268 env->xmm_regs[i].ZMM_Q(0) = ldq_p(&fpu.xmm[i][0]);
2269 env->xmm_regs[i].ZMM_Q(1) = ldq_p(&fpu.xmm[i][8]);
2270 }
2271 env->mxcsr = fpu.mxcsr;
2272
2273 return 0;
2274 }
2275
2276 static int kvm_get_xsave(X86CPU *cpu)
2277 {
2278 CPUX86State *env = &cpu->env;
2279 X86XSaveArea *xsave = env->xsave_buf;
2280 int ret;
2281
2282 if (!has_xsave) {
2283 return kvm_get_fpu(cpu);
2284 }
2285
2286 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_XSAVE, xsave);
2287 if (ret < 0) {
2288 return ret;
2289 }
2290 x86_cpu_xrstor_all_areas(cpu, xsave);
2291
2292 return 0;
2293 }
2294
2295 static int kvm_get_xcrs(X86CPU *cpu)
2296 {
2297 CPUX86State *env = &cpu->env;
2298 int i, ret;
2299 struct kvm_xcrs xcrs;
2300
2301 if (!has_xcrs) {
2302 return 0;
2303 }
2304
2305 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_XCRS, &xcrs);
2306 if (ret < 0) {
2307 return ret;
2308 }
2309
2310 for (i = 0; i < xcrs.nr_xcrs; i++) {
2311 /* Only support xcr0 now */
2312 if (xcrs.xcrs[i].xcr == 0) {
2313 env->xcr0 = xcrs.xcrs[i].value;
2314 break;
2315 }
2316 }
2317 return 0;
2318 }
2319
2320 static int kvm_get_sregs(X86CPU *cpu)
2321 {
2322 CPUX86State *env = &cpu->env;
2323 struct kvm_sregs sregs;
2324 int bit, i, ret;
2325
2326 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
2327 if (ret < 0) {
2328 return ret;
2329 }
2330
2331 /* There can only be one pending IRQ set in the bitmap at a time, so try
2332 to find it and save its number instead (-1 for none). */
2333 env->interrupt_injected = -1;
2334 for (i = 0; i < ARRAY_SIZE(sregs.interrupt_bitmap); i++) {
2335 if (sregs.interrupt_bitmap[i]) {
2336 bit = ctz64(sregs.interrupt_bitmap[i]);
2337 env->interrupt_injected = i * 64 + bit;
2338 break;
2339 }
2340 }
2341
2342 get_seg(&env->segs[R_CS], &sregs.cs);
2343 get_seg(&env->segs[R_DS], &sregs.ds);
2344 get_seg(&env->segs[R_ES], &sregs.es);
2345 get_seg(&env->segs[R_FS], &sregs.fs);
2346 get_seg(&env->segs[R_GS], &sregs.gs);
2347 get_seg(&env->segs[R_SS], &sregs.ss);
2348
2349 get_seg(&env->tr, &sregs.tr);
2350 get_seg(&env->ldt, &sregs.ldt);
2351
2352 env->idt.limit = sregs.idt.limit;
2353 env->idt.base = sregs.idt.base;
2354 env->gdt.limit = sregs.gdt.limit;
2355 env->gdt.base = sregs.gdt.base;
2356
2357 env->cr[0] = sregs.cr0;
2358 env->cr[2] = sregs.cr2;
2359 env->cr[3] = sregs.cr3;
2360 env->cr[4] = sregs.cr4;
2361
2362 env->efer = sregs.efer;
2363
2364 /* changes to apic base and cr8/tpr are read back via kvm_arch_post_run */
2365 x86_update_hflags(env);
2366
2367 return 0;
2368 }
2369
2370 static int kvm_get_msrs(X86CPU *cpu)
2371 {
2372 CPUX86State *env = &cpu->env;
2373 struct kvm_msr_entry *msrs = cpu->kvm_msr_buf->entries;
2374 int ret, i;
2375 uint64_t mtrr_top_bits;
2376
2377 kvm_msr_buf_reset(cpu);
2378
2379 kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_CS, 0);
2380 kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_ESP, 0);
2381 kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_EIP, 0);
2382 kvm_msr_entry_add(cpu, MSR_PAT, 0);
2383 if (has_msr_star) {
2384 kvm_msr_entry_add(cpu, MSR_STAR, 0);
2385 }
2386 if (has_msr_hsave_pa) {
2387 kvm_msr_entry_add(cpu, MSR_VM_HSAVE_PA, 0);
2388 }
2389 if (has_msr_tsc_aux) {
2390 kvm_msr_entry_add(cpu, MSR_TSC_AUX, 0);
2391 }
2392 if (has_msr_tsc_adjust) {
2393 kvm_msr_entry_add(cpu, MSR_TSC_ADJUST, 0);
2394 }
2395 if (has_msr_tsc_deadline) {
2396 kvm_msr_entry_add(cpu, MSR_IA32_TSCDEADLINE, 0);
2397 }
2398 if (has_msr_misc_enable) {
2399 kvm_msr_entry_add(cpu, MSR_IA32_MISC_ENABLE, 0);
2400 }
2401 if (has_msr_smbase) {
2402 kvm_msr_entry_add(cpu, MSR_IA32_SMBASE, 0);
2403 }
2404 if (has_msr_smi_count) {
2405 kvm_msr_entry_add(cpu, MSR_SMI_COUNT, 0);
2406 }
2407 if (has_msr_feature_control) {
2408 kvm_msr_entry_add(cpu, MSR_IA32_FEATURE_CONTROL, 0);
2409 }
2410 if (has_msr_bndcfgs) {
2411 kvm_msr_entry_add(cpu, MSR_IA32_BNDCFGS, 0);
2412 }
2413 if (has_msr_xss) {
2414 kvm_msr_entry_add(cpu, MSR_IA32_XSS, 0);
2415 }
2416 if (has_msr_spec_ctrl) {
2417 kvm_msr_entry_add(cpu, MSR_IA32_SPEC_CTRL, 0);
2418 }
2419 if (has_msr_virt_ssbd) {
2420 kvm_msr_entry_add(cpu, MSR_VIRT_SSBD, 0);
2421 }
2422 if (!env->tsc_valid) {
2423 kvm_msr_entry_add(cpu, MSR_IA32_TSC, 0);
2424 env->tsc_valid = !runstate_is_running();
2425 }
2426
2427 #ifdef TARGET_X86_64
2428 if (lm_capable_kernel) {
2429 kvm_msr_entry_add(cpu, MSR_CSTAR, 0);
2430 kvm_msr_entry_add(cpu, MSR_KERNELGSBASE, 0);
2431 kvm_msr_entry_add(cpu, MSR_FMASK, 0);
2432 kvm_msr_entry_add(cpu, MSR_LSTAR, 0);
2433 }
2434 #endif
2435 kvm_msr_entry_add(cpu, MSR_KVM_SYSTEM_TIME, 0);
2436 kvm_msr_entry_add(cpu, MSR_KVM_WALL_CLOCK, 0);
2437 if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_ASYNC_PF)) {
2438 kvm_msr_entry_add(cpu, MSR_KVM_ASYNC_PF_EN, 0);
2439 }
2440 if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_PV_EOI)) {
2441 kvm_msr_entry_add(cpu, MSR_KVM_PV_EOI_EN, 0);
2442 }
2443 if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_STEAL_TIME)) {
2444 kvm_msr_entry_add(cpu, MSR_KVM_STEAL_TIME, 0);
2445 }
2446 if (has_architectural_pmu_version > 0) {
2447 if (has_architectural_pmu_version > 1) {
2448 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR_CTRL, 0);
2449 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_CTRL, 0);
2450 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_STATUS, 0);
2451 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_OVF_CTRL, 0);
2452 }
2453 for (i = 0; i < num_architectural_pmu_fixed_counters; i++) {
2454 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR0 + i, 0);
2455 }
2456 for (i = 0; i < num_architectural_pmu_gp_counters; i++) {
2457 kvm_msr_entry_add(cpu, MSR_P6_PERFCTR0 + i, 0);
2458 kvm_msr_entry_add(cpu, MSR_P6_EVNTSEL0 + i, 0);
2459 }
2460 }
2461
2462 if (env->mcg_cap) {
2463 kvm_msr_entry_add(cpu, MSR_MCG_STATUS, 0);
2464 kvm_msr_entry_add(cpu, MSR_MCG_CTL, 0);
2465 if (has_msr_mcg_ext_ctl) {
2466 kvm_msr_entry_add(cpu, MSR_MCG_EXT_CTL, 0);
2467 }
2468 for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++) {
2469 kvm_msr_entry_add(cpu, MSR_MC0_CTL + i, 0);
2470 }
2471 }
2472
2473 if (has_msr_hv_hypercall) {
2474 kvm_msr_entry_add(cpu, HV_X64_MSR_HYPERCALL, 0);
2475 kvm_msr_entry_add(cpu, HV_X64_MSR_GUEST_OS_ID, 0);
2476 }
2477 if (cpu->hyperv_vapic) {
2478 kvm_msr_entry_add(cpu, HV_X64_MSR_APIC_ASSIST_PAGE, 0);
2479 }
2480 if (cpu->hyperv_time) {
2481 kvm_msr_entry_add(cpu, HV_X64_MSR_REFERENCE_TSC, 0);
2482 }
2483 if (cpu->hyperv_reenlightenment) {
2484 kvm_msr_entry_add(cpu, HV_X64_MSR_REENLIGHTENMENT_CONTROL, 0);
2485 kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_CONTROL, 0);
2486 kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_STATUS, 0);
2487 }
2488 if (has_msr_hv_crash) {
2489 int j;
2490
2491 for (j = 0; j < HV_CRASH_PARAMS; j++) {
2492 kvm_msr_entry_add(cpu, HV_X64_MSR_CRASH_P0 + j, 0);
2493 }
2494 }
2495 if (has_msr_hv_runtime) {
2496 kvm_msr_entry_add(cpu, HV_X64_MSR_VP_RUNTIME, 0);
2497 }
2498 if (cpu->hyperv_synic) {
2499 uint32_t msr;
2500
2501 kvm_msr_entry_add(cpu, HV_X64_MSR_SCONTROL, 0);
2502 kvm_msr_entry_add(cpu, HV_X64_MSR_SIEFP, 0);
2503 kvm_msr_entry_add(cpu, HV_X64_MSR_SIMP, 0);
2504 for (msr = HV_X64_MSR_SINT0; msr <= HV_X64_MSR_SINT15; msr++) {
2505 kvm_msr_entry_add(cpu, msr, 0);
2506 }
2507 }
2508 if (has_msr_hv_stimer) {
2509 uint32_t msr;
2510
2511 for (msr = HV_X64_MSR_STIMER0_CONFIG; msr <= HV_X64_MSR_STIMER3_COUNT;
2512 msr++) {
2513 kvm_msr_entry_add(cpu, msr, 0);
2514 }
2515 }
2516 if (env->features[FEAT_1_EDX] & CPUID_MTRR) {
2517 kvm_msr_entry_add(cpu, MSR_MTRRdefType, 0);
2518 kvm_msr_entry_add(cpu, MSR_MTRRfix64K_00000, 0);
2519 kvm_msr_entry_add(cpu, MSR_MTRRfix16K_80000, 0);
2520 kvm_msr_entry_add(cpu, MSR_MTRRfix16K_A0000, 0);
2521 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C0000, 0);
2522 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C8000, 0);
2523 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D0000, 0);
2524 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D8000, 0);
2525 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E0000, 0);
2526 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E8000, 0);
2527 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F0000, 0);
2528 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F8000, 0);
2529 for (i = 0; i < MSR_MTRRcap_VCNT; i++) {
2530 kvm_msr_entry_add(cpu, MSR_MTRRphysBase(i), 0);
2531 kvm_msr_entry_add(cpu, MSR_MTRRphysMask(i), 0);
2532 }
2533 }
2534
2535 if (env->features[FEAT_7_0_EBX] & CPUID_7_0_EBX_INTEL_PT) {
2536 int addr_num =
2537 kvm_arch_get_supported_cpuid(kvm_state, 0x14, 1, R_EAX) & 0x7;
2538
2539 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CTL, 0);
2540 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_STATUS, 0);
2541 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_BASE, 0);
2542 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_MASK, 0);
2543 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CR3_MATCH, 0);
2544 for (i = 0; i < addr_num; i++) {
2545 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_ADDR0_A + i, 0);
2546 }
2547 }
2548
2549 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MSRS, cpu->kvm_msr_buf);
2550 if (ret < 0) {
2551 return ret;
2552 }
2553
2554 if (ret < cpu->kvm_msr_buf->nmsrs) {
2555 struct kvm_msr_entry *e = &cpu->kvm_msr_buf->entries[ret];
2556 error_report("error: failed to get MSR 0x%" PRIx32,
2557 (uint32_t)e->index);
2558 }
2559
2560 assert(ret == cpu->kvm_msr_buf->nmsrs);
2561 /*
2562 * MTRR masks: Each mask consists of 5 parts
2563 * a 10..0: must be zero
2564 * b 11 : valid bit
2565 * c n-1.12: actual mask bits
2566 * d 51..n: reserved must be zero
2567 * e 63.52: reserved must be zero
2568 *
2569 * 'n' is the number of physical bits supported by the CPU and is
2570 * apparently always <= 52. We know our 'n' but don't know what
2571 * the destinations 'n' is; it might be smaller, in which case
2572 * it masks (c) on loading. It might be larger, in which case
2573 * we fill 'd' so that d..c is consistent irrespetive of the 'n'
2574 * we're migrating to.
2575 */
2576
2577 if (cpu->fill_mtrr_mask) {
2578 QEMU_BUILD_BUG_ON(TARGET_PHYS_ADDR_SPACE_BITS > 52);
2579 assert(cpu->phys_bits <= TARGET_PHYS_ADDR_SPACE_BITS);
2580 mtrr_top_bits = MAKE_64BIT_MASK(cpu->phys_bits, 52 - cpu->phys_bits);
2581 } else {
2582 mtrr_top_bits = 0;
2583 }
2584
2585 for (i = 0; i < ret; i++) {
2586 uint32_t index = msrs[i].index;
2587 switch (index) {
2588 case MSR_IA32_SYSENTER_CS:
2589 env->sysenter_cs = msrs[i].data;
2590 break;
2591 case MSR_IA32_SYSENTER_ESP:
2592 env->sysenter_esp = msrs[i].data;
2593 break;
2594 case MSR_IA32_SYSENTER_EIP:
2595 env->sysenter_eip = msrs[i].data;
2596 break;
2597 case MSR_PAT:
2598 env->pat = msrs[i].data;
2599 break;
2600 case MSR_STAR:
2601 env->star = msrs[i].data;
2602 break;
2603 #ifdef TARGET_X86_64
2604 case MSR_CSTAR:
2605 env->cstar = msrs[i].data;
2606 break;
2607 case MSR_KERNELGSBASE:
2608 env->kernelgsbase = msrs[i].data;
2609 break;
2610 case MSR_FMASK:
2611 env->fmask = msrs[i].data;
2612 break;
2613 case MSR_LSTAR:
2614 env->lstar = msrs[i].data;
2615 break;
2616 #endif
2617 case MSR_IA32_TSC:
2618 env->tsc = msrs[i].data;
2619 break;
2620 case MSR_TSC_AUX:
2621 env->tsc_aux = msrs[i].data;
2622 break;
2623 case MSR_TSC_ADJUST:
2624 env->tsc_adjust = msrs[i].data;
2625 break;
2626 case MSR_IA32_TSCDEADLINE:
2627 env->tsc_deadline = msrs[i].data;
2628 break;
2629 case MSR_VM_HSAVE_PA:
2630 env->vm_hsave = msrs[i].data;
2631 break;
2632 case MSR_KVM_SYSTEM_TIME:
2633 env->system_time_msr = msrs[i].data;
2634 break;
2635 case MSR_KVM_WALL_CLOCK:
2636 env->wall_clock_msr = msrs[i].data;
2637 break;
2638 case MSR_MCG_STATUS:
2639 env->mcg_status = msrs[i].data;
2640 break;
2641 case MSR_MCG_CTL:
2642 env->mcg_ctl = msrs[i].data;
2643 break;
2644 case MSR_MCG_EXT_CTL:
2645 env->mcg_ext_ctl = msrs[i].data;
2646 break;
2647 case MSR_IA32_MISC_ENABLE:
2648 env->msr_ia32_misc_enable = msrs[i].data;
2649 break;
2650 case MSR_IA32_SMBASE:
2651 env->smbase = msrs[i].data;
2652 break;
2653 case MSR_SMI_COUNT:
2654 env->msr_smi_count = msrs[i].data;
2655 break;
2656 case MSR_IA32_FEATURE_CONTROL:
2657 env->msr_ia32_feature_control = msrs[i].data;
2658 break;
2659 case MSR_IA32_BNDCFGS:
2660 env->msr_bndcfgs = msrs[i].data;
2661 break;
2662 case MSR_IA32_XSS:
2663 env->xss = msrs[i].data;
2664 break;
2665 default:
2666 if (msrs[i].index >= MSR_MC0_CTL &&
2667 msrs[i].index < MSR_MC0_CTL + (env->mcg_cap & 0xff) * 4) {
2668 env->mce_banks[msrs[i].index - MSR_MC0_CTL] = msrs[i].data;
2669 }
2670 break;
2671 case MSR_KVM_ASYNC_PF_EN:
2672 env->async_pf_en_msr = msrs[i].data;
2673 break;
2674 case MSR_KVM_PV_EOI_EN:
2675 env->pv_eoi_en_msr = msrs[i].data;
2676 break;
2677 case MSR_KVM_STEAL_TIME:
2678 env->steal_time_msr = msrs[i].data;
2679 break;
2680 case MSR_CORE_PERF_FIXED_CTR_CTRL:
2681 env->msr_fixed_ctr_ctrl = msrs[i].data;
2682 break;
2683 case MSR_CORE_PERF_GLOBAL_CTRL:
2684 env->msr_global_ctrl = msrs[i].data;
2685 break;
2686 case MSR_CORE_PERF_GLOBAL_STATUS:
2687 env->msr_global_status = msrs[i].data;
2688 break;
2689 case MSR_CORE_PERF_GLOBAL_OVF_CTRL:
2690 env->msr_global_ovf_ctrl = msrs[i].data;
2691 break;
2692 case MSR_CORE_PERF_FIXED_CTR0 ... MSR_CORE_PERF_FIXED_CTR0 + MAX_FIXED_COUNTERS - 1:
2693 env->msr_fixed_counters[index - MSR_CORE_PERF_FIXED_CTR0] = msrs[i].data;
2694 break;
2695 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR0 + MAX_GP_COUNTERS - 1:
2696 env->msr_gp_counters[index - MSR_P6_PERFCTR0] = msrs[i].data;
2697 break;
2698 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL0 + MAX_GP_COUNTERS - 1:
2699 env->msr_gp_evtsel[index - MSR_P6_EVNTSEL0] = msrs[i].data;
2700 break;
2701 case HV_X64_MSR_HYPERCALL:
2702 env->msr_hv_hypercall = msrs[i].data;
2703 break;
2704 case HV_X64_MSR_GUEST_OS_ID:
2705 env->msr_hv_guest_os_id = msrs[i].data;
2706 break;
2707 case HV_X64_MSR_APIC_ASSIST_PAGE:
2708 env->msr_hv_vapic = msrs[i].data;
2709 break;
2710 case HV_X64_MSR_REFERENCE_TSC:
2711 env->msr_hv_tsc = msrs[i].data;
2712 break;
2713 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2714 env->msr_hv_crash_params[index - HV_X64_MSR_CRASH_P0] = msrs[i].data;
2715 break;
2716 case HV_X64_MSR_VP_RUNTIME:
2717 env->msr_hv_runtime = msrs[i].data;
2718 break;
2719 case HV_X64_MSR_SCONTROL:
2720 env->msr_hv_synic_control = msrs[i].data;
2721 break;
2722 case HV_X64_MSR_SIEFP:
2723 env->msr_hv_synic_evt_page = msrs[i].data;
2724 break;
2725 case HV_X64_MSR_SIMP:
2726 env->msr_hv_synic_msg_page = msrs[i].data;
2727 break;
2728 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
2729 env->msr_hv_synic_sint[index - HV_X64_MSR_SINT0] = msrs[i].data;
2730 break;
2731 case HV_X64_MSR_STIMER0_CONFIG:
2732 case HV_X64_MSR_STIMER1_CONFIG:
2733 case HV_X64_MSR_STIMER2_CONFIG:
2734 case HV_X64_MSR_STIMER3_CONFIG:
2735 env->msr_hv_stimer_config[(index - HV_X64_MSR_STIMER0_CONFIG)/2] =
2736 msrs[i].data;
2737 break;
2738 case HV_X64_MSR_STIMER0_COUNT:
2739 case HV_X64_MSR_STIMER1_COUNT:
2740 case HV_X64_MSR_STIMER2_COUNT:
2741 case HV_X64_MSR_STIMER3_COUNT:
2742 env->msr_hv_stimer_count[(index - HV_X64_MSR_STIMER0_COUNT)/2] =
2743 msrs[i].data;
2744 break;
2745 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2746 env->msr_hv_reenlightenment_control = msrs[i].data;
2747 break;
2748 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2749 env->msr_hv_tsc_emulation_control = msrs[i].data;
2750 break;
2751 case HV_X64_MSR_TSC_EMULATION_STATUS:
2752 env->msr_hv_tsc_emulation_status = msrs[i].data;
2753 break;
2754 case MSR_MTRRdefType:
2755 env->mtrr_deftype = msrs[i].data;
2756 break;
2757 case MSR_MTRRfix64K_00000:
2758 env->mtrr_fixed[0] = msrs[i].data;
2759 break;
2760 case MSR_MTRRfix16K_80000:
2761 env->mtrr_fixed[1] = msrs[i].data;
2762 break;
2763 case MSR_MTRRfix16K_A0000:
2764 env->mtrr_fixed[2] = msrs[i].data;
2765 break;
2766 case MSR_MTRRfix4K_C0000:
2767 env->mtrr_fixed[3] = msrs[i].data;
2768 break;
2769 case MSR_MTRRfix4K_C8000:
2770 env->mtrr_fixed[4] = msrs[i].data;
2771 break;
2772 case MSR_MTRRfix4K_D0000:
2773 env->mtrr_fixed[5] = msrs[i].data;
2774 break;
2775 case MSR_MTRRfix4K_D8000:
2776 env->mtrr_fixed[6] = msrs[i].data;
2777 break;
2778 case MSR_MTRRfix4K_E0000:
2779 env->mtrr_fixed[7] = msrs[i].data;
2780 break;
2781 case MSR_MTRRfix4K_E8000:
2782 env->mtrr_fixed[8] = msrs[i].data;
2783 break;
2784 case MSR_MTRRfix4K_F0000:
2785 env->mtrr_fixed[9] = msrs[i].data;
2786 break;
2787 case MSR_MTRRfix4K_F8000:
2788 env->mtrr_fixed[10] = msrs[i].data;
2789 break;
2790 case MSR_MTRRphysBase(0) ... MSR_MTRRphysMask(MSR_MTRRcap_VCNT - 1):
2791 if (index & 1) {
2792 env->mtrr_var[MSR_MTRRphysIndex(index)].mask = msrs[i].data |
2793 mtrr_top_bits;
2794 } else {
2795 env->mtrr_var[MSR_MTRRphysIndex(index)].base = msrs[i].data;
2796 }
2797 break;
2798 case MSR_IA32_SPEC_CTRL:
2799 env->spec_ctrl = msrs[i].data;
2800 break;
2801 case MSR_VIRT_SSBD:
2802 env->virt_ssbd = msrs[i].data;
2803 break;
2804 case MSR_IA32_RTIT_CTL:
2805 env->msr_rtit_ctrl = msrs[i].data;
2806 break;
2807 case MSR_IA32_RTIT_STATUS:
2808 env->msr_rtit_status = msrs[i].data;
2809 break;
2810 case MSR_IA32_RTIT_OUTPUT_BASE:
2811 env->msr_rtit_output_base = msrs[i].data;
2812 break;
2813 case MSR_IA32_RTIT_OUTPUT_MASK:
2814 env->msr_rtit_output_mask = msrs[i].data;
2815 break;
2816 case MSR_IA32_RTIT_CR3_MATCH:
2817 env->msr_rtit_cr3_match = msrs[i].data;
2818 break;
2819 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2820 env->msr_rtit_addrs[index - MSR_IA32_RTIT_ADDR0_A] = msrs[i].data;
2821 break;
2822 }
2823 }
2824
2825 return 0;
2826 }
2827
2828 static int kvm_put_mp_state(X86CPU *cpu)
2829 {
2830 struct kvm_mp_state mp_state = { .mp_state = cpu->env.mp_state };
2831
2832 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
2833 }
2834
2835 static int kvm_get_mp_state(X86CPU *cpu)
2836 {
2837 CPUState *cs = CPU(cpu);
2838 CPUX86State *env = &cpu->env;
2839 struct kvm_mp_state mp_state;
2840 int ret;
2841
2842 ret = kvm_vcpu_ioctl(cs, KVM_GET_MP_STATE, &mp_state);
2843 if (ret < 0) {
2844 return ret;
2845 }
2846 env->mp_state = mp_state.mp_state;
2847 if (kvm_irqchip_in_kernel()) {
2848 cs->halted = (mp_state.mp_state == KVM_MP_STATE_HALTED);
2849 }
2850 return 0;
2851 }
2852
2853 static int kvm_get_apic(X86CPU *cpu)
2854 {
2855 DeviceState *apic = cpu->apic_state;
2856 struct kvm_lapic_state kapic;
2857 int ret;
2858
2859 if (apic && kvm_irqchip_in_kernel()) {
2860 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_LAPIC, &kapic);
2861 if (ret < 0) {
2862 return ret;
2863 }
2864
2865 kvm_get_apic_state(apic, &kapic);
2866 }
2867 return 0;
2868 }
2869
2870 static int kvm_put_vcpu_events(X86CPU *cpu, int level)
2871 {
2872 CPUState *cs = CPU(cpu);
2873 CPUX86State *env = &cpu->env;
2874 struct kvm_vcpu_events events = {};
2875
2876 if (!kvm_has_vcpu_events()) {
2877 return 0;
2878 }
2879
2880 events.exception.injected = (env->exception_injected >= 0);
2881 events.exception.nr = env->exception_injected;
2882 events.exception.has_error_code = env->has_error_code;
2883 events.exception.error_code = env->error_code;
2884
2885 events.interrupt.injected = (env->interrupt_injected >= 0);
2886 events.interrupt.nr = env->interrupt_injected;
2887 events.interrupt.soft = env->soft_interrupt;
2888
2889 events.nmi.injected = env->nmi_injected;
2890 events.nmi.pending = env->nmi_pending;
2891 events.nmi.masked = !!(env->hflags2 & HF2_NMI_MASK);
2892
2893 events.sipi_vector = env->sipi_vector;
2894 events.flags = 0;
2895
2896 if (has_msr_smbase) {
2897 events.smi.smm = !!(env->hflags & HF_SMM_MASK);
2898 events.smi.smm_inside_nmi = !!(env->hflags2 & HF2_SMM_INSIDE_NMI_MASK);
2899 if (kvm_irqchip_in_kernel()) {
2900 /* As soon as these are moved to the kernel, remove them
2901 * from cs->interrupt_request.
2902 */
2903 events.smi.pending = cs->interrupt_request & CPU_INTERRUPT_SMI;
2904 events.smi.latched_init = cs->interrupt_request & CPU_INTERRUPT_INIT;
2905 cs->interrupt_request &= ~(CPU_INTERRUPT_INIT | CPU_INTERRUPT_SMI);
2906 } else {
2907 /* Keep these in cs->interrupt_request. */
2908 events.smi.pending = 0;
2909 events.smi.latched_init = 0;
2910 }
2911 /* Stop SMI delivery on old machine types to avoid a reboot
2912 * on an inward migration of an old VM.
2913 */
2914 if (!cpu->kvm_no_smi_migration) {
2915 events.flags |= KVM_VCPUEVENT_VALID_SMM;
2916 }
2917 }
2918
2919 if (level >= KVM_PUT_RESET_STATE) {
2920 events.flags |= KVM_VCPUEVENT_VALID_NMI_PENDING;
2921 if (env->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
2922 events.flags |= KVM_VCPUEVENT_VALID_SIPI_VECTOR;
2923 }
2924 }
2925
2926 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_VCPU_EVENTS, &events);
2927 }
2928
2929 static int kvm_get_vcpu_events(X86CPU *cpu)
2930 {
2931 CPUX86State *env = &cpu->env;
2932 struct kvm_vcpu_events events;
2933 int ret;
2934
2935 if (!kvm_has_vcpu_events()) {
2936 return 0;
2937 }
2938
2939 memset(&events, 0, sizeof(events));
2940 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_VCPU_EVENTS, &events);
2941 if (ret < 0) {
2942 return ret;
2943 }
2944 env->exception_injected =
2945 events.exception.injected ? events.exception.nr : -1;
2946 env->has_error_code = events.exception.has_error_code;
2947 env->error_code = events.exception.error_code;
2948
2949 env->interrupt_injected =
2950 events.interrupt.injected ? events.interrupt.nr : -1;
2951 env->soft_interrupt = events.interrupt.soft;
2952
2953 env->nmi_injected = events.nmi.injected;
2954 env->nmi_pending = events.nmi.pending;
2955 if (events.nmi.masked) {
2956 env->hflags2 |= HF2_NMI_MASK;
2957 } else {
2958 env->hflags2 &= ~HF2_NMI_MASK;
2959 }
2960
2961 if (events.flags & KVM_VCPUEVENT_VALID_SMM) {
2962 if (events.smi.smm) {
2963 env->hflags |= HF_SMM_MASK;
2964 } else {
2965 env->hflags &= ~HF_SMM_MASK;
2966 }
2967 if (events.smi.pending) {
2968 cpu_interrupt(CPU(cpu), CPU_INTERRUPT_SMI);
2969 } else {
2970 cpu_reset_interrupt(CPU(cpu), CPU_INTERRUPT_SMI);
2971 }
2972 if (events.smi.smm_inside_nmi) {
2973 env->hflags2 |= HF2_SMM_INSIDE_NMI_MASK;
2974 } else {
2975 env->hflags2 &= ~HF2_SMM_INSIDE_NMI_MASK;
2976 }
2977 if (events.smi.latched_init) {
2978 cpu_interrupt(CPU(cpu), CPU_INTERRUPT_INIT);
2979 } else {
2980 cpu_reset_interrupt(CPU(cpu), CPU_INTERRUPT_INIT);
2981 }
2982 }
2983
2984 env->sipi_vector = events.sipi_vector;
2985
2986 return 0;
2987 }
2988
2989 static int kvm_guest_debug_workarounds(X86CPU *cpu)
2990 {
2991 CPUState *cs = CPU(cpu);
2992 CPUX86State *env = &cpu->env;
2993 int ret = 0;
2994 unsigned long reinject_trap = 0;
2995
2996 if (!kvm_has_vcpu_events()) {
2997 if (env->exception_injected == 1) {
2998 reinject_trap = KVM_GUESTDBG_INJECT_DB;
2999 } else if (env->exception_injected == 3) {
3000 reinject_trap = KVM_GUESTDBG_INJECT_BP;
3001 }
3002 env->exception_injected = -1;
3003 }
3004
3005 /*
3006 * Kernels before KVM_CAP_X86_ROBUST_SINGLESTEP overwrote flags.TF
3007 * injected via SET_GUEST_DEBUG while updating GP regs. Work around this
3008 * by updating the debug state once again if single-stepping is on.
3009 * Another reason to call kvm_update_guest_debug here is a pending debug
3010 * trap raise by the guest. On kernels without SET_VCPU_EVENTS we have to
3011 * reinject them via SET_GUEST_DEBUG.
3012 */
3013 if (reinject_trap ||
3014 (!kvm_has_robust_singlestep() && cs->singlestep_enabled)) {
3015 ret = kvm_update_guest_debug(cs, reinject_trap);
3016 }
3017 return ret;
3018 }
3019
3020 static int kvm_put_debugregs(X86CPU *cpu)
3021 {
3022 CPUX86State *env = &cpu->env;
3023 struct kvm_debugregs dbgregs;
3024 int i;
3025
3026 if (!kvm_has_debugregs()) {
3027 return 0;
3028 }
3029
3030 for (i = 0; i < 4; i++) {
3031 dbgregs.db[i] = env->dr[i];
3032 }
3033 dbgregs.dr6 = env->dr[6];
3034 dbgregs.dr7 = env->dr[7];
3035 dbgregs.flags = 0;
3036
3037 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_DEBUGREGS, &dbgregs);
3038 }
3039
3040 static int kvm_get_debugregs(X86CPU *cpu)
3041 {
3042 CPUX86State *env = &cpu->env;
3043 struct kvm_debugregs dbgregs;
3044 int i, ret;
3045
3046 if (!kvm_has_debugregs()) {
3047 return 0;
3048 }
3049
3050 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_DEBUGREGS, &dbgregs);
3051 if (ret < 0) {
3052 return ret;
3053 }
3054 for (i = 0; i < 4; i++) {
3055 env->dr[i] = dbgregs.db[i];
3056 }
3057 env->dr[4] = env->dr[6] = dbgregs.dr6;
3058 env->dr[5] = env->dr[7] = dbgregs.dr7;
3059
3060 return 0;
3061 }
3062
3063 int kvm_arch_put_registers(CPUState *cpu, int level)
3064 {
3065 X86CPU *x86_cpu = X86_CPU(cpu);
3066 int ret;
3067
3068 assert(cpu_is_stopped(cpu) || qemu_cpu_is_self(cpu));
3069
3070 if (level >= KVM_PUT_RESET_STATE) {
3071 ret = kvm_put_msr_feature_control(x86_cpu);
3072 if (ret < 0) {
3073 return ret;
3074 }
3075 }
3076
3077 if (level == KVM_PUT_FULL_STATE) {
3078 /* We don't check for kvm_arch_set_tsc_khz() errors here,
3079 * because TSC frequency mismatch shouldn't abort migration,
3080 * unless the user explicitly asked for a more strict TSC
3081 * setting (e.g. using an explicit "tsc-freq" option).
3082 */
3083 kvm_arch_set_tsc_khz(cpu);
3084 }
3085
3086 ret = kvm_getput_regs(x86_cpu, 1);
3087 if (ret < 0) {
3088 return ret;
3089 }
3090 ret = kvm_put_xsave(x86_cpu);
3091 if (ret < 0) {
3092 return ret;
3093 }
3094 ret = kvm_put_xcrs(x86_cpu);
3095 if (ret < 0) {
3096 return ret;
3097 }
3098 ret = kvm_put_sregs(x86_cpu);
3099 if (ret < 0) {
3100 return ret;
3101 }
3102 /* must be before kvm_put_msrs */
3103 ret = kvm_inject_mce_oldstyle(x86_cpu);
3104 if (ret < 0) {
3105 return ret;
3106 }
3107 ret = kvm_put_msrs(x86_cpu, level);
3108 if (ret < 0) {
3109 return ret;
3110 }
3111 ret = kvm_put_vcpu_events(x86_cpu, level);
3112 if (ret < 0) {
3113 return ret;
3114 }
3115 if (level >= KVM_PUT_RESET_STATE) {
3116 ret = kvm_put_mp_state(x86_cpu);
3117 if (ret < 0) {
3118 return ret;
3119 }
3120 }
3121
3122 ret = kvm_put_tscdeadline_msr(x86_cpu);
3123 if (ret < 0) {
3124 return ret;
3125 }
3126 ret = kvm_put_debugregs(x86_cpu);
3127 if (ret < 0) {
3128 return ret;
3129 }
3130 /* must be last */
3131 ret = kvm_guest_debug_workarounds(x86_cpu);
3132 if (ret < 0) {
3133 return ret;
3134 }
3135 return 0;
3136 }
3137
3138 int kvm_arch_get_registers(CPUState *cs)
3139 {
3140 X86CPU *cpu = X86_CPU(cs);
3141 int ret;
3142
3143 assert(cpu_is_stopped(cs) || qemu_cpu_is_self(cs));
3144
3145 ret = kvm_get_vcpu_events(cpu);
3146 if (ret < 0) {
3147 goto out;
3148 }
3149 /*
3150 * KVM_GET_MPSTATE can modify CS and RIP, call it before
3151 * KVM_GET_REGS and KVM_GET_SREGS.
3152 */
3153 ret = kvm_get_mp_state(cpu);
3154 if (ret < 0) {
3155 goto out;
3156 }
3157 ret = kvm_getput_regs(cpu, 0);
3158 if (ret < 0) {
3159 goto out;
3160 }
3161 ret = kvm_get_xsave(cpu);
3162 if (ret < 0) {
3163 goto out;
3164 }
3165 ret = kvm_get_xcrs(cpu);
3166 if (ret < 0) {
3167 goto out;
3168 }
3169 ret = kvm_get_sregs(cpu);
3170 if (ret < 0) {
3171 goto out;
3172 }
3173 ret = kvm_get_msrs(cpu);
3174 if (ret < 0) {
3175 goto out;
3176 }
3177 ret = kvm_get_apic(cpu);
3178 if (ret < 0) {
3179 goto out;
3180 }
3181 ret = kvm_get_debugregs(cpu);
3182 if (ret < 0) {
3183 goto out;
3184 }
3185 ret = 0;
3186 out:
3187 cpu_sync_bndcs_hflags(&cpu->env);
3188 return ret;
3189 }
3190
3191 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
3192 {
3193 X86CPU *x86_cpu = X86_CPU(cpu);
3194 CPUX86State *env = &x86_cpu->env;
3195 int ret;
3196
3197 /* Inject NMI */
3198 if (cpu->interrupt_request & (CPU_INTERRUPT_NMI | CPU_INTERRUPT_SMI)) {
3199 if (cpu->interrupt_request & CPU_INTERRUPT_NMI) {
3200 qemu_mutex_lock_iothread();
3201 cpu->interrupt_request &= ~CPU_INTERRUPT_NMI;
3202 qemu_mutex_unlock_iothread();
3203 DPRINTF("injected NMI\n");
3204 ret = kvm_vcpu_ioctl(cpu, KVM_NMI);
3205 if (ret < 0) {
3206 fprintf(stderr, "KVM: injection failed, NMI lost (%s)\n",
3207 strerror(-ret));
3208 }
3209 }
3210 if (cpu->interrupt_request & CPU_INTERRUPT_SMI) {
3211 qemu_mutex_lock_iothread();
3212 cpu->interrupt_request &= ~CPU_INTERRUPT_SMI;
3213 qemu_mutex_unlock_iothread();
3214 DPRINTF("injected SMI\n");
3215 ret = kvm_vcpu_ioctl(cpu, KVM_SMI);
3216 if (ret < 0) {
3217 fprintf(stderr, "KVM: injection failed, SMI lost (%s)\n",
3218 strerror(-ret));
3219 }
3220 }
3221 }
3222
3223 if (!kvm_pic_in_kernel()) {
3224 qemu_mutex_lock_iothread();
3225 }
3226
3227 /* Force the VCPU out of its inner loop to process any INIT requests
3228 * or (for userspace APIC, but it is cheap to combine the checks here)
3229 * pending TPR access reports.
3230 */
3231 if (cpu->interrupt_request & (CPU_INTERRUPT_INIT | CPU_INTERRUPT_TPR)) {
3232 if ((cpu->interrupt_request & CPU_INTERRUPT_INIT) &&
3233 !(env->hflags & HF_SMM_MASK)) {
3234 cpu->exit_request = 1;
3235 }
3236 if (cpu->interrupt_request & CPU_INTERRUPT_TPR) {
3237 cpu->exit_request = 1;
3238 }
3239 }
3240
3241 if (!kvm_pic_in_kernel()) {
3242 /* Try to inject an interrupt if the guest can accept it */
3243 if (run->ready_for_interrupt_injection &&
3244 (cpu->interrupt_request & CPU_INTERRUPT_HARD) &&
3245 (env->eflags & IF_MASK)) {
3246 int irq;
3247
3248 cpu->interrupt_request &= ~CPU_INTERRUPT_HARD;
3249 irq = cpu_get_pic_interrupt(env);
3250 if (irq >= 0) {
3251 struct kvm_interrupt intr;
3252
3253 intr.irq = irq;
3254 DPRINTF("injected interrupt %d\n", irq);
3255 ret = kvm_vcpu_ioctl(cpu, KVM_INTERRUPT, &intr);
3256 if (ret < 0) {
3257 fprintf(stderr,
3258 "KVM: injection failed, interrupt lost (%s)\n",
3259 strerror(-ret));
3260 }
3261 }
3262 }
3263
3264 /* If we have an interrupt but the guest is not ready to receive an
3265 * interrupt, request an interrupt window exit. This will
3266 * cause a return to userspace as soon as the guest is ready to
3267 * receive interrupts. */
3268 if ((cpu->interrupt_request & CPU_INTERRUPT_HARD)) {
3269 run->request_interrupt_window = 1;
3270 } else {
3271 run->request_interrupt_window = 0;
3272 }
3273
3274 DPRINTF("setting tpr\n");
3275 run->cr8 = cpu_get_apic_tpr(x86_cpu->apic_state);
3276
3277 qemu_mutex_unlock_iothread();
3278 }
3279 }
3280
3281 MemTxAttrs kvm_arch_post_run(CPUState *cpu, struct kvm_run *run)
3282 {
3283 X86CPU *x86_cpu = X86_CPU(cpu);
3284 CPUX86State *env = &x86_cpu->env;
3285
3286 if (run->flags & KVM_RUN_X86_SMM) {
3287 env->hflags |= HF_SMM_MASK;
3288 } else {
3289 env->hflags &= ~HF_SMM_MASK;
3290 }
3291 if (run->if_flag) {
3292 env->eflags |= IF_MASK;
3293 } else {
3294 env->eflags &= ~IF_MASK;
3295 }
3296
3297 /* We need to protect the apic state against concurrent accesses from
3298 * different threads in case the userspace irqchip is used. */
3299 if (!kvm_irqchip_in_kernel()) {
3300 qemu_mutex_lock_iothread();
3301 }
3302 cpu_set_apic_tpr(x86_cpu->apic_state, run->cr8);
3303 cpu_set_apic_base(x86_cpu->apic_state, run->apic_base);
3304 if (!kvm_irqchip_in_kernel()) {
3305 qemu_mutex_unlock_iothread();
3306 }
3307 return cpu_get_mem_attrs(env);
3308 }
3309
3310 int kvm_arch_process_async_events(CPUState *cs)
3311 {
3312 X86CPU *cpu = X86_CPU(cs);
3313 CPUX86State *env = &cpu->env;
3314
3315 if (cs->interrupt_request & CPU_INTERRUPT_MCE) {
3316 /* We must not raise CPU_INTERRUPT_MCE if it's not supported. */
3317 assert(env->mcg_cap);
3318
3319 cs->interrupt_request &= ~CPU_INTERRUPT_MCE;
3320
3321 kvm_cpu_synchronize_state(cs);
3322
3323 if (env->exception_injected == EXCP08_DBLE) {
3324 /* this means triple fault */
3325 qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET);
3326 cs->exit_request = 1;
3327 return 0;
3328 }
3329 env->exception_injected = EXCP12_MCHK;
3330 env->has_error_code = 0;
3331
3332 cs->halted = 0;
3333 if (kvm_irqchip_in_kernel() && env->mp_state == KVM_MP_STATE_HALTED) {
3334 env->mp_state = KVM_MP_STATE_RUNNABLE;
3335 }
3336 }
3337
3338 if ((cs->interrupt_request & CPU_INTERRUPT_INIT) &&
3339 !(env->hflags & HF_SMM_MASK)) {
3340 kvm_cpu_synchronize_state(cs);
3341 do_cpu_init(cpu);
3342 }
3343
3344 if (kvm_irqchip_in_kernel()) {
3345 return 0;
3346 }
3347
3348 if (cs->interrupt_request & CPU_INTERRUPT_POLL) {
3349 cs->interrupt_request &= ~CPU_INTERRUPT_POLL;
3350 apic_poll_irq(cpu->apic_state);
3351 }
3352 if (((cs->interrupt_request & CPU_INTERRUPT_HARD) &&
3353 (env->eflags & IF_MASK)) ||
3354 (cs->interrupt_request & CPU_INTERRUPT_NMI)) {
3355 cs->halted = 0;
3356 }
3357 if (cs->interrupt_request & CPU_INTERRUPT_SIPI) {
3358 kvm_cpu_synchronize_state(cs);
3359 do_cpu_sipi(cpu);
3360 }
3361 if (cs->interrupt_request & CPU_INTERRUPT_TPR) {
3362 cs->interrupt_request &= ~CPU_INTERRUPT_TPR;
3363 kvm_cpu_synchronize_state(cs);
3364 apic_handle_tpr_access_report(cpu->apic_state, env->eip,
3365 env->tpr_access_type);
3366 }
3367
3368 return cs->halted;
3369 }
3370
3371 static int kvm_handle_halt(X86CPU *cpu)
3372 {
3373 CPUState *cs = CPU(cpu);
3374 CPUX86State *env = &cpu->env;
3375
3376 if (!((cs->interrupt_request & CPU_INTERRUPT_HARD) &&
3377 (env->eflags & IF_MASK)) &&
3378 !(cs->interrupt_request & CPU_INTERRUPT_NMI)) {
3379 cs->halted = 1;
3380 return EXCP_HLT;
3381 }
3382
3383 return 0;
3384 }
3385
3386 static int kvm_handle_tpr_access(X86CPU *cpu)
3387 {
3388 CPUState *cs = CPU(cpu);
3389 struct kvm_run *run = cs->kvm_run;
3390
3391 apic_handle_tpr_access_report(cpu->apic_state, run->tpr_access.rip,
3392 run->tpr_access.is_write ? TPR_ACCESS_WRITE
3393 : TPR_ACCESS_READ);
3394 return 1;
3395 }
3396
3397 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
3398 {
3399 static const uint8_t int3 = 0xcc;
3400
3401 if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 1, 0) ||
3402 cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&int3, 1, 1)) {
3403 return -EINVAL;
3404 }
3405 return 0;
3406 }
3407
3408 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
3409 {
3410 uint8_t int3;
3411
3412 if (cpu_memory_rw_debug(cs, bp->pc, &int3, 1, 0) || int3 != 0xcc ||
3413 cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 1, 1)) {
3414 return -EINVAL;
3415 }
3416 return 0;
3417 }
3418
3419 static struct {
3420 target_ulong addr;
3421 int len;
3422 int type;
3423 } hw_breakpoint[4];
3424
3425 static int nb_hw_breakpoint;
3426
3427 static int find_hw_breakpoint(target_ulong addr, int len, int type)
3428 {
3429 int n;
3430
3431 for (n = 0; n < nb_hw_breakpoint; n++) {
3432 if (hw_breakpoint[n].addr == addr && hw_breakpoint[n].type == type &&
3433 (hw_breakpoint[n].len == len || len == -1)) {
3434 return n;
3435 }
3436 }
3437 return -1;
3438 }
3439
3440 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
3441 target_ulong len, int type)
3442 {
3443 switch (type) {
3444 case GDB_BREAKPOINT_HW:
3445 len = 1;
3446 break;
3447 case GDB_WATCHPOINT_WRITE:
3448 case GDB_WATCHPOINT_ACCESS:
3449 switch (len) {
3450 case 1:
3451 break;
3452 case 2:
3453 case 4:
3454 case 8:
3455 if (addr & (len - 1)) {
3456 return -EINVAL;
3457 }
3458 break;
3459 default:
3460 return -EINVAL;
3461 }
3462 break;
3463 default:
3464 return -ENOSYS;
3465 }
3466
3467 if (nb_hw_breakpoint == 4) {
3468 return -ENOBUFS;
3469 }
3470 if (find_hw_breakpoint(addr, len, type) >= 0) {
3471 return -EEXIST;
3472 }
3473 hw_breakpoint[nb_hw_breakpoint].addr = addr;
3474 hw_breakpoint[nb_hw_breakpoint].len = len;
3475 hw_breakpoint[nb_hw_breakpoint].type = type;
3476 nb_hw_breakpoint++;
3477
3478 return 0;
3479 }
3480
3481 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
3482 target_ulong len, int type)
3483 {
3484 int n;
3485
3486 n = find_hw_breakpoint(addr, (type == GDB_BREAKPOINT_HW) ? 1 : len, type);
3487 if (n < 0) {
3488 return -ENOENT;
3489 }
3490 nb_hw_breakpoint--;
3491 hw_breakpoint[n] = hw_breakpoint[nb_hw_breakpoint];
3492
3493 return 0;
3494 }
3495
3496 void kvm_arch_remove_all_hw_breakpoints(void)
3497 {
3498 nb_hw_breakpoint = 0;
3499 }
3500
3501 static CPUWatchpoint hw_watchpoint;
3502
3503 static int kvm_handle_debug(X86CPU *cpu,
3504 struct kvm_debug_exit_arch *arch_info)
3505 {
3506 CPUState *cs = CPU(cpu);
3507 CPUX86State *env = &cpu->env;
3508 int ret = 0;
3509 int n;
3510
3511 if (arch_info->exception == 1) {
3512 if (arch_info->dr6 & (1 << 14)) {
3513 if (cs->singlestep_enabled) {
3514 ret = EXCP_DEBUG;
3515 }
3516 } else {
3517 for (n = 0; n < 4; n++) {
3518 if (arch_info->dr6 & (1 << n)) {
3519 switch ((arch_info->dr7 >> (16 + n*4)) & 0x3) {
3520 case 0x0:
3521 ret = EXCP_DEBUG;
3522 break;
3523 case 0x1:
3524 ret = EXCP_DEBUG;
3525 cs->watchpoint_hit = &hw_watchpoint;
3526 hw_watchpoint.vaddr = hw_breakpoint[n].addr;
3527 hw_watchpoint.flags = BP_MEM_WRITE;
3528 break;
3529 case 0x3:
3530 ret = EXCP_DEBUG;
3531 cs->watchpoint_hit = &hw_watchpoint;
3532 hw_watchpoint.vaddr = hw_breakpoint[n].addr;
3533 hw_watchpoint.flags = BP_MEM_ACCESS;
3534 break;
3535 }
3536 }
3537 }
3538 }
3539 } else if (kvm_find_sw_breakpoint(cs, arch_info->pc)) {
3540 ret = EXCP_DEBUG;
3541 }
3542 if (ret == 0) {
3543 cpu_synchronize_state(cs);
3544 assert(env->exception_injected == -1);
3545
3546 /* pass to guest */
3547 env->exception_injected = arch_info->exception;
3548 env->has_error_code = 0;
3549 }
3550
3551 return ret;
3552 }
3553
3554 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
3555 {
3556 const uint8_t type_code[] = {
3557 [GDB_BREAKPOINT_HW] = 0x0,
3558 [GDB_WATCHPOINT_WRITE] = 0x1,
3559 [GDB_WATCHPOINT_ACCESS] = 0x3
3560 };
3561 const uint8_t len_code[] = {
3562 [1] = 0x0, [2] = 0x1, [4] = 0x3, [8] = 0x2
3563 };
3564 int n;
3565
3566 if (kvm_sw_breakpoints_active(cpu)) {
3567 dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
3568 }
3569 if (nb_hw_breakpoint > 0) {
3570 dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
3571 dbg->arch.debugreg[7] = 0x0600;
3572 for (n = 0; n < nb_hw_breakpoint; n++) {
3573 dbg->arch.debugreg[n] = hw_breakpoint[n].addr;
3574 dbg->arch.debugreg[7] |= (2 << (n * 2)) |
3575 (type_code[hw_breakpoint[n].type] << (16 + n*4)) |
3576 ((uint32_t)len_code[hw_breakpoint[n].len] << (18 + n*4));
3577 }
3578 }
3579 }
3580
3581 static bool host_supports_vmx(void)
3582 {
3583 uint32_t ecx, unused;
3584
3585 host_cpuid(1, 0, &unused, &unused, &ecx, &unused);
3586 return ecx & CPUID_EXT_VMX;
3587 }
3588
3589 #define VMX_INVALID_GUEST_STATE 0x80000021
3590
3591 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
3592 {
3593 X86CPU *cpu = X86_CPU(cs);
3594 uint64_t code;
3595 int ret;
3596
3597 switch (run->exit_reason) {
3598 case KVM_EXIT_HLT:
3599 DPRINTF("handle_hlt\n");
3600 qemu_mutex_lock_iothread();
3601 ret = kvm_handle_halt(cpu);
3602 qemu_mutex_unlock_iothread();
3603 break;
3604 case KVM_EXIT_SET_TPR:
3605 ret = 0;
3606 break;
3607 case KVM_EXIT_TPR_ACCESS:
3608 qemu_mutex_lock_iothread();
3609 ret = kvm_handle_tpr_access(cpu);
3610 qemu_mutex_unlock_iothread();
3611 break;
3612 case KVM_EXIT_FAIL_ENTRY:
3613 code = run->fail_entry.hardware_entry_failure_reason;
3614 fprintf(stderr, "KVM: entry failed, hardware error 0x%" PRIx64 "\n",
3615 code);
3616 if (host_supports_vmx() && code == VMX_INVALID_GUEST_STATE) {
3617 fprintf(stderr,
3618 "\nIf you're running a guest on an Intel machine without "
3619 "unrestricted mode\n"
3620 "support, the failure can be most likely due to the guest "
3621 "entering an invalid\n"
3622 "state for Intel VT. For example, the guest maybe running "
3623 "in big real mode\n"
3624 "which is not supported on less recent Intel processors."
3625 "\n\n");
3626 }
3627 ret = -1;
3628 break;
3629 case KVM_EXIT_EXCEPTION:
3630 fprintf(stderr, "KVM: exception %d exit (error code 0x%x)\n",
3631 run->ex.exception, run->ex.error_code);
3632 ret = -1;
3633 break;
3634 case KVM_EXIT_DEBUG:
3635 DPRINTF("kvm_exit_debug\n");
3636 qemu_mutex_lock_iothread();
3637 ret = kvm_handle_debug(cpu, &run->debug.arch);
3638 qemu_mutex_unlock_iothread();
3639 break;
3640 case KVM_EXIT_HYPERV:
3641 ret = kvm_hv_handle_exit(cpu, &run->hyperv);
3642 break;
3643 case KVM_EXIT_IOAPIC_EOI:
3644 ioapic_eoi_broadcast(run->eoi.vector);
3645 ret = 0;
3646 break;
3647 default:
3648 fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
3649 ret = -1;
3650 break;
3651 }
3652
3653 return ret;
3654 }
3655
3656 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
3657 {
3658 X86CPU *cpu = X86_CPU(cs);
3659 CPUX86State *env = &cpu->env;
3660
3661 kvm_cpu_synchronize_state(cs);
3662 return !(env->cr[0] & CR0_PE_MASK) ||
3663 ((env->segs[R_CS].selector & 3) != 3);
3664 }
3665
3666 void kvm_arch_init_irq_routing(KVMState *s)
3667 {
3668 if (!kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
3669 /* If kernel can't do irq routing, interrupt source
3670 * override 0->2 cannot be set up as required by HPET.
3671 * So we have to disable it.
3672 */
3673 no_hpet = 1;
3674 }
3675 /* We know at this point that we're using the in-kernel
3676 * irqchip, so we can use irqfds, and on x86 we know
3677 * we can use msi via irqfd and GSI routing.
3678 */
3679 kvm_msi_via_irqfd_allowed = true;
3680 kvm_gsi_routing_allowed = true;
3681
3682 if (kvm_irqchip_is_split()) {
3683 int i;
3684
3685 /* If the ioapic is in QEMU and the lapics are in KVM, reserve
3686 MSI routes for signaling interrupts to the local apics. */
3687 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
3688 if (kvm_irqchip_add_msi_route(s, 0, NULL) < 0) {
3689 error_report("Could not enable split IRQ mode.");
3690 exit(1);
3691 }
3692 }
3693 }
3694 }
3695
3696 int kvm_arch_irqchip_create(MachineState *ms, KVMState *s)
3697 {
3698 int ret;
3699 if (machine_kernel_irqchip_split(ms)) {
3700 ret = kvm_vm_enable_cap(s, KVM_CAP_SPLIT_IRQCHIP, 0, 24);
3701 if (ret) {
3702 error_report("Could not enable split irqchip mode: %s",
3703 strerror(-ret));
3704 exit(1);
3705 } else {
3706 DPRINTF("Enabled KVM_CAP_SPLIT_IRQCHIP\n");
3707 kvm_split_irqchip = true;
3708 return 1;
3709 }
3710 } else {
3711 return 0;
3712 }
3713 }
3714
3715 /* Classic KVM device assignment interface. Will remain x86 only. */
3716 int kvm_device_pci_assign(KVMState *s, PCIHostDeviceAddress *dev_addr,
3717 uint32_t flags, uint32_t *dev_id)
3718 {
3719 struct kvm_assigned_pci_dev dev_data = {
3720 .segnr = dev_addr->domain,
3721 .busnr = dev_addr->bus,
3722 .devfn = PCI_DEVFN(dev_addr->slot, dev_addr->function),
3723 .flags = flags,
3724 };
3725 int ret;
3726
3727 dev_data.assigned_dev_id =
3728 (dev_addr->domain << 16) | (dev_addr->bus << 8) | dev_data.devfn;
3729
3730 ret = kvm_vm_ioctl(s, KVM_ASSIGN_PCI_DEVICE, &dev_data);
3731 if (ret < 0) {
3732 return ret;
3733 }
3734
3735 *dev_id = dev_data.assigned_dev_id;
3736
3737 return 0;
3738 }
3739
3740 int kvm_device_pci_deassign(KVMState *s, uint32_t dev_id)
3741 {
3742 struct kvm_assigned_pci_dev dev_data = {
3743 .assigned_dev_id = dev_id,
3744 };
3745
3746 return kvm_vm_ioctl(s, KVM_DEASSIGN_PCI_DEVICE, &dev_data);
3747 }
3748
3749 static int kvm_assign_irq_internal(KVMState *s, uint32_t dev_id,
3750 uint32_t irq_type, uint32_t guest_irq)
3751 {
3752 struct kvm_assigned_irq assigned_irq = {
3753 .assigned_dev_id = dev_id,
3754 .guest_irq = guest_irq,
3755 .flags = irq_type,
3756 };
3757
3758 if (kvm_check_extension(s, KVM_CAP_ASSIGN_DEV_IRQ)) {
3759 return kvm_vm_ioctl(s, KVM_ASSIGN_DEV_IRQ, &assigned_irq);
3760 } else {
3761 return kvm_vm_ioctl(s, KVM_ASSIGN_IRQ, &assigned_irq);
3762 }
3763 }
3764
3765 int kvm_device_intx_assign(KVMState *s, uint32_t dev_id, bool use_host_msi,
3766 uint32_t guest_irq)
3767 {
3768 uint32_t irq_type = KVM_DEV_IRQ_GUEST_INTX |
3769 (use_host_msi ? KVM_DEV_IRQ_HOST_MSI : KVM_DEV_IRQ_HOST_INTX);
3770
3771 return kvm_assign_irq_internal(s, dev_id, irq_type, guest_irq);
3772 }
3773
3774 int kvm_device_intx_set_mask(KVMState *s, uint32_t dev_id, bool masked)
3775 {
3776 struct kvm_assigned_pci_dev dev_data = {
3777 .assigned_dev_id = dev_id,
3778 .flags = masked ? KVM_DEV_ASSIGN_MASK_INTX : 0,
3779 };
3780
3781 return kvm_vm_ioctl(s, KVM_ASSIGN_SET_INTX_MASK, &dev_data);
3782 }
3783
3784 static int kvm_deassign_irq_internal(KVMState *s, uint32_t dev_id,
3785 uint32_t type)
3786 {
3787 struct kvm_assigned_irq assigned_irq = {
3788 .assigned_dev_id = dev_id,
3789 .flags = type,
3790 };
3791
3792 return kvm_vm_ioctl(s, KVM_DEASSIGN_DEV_IRQ, &assigned_irq);
3793 }
3794
3795 int kvm_device_intx_deassign(KVMState *s, uint32_t dev_id, bool use_host_msi)
3796 {
3797 return kvm_deassign_irq_internal(s, dev_id, KVM_DEV_IRQ_GUEST_INTX |
3798 (use_host_msi ? KVM_DEV_IRQ_HOST_MSI : KVM_DEV_IRQ_HOST_INTX));
3799 }
3800
3801 int kvm_device_msi_assign(KVMState *s, uint32_t dev_id, int virq)
3802 {
3803 return kvm_assign_irq_internal(s, dev_id, KVM_DEV_IRQ_HOST_MSI |
3804 KVM_DEV_IRQ_GUEST_MSI, virq);
3805 }
3806
3807 int kvm_device_msi_deassign(KVMState *s, uint32_t dev_id)
3808 {
3809 return kvm_deassign_irq_internal(s, dev_id, KVM_DEV_IRQ_GUEST_MSI |
3810 KVM_DEV_IRQ_HOST_MSI);
3811 }
3812
3813 bool kvm_device_msix_supported(KVMState *s)
3814 {
3815 /* The kernel lacks a corresponding KVM_CAP, so we probe by calling
3816 * KVM_ASSIGN_SET_MSIX_NR with an invalid parameter. */
3817 return kvm_vm_ioctl(s, KVM_ASSIGN_SET_MSIX_NR, NULL) == -EFAULT;
3818 }
3819
3820 int kvm_device_msix_init_vectors(KVMState *s, uint32_t dev_id,
3821 uint32_t nr_vectors)
3822 {
3823 struct kvm_assigned_msix_nr msix_nr = {
3824 .assigned_dev_id = dev_id,
3825 .entry_nr = nr_vectors,
3826 };
3827
3828 return kvm_vm_ioctl(s, KVM_ASSIGN_SET_MSIX_NR, &msix_nr);
3829 }
3830
3831 int kvm_device_msix_set_vector(KVMState *s, uint32_t dev_id, uint32_t vector,
3832 int virq)
3833 {
3834 struct kvm_assigned_msix_entry msix_entry = {
3835 .assigned_dev_id = dev_id,
3836 .gsi = virq,
3837 .entry = vector,
3838 };
3839
3840 return kvm_vm_ioctl(s, KVM_ASSIGN_SET_MSIX_ENTRY, &msix_entry);
3841 }
3842
3843 int kvm_device_msix_assign(KVMState *s, uint32_t dev_id)
3844 {
3845 return kvm_assign_irq_internal(s, dev_id, KVM_DEV_IRQ_HOST_MSIX |
3846 KVM_DEV_IRQ_GUEST_MSIX, 0);
3847 }
3848
3849 int kvm_device_msix_deassign(KVMState *s, uint32_t dev_id)
3850 {
3851 return kvm_deassign_irq_internal(s, dev_id, KVM_DEV_IRQ_GUEST_MSIX |
3852 KVM_DEV_IRQ_HOST_MSIX);
3853 }
3854
3855 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
3856 uint64_t address, uint32_t data, PCIDevice *dev)
3857 {
3858 X86IOMMUState *iommu = x86_iommu_get_default();
3859
3860 if (iommu) {
3861 int ret;
3862 MSIMessage src, dst;
3863 X86IOMMUClass *class = X86_IOMMU_GET_CLASS(iommu);
3864
3865 if (!class->int_remap) {
3866 return 0;
3867 }
3868
3869 src.address = route->u.msi.address_hi;
3870 src.address <<= VTD_MSI_ADDR_HI_SHIFT;
3871 src.address |= route->u.msi.address_lo;
3872 src.data = route->u.msi.data;
3873
3874 ret = class->int_remap(iommu, &src, &dst, dev ? \
3875 pci_requester_id(dev) : \
3876 X86_IOMMU_SID_INVALID);
3877 if (ret) {
3878 trace_kvm_x86_fixup_msi_error(route->gsi);
3879 return 1;
3880 }
3881
3882 route->u.msi.address_hi = dst.address >> VTD_MSI_ADDR_HI_SHIFT;
3883 route->u.msi.address_lo = dst.address & VTD_MSI_ADDR_LO_MASK;
3884 route->u.msi.data = dst.data;
3885 }
3886
3887 return 0;
3888 }
3889
3890 typedef struct MSIRouteEntry MSIRouteEntry;
3891
3892 struct MSIRouteEntry {
3893 PCIDevice *dev; /* Device pointer */
3894 int vector; /* MSI/MSIX vector index */
3895 int virq; /* Virtual IRQ index */
3896 QLIST_ENTRY(MSIRouteEntry) list;
3897 };
3898
3899 /* List of used GSI routes */
3900 static QLIST_HEAD(, MSIRouteEntry) msi_route_list = \
3901 QLIST_HEAD_INITIALIZER(msi_route_list);
3902
3903 static void kvm_update_msi_routes_all(void *private, bool global,
3904 uint32_t index, uint32_t mask)
3905 {
3906 int cnt = 0, vector;
3907 MSIRouteEntry *entry;
3908 MSIMessage msg;
3909 PCIDevice *dev;
3910
3911 /* TODO: explicit route update */
3912 QLIST_FOREACH(entry, &msi_route_list, list) {
3913 cnt++;
3914 vector = entry->vector;
3915 dev = entry->dev;
3916 if (msix_enabled(dev) && !msix_is_masked(dev, vector)) {
3917 msg = msix_get_message(dev, vector);
3918 } else if (msi_enabled(dev) && !msi_is_masked(dev, vector)) {
3919 msg = msi_get_message(dev, vector);
3920 } else {
3921 /*
3922 * Either MSI/MSIX is disabled for the device, or the
3923 * specific message was masked out. Skip this one.
3924 */
3925 continue;
3926 }
3927 kvm_irqchip_update_msi_route(kvm_state, entry->virq, msg, dev);
3928 }
3929 kvm_irqchip_commit_routes(kvm_state);
3930 trace_kvm_x86_update_msi_routes(cnt);
3931 }
3932
3933 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
3934 int vector, PCIDevice *dev)
3935 {
3936 static bool notify_list_inited = false;
3937 MSIRouteEntry *entry;
3938
3939 if (!dev) {
3940 /* These are (possibly) IOAPIC routes only used for split
3941 * kernel irqchip mode, while what we are housekeeping are
3942 * PCI devices only. */
3943 return 0;
3944 }
3945
3946 entry = g_new0(MSIRouteEntry, 1);
3947 entry->dev = dev;
3948 entry->vector = vector;
3949 entry->virq = route->gsi;
3950 QLIST_INSERT_HEAD(&msi_route_list, entry, list);
3951
3952 trace_kvm_x86_add_msi_route(route->gsi);
3953
3954 if (!notify_list_inited) {
3955 /* For the first time we do add route, add ourselves into
3956 * IOMMU's IEC notify list if needed. */
3957 X86IOMMUState *iommu = x86_iommu_get_default();
3958 if (iommu) {
3959 x86_iommu_iec_register_notifier(iommu,
3960 kvm_update_msi_routes_all,
3961 NULL);
3962 }
3963 notify_list_inited = true;
3964 }
3965 return 0;
3966 }
3967
3968 int kvm_arch_release_virq_post(int virq)
3969 {
3970 MSIRouteEntry *entry, *next;
3971 QLIST_FOREACH_SAFE(entry, &msi_route_list, list, next) {
3972 if (entry->virq == virq) {
3973 trace_kvm_x86_remove_msi_route(virq);
3974 QLIST_REMOVE(entry, list);
3975 g_free(entry);
3976 break;
3977 }
3978 }
3979 return 0;
3980 }
3981
3982 int kvm_arch_msi_data_to_gsi(uint32_t data)
3983 {
3984 abort();
3985 }
QEmu