?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
qapi: drop the sentinel in enum array
[qemu/armbru.git] / target / i386 / kvm.c
blob6db7783edca2c3708a7a0ebc08a7b2418e652e59
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 <linux/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
31 #include "exec/gdbstub.h"
32 #include "qemu/host-utils.h"
33 #include "qemu/config-file.h"
34 #include "qemu/error-report.h"
35 #include "hw/i386/pc.h"
36 #include "hw/i386/apic.h"
37 #include "hw/i386/apic_internal.h"
38 #include "hw/i386/apic-msidef.h"
39 #include "hw/i386/intel_iommu.h"
40 #include "hw/i386/x86-iommu.h"
41
42 #include "exec/ioport.h"
43 #include "standard-headers/asm-x86/hyperv.h"
44 #include "hw/pci/pci.h"
45 #include "hw/pci/msi.h"
46 #include "hw/pci/msix.h"
47 #include "migration/blocker.h"
48 #include "exec/memattrs.h"
49 #include "trace.h"
50
51 //#define DEBUG_KVM
52
53 #ifdef DEBUG_KVM
54 #define DPRINTF(fmt, ...) \
55 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
56 #else
57 #define DPRINTF(fmt, ...) \
58 do { } while (0)
59 #endif
60
61 #define MSR_KVM_WALL_CLOCK 0x11
62 #define MSR_KVM_SYSTEM_TIME 0x12
63
64 /* A 4096-byte buffer can hold the 8-byte kvm_msrs header, plus
65 * 255 kvm_msr_entry structs */
66 #define MSR_BUF_SIZE 4096
67
68 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
69 KVM_CAP_INFO(SET_TSS_ADDR),
70 KVM_CAP_INFO(EXT_CPUID),
71 KVM_CAP_INFO(MP_STATE),
72 KVM_CAP_LAST_INFO
73 };
74
75 static bool has_msr_star;
76 static bool has_msr_hsave_pa;
77 static bool has_msr_tsc_aux;
78 static bool has_msr_tsc_adjust;
79 static bool has_msr_tsc_deadline;
80 static bool has_msr_feature_control;
81 static bool has_msr_misc_enable;
82 static bool has_msr_smbase;
83 static bool has_msr_bndcfgs;
84 static int lm_capable_kernel;
85 static bool has_msr_hv_hypercall;
86 static bool has_msr_hv_crash;
87 static bool has_msr_hv_reset;
88 static bool has_msr_hv_vpindex;
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_xss;
93
94 static bool has_msr_architectural_pmu;
95 static uint32_t num_architectural_pmu_counters;
96
97 static int has_xsave;
98 static int has_xcrs;
99 static int has_pit_state2;
100
101 static bool has_msr_mcg_ext_ctl;
102
103 static struct kvm_cpuid2 *cpuid_cache;
104
105 int kvm_has_pit_state2(void)
106 {
107 return has_pit_state2;
108 }
109
110 bool kvm_has_smm(void)
111 {
112 return kvm_check_extension(kvm_state, KVM_CAP_X86_SMM);
113 }
114
115 bool kvm_has_adjust_clock_stable(void)
116 {
117 int ret = kvm_check_extension(kvm_state, KVM_CAP_ADJUST_CLOCK);
118
119 return (ret == KVM_CLOCK_TSC_STABLE);
120 }
121
122 bool kvm_allows_irq0_override(void)
123 {
124 return !kvm_irqchip_in_kernel() || kvm_has_gsi_routing();
125 }
126
127 static bool kvm_x2apic_api_set_flags(uint64_t flags)
128 {
129 KVMState *s = KVM_STATE(current_machine->accelerator);
130
131 return !kvm_vm_enable_cap(s, KVM_CAP_X2APIC_API, 0, flags);
132 }
133
134 #define MEMORIZE(fn, _result) \
135 ({ \
136 static bool _memorized; \
137 \
138 if (_memorized) { \
139 return _result; \
140 } \
141 _memorized = true; \
142 _result = fn; \
143 })
144
145 static bool has_x2apic_api;
146
147 bool kvm_has_x2apic_api(void)
148 {
149 return has_x2apic_api;
150 }
151
152 bool kvm_enable_x2apic(void)
153 {
154 return MEMORIZE(
155 kvm_x2apic_api_set_flags(KVM_X2APIC_API_USE_32BIT_IDS |
156 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK),
157 has_x2apic_api);
158 }
159
160 static int kvm_get_tsc(CPUState *cs)
161 {
162 X86CPU *cpu = X86_CPU(cs);
163 CPUX86State *env = &cpu->env;
164 struct {
165 struct kvm_msrs info;
166 struct kvm_msr_entry entries[1];
167 } msr_data;
168 int ret;
169
170 if (env->tsc_valid) {
171 return 0;
172 }
173
174 msr_data.info.nmsrs = 1;
175 msr_data.entries[0].index = MSR_IA32_TSC;
176 env->tsc_valid = !runstate_is_running();
177
178 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MSRS, &msr_data);
179 if (ret < 0) {
180 return ret;
181 }
182
183 assert(ret == 1);
184 env->tsc = msr_data.entries[0].data;
185 return 0;
186 }
187
188 static inline void do_kvm_synchronize_tsc(CPUState *cpu, run_on_cpu_data arg)
189 {
190 kvm_get_tsc(cpu);
191 }
192
193 void kvm_synchronize_all_tsc(void)
194 {
195 CPUState *cpu;
196
197 if (kvm_enabled()) {
198 CPU_FOREACH(cpu) {
199 run_on_cpu(cpu, do_kvm_synchronize_tsc, RUN_ON_CPU_NULL);
200 }
201 }
202 }
203
204 static struct kvm_cpuid2 *try_get_cpuid(KVMState *s, int max)
205 {
206 struct kvm_cpuid2 *cpuid;
207 int r, size;
208
209 size = sizeof(*cpuid) + max * sizeof(*cpuid->entries);
210 cpuid = g_malloc0(size);
211 cpuid->nent = max;
212 r = kvm_ioctl(s, KVM_GET_SUPPORTED_CPUID, cpuid);
213 if (r == 0 && cpuid->nent >= max) {
214 r = -E2BIG;
215 }
216 if (r < 0) {
217 if (r == -E2BIG) {
218 g_free(cpuid);
219 return NULL;
220 } else {
221 fprintf(stderr, "KVM_GET_SUPPORTED_CPUID failed: %s\n",
222 strerror(-r));
223 exit(1);
224 }
225 }
226 return cpuid;
227 }
228
229 /* Run KVM_GET_SUPPORTED_CPUID ioctl(), allocating a buffer large enough
230 * for all entries.
231 */
232 static struct kvm_cpuid2 *get_supported_cpuid(KVMState *s)
233 {
234 struct kvm_cpuid2 *cpuid;
235 int max = 1;
236
237 if (cpuid_cache != NULL) {
238 return cpuid_cache;
239 }
240 while ((cpuid = try_get_cpuid(s, max)) == NULL) {
241 max *= 2;
242 }
243 cpuid_cache = cpuid;
244 return cpuid;
245 }
246
247 static const struct kvm_para_features {
248 int cap;
249 int feature;
250 } para_features[] = {
251 { KVM_CAP_CLOCKSOURCE, KVM_FEATURE_CLOCKSOURCE },
252 { KVM_CAP_NOP_IO_DELAY, KVM_FEATURE_NOP_IO_DELAY },
253 { KVM_CAP_PV_MMU, KVM_FEATURE_MMU_OP },
254 { KVM_CAP_ASYNC_PF, KVM_FEATURE_ASYNC_PF },
255 };
256
257 static int get_para_features(KVMState *s)
258 {
259 int i, features = 0;
260
261 for (i = 0; i < ARRAY_SIZE(para_features); i++) {
262 if (kvm_check_extension(s, para_features[i].cap)) {
263 features |= (1 << para_features[i].feature);
264 }
265 }
266
267 return features;
268 }
269
270 static bool host_tsx_blacklisted(void)
271 {
272 int family, model, stepping;\
273 char vendor[CPUID_VENDOR_SZ + 1];
274
275 host_vendor_fms(vendor, &family, &model, &stepping);
276
277 /* Check if we are running on a Haswell host known to have broken TSX */
278 return !strcmp(vendor, CPUID_VENDOR_INTEL) &&
279 (family == 6) &&
280 ((model == 63 && stepping < 4) ||
281 model == 60 || model == 69 || model == 70);
282 }
283
284 /* Returns the value for a specific register on the cpuid entry
285 */
286 static uint32_t cpuid_entry_get_reg(struct kvm_cpuid_entry2 *entry, int reg)
287 {
288 uint32_t ret = 0;
289 switch (reg) {
290 case R_EAX:
291 ret = entry->eax;
292 break;
293 case R_EBX:
294 ret = entry->ebx;
295 break;
296 case R_ECX:
297 ret = entry->ecx;
298 break;
299 case R_EDX:
300 ret = entry->edx;
301 break;
302 }
303 return ret;
304 }
305
306 /* Find matching entry for function/index on kvm_cpuid2 struct
307 */
308 static struct kvm_cpuid_entry2 *cpuid_find_entry(struct kvm_cpuid2 *cpuid,
309 uint32_t function,
310 uint32_t index)
311 {
312 int i;
313 for (i = 0; i < cpuid->nent; ++i) {
314 if (cpuid->entries[i].function == function &&
315 cpuid->entries[i].index == index) {
316 return &cpuid->entries[i];
317 }
318 }
319 /* not found: */
320 return NULL;
321 }
322
323 uint32_t kvm_arch_get_supported_cpuid(KVMState *s, uint32_t function,
324 uint32_t index, int reg)
325 {
326 struct kvm_cpuid2 *cpuid;
327 uint32_t ret = 0;
328 uint32_t cpuid_1_edx;
329 bool found = false;
330
331 cpuid = get_supported_cpuid(s);
332
333 struct kvm_cpuid_entry2 *entry = cpuid_find_entry(cpuid, function, index);
334 if (entry) {
335 found = true;
336 ret = cpuid_entry_get_reg(entry, reg);
337 }
338
339 /* Fixups for the data returned by KVM, below */
340
341 if (function == 1 && reg == R_EDX) {
342 /* KVM before 2.6.30 misreports the following features */
343 ret |= CPUID_MTRR | CPUID_PAT | CPUID_MCE | CPUID_MCA;
344 } else if (function == 1 && reg == R_ECX) {
345 /* We can set the hypervisor flag, even if KVM does not return it on
346 * GET_SUPPORTED_CPUID
347 */
348 ret |= CPUID_EXT_HYPERVISOR;
349 /* tsc-deadline flag is not returned by GET_SUPPORTED_CPUID, but it
350 * can be enabled if the kernel has KVM_CAP_TSC_DEADLINE_TIMER,
351 * and the irqchip is in the kernel.
352 */
353 if (kvm_irqchip_in_kernel() &&
354 kvm_check_extension(s, KVM_CAP_TSC_DEADLINE_TIMER)) {
355 ret |= CPUID_EXT_TSC_DEADLINE_TIMER;
356 }
357
358 /* x2apic is reported by GET_SUPPORTED_CPUID, but it can't be enabled
359 * without the in-kernel irqchip
360 */
361 if (!kvm_irqchip_in_kernel()) {
362 ret &= ~CPUID_EXT_X2APIC;
363 }
364 } else if (function == 6 && reg == R_EAX) {
365 ret |= CPUID_6_EAX_ARAT; /* safe to allow because of emulated APIC */
366 } else if (function == 7 && index == 0 && reg == R_EBX) {
367 if (host_tsx_blacklisted()) {
368 ret &= ~(CPUID_7_0_EBX_RTM | CPUID_7_0_EBX_HLE);
369 }
370 } else if (function == 0x80000001 && reg == R_EDX) {
371 /* On Intel, kvm returns cpuid according to the Intel spec,
372 * so add missing bits according to the AMD spec:
373 */
374 cpuid_1_edx = kvm_arch_get_supported_cpuid(s, 1, 0, R_EDX);
375 ret |= cpuid_1_edx & CPUID_EXT2_AMD_ALIASES;
376 } else if (function == KVM_CPUID_FEATURES && reg == R_EAX) {
377 /* kvm_pv_unhalt is reported by GET_SUPPORTED_CPUID, but it can't
378 * be enabled without the in-kernel irqchip
379 */
380 if (!kvm_irqchip_in_kernel()) {
381 ret &= ~(1U << KVM_FEATURE_PV_UNHALT);
382 }
383 }
384
385 /* fallback for older kernels */
386 if ((function == KVM_CPUID_FEATURES) && !found) {
387 ret = get_para_features(s);
388 }
389
390 return ret;
391 }
392
393 typedef struct HWPoisonPage {
394 ram_addr_t ram_addr;
395 QLIST_ENTRY(HWPoisonPage) list;
396 } HWPoisonPage;
397
398 static QLIST_HEAD(, HWPoisonPage) hwpoison_page_list =
399 QLIST_HEAD_INITIALIZER(hwpoison_page_list);
400
401 static void kvm_unpoison_all(void *param)
402 {
403 HWPoisonPage *page, *next_page;
404
405 QLIST_FOREACH_SAFE(page, &hwpoison_page_list, list, next_page) {
406 QLIST_REMOVE(page, list);
407 qemu_ram_remap(page->ram_addr, TARGET_PAGE_SIZE);
408 g_free(page);
409 }
410 }
411
412 static void kvm_hwpoison_page_add(ram_addr_t ram_addr)
413 {
414 HWPoisonPage *page;
415
416 QLIST_FOREACH(page, &hwpoison_page_list, list) {
417 if (page->ram_addr == ram_addr) {
418 return;
419 }
420 }
421 page = g_new(HWPoisonPage, 1);
422 page->ram_addr = ram_addr;
423 QLIST_INSERT_HEAD(&hwpoison_page_list, page, list);
424 }
425
426 static int kvm_get_mce_cap_supported(KVMState *s, uint64_t *mce_cap,
427 int *max_banks)
428 {
429 int r;
430
431 r = kvm_check_extension(s, KVM_CAP_MCE);
432 if (r > 0) {
433 *max_banks = r;
434 return kvm_ioctl(s, KVM_X86_GET_MCE_CAP_SUPPORTED, mce_cap);
435 }
436 return -ENOSYS;
437 }
438
439 static void kvm_mce_inject(X86CPU *cpu, hwaddr paddr, int code)
440 {
441 CPUState *cs = CPU(cpu);
442 CPUX86State *env = &cpu->env;
443 uint64_t status = MCI_STATUS_VAL | MCI_STATUS_UC | MCI_STATUS_EN |
444 MCI_STATUS_MISCV | MCI_STATUS_ADDRV | MCI_STATUS_S;
445 uint64_t mcg_status = MCG_STATUS_MCIP;
446 int flags = 0;
447
448 if (code == BUS_MCEERR_AR) {
449 status |= MCI_STATUS_AR | 0x134;
450 mcg_status |= MCG_STATUS_EIPV;
451 } else {
452 status |= 0xc0;
453 mcg_status |= MCG_STATUS_RIPV;
454 }
455
456 flags = cpu_x86_support_mca_broadcast(env) ? MCE_INJECT_BROADCAST : 0;
457 /* We need to read back the value of MSR_EXT_MCG_CTL that was set by the
458 * guest kernel back into env->mcg_ext_ctl.
459 */
460 cpu_synchronize_state(cs);
461 if (env->mcg_ext_ctl & MCG_EXT_CTL_LMCE_EN) {
462 mcg_status |= MCG_STATUS_LMCE;
463 flags = 0;
464 }
465
466 cpu_x86_inject_mce(NULL, cpu, 9, status, mcg_status, paddr,
467 (MCM_ADDR_PHYS << 6) | 0xc, flags);
468 }
469
470 static void hardware_memory_error(void)
471 {
472 fprintf(stderr, "Hardware memory error!\n");
473 exit(1);
474 }
475
476 void kvm_arch_on_sigbus_vcpu(CPUState *c, int code, void *addr)
477 {
478 X86CPU *cpu = X86_CPU(c);
479 CPUX86State *env = &cpu->env;
480 ram_addr_t ram_addr;
481 hwaddr paddr;
482
483 /* If we get an action required MCE, it has been injected by KVM
484 * while the VM was running. An action optional MCE instead should
485 * be coming from the main thread, which qemu_init_sigbus identifies
486 * as the "early kill" thread.
487 */
488 assert(code == BUS_MCEERR_AR || code == BUS_MCEERR_AO);
489
490 if ((env->mcg_cap & MCG_SER_P) && addr) {
491 ram_addr = qemu_ram_addr_from_host(addr);
492 if (ram_addr != RAM_ADDR_INVALID &&
493 kvm_physical_memory_addr_from_host(c->kvm_state, addr, &paddr)) {
494 kvm_hwpoison_page_add(ram_addr);
495 kvm_mce_inject(cpu, paddr, code);
496 return;
497 }
498
499 fprintf(stderr, "Hardware memory error for memory used by "
500 "QEMU itself instead of guest system!\n");
501 }
502
503 if (code == BUS_MCEERR_AR) {
504 hardware_memory_error();
505 }
506
507 /* Hope we are lucky for AO MCE */
508 }
509
510 static int kvm_inject_mce_oldstyle(X86CPU *cpu)
511 {
512 CPUX86State *env = &cpu->env;
513
514 if (!kvm_has_vcpu_events() && env->exception_injected == EXCP12_MCHK) {
515 unsigned int bank, bank_num = env->mcg_cap & 0xff;
516 struct kvm_x86_mce mce;
517
518 env->exception_injected = -1;
519
520 /*
521 * There must be at least one bank in use if an MCE is pending.
522 * Find it and use its values for the event injection.
523 */
524 for (bank = 0; bank < bank_num; bank++) {
525 if (env->mce_banks[bank * 4 + 1] & MCI_STATUS_VAL) {
526 break;
527 }
528 }
529 assert(bank < bank_num);
530
531 mce.bank = bank;
532 mce.status = env->mce_banks[bank * 4 + 1];
533 mce.mcg_status = env->mcg_status;
534 mce.addr = env->mce_banks[bank * 4 + 2];
535 mce.misc = env->mce_banks[bank * 4 + 3];
536
537 return kvm_vcpu_ioctl(CPU(cpu), KVM_X86_SET_MCE, &mce);
538 }
539 return 0;
540 }
541
542 static void cpu_update_state(void *opaque, int running, RunState state)
543 {
544 CPUX86State *env = opaque;
545
546 if (running) {
547 env->tsc_valid = false;
548 }
549 }
550
551 unsigned long kvm_arch_vcpu_id(CPUState *cs)
552 {
553 X86CPU *cpu = X86_CPU(cs);
554 return cpu->apic_id;
555 }
556
557 #ifndef KVM_CPUID_SIGNATURE_NEXT
558 #define KVM_CPUID_SIGNATURE_NEXT 0x40000100
559 #endif
560
561 static bool hyperv_hypercall_available(X86CPU *cpu)
562 {
563 return cpu->hyperv_vapic ||
564 (cpu->hyperv_spinlock_attempts != HYPERV_SPINLOCK_NEVER_RETRY);
565 }
566
567 static bool hyperv_enabled(X86CPU *cpu)
568 {
569 CPUState *cs = CPU(cpu);
570 return kvm_check_extension(cs->kvm_state, KVM_CAP_HYPERV) > 0 &&
571 (hyperv_hypercall_available(cpu) ||
572 cpu->hyperv_time ||
573 cpu->hyperv_relaxed_timing ||
574 cpu->hyperv_crash ||
575 cpu->hyperv_reset ||
576 cpu->hyperv_vpindex ||
577 cpu->hyperv_runtime ||
578 cpu->hyperv_synic ||
579 cpu->hyperv_stimer);
580 }
581
582 static int kvm_arch_set_tsc_khz(CPUState *cs)
583 {
584 X86CPU *cpu = X86_CPU(cs);
585 CPUX86State *env = &cpu->env;
586 int r;
587
588 if (!env->tsc_khz) {
589 return 0;
590 }
591
592 r = kvm_check_extension(cs->kvm_state, KVM_CAP_TSC_CONTROL) ?
593 kvm_vcpu_ioctl(cs, KVM_SET_TSC_KHZ, env->tsc_khz) :
594 -ENOTSUP;
595 if (r < 0) {
596 /* When KVM_SET_TSC_KHZ fails, it's an error only if the current
597 * TSC frequency doesn't match the one we want.
598 */
599 int cur_freq = kvm_check_extension(cs->kvm_state, KVM_CAP_GET_TSC_KHZ) ?
600 kvm_vcpu_ioctl(cs, KVM_GET_TSC_KHZ) :
601 -ENOTSUP;
602 if (cur_freq <= 0 || cur_freq != env->tsc_khz) {
603 warn_report("TSC frequency mismatch between "
604 "VM (%" PRId64 " kHz) and host (%d kHz), "
605 "and TSC scaling unavailable",
606 env->tsc_khz, cur_freq);
607 return r;
608 }
609 }
610
611 return 0;
612 }
613
614 static int hyperv_handle_properties(CPUState *cs)
615 {
616 X86CPU *cpu = X86_CPU(cs);
617 CPUX86State *env = &cpu->env;
618
619 if (cpu->hyperv_time &&
620 kvm_check_extension(cs->kvm_state, KVM_CAP_HYPERV_TIME) <= 0) {
621 cpu->hyperv_time = false;
622 }
623
624 if (cpu->hyperv_relaxed_timing) {
625 env->features[FEAT_HYPERV_EAX] |= HV_X64_MSR_HYPERCALL_AVAILABLE;
626 }
627 if (cpu->hyperv_vapic) {
628 env->features[FEAT_HYPERV_EAX] |= HV_X64_MSR_HYPERCALL_AVAILABLE;
629 env->features[FEAT_HYPERV_EAX] |= HV_X64_MSR_APIC_ACCESS_AVAILABLE;
630 }
631 if (cpu->hyperv_time) {
632 env->features[FEAT_HYPERV_EAX] |= HV_X64_MSR_HYPERCALL_AVAILABLE;
633 env->features[FEAT_HYPERV_EAX] |= HV_X64_MSR_TIME_REF_COUNT_AVAILABLE;
634 env->features[FEAT_HYPERV_EAX] |= 0x200;
635 }
636 if (cpu->hyperv_crash && has_msr_hv_crash) {
637 env->features[FEAT_HYPERV_EDX] |= HV_X64_GUEST_CRASH_MSR_AVAILABLE;
638 }
639 env->features[FEAT_HYPERV_EDX] |= HV_X64_CPU_DYNAMIC_PARTITIONING_AVAILABLE;
640 if (cpu->hyperv_reset && has_msr_hv_reset) {
641 env->features[FEAT_HYPERV_EAX] |= HV_X64_MSR_RESET_AVAILABLE;
642 }
643 if (cpu->hyperv_vpindex && has_msr_hv_vpindex) {
644 env->features[FEAT_HYPERV_EAX] |= HV_X64_MSR_VP_INDEX_AVAILABLE;
645 }
646 if (cpu->hyperv_runtime && has_msr_hv_runtime) {
647 env->features[FEAT_HYPERV_EAX] |= HV_X64_MSR_VP_RUNTIME_AVAILABLE;
648 }
649 if (cpu->hyperv_synic) {
650 int sint;
651
652 if (!has_msr_hv_synic ||
653 kvm_vcpu_enable_cap(cs, KVM_CAP_HYPERV_SYNIC, 0)) {
654 fprintf(stderr, "Hyper-V SynIC is not supported by kernel\n");
655 return -ENOSYS;
656 }
657
658 env->features[FEAT_HYPERV_EAX] |= HV_X64_MSR_SYNIC_AVAILABLE;
659 env->msr_hv_synic_version = HV_SYNIC_VERSION_1;
660 for (sint = 0; sint < ARRAY_SIZE(env->msr_hv_synic_sint); sint++) {
661 env->msr_hv_synic_sint[sint] = HV_SYNIC_SINT_MASKED;
662 }
663 }
664 if (cpu->hyperv_stimer) {
665 if (!has_msr_hv_stimer) {
666 fprintf(stderr, "Hyper-V timers aren't supported by kernel\n");
667 return -ENOSYS;
668 }
669 env->features[FEAT_HYPERV_EAX] |= HV_X64_MSR_SYNTIMER_AVAILABLE;
670 }
671 return 0;
672 }
673
674 static Error *invtsc_mig_blocker;
675
676 #define KVM_MAX_CPUID_ENTRIES 100
677
678 int kvm_arch_init_vcpu(CPUState *cs)
679 {
680 struct {
681 struct kvm_cpuid2 cpuid;
682 struct kvm_cpuid_entry2 entries[KVM_MAX_CPUID_ENTRIES];
683 } QEMU_PACKED cpuid_data;
684 X86CPU *cpu = X86_CPU(cs);
685 CPUX86State *env = &cpu->env;
686 uint32_t limit, i, j, cpuid_i;
687 uint32_t unused;
688 struct kvm_cpuid_entry2 *c;
689 uint32_t signature[3];
690 int kvm_base = KVM_CPUID_SIGNATURE;
691 int r;
692 Error *local_err = NULL;
693
694 memset(&cpuid_data, 0, sizeof(cpuid_data));
695
696 cpuid_i = 0;
697
698 /* Paravirtualization CPUIDs */
699 if (hyperv_enabled(cpu)) {
700 c = &cpuid_data.entries[cpuid_i++];
701 c->function = HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS;
702 if (!cpu->hyperv_vendor_id) {
703 memcpy(signature, "Microsoft Hv", 12);
704 } else {
705 size_t len = strlen(cpu->hyperv_vendor_id);
706
707 if (len > 12) {
708 error_report("hv-vendor-id truncated to 12 characters");
709 len = 12;
710 }
711 memset(signature, 0, 12);
712 memcpy(signature, cpu->hyperv_vendor_id, len);
713 }
714 c->eax = HYPERV_CPUID_MIN;
715 c->ebx = signature[0];
716 c->ecx = signature[1];
717 c->edx = signature[2];
718
719 c = &cpuid_data.entries[cpuid_i++];
720 c->function = HYPERV_CPUID_INTERFACE;
721 memcpy(signature, "Hv#1\0\0\0\0\0\0\0\0", 12);
722 c->eax = signature[0];
723 c->ebx = 0;
724 c->ecx = 0;
725 c->edx = 0;
726
727 c = &cpuid_data.entries[cpuid_i++];
728 c->function = HYPERV_CPUID_VERSION;
729 c->eax = 0x00001bbc;
730 c->ebx = 0x00060001;
731
732 c = &cpuid_data.entries[cpuid_i++];
733 c->function = HYPERV_CPUID_FEATURES;
734 r = hyperv_handle_properties(cs);
735 if (r) {
736 return r;
737 }
738 c->eax = env->features[FEAT_HYPERV_EAX];
739 c->ebx = env->features[FEAT_HYPERV_EBX];
740 c->edx = env->features[FEAT_HYPERV_EDX];
741
742 c = &cpuid_data.entries[cpuid_i++];
743 c->function = HYPERV_CPUID_ENLIGHTMENT_INFO;
744 if (cpu->hyperv_relaxed_timing) {
745 c->eax |= HV_X64_RELAXED_TIMING_RECOMMENDED;
746 }
747 if (cpu->hyperv_vapic) {
748 c->eax |= HV_X64_APIC_ACCESS_RECOMMENDED;
749 }
750 c->ebx = cpu->hyperv_spinlock_attempts;
751
752 c = &cpuid_data.entries[cpuid_i++];
753 c->function = HYPERV_CPUID_IMPLEMENT_LIMITS;
754 c->eax = 0x40;
755 c->ebx = 0x40;
756
757 kvm_base = KVM_CPUID_SIGNATURE_NEXT;
758 has_msr_hv_hypercall = true;
759 }
760
761 if (cpu->expose_kvm) {
762 memcpy(signature, "KVMKVMKVM\0\0\0", 12);
763 c = &cpuid_data.entries[cpuid_i++];
764 c->function = KVM_CPUID_SIGNATURE | kvm_base;
765 c->eax = KVM_CPUID_FEATURES | kvm_base;
766 c->ebx = signature[0];
767 c->ecx = signature[1];
768 c->edx = signature[2];
769
770 c = &cpuid_data.entries[cpuid_i++];
771 c->function = KVM_CPUID_FEATURES | kvm_base;
772 c->eax = env->features[FEAT_KVM];
773 }
774
775 cpu_x86_cpuid(env, 0, 0, &limit, &unused, &unused, &unused);
776
777 for (i = 0; i <= limit; i++) {
778 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
779 fprintf(stderr, "unsupported level value: 0x%x\n", limit);
780 abort();
781 }
782 c = &cpuid_data.entries[cpuid_i++];
783
784 switch (i) {
785 case 2: {
786 /* Keep reading function 2 till all the input is received */
787 int times;
788
789 c->function = i;
790 c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC |
791 KVM_CPUID_FLAG_STATE_READ_NEXT;
792 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
793 times = c->eax & 0xff;
794
795 for (j = 1; j < times; ++j) {
796 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
797 fprintf(stderr, "cpuid_data is full, no space for "
798 "cpuid(eax:2):eax & 0xf = 0x%x\n", times);
799 abort();
800 }
801 c = &cpuid_data.entries[cpuid_i++];
802 c->function = i;
803 c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC;
804 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
805 }
806 break;
807 }
808 case 4:
809 case 0xb:
810 case 0xd:
811 for (j = 0; ; j++) {
812 if (i == 0xd && j == 64) {
813 break;
814 }
815 c->function = i;
816 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
817 c->index = j;
818 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
819
820 if (i == 4 && c->eax == 0) {
821 break;
822 }
823 if (i == 0xb && !(c->ecx & 0xff00)) {
824 break;
825 }
826 if (i == 0xd && c->eax == 0) {
827 continue;
828 }
829 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
830 fprintf(stderr, "cpuid_data is full, no space for "
831 "cpuid(eax:0x%x,ecx:0x%x)\n", i, j);
832 abort();
833 }
834 c = &cpuid_data.entries[cpuid_i++];
835 }
836 break;
837 default:
838 c->function = i;
839 c->flags = 0;
840 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
841 break;
842 }
843 }
844
845 if (limit >= 0x0a) {
846 uint32_t ver;
847
848 cpu_x86_cpuid(env, 0x0a, 0, &ver, &unused, &unused, &unused);
849 if ((ver & 0xff) > 0) {
850 has_msr_architectural_pmu = true;
851 num_architectural_pmu_counters = (ver & 0xff00) >> 8;
852
853 /* Shouldn't be more than 32, since that's the number of bits
854 * available in EBX to tell us _which_ counters are available.
855 * Play it safe.
856 */
857 if (num_architectural_pmu_counters > MAX_GP_COUNTERS) {
858 num_architectural_pmu_counters = MAX_GP_COUNTERS;
859 }
860 }
861 }
862
863 cpu_x86_cpuid(env, 0x80000000, 0, &limit, &unused, &unused, &unused);
864
865 for (i = 0x80000000; i <= limit; i++) {
866 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
867 fprintf(stderr, "unsupported xlevel value: 0x%x\n", limit);
868 abort();
869 }
870 c = &cpuid_data.entries[cpuid_i++];
871
872 c->function = i;
873 c->flags = 0;
874 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
875 }
876
877 /* Call Centaur's CPUID instructions they are supported. */
878 if (env->cpuid_xlevel2 > 0) {
879 cpu_x86_cpuid(env, 0xC0000000, 0, &limit, &unused, &unused, &unused);
880
881 for (i = 0xC0000000; i <= limit; i++) {
882 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
883 fprintf(stderr, "unsupported xlevel2 value: 0x%x\n", limit);
884 abort();
885 }
886 c = &cpuid_data.entries[cpuid_i++];
887
888 c->function = i;
889 c->flags = 0;
890 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
891 }
892 }
893
894 cpuid_data.cpuid.nent = cpuid_i;
895
896 if (((env->cpuid_version >> 8)&0xF) >= 6
897 && (env->features[FEAT_1_EDX] & (CPUID_MCE | CPUID_MCA)) ==
898 (CPUID_MCE | CPUID_MCA)
899 && kvm_check_extension(cs->kvm_state, KVM_CAP_MCE) > 0) {
900 uint64_t mcg_cap, unsupported_caps;
901 int banks;
902 int ret;
903
904 ret = kvm_get_mce_cap_supported(cs->kvm_state, &mcg_cap, &banks);
905 if (ret < 0) {
906 fprintf(stderr, "kvm_get_mce_cap_supported: %s", strerror(-ret));
907 return ret;
908 }
909
910 if (banks < (env->mcg_cap & MCG_CAP_BANKS_MASK)) {
911 error_report("kvm: Unsupported MCE bank count (QEMU = %d, KVM = %d)",
912 (int)(env->mcg_cap & MCG_CAP_BANKS_MASK), banks);
913 return -ENOTSUP;
914 }
915
916 unsupported_caps = env->mcg_cap & ~(mcg_cap | MCG_CAP_BANKS_MASK);
917 if (unsupported_caps) {
918 if (unsupported_caps & MCG_LMCE_P) {
919 error_report("kvm: LMCE not supported");
920 return -ENOTSUP;
921 }
922 warn_report("Unsupported MCG_CAP bits: 0x%" PRIx64,
923 unsupported_caps);
924 }
925
926 env->mcg_cap &= mcg_cap | MCG_CAP_BANKS_MASK;
927 ret = kvm_vcpu_ioctl(cs, KVM_X86_SETUP_MCE, &env->mcg_cap);
928 if (ret < 0) {
929 fprintf(stderr, "KVM_X86_SETUP_MCE: %s", strerror(-ret));
930 return ret;
931 }
932 }
933
934 qemu_add_vm_change_state_handler(cpu_update_state, env);
935
936 c = cpuid_find_entry(&cpuid_data.cpuid, 1, 0);
937 if (c) {
938 has_msr_feature_control = !!(c->ecx & CPUID_EXT_VMX) ||
939 !!(c->ecx & CPUID_EXT_SMX);
940 }
941
942 if (env->mcg_cap & MCG_LMCE_P) {
943 has_msr_mcg_ext_ctl = has_msr_feature_control = true;
944 }
945
946 if (!env->user_tsc_khz) {
947 if ((env->features[FEAT_8000_0007_EDX] & CPUID_APM_INVTSC) &&
948 invtsc_mig_blocker == NULL) {
949 /* for migration */
950 error_setg(&invtsc_mig_blocker,
951 "State blocked by non-migratable CPU device"
952 " (invtsc flag)");
953 r = migrate_add_blocker(invtsc_mig_blocker, &local_err);
954 if (local_err) {
955 error_report_err(local_err);
956 error_free(invtsc_mig_blocker);
957 goto fail;
958 }
959 /* for savevm */
960 vmstate_x86_cpu.unmigratable = 1;
961 }
962 }
963
964 r = kvm_arch_set_tsc_khz(cs);
965 if (r < 0) {
966 goto fail;
967 }
968
969 /* vcpu's TSC frequency is either specified by user, or following
970 * the value used by KVM if the former is not present. In the
971 * latter case, we query it from KVM and record in env->tsc_khz,
972 * so that vcpu's TSC frequency can be migrated later via this field.
973 */
974 if (!env->tsc_khz) {
975 r = kvm_check_extension(cs->kvm_state, KVM_CAP_GET_TSC_KHZ) ?
976 kvm_vcpu_ioctl(cs, KVM_GET_TSC_KHZ) :
977 -ENOTSUP;
978 if (r > 0) {
979 env->tsc_khz = r;
980 }
981 }
982
983 if (cpu->vmware_cpuid_freq
984 /* Guests depend on 0x40000000 to detect this feature, so only expose
985 * it if KVM exposes leaf 0x40000000. (Conflicts with Hyper-V) */
986 && cpu->expose_kvm
987 && kvm_base == KVM_CPUID_SIGNATURE
988 /* TSC clock must be stable and known for this feature. */
989 && ((env->features[FEAT_8000_0007_EDX] & CPUID_APM_INVTSC)
990 || env->user_tsc_khz != 0)
991 && env->tsc_khz != 0) {
992
993 c = &cpuid_data.entries[cpuid_i++];
994 c->function = KVM_CPUID_SIGNATURE | 0x10;
995 c->eax = env->tsc_khz;
996 /* LAPIC resolution of 1ns (freq: 1GHz) is hardcoded in KVM's
997 * APIC_BUS_CYCLE_NS */
998 c->ebx = 1000000;
999 c->ecx = c->edx = 0;
1000
1001 c = cpuid_find_entry(&cpuid_data.cpuid, kvm_base, 0);
1002 c->eax = MAX(c->eax, KVM_CPUID_SIGNATURE | 0x10);
1003 }
1004
1005 cpuid_data.cpuid.nent = cpuid_i;
1006
1007 cpuid_data.cpuid.padding = 0;
1008 r = kvm_vcpu_ioctl(cs, KVM_SET_CPUID2, &cpuid_data);
1009 if (r) {
1010 goto fail;
1011 }
1012
1013 if (has_xsave) {
1014 env->kvm_xsave_buf = qemu_memalign(4096, sizeof(struct kvm_xsave));
1015 }
1016 cpu->kvm_msr_buf = g_malloc0(MSR_BUF_SIZE);
1017
1018 if (!(env->features[FEAT_8000_0001_EDX] & CPUID_EXT2_RDTSCP)) {
1019 has_msr_tsc_aux = false;
1020 }
1021
1022 return 0;
1023
1024 fail:
1025 migrate_del_blocker(invtsc_mig_blocker);
1026 return r;
1027 }
1028
1029 void kvm_arch_reset_vcpu(X86CPU *cpu)
1030 {
1031 CPUX86State *env = &cpu->env;
1032
1033 env->exception_injected = -1;
1034 env->interrupt_injected = -1;
1035 env->xcr0 = 1;
1036 if (kvm_irqchip_in_kernel()) {
1037 env->mp_state = cpu_is_bsp(cpu) ? KVM_MP_STATE_RUNNABLE :
1038 KVM_MP_STATE_UNINITIALIZED;
1039 } else {
1040 env->mp_state = KVM_MP_STATE_RUNNABLE;
1041 }
1042 }
1043
1044 void kvm_arch_do_init_vcpu(X86CPU *cpu)
1045 {
1046 CPUX86State *env = &cpu->env;
1047
1048 /* APs get directly into wait-for-SIPI state. */
1049 if (env->mp_state == KVM_MP_STATE_UNINITIALIZED) {
1050 env->mp_state = KVM_MP_STATE_INIT_RECEIVED;
1051 }
1052 }
1053
1054 static int kvm_get_supported_msrs(KVMState *s)
1055 {
1056 static int kvm_supported_msrs;
1057 int ret = 0;
1058
1059 /* first time */
1060 if (kvm_supported_msrs == 0) {
1061 struct kvm_msr_list msr_list, *kvm_msr_list;
1062
1063 kvm_supported_msrs = -1;
1064
1065 /* Obtain MSR list from KVM. These are the MSRs that we must
1066 * save/restore */
1067 msr_list.nmsrs = 0;
1068 ret = kvm_ioctl(s, KVM_GET_MSR_INDEX_LIST, &msr_list);
1069 if (ret < 0 && ret != -E2BIG) {
1070 return ret;
1071 }
1072 /* Old kernel modules had a bug and could write beyond the provided
1073 memory. Allocate at least a safe amount of 1K. */
1074 kvm_msr_list = g_malloc0(MAX(1024, sizeof(msr_list) +
1075 msr_list.nmsrs *
1076 sizeof(msr_list.indices[0])));
1077
1078 kvm_msr_list->nmsrs = msr_list.nmsrs;
1079 ret = kvm_ioctl(s, KVM_GET_MSR_INDEX_LIST, kvm_msr_list);
1080 if (ret >= 0) {
1081 int i;
1082
1083 for (i = 0; i < kvm_msr_list->nmsrs; i++) {
1084 if (kvm_msr_list->indices[i] == MSR_STAR) {
1085 has_msr_star = true;
1086 continue;
1087 }
1088 if (kvm_msr_list->indices[i] == MSR_VM_HSAVE_PA) {
1089 has_msr_hsave_pa = true;
1090 continue;
1091 }
1092 if (kvm_msr_list->indices[i] == MSR_TSC_AUX) {
1093 has_msr_tsc_aux = true;
1094 continue;
1095 }
1096 if (kvm_msr_list->indices[i] == MSR_TSC_ADJUST) {
1097 has_msr_tsc_adjust = true;
1098 continue;
1099 }
1100 if (kvm_msr_list->indices[i] == MSR_IA32_TSCDEADLINE) {
1101 has_msr_tsc_deadline = true;
1102 continue;
1103 }
1104 if (kvm_msr_list->indices[i] == MSR_IA32_SMBASE) {
1105 has_msr_smbase = true;
1106 continue;
1107 }
1108 if (kvm_msr_list->indices[i] == MSR_IA32_MISC_ENABLE) {
1109 has_msr_misc_enable = true;
1110 continue;
1111 }
1112 if (kvm_msr_list->indices[i] == MSR_IA32_BNDCFGS) {
1113 has_msr_bndcfgs = true;
1114 continue;
1115 }
1116 if (kvm_msr_list->indices[i] == MSR_IA32_XSS) {
1117 has_msr_xss = true;
1118 continue;
1119 }
1120 if (kvm_msr_list->indices[i] == HV_X64_MSR_CRASH_CTL) {
1121 has_msr_hv_crash = true;
1122 continue;
1123 }
1124 if (kvm_msr_list->indices[i] == HV_X64_MSR_RESET) {
1125 has_msr_hv_reset = true;
1126 continue;
1127 }
1128 if (kvm_msr_list->indices[i] == HV_X64_MSR_VP_INDEX) {
1129 has_msr_hv_vpindex = true;
1130 continue;
1131 }
1132 if (kvm_msr_list->indices[i] == HV_X64_MSR_VP_RUNTIME) {
1133 has_msr_hv_runtime = true;
1134 continue;
1135 }
1136 if (kvm_msr_list->indices[i] == HV_X64_MSR_SCONTROL) {
1137 has_msr_hv_synic = true;
1138 continue;
1139 }
1140 if (kvm_msr_list->indices[i] == HV_X64_MSR_STIMER0_CONFIG) {
1141 has_msr_hv_stimer = true;
1142 continue;
1143 }
1144 }
1145 }
1146
1147 g_free(kvm_msr_list);
1148 }
1149
1150 return ret;
1151 }
1152
1153 static Notifier smram_machine_done;
1154 static KVMMemoryListener smram_listener;
1155 static AddressSpace smram_address_space;
1156 static MemoryRegion smram_as_root;
1157 static MemoryRegion smram_as_mem;
1158
1159 static void register_smram_listener(Notifier *n, void *unused)
1160 {
1161 MemoryRegion *smram =
1162 (MemoryRegion *) object_resolve_path("/machine/smram", NULL);
1163
1164 /* Outer container... */
1165 memory_region_init(&smram_as_root, OBJECT(kvm_state), "mem-container-smram", ~0ull);
1166 memory_region_set_enabled(&smram_as_root, true);
1167
1168 /* ... with two regions inside: normal system memory with low
1169 * priority, and...
1170 */
1171 memory_region_init_alias(&smram_as_mem, OBJECT(kvm_state), "mem-smram",
1172 get_system_memory(), 0, ~0ull);
1173 memory_region_add_subregion_overlap(&smram_as_root, 0, &smram_as_mem, 0);
1174 memory_region_set_enabled(&smram_as_mem, true);
1175
1176 if (smram) {
1177 /* ... SMRAM with higher priority */
1178 memory_region_add_subregion_overlap(&smram_as_root, 0, smram, 10);
1179 memory_region_set_enabled(smram, true);
1180 }
1181
1182 address_space_init(&smram_address_space, &smram_as_root, "KVM-SMRAM");
1183 kvm_memory_listener_register(kvm_state, &smram_listener,
1184 &smram_address_space, 1);
1185 }
1186
1187 int kvm_arch_init(MachineState *ms, KVMState *s)
1188 {
1189 uint64_t identity_base = 0xfffbc000;
1190 uint64_t shadow_mem;
1191 int ret;
1192 struct utsname utsname;
1193
1194 #ifdef KVM_CAP_XSAVE
1195 has_xsave = kvm_check_extension(s, KVM_CAP_XSAVE);
1196 #endif
1197
1198 #ifdef KVM_CAP_XCRS
1199 has_xcrs = kvm_check_extension(s, KVM_CAP_XCRS);
1200 #endif
1201
1202 #ifdef KVM_CAP_PIT_STATE2
1203 has_pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2);
1204 #endif
1205
1206 ret = kvm_get_supported_msrs(s);
1207 if (ret < 0) {
1208 return ret;
1209 }
1210
1211 uname(&utsname);
1212 lm_capable_kernel = strcmp(utsname.machine, "x86_64") == 0;
1213
1214 /*
1215 * On older Intel CPUs, KVM uses vm86 mode to emulate 16-bit code directly.
1216 * In order to use vm86 mode, an EPT identity map and a TSS are needed.
1217 * Since these must be part of guest physical memory, we need to allocate
1218 * them, both by setting their start addresses in the kernel and by
1219 * creating a corresponding e820 entry. We need 4 pages before the BIOS.
1220 *
1221 * Older KVM versions may not support setting the identity map base. In
1222 * that case we need to stick with the default, i.e. a 256K maximum BIOS
1223 * size.
1224 */
1225 if (kvm_check_extension(s, KVM_CAP_SET_IDENTITY_MAP_ADDR)) {
1226 /* Allows up to 16M BIOSes. */
1227 identity_base = 0xfeffc000;
1228
1229 ret = kvm_vm_ioctl(s, KVM_SET_IDENTITY_MAP_ADDR, &identity_base);
1230 if (ret < 0) {
1231 return ret;
1232 }
1233 }
1234
1235 /* Set TSS base one page after EPT identity map. */
1236 ret = kvm_vm_ioctl(s, KVM_SET_TSS_ADDR, identity_base + 0x1000);
1237 if (ret < 0) {
1238 return ret;
1239 }
1240
1241 /* Tell fw_cfg to notify the BIOS to reserve the range. */
1242 ret = e820_add_entry(identity_base, 0x4000, E820_RESERVED);
1243 if (ret < 0) {
1244 fprintf(stderr, "e820_add_entry() table is full\n");
1245 return ret;
1246 }
1247 qemu_register_reset(kvm_unpoison_all, NULL);
1248
1249 shadow_mem = machine_kvm_shadow_mem(ms);
1250 if (shadow_mem != -1) {
1251 shadow_mem /= 4096;
1252 ret = kvm_vm_ioctl(s, KVM_SET_NR_MMU_PAGES, shadow_mem);
1253 if (ret < 0) {
1254 return ret;
1255 }
1256 }
1257
1258 if (kvm_check_extension(s, KVM_CAP_X86_SMM) &&
1259 object_dynamic_cast(OBJECT(ms), TYPE_PC_MACHINE) &&
1260 pc_machine_is_smm_enabled(PC_MACHINE(ms))) {
1261 smram_machine_done.notify = register_smram_listener;
1262 qemu_add_machine_init_done_notifier(&smram_machine_done);
1263 }
1264 return 0;
1265 }
1266
1267 static void set_v8086_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
1268 {
1269 lhs->selector = rhs->selector;
1270 lhs->base = rhs->base;
1271 lhs->limit = rhs->limit;
1272 lhs->type = 3;
1273 lhs->present = 1;
1274 lhs->dpl = 3;
1275 lhs->db = 0;
1276 lhs->s = 1;
1277 lhs->l = 0;
1278 lhs->g = 0;
1279 lhs->avl = 0;
1280 lhs->unusable = 0;
1281 }
1282
1283 static void set_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
1284 {
1285 unsigned flags = rhs->flags;
1286 lhs->selector = rhs->selector;
1287 lhs->base = rhs->base;
1288 lhs->limit = rhs->limit;
1289 lhs->type = (flags >> DESC_TYPE_SHIFT) & 15;
1290 lhs->present = (flags & DESC_P_MASK) != 0;
1291 lhs->dpl = (flags >> DESC_DPL_SHIFT) & 3;
1292 lhs->db = (flags >> DESC_B_SHIFT) & 1;
1293 lhs->s = (flags & DESC_S_MASK) != 0;
1294 lhs->l = (flags >> DESC_L_SHIFT) & 1;
1295 lhs->g = (flags & DESC_G_MASK) != 0;
1296 lhs->avl = (flags & DESC_AVL_MASK) != 0;
1297 lhs->unusable = !lhs->present;
1298 lhs->padding = 0;
1299 }
1300
1301 static void get_seg(SegmentCache *lhs, const struct kvm_segment *rhs)
1302 {
1303 lhs->selector = rhs->selector;
1304 lhs->base = rhs->base;
1305 lhs->limit = rhs->limit;
1306 lhs->flags = (rhs->type << DESC_TYPE_SHIFT) |
1307 ((rhs->present && !rhs->unusable) * DESC_P_MASK) |
1308 (rhs->dpl << DESC_DPL_SHIFT) |
1309 (rhs->db << DESC_B_SHIFT) |
1310 (rhs->s * DESC_S_MASK) |
1311 (rhs->l << DESC_L_SHIFT) |
1312 (rhs->g * DESC_G_MASK) |
1313 (rhs->avl * DESC_AVL_MASK);
1314 }
1315
1316 static void kvm_getput_reg(__u64 *kvm_reg, target_ulong *qemu_reg, int set)
1317 {
1318 if (set) {
1319 *kvm_reg = *qemu_reg;
1320 } else {
1321 *qemu_reg = *kvm_reg;
1322 }
1323 }
1324
1325 static int kvm_getput_regs(X86CPU *cpu, int set)
1326 {
1327 CPUX86State *env = &cpu->env;
1328 struct kvm_regs regs;
1329 int ret = 0;
1330
1331 if (!set) {
1332 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_REGS, &regs);
1333 if (ret < 0) {
1334 return ret;
1335 }
1336 }
1337
1338 kvm_getput_reg(&regs.rax, &env->regs[R_EAX], set);
1339 kvm_getput_reg(&regs.rbx, &env->regs[R_EBX], set);
1340 kvm_getput_reg(&regs.rcx, &env->regs[R_ECX], set);
1341 kvm_getput_reg(&regs.rdx, &env->regs[R_EDX], set);
1342 kvm_getput_reg(&regs.rsi, &env->regs[R_ESI], set);
1343 kvm_getput_reg(&regs.rdi, &env->regs[R_EDI], set);
1344 kvm_getput_reg(&regs.rsp, &env->regs[R_ESP], set);
1345 kvm_getput_reg(&regs.rbp, &env->regs[R_EBP], set);
1346 #ifdef TARGET_X86_64
1347 kvm_getput_reg(&regs.r8, &env->regs[8], set);
1348 kvm_getput_reg(&regs.r9, &env->regs[9], set);
1349 kvm_getput_reg(&regs.r10, &env->regs[10], set);
1350 kvm_getput_reg(&regs.r11, &env->regs[11], set);
1351 kvm_getput_reg(&regs.r12, &env->regs[12], set);
1352 kvm_getput_reg(&regs.r13, &env->regs[13], set);
1353 kvm_getput_reg(&regs.r14, &env->regs[14], set);
1354 kvm_getput_reg(&regs.r15, &env->regs[15], set);
1355 #endif
1356
1357 kvm_getput_reg(&regs.rflags, &env->eflags, set);
1358 kvm_getput_reg(&regs.rip, &env->eip, set);
1359
1360 if (set) {
1361 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_REGS, &regs);
1362 }
1363
1364 return ret;
1365 }
1366
1367 static int kvm_put_fpu(X86CPU *cpu)
1368 {
1369 CPUX86State *env = &cpu->env;
1370 struct kvm_fpu fpu;
1371 int i;
1372
1373 memset(&fpu, 0, sizeof fpu);
1374 fpu.fsw = env->fpus & ~(7 << 11);
1375 fpu.fsw |= (env->fpstt & 7) << 11;
1376 fpu.fcw = env->fpuc;
1377 fpu.last_opcode = env->fpop;
1378 fpu.last_ip = env->fpip;
1379 fpu.last_dp = env->fpdp;
1380 for (i = 0; i < 8; ++i) {
1381 fpu.ftwx |= (!env->fptags[i]) << i;
1382 }
1383 memcpy(fpu.fpr, env->fpregs, sizeof env->fpregs);
1384 for (i = 0; i < CPU_NB_REGS; i++) {
1385 stq_p(&fpu.xmm[i][0], env->xmm_regs[i].ZMM_Q(0));
1386 stq_p(&fpu.xmm[i][8], env->xmm_regs[i].ZMM_Q(1));
1387 }
1388 fpu.mxcsr = env->mxcsr;
1389
1390 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_FPU, &fpu);
1391 }
1392
1393 #define XSAVE_FCW_FSW 0
1394 #define XSAVE_FTW_FOP 1
1395 #define XSAVE_CWD_RIP 2
1396 #define XSAVE_CWD_RDP 4
1397 #define XSAVE_MXCSR 6
1398 #define XSAVE_ST_SPACE 8
1399 #define XSAVE_XMM_SPACE 40
1400 #define XSAVE_XSTATE_BV 128
1401 #define XSAVE_YMMH_SPACE 144
1402 #define XSAVE_BNDREGS 240
1403 #define XSAVE_BNDCSR 256
1404 #define XSAVE_OPMASK 272
1405 #define XSAVE_ZMM_Hi256 288
1406 #define XSAVE_Hi16_ZMM 416
1407 #define XSAVE_PKRU 672
1408
1409 #define XSAVE_BYTE_OFFSET(word_offset) \
1410 ((word_offset) * sizeof(((struct kvm_xsave *)0)->region[0]))
1411
1412 #define ASSERT_OFFSET(word_offset, field) \
1413 QEMU_BUILD_BUG_ON(XSAVE_BYTE_OFFSET(word_offset) != \
1414 offsetof(X86XSaveArea, field))
1415
1416 ASSERT_OFFSET(XSAVE_FCW_FSW, legacy.fcw);
1417 ASSERT_OFFSET(XSAVE_FTW_FOP, legacy.ftw);
1418 ASSERT_OFFSET(XSAVE_CWD_RIP, legacy.fpip);
1419 ASSERT_OFFSET(XSAVE_CWD_RDP, legacy.fpdp);
1420 ASSERT_OFFSET(XSAVE_MXCSR, legacy.mxcsr);
1421 ASSERT_OFFSET(XSAVE_ST_SPACE, legacy.fpregs);
1422 ASSERT_OFFSET(XSAVE_XMM_SPACE, legacy.xmm_regs);
1423 ASSERT_OFFSET(XSAVE_XSTATE_BV, header.xstate_bv);
1424 ASSERT_OFFSET(XSAVE_YMMH_SPACE, avx_state);
1425 ASSERT_OFFSET(XSAVE_BNDREGS, bndreg_state);
1426 ASSERT_OFFSET(XSAVE_BNDCSR, bndcsr_state);
1427 ASSERT_OFFSET(XSAVE_OPMASK, opmask_state);
1428 ASSERT_OFFSET(XSAVE_ZMM_Hi256, zmm_hi256_state);
1429 ASSERT_OFFSET(XSAVE_Hi16_ZMM, hi16_zmm_state);
1430 ASSERT_OFFSET(XSAVE_PKRU, pkru_state);
1431
1432 static int kvm_put_xsave(X86CPU *cpu)
1433 {
1434 CPUX86State *env = &cpu->env;
1435 X86XSaveArea *xsave = env->kvm_xsave_buf;
1436
1437 if (!has_xsave) {
1438 return kvm_put_fpu(cpu);
1439 }
1440 x86_cpu_xsave_all_areas(cpu, xsave);
1441
1442 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_XSAVE, xsave);
1443 }
1444
1445 static int kvm_put_xcrs(X86CPU *cpu)
1446 {
1447 CPUX86State *env = &cpu->env;
1448 struct kvm_xcrs xcrs = {};
1449
1450 if (!has_xcrs) {
1451 return 0;
1452 }
1453
1454 xcrs.nr_xcrs = 1;
1455 xcrs.flags = 0;
1456 xcrs.xcrs[0].xcr = 0;
1457 xcrs.xcrs[0].value = env->xcr0;
1458 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_XCRS, &xcrs);
1459 }
1460
1461 static int kvm_put_sregs(X86CPU *cpu)
1462 {
1463 CPUX86State *env = &cpu->env;
1464 struct kvm_sregs sregs;
1465
1466 memset(sregs.interrupt_bitmap, 0, sizeof(sregs.interrupt_bitmap));
1467 if (env->interrupt_injected >= 0) {
1468 sregs.interrupt_bitmap[env->interrupt_injected / 64] |=
1469 (uint64_t)1 << (env->interrupt_injected % 64);
1470 }
1471
1472 if ((env->eflags & VM_MASK)) {
1473 set_v8086_seg(&sregs.cs, &env->segs[R_CS]);
1474 set_v8086_seg(&sregs.ds, &env->segs[R_DS]);
1475 set_v8086_seg(&sregs.es, &env->segs[R_ES]);
1476 set_v8086_seg(&sregs.fs, &env->segs[R_FS]);
1477 set_v8086_seg(&sregs.gs, &env->segs[R_GS]);
1478 set_v8086_seg(&sregs.ss, &env->segs[R_SS]);
1479 } else {
1480 set_seg(&sregs.cs, &env->segs[R_CS]);
1481 set_seg(&sregs.ds, &env->segs[R_DS]);
1482 set_seg(&sregs.es, &env->segs[R_ES]);
1483 set_seg(&sregs.fs, &env->segs[R_FS]);
1484 set_seg(&sregs.gs, &env->segs[R_GS]);
1485 set_seg(&sregs.ss, &env->segs[R_SS]);
1486 }
1487
1488 set_seg(&sregs.tr, &env->tr);
1489 set_seg(&sregs.ldt, &env->ldt);
1490
1491 sregs.idt.limit = env->idt.limit;
1492 sregs.idt.base = env->idt.base;
1493 memset(sregs.idt.padding, 0, sizeof sregs.idt.padding);
1494 sregs.gdt.limit = env->gdt.limit;
1495 sregs.gdt.base = env->gdt.base;
1496 memset(sregs.gdt.padding, 0, sizeof sregs.gdt.padding);
1497
1498 sregs.cr0 = env->cr[0];
1499 sregs.cr2 = env->cr[2];
1500 sregs.cr3 = env->cr[3];
1501 sregs.cr4 = env->cr[4];
1502
1503 sregs.cr8 = cpu_get_apic_tpr(cpu->apic_state);
1504 sregs.apic_base = cpu_get_apic_base(cpu->apic_state);
1505
1506 sregs.efer = env->efer;
1507
1508 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_SREGS, &sregs);
1509 }
1510
1511 static void kvm_msr_buf_reset(X86CPU *cpu)
1512 {
1513 memset(cpu->kvm_msr_buf, 0, MSR_BUF_SIZE);
1514 }
1515
1516 static void kvm_msr_entry_add(X86CPU *cpu, uint32_t index, uint64_t value)
1517 {
1518 struct kvm_msrs *msrs = cpu->kvm_msr_buf;
1519 void *limit = ((void *)msrs) + MSR_BUF_SIZE;
1520 struct kvm_msr_entry *entry = &msrs->entries[msrs->nmsrs];
1521
1522 assert((void *)(entry + 1) <= limit);
1523
1524 entry->index = index;
1525 entry->reserved = 0;
1526 entry->data = value;
1527 msrs->nmsrs++;
1528 }
1529
1530 static int kvm_put_one_msr(X86CPU *cpu, int index, uint64_t value)
1531 {
1532 kvm_msr_buf_reset(cpu);
1533 kvm_msr_entry_add(cpu, index, value);
1534
1535 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MSRS, cpu->kvm_msr_buf);
1536 }
1537
1538 void kvm_put_apicbase(X86CPU *cpu, uint64_t value)
1539 {
1540 int ret;
1541
1542 ret = kvm_put_one_msr(cpu, MSR_IA32_APICBASE, value);
1543 assert(ret == 1);
1544 }
1545
1546 static int kvm_put_tscdeadline_msr(X86CPU *cpu)
1547 {
1548 CPUX86State *env = &cpu->env;
1549 int ret;
1550
1551 if (!has_msr_tsc_deadline) {
1552 return 0;
1553 }
1554
1555 ret = kvm_put_one_msr(cpu, MSR_IA32_TSCDEADLINE, env->tsc_deadline);
1556 if (ret < 0) {
1557 return ret;
1558 }
1559
1560 assert(ret == 1);
1561 return 0;
1562 }
1563
1564 /*
1565 * Provide a separate write service for the feature control MSR in order to
1566 * kick the VCPU out of VMXON or even guest mode on reset. This has to be done
1567 * before writing any other state because forcibly leaving nested mode
1568 * invalidates the VCPU state.
1569 */
1570 static int kvm_put_msr_feature_control(X86CPU *cpu)
1571 {
1572 int ret;
1573
1574 if (!has_msr_feature_control) {
1575 return 0;
1576 }
1577
1578 ret = kvm_put_one_msr(cpu, MSR_IA32_FEATURE_CONTROL,
1579 cpu->env.msr_ia32_feature_control);
1580 if (ret < 0) {
1581 return ret;
1582 }
1583
1584 assert(ret == 1);
1585 return 0;
1586 }
1587
1588 static int kvm_put_msrs(X86CPU *cpu, int level)
1589 {
1590 CPUX86State *env = &cpu->env;
1591 int i;
1592 int ret;
1593
1594 kvm_msr_buf_reset(cpu);
1595
1596 kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_CS, env->sysenter_cs);
1597 kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_ESP, env->sysenter_esp);
1598 kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_EIP, env->sysenter_eip);
1599 kvm_msr_entry_add(cpu, MSR_PAT, env->pat);
1600 if (has_msr_star) {
1601 kvm_msr_entry_add(cpu, MSR_STAR, env->star);
1602 }
1603 if (has_msr_hsave_pa) {
1604 kvm_msr_entry_add(cpu, MSR_VM_HSAVE_PA, env->vm_hsave);
1605 }
1606 if (has_msr_tsc_aux) {
1607 kvm_msr_entry_add(cpu, MSR_TSC_AUX, env->tsc_aux);
1608 }
1609 if (has_msr_tsc_adjust) {
1610 kvm_msr_entry_add(cpu, MSR_TSC_ADJUST, env->tsc_adjust);
1611 }
1612 if (has_msr_misc_enable) {
1613 kvm_msr_entry_add(cpu, MSR_IA32_MISC_ENABLE,
1614 env->msr_ia32_misc_enable);
1615 }
1616 if (has_msr_smbase) {
1617 kvm_msr_entry_add(cpu, MSR_IA32_SMBASE, env->smbase);
1618 }
1619 if (has_msr_bndcfgs) {
1620 kvm_msr_entry_add(cpu, MSR_IA32_BNDCFGS, env->msr_bndcfgs);
1621 }
1622 if (has_msr_xss) {
1623 kvm_msr_entry_add(cpu, MSR_IA32_XSS, env->xss);
1624 }
1625 #ifdef TARGET_X86_64
1626 if (lm_capable_kernel) {
1627 kvm_msr_entry_add(cpu, MSR_CSTAR, env->cstar);
1628 kvm_msr_entry_add(cpu, MSR_KERNELGSBASE, env->kernelgsbase);
1629 kvm_msr_entry_add(cpu, MSR_FMASK, env->fmask);
1630 kvm_msr_entry_add(cpu, MSR_LSTAR, env->lstar);
1631 }
1632 #endif
1633 /*
1634 * The following MSRs have side effects on the guest or are too heavy
1635 * for normal writeback. Limit them to reset or full state updates.
1636 */
1637 if (level >= KVM_PUT_RESET_STATE) {
1638 kvm_msr_entry_add(cpu, MSR_IA32_TSC, env->tsc);
1639 kvm_msr_entry_add(cpu, MSR_KVM_SYSTEM_TIME, env->system_time_msr);
1640 kvm_msr_entry_add(cpu, MSR_KVM_WALL_CLOCK, env->wall_clock_msr);
1641 if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_ASYNC_PF)) {
1642 kvm_msr_entry_add(cpu, MSR_KVM_ASYNC_PF_EN, env->async_pf_en_msr);
1643 }
1644 if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_PV_EOI)) {
1645 kvm_msr_entry_add(cpu, MSR_KVM_PV_EOI_EN, env->pv_eoi_en_msr);
1646 }
1647 if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_STEAL_TIME)) {
1648 kvm_msr_entry_add(cpu, MSR_KVM_STEAL_TIME, env->steal_time_msr);
1649 }
1650 if (has_msr_architectural_pmu) {
1651 /* Stop the counter. */
1652 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR_CTRL, 0);
1653 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_CTRL, 0);
1654
1655 /* Set the counter values. */
1656 for (i = 0; i < MAX_FIXED_COUNTERS; i++) {
1657 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR0 + i,
1658 env->msr_fixed_counters[i]);
1659 }
1660 for (i = 0; i < num_architectural_pmu_counters; i++) {
1661 kvm_msr_entry_add(cpu, MSR_P6_PERFCTR0 + i,
1662 env->msr_gp_counters[i]);
1663 kvm_msr_entry_add(cpu, MSR_P6_EVNTSEL0 + i,
1664 env->msr_gp_evtsel[i]);
1665 }
1666 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_STATUS,
1667 env->msr_global_status);
1668 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1669 env->msr_global_ovf_ctrl);
1670
1671 /* Now start the PMU. */
1672 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR_CTRL,
1673 env->msr_fixed_ctr_ctrl);
1674 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_CTRL,
1675 env->msr_global_ctrl);
1676 }
1677 if (has_msr_hv_hypercall) {
1678 kvm_msr_entry_add(cpu, HV_X64_MSR_GUEST_OS_ID,
1679 env->msr_hv_guest_os_id);
1680 kvm_msr_entry_add(cpu, HV_X64_MSR_HYPERCALL,
1681 env->msr_hv_hypercall);
1682 }
1683 if (cpu->hyperv_vapic) {
1684 kvm_msr_entry_add(cpu, HV_X64_MSR_APIC_ASSIST_PAGE,
1685 env->msr_hv_vapic);
1686 }
1687 if (cpu->hyperv_time) {
1688 kvm_msr_entry_add(cpu, HV_X64_MSR_REFERENCE_TSC, env->msr_hv_tsc);
1689 }
1690 if (has_msr_hv_crash) {
1691 int j;
1692
1693 for (j = 0; j < HV_X64_MSR_CRASH_PARAMS; j++)
1694 kvm_msr_entry_add(cpu, HV_X64_MSR_CRASH_P0 + j,
1695 env->msr_hv_crash_params[j]);
1696
1697 kvm_msr_entry_add(cpu, HV_X64_MSR_CRASH_CTL,
1698 HV_X64_MSR_CRASH_CTL_NOTIFY);
1699 }
1700 if (has_msr_hv_runtime) {
1701 kvm_msr_entry_add(cpu, HV_X64_MSR_VP_RUNTIME, env->msr_hv_runtime);
1702 }
1703 if (cpu->hyperv_synic) {
1704 int j;
1705
1706 kvm_msr_entry_add(cpu, HV_X64_MSR_SCONTROL,
1707 env->msr_hv_synic_control);
1708 kvm_msr_entry_add(cpu, HV_X64_MSR_SVERSION,
1709 env->msr_hv_synic_version);
1710 kvm_msr_entry_add(cpu, HV_X64_MSR_SIEFP,
1711 env->msr_hv_synic_evt_page);
1712 kvm_msr_entry_add(cpu, HV_X64_MSR_SIMP,
1713 env->msr_hv_synic_msg_page);
1714
1715 for (j = 0; j < ARRAY_SIZE(env->msr_hv_synic_sint); j++) {
1716 kvm_msr_entry_add(cpu, HV_X64_MSR_SINT0 + j,
1717 env->msr_hv_synic_sint[j]);
1718 }
1719 }
1720 if (has_msr_hv_stimer) {
1721 int j;
1722
1723 for (j = 0; j < ARRAY_SIZE(env->msr_hv_stimer_config); j++) {
1724 kvm_msr_entry_add(cpu, HV_X64_MSR_STIMER0_CONFIG + j * 2,
1725 env->msr_hv_stimer_config[j]);
1726 }
1727
1728 for (j = 0; j < ARRAY_SIZE(env->msr_hv_stimer_count); j++) {
1729 kvm_msr_entry_add(cpu, HV_X64_MSR_STIMER0_COUNT + j * 2,
1730 env->msr_hv_stimer_count[j]);
1731 }
1732 }
1733 if (env->features[FEAT_1_EDX] & CPUID_MTRR) {
1734 uint64_t phys_mask = MAKE_64BIT_MASK(0, cpu->phys_bits);
1735
1736 kvm_msr_entry_add(cpu, MSR_MTRRdefType, env->mtrr_deftype);
1737 kvm_msr_entry_add(cpu, MSR_MTRRfix64K_00000, env->mtrr_fixed[0]);
1738 kvm_msr_entry_add(cpu, MSR_MTRRfix16K_80000, env->mtrr_fixed[1]);
1739 kvm_msr_entry_add(cpu, MSR_MTRRfix16K_A0000, env->mtrr_fixed[2]);
1740 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C0000, env->mtrr_fixed[3]);
1741 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C8000, env->mtrr_fixed[4]);
1742 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D0000, env->mtrr_fixed[5]);
1743 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D8000, env->mtrr_fixed[6]);
1744 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E0000, env->mtrr_fixed[7]);
1745 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E8000, env->mtrr_fixed[8]);
1746 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F0000, env->mtrr_fixed[9]);
1747 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F8000, env->mtrr_fixed[10]);
1748 for (i = 0; i < MSR_MTRRcap_VCNT; i++) {
1749 /* The CPU GPs if we write to a bit above the physical limit of
1750 * the host CPU (and KVM emulates that)
1751 */
1752 uint64_t mask = env->mtrr_var[i].mask;
1753 mask &= phys_mask;
1754
1755 kvm_msr_entry_add(cpu, MSR_MTRRphysBase(i),
1756 env->mtrr_var[i].base);
1757 kvm_msr_entry_add(cpu, MSR_MTRRphysMask(i), mask);
1758 }
1759 }
1760
1761 /* Note: MSR_IA32_FEATURE_CONTROL is written separately, see
1762 * kvm_put_msr_feature_control. */
1763 }
1764 if (env->mcg_cap) {
1765 int i;
1766
1767 kvm_msr_entry_add(cpu, MSR_MCG_STATUS, env->mcg_status);
1768 kvm_msr_entry_add(cpu, MSR_MCG_CTL, env->mcg_ctl);
1769 if (has_msr_mcg_ext_ctl) {
1770 kvm_msr_entry_add(cpu, MSR_MCG_EXT_CTL, env->mcg_ext_ctl);
1771 }
1772 for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++) {
1773 kvm_msr_entry_add(cpu, MSR_MC0_CTL + i, env->mce_banks[i]);
1774 }
1775 }
1776
1777 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MSRS, cpu->kvm_msr_buf);
1778 if (ret < 0) {
1779 return ret;
1780 }
1781
1782 if (ret < cpu->kvm_msr_buf->nmsrs) {
1783 struct kvm_msr_entry *e = &cpu->kvm_msr_buf->entries[ret];
1784 error_report("error: failed to set MSR 0x%" PRIx32 " to 0x%" PRIx64,
1785 (uint32_t)e->index, (uint64_t)e->data);
1786 }
1787
1788 assert(ret == cpu->kvm_msr_buf->nmsrs);
1789 return 0;
1790 }
1791
1792
1793 static int kvm_get_fpu(X86CPU *cpu)
1794 {
1795 CPUX86State *env = &cpu->env;
1796 struct kvm_fpu fpu;
1797 int i, ret;
1798
1799 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_FPU, &fpu);
1800 if (ret < 0) {
1801 return ret;
1802 }
1803
1804 env->fpstt = (fpu.fsw >> 11) & 7;
1805 env->fpus = fpu.fsw;
1806 env->fpuc = fpu.fcw;
1807 env->fpop = fpu.last_opcode;
1808 env->fpip = fpu.last_ip;
1809 env->fpdp = fpu.last_dp;
1810 for (i = 0; i < 8; ++i) {
1811 env->fptags[i] = !((fpu.ftwx >> i) & 1);
1812 }
1813 memcpy(env->fpregs, fpu.fpr, sizeof env->fpregs);
1814 for (i = 0; i < CPU_NB_REGS; i++) {
1815 env->xmm_regs[i].ZMM_Q(0) = ldq_p(&fpu.xmm[i][0]);
1816 env->xmm_regs[i].ZMM_Q(1) = ldq_p(&fpu.xmm[i][8]);
1817 }
1818 env->mxcsr = fpu.mxcsr;
1819
1820 return 0;
1821 }
1822
1823 static int kvm_get_xsave(X86CPU *cpu)
1824 {
1825 CPUX86State *env = &cpu->env;
1826 X86XSaveArea *xsave = env->kvm_xsave_buf;
1827 int ret;
1828
1829 if (!has_xsave) {
1830 return kvm_get_fpu(cpu);
1831 }
1832
1833 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_XSAVE, xsave);
1834 if (ret < 0) {
1835 return ret;
1836 }
1837 x86_cpu_xrstor_all_areas(cpu, xsave);
1838
1839 return 0;
1840 }
1841
1842 static int kvm_get_xcrs(X86CPU *cpu)
1843 {
1844 CPUX86State *env = &cpu->env;
1845 int i, ret;
1846 struct kvm_xcrs xcrs;
1847
1848 if (!has_xcrs) {
1849 return 0;
1850 }
1851
1852 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_XCRS, &xcrs);
1853 if (ret < 0) {
1854 return ret;
1855 }
1856
1857 for (i = 0; i < xcrs.nr_xcrs; i++) {
1858 /* Only support xcr0 now */
1859 if (xcrs.xcrs[i].xcr == 0) {
1860 env->xcr0 = xcrs.xcrs[i].value;
1861 break;
1862 }
1863 }
1864 return 0;
1865 }
1866
1867 static int kvm_get_sregs(X86CPU *cpu)
1868 {
1869 CPUX86State *env = &cpu->env;
1870 struct kvm_sregs sregs;
1871 uint32_t hflags;
1872 int bit, i, ret;
1873
1874 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
1875 if (ret < 0) {
1876 return ret;
1877 }
1878
1879 /* There can only be one pending IRQ set in the bitmap at a time, so try
1880 to find it and save its number instead (-1 for none). */
1881 env->interrupt_injected = -1;
1882 for (i = 0; i < ARRAY_SIZE(sregs.interrupt_bitmap); i++) {
1883 if (sregs.interrupt_bitmap[i]) {
1884 bit = ctz64(sregs.interrupt_bitmap[i]);
1885 env->interrupt_injected = i * 64 + bit;
1886 break;
1887 }
1888 }
1889
1890 get_seg(&env->segs[R_CS], &sregs.cs);
1891 get_seg(&env->segs[R_DS], &sregs.ds);
1892 get_seg(&env->segs[R_ES], &sregs.es);
1893 get_seg(&env->segs[R_FS], &sregs.fs);
1894 get_seg(&env->segs[R_GS], &sregs.gs);
1895 get_seg(&env->segs[R_SS], &sregs.ss);
1896
1897 get_seg(&env->tr, &sregs.tr);
1898 get_seg(&env->ldt, &sregs.ldt);
1899
1900 env->idt.limit = sregs.idt.limit;
1901 env->idt.base = sregs.idt.base;
1902 env->gdt.limit = sregs.gdt.limit;
1903 env->gdt.base = sregs.gdt.base;
1904
1905 env->cr[0] = sregs.cr0;
1906 env->cr[2] = sregs.cr2;
1907 env->cr[3] = sregs.cr3;
1908 env->cr[4] = sregs.cr4;
1909
1910 env->efer = sregs.efer;
1911
1912 /* changes to apic base and cr8/tpr are read back via kvm_arch_post_run */
1913
1914 #define HFLAG_COPY_MASK \
1915 ~( HF_CPL_MASK | HF_PE_MASK | HF_MP_MASK | HF_EM_MASK | \
1916 HF_TS_MASK | HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK | \
1917 HF_OSFXSR_MASK | HF_LMA_MASK | HF_CS32_MASK | \
1918 HF_SS32_MASK | HF_CS64_MASK | HF_ADDSEG_MASK)
1919
1920 hflags = env->hflags & HFLAG_COPY_MASK;
1921 hflags |= (env->segs[R_SS].flags >> DESC_DPL_SHIFT) & HF_CPL_MASK;
1922 hflags |= (env->cr[0] & CR0_PE_MASK) << (HF_PE_SHIFT - CR0_PE_SHIFT);
1923 hflags |= (env->cr[0] << (HF_MP_SHIFT - CR0_MP_SHIFT)) &
1924 (HF_MP_MASK | HF_EM_MASK | HF_TS_MASK);
1925 hflags |= (env->eflags & (HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK));
1926
1927 if (env->cr[4] & CR4_OSFXSR_MASK) {
1928 hflags |= HF_OSFXSR_MASK;
1929 }
1930
1931 if (env->efer & MSR_EFER_LMA) {
1932 hflags |= HF_LMA_MASK;
1933 }
1934
1935 if ((hflags & HF_LMA_MASK) && (env->segs[R_CS].flags & DESC_L_MASK)) {
1936 hflags |= HF_CS32_MASK | HF_SS32_MASK | HF_CS64_MASK;
1937 } else {
1938 hflags |= (env->segs[R_CS].flags & DESC_B_MASK) >>
1939 (DESC_B_SHIFT - HF_CS32_SHIFT);
1940 hflags |= (env->segs[R_SS].flags & DESC_B_MASK) >>
1941 (DESC_B_SHIFT - HF_SS32_SHIFT);
1942 if (!(env->cr[0] & CR0_PE_MASK) || (env->eflags & VM_MASK) ||
1943 !(hflags & HF_CS32_MASK)) {
1944 hflags |= HF_ADDSEG_MASK;
1945 } else {
1946 hflags |= ((env->segs[R_DS].base | env->segs[R_ES].base |
1947 env->segs[R_SS].base) != 0) << HF_ADDSEG_SHIFT;
1948 }
1949 }
1950 env->hflags = hflags;
1951
1952 return 0;
1953 }
1954
1955 static int kvm_get_msrs(X86CPU *cpu)
1956 {
1957 CPUX86State *env = &cpu->env;
1958 struct kvm_msr_entry *msrs = cpu->kvm_msr_buf->entries;
1959 int ret, i;
1960 uint64_t mtrr_top_bits;
1961
1962 kvm_msr_buf_reset(cpu);
1963
1964 kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_CS, 0);
1965 kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_ESP, 0);
1966 kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_EIP, 0);
1967 kvm_msr_entry_add(cpu, MSR_PAT, 0);
1968 if (has_msr_star) {
1969 kvm_msr_entry_add(cpu, MSR_STAR, 0);
1970 }
1971 if (has_msr_hsave_pa) {
1972 kvm_msr_entry_add(cpu, MSR_VM_HSAVE_PA, 0);
1973 }
1974 if (has_msr_tsc_aux) {
1975 kvm_msr_entry_add(cpu, MSR_TSC_AUX, 0);
1976 }
1977 if (has_msr_tsc_adjust) {
1978 kvm_msr_entry_add(cpu, MSR_TSC_ADJUST, 0);
1979 }
1980 if (has_msr_tsc_deadline) {
1981 kvm_msr_entry_add(cpu, MSR_IA32_TSCDEADLINE, 0);
1982 }
1983 if (has_msr_misc_enable) {
1984 kvm_msr_entry_add(cpu, MSR_IA32_MISC_ENABLE, 0);
1985 }
1986 if (has_msr_smbase) {
1987 kvm_msr_entry_add(cpu, MSR_IA32_SMBASE, 0);
1988 }
1989 if (has_msr_feature_control) {
1990 kvm_msr_entry_add(cpu, MSR_IA32_FEATURE_CONTROL, 0);
1991 }
1992 if (has_msr_bndcfgs) {
1993 kvm_msr_entry_add(cpu, MSR_IA32_BNDCFGS, 0);
1994 }
1995 if (has_msr_xss) {
1996 kvm_msr_entry_add(cpu, MSR_IA32_XSS, 0);
1997 }
1998
1999
2000 if (!env->tsc_valid) {
2001 kvm_msr_entry_add(cpu, MSR_IA32_TSC, 0);
2002 env->tsc_valid = !runstate_is_running();
2003 }
2004
2005 #ifdef TARGET_X86_64
2006 if (lm_capable_kernel) {
2007 kvm_msr_entry_add(cpu, MSR_CSTAR, 0);
2008 kvm_msr_entry_add(cpu, MSR_KERNELGSBASE, 0);
2009 kvm_msr_entry_add(cpu, MSR_FMASK, 0);
2010 kvm_msr_entry_add(cpu, MSR_LSTAR, 0);
2011 }
2012 #endif
2013 kvm_msr_entry_add(cpu, MSR_KVM_SYSTEM_TIME, 0);
2014 kvm_msr_entry_add(cpu, MSR_KVM_WALL_CLOCK, 0);
2015 if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_ASYNC_PF)) {
2016 kvm_msr_entry_add(cpu, MSR_KVM_ASYNC_PF_EN, 0);
2017 }
2018 if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_PV_EOI)) {
2019 kvm_msr_entry_add(cpu, MSR_KVM_PV_EOI_EN, 0);
2020 }
2021 if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_STEAL_TIME)) {
2022 kvm_msr_entry_add(cpu, MSR_KVM_STEAL_TIME, 0);
2023 }
2024 if (has_msr_architectural_pmu) {
2025 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR_CTRL, 0);
2026 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_CTRL, 0);
2027 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_STATUS, 0);
2028 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_OVF_CTRL, 0);
2029 for (i = 0; i < MAX_FIXED_COUNTERS; i++) {
2030 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR0 + i, 0);
2031 }
2032 for (i = 0; i < num_architectural_pmu_counters; i++) {
2033 kvm_msr_entry_add(cpu, MSR_P6_PERFCTR0 + i, 0);
2034 kvm_msr_entry_add(cpu, MSR_P6_EVNTSEL0 + i, 0);
2035 }
2036 }
2037
2038 if (env->mcg_cap) {
2039 kvm_msr_entry_add(cpu, MSR_MCG_STATUS, 0);
2040 kvm_msr_entry_add(cpu, MSR_MCG_CTL, 0);
2041 if (has_msr_mcg_ext_ctl) {
2042 kvm_msr_entry_add(cpu, MSR_MCG_EXT_CTL, 0);
2043 }
2044 for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++) {
2045 kvm_msr_entry_add(cpu, MSR_MC0_CTL + i, 0);
2046 }
2047 }
2048
2049 if (has_msr_hv_hypercall) {
2050 kvm_msr_entry_add(cpu, HV_X64_MSR_HYPERCALL, 0);
2051 kvm_msr_entry_add(cpu, HV_X64_MSR_GUEST_OS_ID, 0);
2052 }
2053 if (cpu->hyperv_vapic) {
2054 kvm_msr_entry_add(cpu, HV_X64_MSR_APIC_ASSIST_PAGE, 0);
2055 }
2056 if (cpu->hyperv_time) {
2057 kvm_msr_entry_add(cpu, HV_X64_MSR_REFERENCE_TSC, 0);
2058 }
2059 if (has_msr_hv_crash) {
2060 int j;
2061
2062 for (j = 0; j < HV_X64_MSR_CRASH_PARAMS; j++) {
2063 kvm_msr_entry_add(cpu, HV_X64_MSR_CRASH_P0 + j, 0);
2064 }
2065 }
2066 if (has_msr_hv_runtime) {
2067 kvm_msr_entry_add(cpu, HV_X64_MSR_VP_RUNTIME, 0);
2068 }
2069 if (cpu->hyperv_synic) {
2070 uint32_t msr;
2071
2072 kvm_msr_entry_add(cpu, HV_X64_MSR_SCONTROL, 0);
2073 kvm_msr_entry_add(cpu, HV_X64_MSR_SVERSION, 0);
2074 kvm_msr_entry_add(cpu, HV_X64_MSR_SIEFP, 0);
2075 kvm_msr_entry_add(cpu, HV_X64_MSR_SIMP, 0);
2076 for (msr = HV_X64_MSR_SINT0; msr <= HV_X64_MSR_SINT15; msr++) {
2077 kvm_msr_entry_add(cpu, msr, 0);
2078 }
2079 }
2080 if (has_msr_hv_stimer) {
2081 uint32_t msr;
2082
2083 for (msr = HV_X64_MSR_STIMER0_CONFIG; msr <= HV_X64_MSR_STIMER3_COUNT;
2084 msr++) {
2085 kvm_msr_entry_add(cpu, msr, 0);
2086 }
2087 }
2088 if (env->features[FEAT_1_EDX] & CPUID_MTRR) {
2089 kvm_msr_entry_add(cpu, MSR_MTRRdefType, 0);
2090 kvm_msr_entry_add(cpu, MSR_MTRRfix64K_00000, 0);
2091 kvm_msr_entry_add(cpu, MSR_MTRRfix16K_80000, 0);
2092 kvm_msr_entry_add(cpu, MSR_MTRRfix16K_A0000, 0);
2093 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C0000, 0);
2094 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C8000, 0);
2095 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D0000, 0);
2096 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D8000, 0);
2097 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E0000, 0);
2098 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E8000, 0);
2099 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F0000, 0);
2100 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F8000, 0);
2101 for (i = 0; i < MSR_MTRRcap_VCNT; i++) {
2102 kvm_msr_entry_add(cpu, MSR_MTRRphysBase(i), 0);
2103 kvm_msr_entry_add(cpu, MSR_MTRRphysMask(i), 0);
2104 }
2105 }
2106
2107 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MSRS, cpu->kvm_msr_buf);
2108 if (ret < 0) {
2109 return ret;
2110 }
2111
2112 if (ret < cpu->kvm_msr_buf->nmsrs) {
2113 struct kvm_msr_entry *e = &cpu->kvm_msr_buf->entries[ret];
2114 error_report("error: failed to get MSR 0x%" PRIx32,
2115 (uint32_t)e->index);
2116 }
2117
2118 assert(ret == cpu->kvm_msr_buf->nmsrs);
2119 /*
2120 * MTRR masks: Each mask consists of 5 parts
2121 * a 10..0: must be zero
2122 * b 11 : valid bit
2123 * c n-1.12: actual mask bits
2124 * d 51..n: reserved must be zero
2125 * e 63.52: reserved must be zero
2126 *
2127 * 'n' is the number of physical bits supported by the CPU and is
2128 * apparently always <= 52. We know our 'n' but don't know what
2129 * the destinations 'n' is; it might be smaller, in which case
2130 * it masks (c) on loading. It might be larger, in which case
2131 * we fill 'd' so that d..c is consistent irrespetive of the 'n'
2132 * we're migrating to.
2133 */
2134
2135 if (cpu->fill_mtrr_mask) {
2136 QEMU_BUILD_BUG_ON(TARGET_PHYS_ADDR_SPACE_BITS > 52);
2137 assert(cpu->phys_bits <= TARGET_PHYS_ADDR_SPACE_BITS);
2138 mtrr_top_bits = MAKE_64BIT_MASK(cpu->phys_bits, 52 - cpu->phys_bits);
2139 } else {
2140 mtrr_top_bits = 0;
2141 }
2142
2143 for (i = 0; i < ret; i++) {
2144 uint32_t index = msrs[i].index;
2145 switch (index) {
2146 case MSR_IA32_SYSENTER_CS:
2147 env->sysenter_cs = msrs[i].data;
2148 break;
2149 case MSR_IA32_SYSENTER_ESP:
2150 env->sysenter_esp = msrs[i].data;
2151 break;
2152 case MSR_IA32_SYSENTER_EIP:
2153 env->sysenter_eip = msrs[i].data;
2154 break;
2155 case MSR_PAT:
2156 env->pat = msrs[i].data;
2157 break;
2158 case MSR_STAR:
2159 env->star = msrs[i].data;
2160 break;
2161 #ifdef TARGET_X86_64
2162 case MSR_CSTAR:
2163 env->cstar = msrs[i].data;
2164 break;
2165 case MSR_KERNELGSBASE:
2166 env->kernelgsbase = msrs[i].data;
2167 break;
2168 case MSR_FMASK:
2169 env->fmask = msrs[i].data;
2170 break;
2171 case MSR_LSTAR:
2172 env->lstar = msrs[i].data;
2173 break;
2174 #endif
2175 case MSR_IA32_TSC:
2176 env->tsc = msrs[i].data;
2177 break;
2178 case MSR_TSC_AUX:
2179 env->tsc_aux = msrs[i].data;
2180 break;
2181 case MSR_TSC_ADJUST:
2182 env->tsc_adjust = msrs[i].data;
2183 break;
2184 case MSR_IA32_TSCDEADLINE:
2185 env->tsc_deadline = msrs[i].data;
2186 break;
2187 case MSR_VM_HSAVE_PA:
2188 env->vm_hsave = msrs[i].data;
2189 break;
2190 case MSR_KVM_SYSTEM_TIME:
2191 env->system_time_msr = msrs[i].data;
2192 break;
2193 case MSR_KVM_WALL_CLOCK:
2194 env->wall_clock_msr = msrs[i].data;
2195 break;
2196 case MSR_MCG_STATUS:
2197 env->mcg_status = msrs[i].data;
2198 break;
2199 case MSR_MCG_CTL:
2200 env->mcg_ctl = msrs[i].data;
2201 break;
2202 case MSR_MCG_EXT_CTL:
2203 env->mcg_ext_ctl = msrs[i].data;
2204 break;
2205 case MSR_IA32_MISC_ENABLE:
2206 env->msr_ia32_misc_enable = msrs[i].data;
2207 break;
2208 case MSR_IA32_SMBASE:
2209 env->smbase = msrs[i].data;
2210 break;
2211 case MSR_IA32_FEATURE_CONTROL:
2212 env->msr_ia32_feature_control = msrs[i].data;
2213 break;
2214 case MSR_IA32_BNDCFGS:
2215 env->msr_bndcfgs = msrs[i].data;
2216 break;
2217 case MSR_IA32_XSS:
2218 env->xss = msrs[i].data;
2219 break;
2220 default:
2221 if (msrs[i].index >= MSR_MC0_CTL &&
2222 msrs[i].index < MSR_MC0_CTL + (env->mcg_cap & 0xff) * 4) {
2223 env->mce_banks[msrs[i].index - MSR_MC0_CTL] = msrs[i].data;
2224 }
2225 break;
2226 case MSR_KVM_ASYNC_PF_EN:
2227 env->async_pf_en_msr = msrs[i].data;
2228 break;
2229 case MSR_KVM_PV_EOI_EN:
2230 env->pv_eoi_en_msr = msrs[i].data;
2231 break;
2232 case MSR_KVM_STEAL_TIME:
2233 env->steal_time_msr = msrs[i].data;
2234 break;
2235 case MSR_CORE_PERF_FIXED_CTR_CTRL:
2236 env->msr_fixed_ctr_ctrl = msrs[i].data;
2237 break;
2238 case MSR_CORE_PERF_GLOBAL_CTRL:
2239 env->msr_global_ctrl = msrs[i].data;
2240 break;
2241 case MSR_CORE_PERF_GLOBAL_STATUS:
2242 env->msr_global_status = msrs[i].data;
2243 break;
2244 case MSR_CORE_PERF_GLOBAL_OVF_CTRL:
2245 env->msr_global_ovf_ctrl = msrs[i].data;
2246 break;
2247 case MSR_CORE_PERF_FIXED_CTR0 ... MSR_CORE_PERF_FIXED_CTR0 + MAX_FIXED_COUNTERS - 1:
2248 env->msr_fixed_counters[index - MSR_CORE_PERF_FIXED_CTR0] = msrs[i].data;
2249 break;
2250 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR0 + MAX_GP_COUNTERS - 1:
2251 env->msr_gp_counters[index - MSR_P6_PERFCTR0] = msrs[i].data;
2252 break;
2253 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL0 + MAX_GP_COUNTERS - 1:
2254 env->msr_gp_evtsel[index - MSR_P6_EVNTSEL0] = msrs[i].data;
2255 break;
2256 case HV_X64_MSR_HYPERCALL:
2257 env->msr_hv_hypercall = msrs[i].data;
2258 break;
2259 case HV_X64_MSR_GUEST_OS_ID:
2260 env->msr_hv_guest_os_id = msrs[i].data;
2261 break;
2262 case HV_X64_MSR_APIC_ASSIST_PAGE:
2263 env->msr_hv_vapic = msrs[i].data;
2264 break;
2265 case HV_X64_MSR_REFERENCE_TSC:
2266 env->msr_hv_tsc = msrs[i].data;
2267 break;
2268 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2269 env->msr_hv_crash_params[index - HV_X64_MSR_CRASH_P0] = msrs[i].data;
2270 break;
2271 case HV_X64_MSR_VP_RUNTIME:
2272 env->msr_hv_runtime = msrs[i].data;
2273 break;
2274 case HV_X64_MSR_SCONTROL:
2275 env->msr_hv_synic_control = msrs[i].data;
2276 break;
2277 case HV_X64_MSR_SVERSION:
2278 env->msr_hv_synic_version = msrs[i].data;
2279 break;
2280 case HV_X64_MSR_SIEFP:
2281 env->msr_hv_synic_evt_page = msrs[i].data;
2282 break;
2283 case HV_X64_MSR_SIMP:
2284 env->msr_hv_synic_msg_page = msrs[i].data;
2285 break;
2286 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
2287 env->msr_hv_synic_sint[index - HV_X64_MSR_SINT0] = msrs[i].data;
2288 break;
2289 case HV_X64_MSR_STIMER0_CONFIG:
2290 case HV_X64_MSR_STIMER1_CONFIG:
2291 case HV_X64_MSR_STIMER2_CONFIG:
2292 case HV_X64_MSR_STIMER3_CONFIG:
2293 env->msr_hv_stimer_config[(index - HV_X64_MSR_STIMER0_CONFIG)/2] =
2294 msrs[i].data;
2295 break;
2296 case HV_X64_MSR_STIMER0_COUNT:
2297 case HV_X64_MSR_STIMER1_COUNT:
2298 case HV_X64_MSR_STIMER2_COUNT:
2299 case HV_X64_MSR_STIMER3_COUNT:
2300 env->msr_hv_stimer_count[(index - HV_X64_MSR_STIMER0_COUNT)/2] =
2301 msrs[i].data;
2302 break;
2303 case MSR_MTRRdefType:
2304 env->mtrr_deftype = msrs[i].data;
2305 break;
2306 case MSR_MTRRfix64K_00000:
2307 env->mtrr_fixed[0] = msrs[i].data;
2308 break;
2309 case MSR_MTRRfix16K_80000:
2310 env->mtrr_fixed[1] = msrs[i].data;
2311 break;
2312 case MSR_MTRRfix16K_A0000:
2313 env->mtrr_fixed[2] = msrs[i].data;
2314 break;
2315 case MSR_MTRRfix4K_C0000:
2316 env->mtrr_fixed[3] = msrs[i].data;
2317 break;
2318 case MSR_MTRRfix4K_C8000:
2319 env->mtrr_fixed[4] = msrs[i].data;
2320 break;
2321 case MSR_MTRRfix4K_D0000:
2322 env->mtrr_fixed[5] = msrs[i].data;
2323 break;
2324 case MSR_MTRRfix4K_D8000:
2325 env->mtrr_fixed[6] = msrs[i].data;
2326 break;
2327 case MSR_MTRRfix4K_E0000:
2328 env->mtrr_fixed[7] = msrs[i].data;
2329 break;
2330 case MSR_MTRRfix4K_E8000:
2331 env->mtrr_fixed[8] = msrs[i].data;
2332 break;
2333 case MSR_MTRRfix4K_F0000:
2334 env->mtrr_fixed[9] = msrs[i].data;
2335 break;
2336 case MSR_MTRRfix4K_F8000:
2337 env->mtrr_fixed[10] = msrs[i].data;
2338 break;
2339 case MSR_MTRRphysBase(0) ... MSR_MTRRphysMask(MSR_MTRRcap_VCNT - 1):
2340 if (index & 1) {
2341 env->mtrr_var[MSR_MTRRphysIndex(index)].mask = msrs[i].data |
2342 mtrr_top_bits;
2343 } else {
2344 env->mtrr_var[MSR_MTRRphysIndex(index)].base = msrs[i].data;
2345 }
2346 break;
2347 }
2348 }
2349
2350 return 0;
2351 }
2352
2353 static int kvm_put_mp_state(X86CPU *cpu)
2354 {
2355 struct kvm_mp_state mp_state = { .mp_state = cpu->env.mp_state };
2356
2357 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
2358 }
2359
2360 static int kvm_get_mp_state(X86CPU *cpu)
2361 {
2362 CPUState *cs = CPU(cpu);
2363 CPUX86State *env = &cpu->env;
2364 struct kvm_mp_state mp_state;
2365 int ret;
2366
2367 ret = kvm_vcpu_ioctl(cs, KVM_GET_MP_STATE, &mp_state);
2368 if (ret < 0) {
2369 return ret;
2370 }
2371 env->mp_state = mp_state.mp_state;
2372 if (kvm_irqchip_in_kernel()) {
2373 cs->halted = (mp_state.mp_state == KVM_MP_STATE_HALTED);
2374 }
2375 return 0;
2376 }
2377
2378 static int kvm_get_apic(X86CPU *cpu)
2379 {
2380 DeviceState *apic = cpu->apic_state;
2381 struct kvm_lapic_state kapic;
2382 int ret;
2383
2384 if (apic && kvm_irqchip_in_kernel()) {
2385 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_LAPIC, &kapic);
2386 if (ret < 0) {
2387 return ret;
2388 }
2389
2390 kvm_get_apic_state(apic, &kapic);
2391 }
2392 return 0;
2393 }
2394
2395 static int kvm_put_vcpu_events(X86CPU *cpu, int level)
2396 {
2397 CPUState *cs = CPU(cpu);
2398 CPUX86State *env = &cpu->env;
2399 struct kvm_vcpu_events events = {};
2400
2401 if (!kvm_has_vcpu_events()) {
2402 return 0;
2403 }
2404
2405 events.exception.injected = (env->exception_injected >= 0);
2406 events.exception.nr = env->exception_injected;
2407 events.exception.has_error_code = env->has_error_code;
2408 events.exception.error_code = env->error_code;
2409 events.exception.pad = 0;
2410
2411 events.interrupt.injected = (env->interrupt_injected >= 0);
2412 events.interrupt.nr = env->interrupt_injected;
2413 events.interrupt.soft = env->soft_interrupt;
2414
2415 events.nmi.injected = env->nmi_injected;
2416 events.nmi.pending = env->nmi_pending;
2417 events.nmi.masked = !!(env->hflags2 & HF2_NMI_MASK);
2418 events.nmi.pad = 0;
2419
2420 events.sipi_vector = env->sipi_vector;
2421 events.flags = 0;
2422
2423 if (has_msr_smbase) {
2424 events.smi.smm = !!(env->hflags & HF_SMM_MASK);
2425 events.smi.smm_inside_nmi = !!(env->hflags2 & HF2_SMM_INSIDE_NMI_MASK);
2426 if (kvm_irqchip_in_kernel()) {
2427 /* As soon as these are moved to the kernel, remove them
2428 * from cs->interrupt_request.
2429 */
2430 events.smi.pending = cs->interrupt_request & CPU_INTERRUPT_SMI;
2431 events.smi.latched_init = cs->interrupt_request & CPU_INTERRUPT_INIT;
2432 cs->interrupt_request &= ~(CPU_INTERRUPT_INIT | CPU_INTERRUPT_SMI);
2433 } else {
2434 /* Keep these in cs->interrupt_request. */
2435 events.smi.pending = 0;
2436 events.smi.latched_init = 0;
2437 }
2438 /* Stop SMI delivery on old machine types to avoid a reboot
2439 * on an inward migration of an old VM.
2440 */
2441 if (!cpu->kvm_no_smi_migration) {
2442 events.flags |= KVM_VCPUEVENT_VALID_SMM;
2443 }
2444 }
2445
2446 if (level >= KVM_PUT_RESET_STATE) {
2447 events.flags |= KVM_VCPUEVENT_VALID_NMI_PENDING;
2448 if (env->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
2449 events.flags |= KVM_VCPUEVENT_VALID_SIPI_VECTOR;
2450 }
2451 }
2452
2453 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_VCPU_EVENTS, &events);
2454 }
2455
2456 static int kvm_get_vcpu_events(X86CPU *cpu)
2457 {
2458 CPUX86State *env = &cpu->env;
2459 struct kvm_vcpu_events events;
2460 int ret;
2461
2462 if (!kvm_has_vcpu_events()) {
2463 return 0;
2464 }
2465
2466 memset(&events, 0, sizeof(events));
2467 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_VCPU_EVENTS, &events);
2468 if (ret < 0) {
2469 return ret;
2470 }
2471 env->exception_injected =
2472 events.exception.injected ? events.exception.nr : -1;
2473 env->has_error_code = events.exception.has_error_code;
2474 env->error_code = events.exception.error_code;
2475
2476 env->interrupt_injected =
2477 events.interrupt.injected ? events.interrupt.nr : -1;
2478 env->soft_interrupt = events.interrupt.soft;
2479
2480 env->nmi_injected = events.nmi.injected;
2481 env->nmi_pending = events.nmi.pending;
2482 if (events.nmi.masked) {
2483 env->hflags2 |= HF2_NMI_MASK;
2484 } else {
2485 env->hflags2 &= ~HF2_NMI_MASK;
2486 }
2487
2488 if (events.flags & KVM_VCPUEVENT_VALID_SMM) {
2489 if (events.smi.smm) {
2490 env->hflags |= HF_SMM_MASK;
2491 } else {
2492 env->hflags &= ~HF_SMM_MASK;
2493 }
2494 if (events.smi.pending) {
2495 cpu_interrupt(CPU(cpu), CPU_INTERRUPT_SMI);
2496 } else {
2497 cpu_reset_interrupt(CPU(cpu), CPU_INTERRUPT_SMI);
2498 }
2499 if (events.smi.smm_inside_nmi) {
2500 env->hflags2 |= HF2_SMM_INSIDE_NMI_MASK;
2501 } else {
2502 env->hflags2 &= ~HF2_SMM_INSIDE_NMI_MASK;
2503 }
2504 if (events.smi.latched_init) {
2505 cpu_interrupt(CPU(cpu), CPU_INTERRUPT_INIT);
2506 } else {
2507 cpu_reset_interrupt(CPU(cpu), CPU_INTERRUPT_INIT);
2508 }
2509 }
2510
2511 env->sipi_vector = events.sipi_vector;
2512
2513 return 0;
2514 }
2515
2516 static int kvm_guest_debug_workarounds(X86CPU *cpu)
2517 {
2518 CPUState *cs = CPU(cpu);
2519 CPUX86State *env = &cpu->env;
2520 int ret = 0;
2521 unsigned long reinject_trap = 0;
2522
2523 if (!kvm_has_vcpu_events()) {
2524 if (env->exception_injected == 1) {
2525 reinject_trap = KVM_GUESTDBG_INJECT_DB;
2526 } else if (env->exception_injected == 3) {
2527 reinject_trap = KVM_GUESTDBG_INJECT_BP;
2528 }
2529 env->exception_injected = -1;
2530 }
2531
2532 /*
2533 * Kernels before KVM_CAP_X86_ROBUST_SINGLESTEP overwrote flags.TF
2534 * injected via SET_GUEST_DEBUG while updating GP regs. Work around this
2535 * by updating the debug state once again if single-stepping is on.
2536 * Another reason to call kvm_update_guest_debug here is a pending debug
2537 * trap raise by the guest. On kernels without SET_VCPU_EVENTS we have to
2538 * reinject them via SET_GUEST_DEBUG.
2539 */
2540 if (reinject_trap ||
2541 (!kvm_has_robust_singlestep() && cs->singlestep_enabled)) {
2542 ret = kvm_update_guest_debug(cs, reinject_trap);
2543 }
2544 return ret;
2545 }
2546
2547 static int kvm_put_debugregs(X86CPU *cpu)
2548 {
2549 CPUX86State *env = &cpu->env;
2550 struct kvm_debugregs dbgregs;
2551 int i;
2552
2553 if (!kvm_has_debugregs()) {
2554 return 0;
2555 }
2556
2557 for (i = 0; i < 4; i++) {
2558 dbgregs.db[i] = env->dr[i];
2559 }
2560 dbgregs.dr6 = env->dr[6];
2561 dbgregs.dr7 = env->dr[7];
2562 dbgregs.flags = 0;
2563
2564 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_DEBUGREGS, &dbgregs);
2565 }
2566
2567 static int kvm_get_debugregs(X86CPU *cpu)
2568 {
2569 CPUX86State *env = &cpu->env;
2570 struct kvm_debugregs dbgregs;
2571 int i, ret;
2572
2573 if (!kvm_has_debugregs()) {
2574 return 0;
2575 }
2576
2577 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_DEBUGREGS, &dbgregs);
2578 if (ret < 0) {
2579 return ret;
2580 }
2581 for (i = 0; i < 4; i++) {
2582 env->dr[i] = dbgregs.db[i];
2583 }
2584 env->dr[4] = env->dr[6] = dbgregs.dr6;
2585 env->dr[5] = env->dr[7] = dbgregs.dr7;
2586
2587 return 0;
2588 }
2589
2590 int kvm_arch_put_registers(CPUState *cpu, int level)
2591 {
2592 X86CPU *x86_cpu = X86_CPU(cpu);
2593 int ret;
2594
2595 assert(cpu_is_stopped(cpu) || qemu_cpu_is_self(cpu));
2596
2597 if (level >= KVM_PUT_RESET_STATE) {
2598 ret = kvm_put_msr_feature_control(x86_cpu);
2599 if (ret < 0) {
2600 return ret;
2601 }
2602 }
2603
2604 if (level == KVM_PUT_FULL_STATE) {
2605 /* We don't check for kvm_arch_set_tsc_khz() errors here,
2606 * because TSC frequency mismatch shouldn't abort migration,
2607 * unless the user explicitly asked for a more strict TSC
2608 * setting (e.g. using an explicit "tsc-freq" option).
2609 */
2610 kvm_arch_set_tsc_khz(cpu);
2611 }
2612
2613 ret = kvm_getput_regs(x86_cpu, 1);
2614 if (ret < 0) {
2615 return ret;
2616 }
2617 ret = kvm_put_xsave(x86_cpu);
2618 if (ret < 0) {
2619 return ret;
2620 }
2621 ret = kvm_put_xcrs(x86_cpu);
2622 if (ret < 0) {
2623 return ret;
2624 }
2625 ret = kvm_put_sregs(x86_cpu);
2626 if (ret < 0) {
2627 return ret;
2628 }
2629 /* must be before kvm_put_msrs */
2630 ret = kvm_inject_mce_oldstyle(x86_cpu);
2631 if (ret < 0) {
2632 return ret;
2633 }
2634 ret = kvm_put_msrs(x86_cpu, level);
2635 if (ret < 0) {
2636 return ret;
2637 }
2638 ret = kvm_put_vcpu_events(x86_cpu, level);
2639 if (ret < 0) {
2640 return ret;
2641 }
2642 if (level >= KVM_PUT_RESET_STATE) {
2643 ret = kvm_put_mp_state(x86_cpu);
2644 if (ret < 0) {
2645 return ret;
2646 }
2647 }
2648
2649 ret = kvm_put_tscdeadline_msr(x86_cpu);
2650 if (ret < 0) {
2651 return ret;
2652 }
2653 ret = kvm_put_debugregs(x86_cpu);
2654 if (ret < 0) {
2655 return ret;
2656 }
2657 /* must be last */
2658 ret = kvm_guest_debug_workarounds(x86_cpu);
2659 if (ret < 0) {
2660 return ret;
2661 }
2662 return 0;
2663 }
2664
2665 int kvm_arch_get_registers(CPUState *cs)
2666 {
2667 X86CPU *cpu = X86_CPU(cs);
2668 int ret;
2669
2670 assert(cpu_is_stopped(cs) || qemu_cpu_is_self(cs));
2671
2672 ret = kvm_get_vcpu_events(cpu);
2673 if (ret < 0) {
2674 goto out;
2675 }
2676 /*
2677 * KVM_GET_MPSTATE can modify CS and RIP, call it before
2678 * KVM_GET_REGS and KVM_GET_SREGS.
2679 */
2680 ret = kvm_get_mp_state(cpu);
2681 if (ret < 0) {
2682 goto out;
2683 }
2684 ret = kvm_getput_regs(cpu, 0);
2685 if (ret < 0) {
2686 goto out;
2687 }
2688 ret = kvm_get_xsave(cpu);
2689 if (ret < 0) {
2690 goto out;
2691 }
2692 ret = kvm_get_xcrs(cpu);
2693 if (ret < 0) {
2694 goto out;
2695 }
2696 ret = kvm_get_sregs(cpu);
2697 if (ret < 0) {
2698 goto out;
2699 }
2700 ret = kvm_get_msrs(cpu);
2701 if (ret < 0) {
2702 goto out;
2703 }
2704 ret = kvm_get_apic(cpu);
2705 if (ret < 0) {
2706 goto out;
2707 }
2708 ret = kvm_get_debugregs(cpu);
2709 if (ret < 0) {
2710 goto out;
2711 }
2712 ret = 0;
2713 out:
2714 cpu_sync_bndcs_hflags(&cpu->env);
2715 return ret;
2716 }
2717
2718 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
2719 {
2720 X86CPU *x86_cpu = X86_CPU(cpu);
2721 CPUX86State *env = &x86_cpu->env;
2722 int ret;
2723
2724 /* Inject NMI */
2725 if (cpu->interrupt_request & (CPU_INTERRUPT_NMI | CPU_INTERRUPT_SMI)) {
2726 if (cpu->interrupt_request & CPU_INTERRUPT_NMI) {
2727 qemu_mutex_lock_iothread();
2728 cpu->interrupt_request &= ~CPU_INTERRUPT_NMI;
2729 qemu_mutex_unlock_iothread();
2730 DPRINTF("injected NMI\n");
2731 ret = kvm_vcpu_ioctl(cpu, KVM_NMI);
2732 if (ret < 0) {
2733 fprintf(stderr, "KVM: injection failed, NMI lost (%s)\n",
2734 strerror(-ret));
2735 }
2736 }
2737 if (cpu->interrupt_request & CPU_INTERRUPT_SMI) {
2738 qemu_mutex_lock_iothread();
2739 cpu->interrupt_request &= ~CPU_INTERRUPT_SMI;
2740 qemu_mutex_unlock_iothread();
2741 DPRINTF("injected SMI\n");
2742 ret = kvm_vcpu_ioctl(cpu, KVM_SMI);
2743 if (ret < 0) {
2744 fprintf(stderr, "KVM: injection failed, SMI lost (%s)\n",
2745 strerror(-ret));
2746 }
2747 }
2748 }
2749
2750 if (!kvm_pic_in_kernel()) {
2751 qemu_mutex_lock_iothread();
2752 }
2753
2754 /* Force the VCPU out of its inner loop to process any INIT requests
2755 * or (for userspace APIC, but it is cheap to combine the checks here)
2756 * pending TPR access reports.
2757 */
2758 if (cpu->interrupt_request & (CPU_INTERRUPT_INIT | CPU_INTERRUPT_TPR)) {
2759 if ((cpu->interrupt_request & CPU_INTERRUPT_INIT) &&
2760 !(env->hflags & HF_SMM_MASK)) {
2761 cpu->exit_request = 1;
2762 }
2763 if (cpu->interrupt_request & CPU_INTERRUPT_TPR) {
2764 cpu->exit_request = 1;
2765 }
2766 }
2767
2768 if (!kvm_pic_in_kernel()) {
2769 /* Try to inject an interrupt if the guest can accept it */
2770 if (run->ready_for_interrupt_injection &&
2771 (cpu->interrupt_request & CPU_INTERRUPT_HARD) &&
2772 (env->eflags & IF_MASK)) {
2773 int irq;
2774
2775 cpu->interrupt_request &= ~CPU_INTERRUPT_HARD;
2776 irq = cpu_get_pic_interrupt(env);
2777 if (irq >= 0) {
2778 struct kvm_interrupt intr;
2779
2780 intr.irq = irq;
2781 DPRINTF("injected interrupt %d\n", irq);
2782 ret = kvm_vcpu_ioctl(cpu, KVM_INTERRUPT, &intr);
2783 if (ret < 0) {
2784 fprintf(stderr,
2785 "KVM: injection failed, interrupt lost (%s)\n",
2786 strerror(-ret));
2787 }
2788 }
2789 }
2790
2791 /* If we have an interrupt but the guest is not ready to receive an
2792 * interrupt, request an interrupt window exit. This will
2793 * cause a return to userspace as soon as the guest is ready to
2794 * receive interrupts. */
2795 if ((cpu->interrupt_request & CPU_INTERRUPT_HARD)) {
2796 run->request_interrupt_window = 1;
2797 } else {
2798 run->request_interrupt_window = 0;
2799 }
2800
2801 DPRINTF("setting tpr\n");
2802 run->cr8 = cpu_get_apic_tpr(x86_cpu->apic_state);
2803
2804 qemu_mutex_unlock_iothread();
2805 }
2806 }
2807
2808 MemTxAttrs kvm_arch_post_run(CPUState *cpu, struct kvm_run *run)
2809 {
2810 X86CPU *x86_cpu = X86_CPU(cpu);
2811 CPUX86State *env = &x86_cpu->env;
2812
2813 if (run->flags & KVM_RUN_X86_SMM) {
2814 env->hflags |= HF_SMM_MASK;
2815 } else {
2816 env->hflags &= ~HF_SMM_MASK;
2817 }
2818 if (run->if_flag) {
2819 env->eflags |= IF_MASK;
2820 } else {
2821 env->eflags &= ~IF_MASK;
2822 }
2823
2824 /* We need to protect the apic state against concurrent accesses from
2825 * different threads in case the userspace irqchip is used. */
2826 if (!kvm_irqchip_in_kernel()) {
2827 qemu_mutex_lock_iothread();
2828 }
2829 cpu_set_apic_tpr(x86_cpu->apic_state, run->cr8);
2830 cpu_set_apic_base(x86_cpu->apic_state, run->apic_base);
2831 if (!kvm_irqchip_in_kernel()) {
2832 qemu_mutex_unlock_iothread();
2833 }
2834 return cpu_get_mem_attrs(env);
2835 }
2836
2837 int kvm_arch_process_async_events(CPUState *cs)
2838 {
2839 X86CPU *cpu = X86_CPU(cs);
2840 CPUX86State *env = &cpu->env;
2841
2842 if (cs->interrupt_request & CPU_INTERRUPT_MCE) {
2843 /* We must not raise CPU_INTERRUPT_MCE if it's not supported. */
2844 assert(env->mcg_cap);
2845
2846 cs->interrupt_request &= ~CPU_INTERRUPT_MCE;
2847
2848 kvm_cpu_synchronize_state(cs);
2849
2850 if (env->exception_injected == EXCP08_DBLE) {
2851 /* this means triple fault */
2852 qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET);
2853 cs->exit_request = 1;
2854 return 0;
2855 }
2856 env->exception_injected = EXCP12_MCHK;
2857 env->has_error_code = 0;
2858
2859 cs->halted = 0;
2860 if (kvm_irqchip_in_kernel() && env->mp_state == KVM_MP_STATE_HALTED) {
2861 env->mp_state = KVM_MP_STATE_RUNNABLE;
2862 }
2863 }
2864
2865 if ((cs->interrupt_request & CPU_INTERRUPT_INIT) &&
2866 !(env->hflags & HF_SMM_MASK)) {
2867 kvm_cpu_synchronize_state(cs);
2868 do_cpu_init(cpu);
2869 }
2870
2871 if (kvm_irqchip_in_kernel()) {
2872 return 0;
2873 }
2874
2875 if (cs->interrupt_request & CPU_INTERRUPT_POLL) {
2876 cs->interrupt_request &= ~CPU_INTERRUPT_POLL;
2877 apic_poll_irq(cpu->apic_state);
2878 }
2879 if (((cs->interrupt_request & CPU_INTERRUPT_HARD) &&
2880 (env->eflags & IF_MASK)) ||
2881 (cs->interrupt_request & CPU_INTERRUPT_NMI)) {
2882 cs->halted = 0;
2883 }
2884 if (cs->interrupt_request & CPU_INTERRUPT_SIPI) {
2885 kvm_cpu_synchronize_state(cs);
2886 do_cpu_sipi(cpu);
2887 }
2888 if (cs->interrupt_request & CPU_INTERRUPT_TPR) {
2889 cs->interrupt_request &= ~CPU_INTERRUPT_TPR;
2890 kvm_cpu_synchronize_state(cs);
2891 apic_handle_tpr_access_report(cpu->apic_state, env->eip,
2892 env->tpr_access_type);
2893 }
2894
2895 return cs->halted;
2896 }
2897
2898 static int kvm_handle_halt(X86CPU *cpu)
2899 {
2900 CPUState *cs = CPU(cpu);
2901 CPUX86State *env = &cpu->env;
2902
2903 if (!((cs->interrupt_request & CPU_INTERRUPT_HARD) &&
2904 (env->eflags & IF_MASK)) &&
2905 !(cs->interrupt_request & CPU_INTERRUPT_NMI)) {
2906 cs->halted = 1;
2907 return EXCP_HLT;
2908 }
2909
2910 return 0;
2911 }
2912
2913 static int kvm_handle_tpr_access(X86CPU *cpu)
2914 {
2915 CPUState *cs = CPU(cpu);
2916 struct kvm_run *run = cs->kvm_run;
2917
2918 apic_handle_tpr_access_report(cpu->apic_state, run->tpr_access.rip,
2919 run->tpr_access.is_write ? TPR_ACCESS_WRITE
2920 : TPR_ACCESS_READ);
2921 return 1;
2922 }
2923
2924 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
2925 {
2926 static const uint8_t int3 = 0xcc;
2927
2928 if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 1, 0) ||
2929 cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&int3, 1, 1)) {
2930 return -EINVAL;
2931 }
2932 return 0;
2933 }
2934
2935 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
2936 {
2937 uint8_t int3;
2938
2939 if (cpu_memory_rw_debug(cs, bp->pc, &int3, 1, 0) || int3 != 0xcc ||
2940 cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 1, 1)) {
2941 return -EINVAL;
2942 }
2943 return 0;
2944 }
2945
2946 static struct {
2947 target_ulong addr;
2948 int len;
2949 int type;
2950 } hw_breakpoint[4];
2951
2952 static int nb_hw_breakpoint;
2953
2954 static int find_hw_breakpoint(target_ulong addr, int len, int type)
2955 {
2956 int n;
2957
2958 for (n = 0; n < nb_hw_breakpoint; n++) {
2959 if (hw_breakpoint[n].addr == addr && hw_breakpoint[n].type == type &&
2960 (hw_breakpoint[n].len == len || len == -1)) {
2961 return n;
2962 }
2963 }
2964 return -1;
2965 }
2966
2967 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
2968 target_ulong len, int type)
2969 {
2970 switch (type) {
2971 case GDB_BREAKPOINT_HW:
2972 len = 1;
2973 break;
2974 case GDB_WATCHPOINT_WRITE:
2975 case GDB_WATCHPOINT_ACCESS:
2976 switch (len) {
2977 case 1:
2978 break;
2979 case 2:
2980 case 4:
2981 case 8:
2982 if (addr & (len - 1)) {
2983 return -EINVAL;
2984 }
2985 break;
2986 default:
2987 return -EINVAL;
2988 }
2989 break;
2990 default:
2991 return -ENOSYS;
2992 }
2993
2994 if (nb_hw_breakpoint == 4) {
2995 return -ENOBUFS;
2996 }
2997 if (find_hw_breakpoint(addr, len, type) >= 0) {
2998 return -EEXIST;
2999 }
3000 hw_breakpoint[nb_hw_breakpoint].addr = addr;
3001 hw_breakpoint[nb_hw_breakpoint].len = len;
3002 hw_breakpoint[nb_hw_breakpoint].type = type;
3003 nb_hw_breakpoint++;
3004
3005 return 0;
3006 }
3007
3008 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
3009 target_ulong len, int type)
3010 {
3011 int n;
3012
3013 n = find_hw_breakpoint(addr, (type == GDB_BREAKPOINT_HW) ? 1 : len, type);
3014 if (n < 0) {
3015 return -ENOENT;
3016 }
3017 nb_hw_breakpoint--;
3018 hw_breakpoint[n] = hw_breakpoint[nb_hw_breakpoint];
3019
3020 return 0;
3021 }
3022
3023 void kvm_arch_remove_all_hw_breakpoints(void)
3024 {
3025 nb_hw_breakpoint = 0;
3026 }
3027
3028 static CPUWatchpoint hw_watchpoint;
3029
3030 static int kvm_handle_debug(X86CPU *cpu,
3031 struct kvm_debug_exit_arch *arch_info)
3032 {
3033 CPUState *cs = CPU(cpu);
3034 CPUX86State *env = &cpu->env;
3035 int ret = 0;
3036 int n;
3037
3038 if (arch_info->exception == 1) {
3039 if (arch_info->dr6 & (1 << 14)) {
3040 if (cs->singlestep_enabled) {
3041 ret = EXCP_DEBUG;
3042 }
3043 } else {
3044 for (n = 0; n < 4; n++) {
3045 if (arch_info->dr6 & (1 << n)) {
3046 switch ((arch_info->dr7 >> (16 + n*4)) & 0x3) {
3047 case 0x0:
3048 ret = EXCP_DEBUG;
3049 break;
3050 case 0x1:
3051 ret = EXCP_DEBUG;
3052 cs->watchpoint_hit = &hw_watchpoint;
3053 hw_watchpoint.vaddr = hw_breakpoint[n].addr;
3054 hw_watchpoint.flags = BP_MEM_WRITE;
3055 break;
3056 case 0x3:
3057 ret = EXCP_DEBUG;
3058 cs->watchpoint_hit = &hw_watchpoint;
3059 hw_watchpoint.vaddr = hw_breakpoint[n].addr;
3060 hw_watchpoint.flags = BP_MEM_ACCESS;
3061 break;
3062 }
3063 }
3064 }
3065 }
3066 } else if (kvm_find_sw_breakpoint(cs, arch_info->pc)) {
3067 ret = EXCP_DEBUG;
3068 }
3069 if (ret == 0) {
3070 cpu_synchronize_state(cs);
3071 assert(env->exception_injected == -1);
3072
3073 /* pass to guest */
3074 env->exception_injected = arch_info->exception;
3075 env->has_error_code = 0;
3076 }
3077
3078 return ret;
3079 }
3080
3081 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
3082 {
3083 const uint8_t type_code[] = {
3084 [GDB_BREAKPOINT_HW] = 0x0,
3085 [GDB_WATCHPOINT_WRITE] = 0x1,
3086 [GDB_WATCHPOINT_ACCESS] = 0x3
3087 };
3088 const uint8_t len_code[] = {
3089 [1] = 0x0, [2] = 0x1, [4] = 0x3, [8] = 0x2
3090 };
3091 int n;
3092
3093 if (kvm_sw_breakpoints_active(cpu)) {
3094 dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
3095 }
3096 if (nb_hw_breakpoint > 0) {
3097 dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
3098 dbg->arch.debugreg[7] = 0x0600;
3099 for (n = 0; n < nb_hw_breakpoint; n++) {
3100 dbg->arch.debugreg[n] = hw_breakpoint[n].addr;
3101 dbg->arch.debugreg[7] |= (2 << (n * 2)) |
3102 (type_code[hw_breakpoint[n].type] << (16 + n*4)) |
3103 ((uint32_t)len_code[hw_breakpoint[n].len] << (18 + n*4));
3104 }
3105 }
3106 }
3107
3108 static bool host_supports_vmx(void)
3109 {
3110 uint32_t ecx, unused;
3111
3112 host_cpuid(1, 0, &unused, &unused, &ecx, &unused);
3113 return ecx & CPUID_EXT_VMX;
3114 }
3115
3116 #define VMX_INVALID_GUEST_STATE 0x80000021
3117
3118 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
3119 {
3120 X86CPU *cpu = X86_CPU(cs);
3121 uint64_t code;
3122 int ret;
3123
3124 switch (run->exit_reason) {
3125 case KVM_EXIT_HLT:
3126 DPRINTF("handle_hlt\n");
3127 qemu_mutex_lock_iothread();
3128 ret = kvm_handle_halt(cpu);
3129 qemu_mutex_unlock_iothread();
3130 break;
3131 case KVM_EXIT_SET_TPR:
3132 ret = 0;
3133 break;
3134 case KVM_EXIT_TPR_ACCESS:
3135 qemu_mutex_lock_iothread();
3136 ret = kvm_handle_tpr_access(cpu);
3137 qemu_mutex_unlock_iothread();
3138 break;
3139 case KVM_EXIT_FAIL_ENTRY:
3140 code = run->fail_entry.hardware_entry_failure_reason;
3141 fprintf(stderr, "KVM: entry failed, hardware error 0x%" PRIx64 "\n",
3142 code);
3143 if (host_supports_vmx() && code == VMX_INVALID_GUEST_STATE) {
3144 fprintf(stderr,
3145 "\nIf you're running a guest on an Intel machine without "
3146 "unrestricted mode\n"
3147 "support, the failure can be most likely due to the guest "
3148 "entering an invalid\n"
3149 "state for Intel VT. For example, the guest maybe running "
3150 "in big real mode\n"
3151 "which is not supported on less recent Intel processors."
3152 "\n\n");
3153 }
3154 ret = -1;
3155 break;
3156 case KVM_EXIT_EXCEPTION:
3157 fprintf(stderr, "KVM: exception %d exit (error code 0x%x)\n",
3158 run->ex.exception, run->ex.error_code);
3159 ret = -1;
3160 break;
3161 case KVM_EXIT_DEBUG:
3162 DPRINTF("kvm_exit_debug\n");
3163 qemu_mutex_lock_iothread();
3164 ret = kvm_handle_debug(cpu, &run->debug.arch);
3165 qemu_mutex_unlock_iothread();
3166 break;
3167 case KVM_EXIT_HYPERV:
3168 ret = kvm_hv_handle_exit(cpu, &run->hyperv);
3169 break;
3170 case KVM_EXIT_IOAPIC_EOI:
3171 ioapic_eoi_broadcast(run->eoi.vector);
3172 ret = 0;
3173 break;
3174 default:
3175 fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
3176 ret = -1;
3177 break;
3178 }
3179
3180 return ret;
3181 }
3182
3183 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
3184 {
3185 X86CPU *cpu = X86_CPU(cs);
3186 CPUX86State *env = &cpu->env;
3187
3188 kvm_cpu_synchronize_state(cs);
3189 return !(env->cr[0] & CR0_PE_MASK) ||
3190 ((env->segs[R_CS].selector & 3) != 3);
3191 }
3192
3193 void kvm_arch_init_irq_routing(KVMState *s)
3194 {
3195 if (!kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
3196 /* If kernel can't do irq routing, interrupt source
3197 * override 0->2 cannot be set up as required by HPET.
3198 * So we have to disable it.
3199 */
3200 no_hpet = 1;
3201 }
3202 /* We know at this point that we're using the in-kernel
3203 * irqchip, so we can use irqfds, and on x86 we know
3204 * we can use msi via irqfd and GSI routing.
3205 */
3206 kvm_msi_via_irqfd_allowed = true;
3207 kvm_gsi_routing_allowed = true;
3208
3209 if (kvm_irqchip_is_split()) {
3210 int i;
3211
3212 /* If the ioapic is in QEMU and the lapics are in KVM, reserve
3213 MSI routes for signaling interrupts to the local apics. */
3214 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
3215 if (kvm_irqchip_add_msi_route(s, 0, NULL) < 0) {
3216 error_report("Could not enable split IRQ mode.");
3217 exit(1);
3218 }
3219 }
3220 }
3221 }
3222
3223 int kvm_arch_irqchip_create(MachineState *ms, KVMState *s)
3224 {
3225 int ret;
3226 if (machine_kernel_irqchip_split(ms)) {
3227 ret = kvm_vm_enable_cap(s, KVM_CAP_SPLIT_IRQCHIP, 0, 24);
3228 if (ret) {
3229 error_report("Could not enable split irqchip mode: %s",
3230 strerror(-ret));
3231 exit(1);
3232 } else {
3233 DPRINTF("Enabled KVM_CAP_SPLIT_IRQCHIP\n");
3234 kvm_split_irqchip = true;
3235 return 1;
3236 }
3237 } else {
3238 return 0;
3239 }
3240 }
3241
3242 /* Classic KVM device assignment interface. Will remain x86 only. */
3243 int kvm_device_pci_assign(KVMState *s, PCIHostDeviceAddress *dev_addr,
3244 uint32_t flags, uint32_t *dev_id)
3245 {
3246 struct kvm_assigned_pci_dev dev_data = {
3247 .segnr = dev_addr->domain,
3248 .busnr = dev_addr->bus,
3249 .devfn = PCI_DEVFN(dev_addr->slot, dev_addr->function),
3250 .flags = flags,
3251 };
3252 int ret;
3253
3254 dev_data.assigned_dev_id =
3255 (dev_addr->domain << 16) | (dev_addr->bus << 8) | dev_data.devfn;
3256
3257 ret = kvm_vm_ioctl(s, KVM_ASSIGN_PCI_DEVICE, &dev_data);
3258 if (ret < 0) {
3259 return ret;
3260 }
3261
3262 *dev_id = dev_data.assigned_dev_id;
3263
3264 return 0;
3265 }
3266
3267 int kvm_device_pci_deassign(KVMState *s, uint32_t dev_id)
3268 {
3269 struct kvm_assigned_pci_dev dev_data = {
3270 .assigned_dev_id = dev_id,
3271 };
3272
3273 return kvm_vm_ioctl(s, KVM_DEASSIGN_PCI_DEVICE, &dev_data);
3274 }
3275
3276 static int kvm_assign_irq_internal(KVMState *s, uint32_t dev_id,
3277 uint32_t irq_type, uint32_t guest_irq)
3278 {
3279 struct kvm_assigned_irq assigned_irq = {
3280 .assigned_dev_id = dev_id,
3281 .guest_irq = guest_irq,
3282 .flags = irq_type,
3283 };
3284
3285 if (kvm_check_extension(s, KVM_CAP_ASSIGN_DEV_IRQ)) {
3286 return kvm_vm_ioctl(s, KVM_ASSIGN_DEV_IRQ, &assigned_irq);
3287 } else {
3288 return kvm_vm_ioctl(s, KVM_ASSIGN_IRQ, &assigned_irq);
3289 }
3290 }
3291
3292 int kvm_device_intx_assign(KVMState *s, uint32_t dev_id, bool use_host_msi,
3293 uint32_t guest_irq)
3294 {
3295 uint32_t irq_type = KVM_DEV_IRQ_GUEST_INTX |
3296 (use_host_msi ? KVM_DEV_IRQ_HOST_MSI : KVM_DEV_IRQ_HOST_INTX);
3297
3298 return kvm_assign_irq_internal(s, dev_id, irq_type, guest_irq);
3299 }
3300
3301 int kvm_device_intx_set_mask(KVMState *s, uint32_t dev_id, bool masked)
3302 {
3303 struct kvm_assigned_pci_dev dev_data = {
3304 .assigned_dev_id = dev_id,
3305 .flags = masked ? KVM_DEV_ASSIGN_MASK_INTX : 0,
3306 };
3307
3308 return kvm_vm_ioctl(s, KVM_ASSIGN_SET_INTX_MASK, &dev_data);
3309 }
3310
3311 static int kvm_deassign_irq_internal(KVMState *s, uint32_t dev_id,
3312 uint32_t type)
3313 {
3314 struct kvm_assigned_irq assigned_irq = {
3315 .assigned_dev_id = dev_id,
3316 .flags = type,
3317 };
3318
3319 return kvm_vm_ioctl(s, KVM_DEASSIGN_DEV_IRQ, &assigned_irq);
3320 }
3321
3322 int kvm_device_intx_deassign(KVMState *s, uint32_t dev_id, bool use_host_msi)
3323 {
3324 return kvm_deassign_irq_internal(s, dev_id, KVM_DEV_IRQ_GUEST_INTX |
3325 (use_host_msi ? KVM_DEV_IRQ_HOST_MSI : KVM_DEV_IRQ_HOST_INTX));
3326 }
3327
3328 int kvm_device_msi_assign(KVMState *s, uint32_t dev_id, int virq)
3329 {
3330 return kvm_assign_irq_internal(s, dev_id, KVM_DEV_IRQ_HOST_MSI |
3331 KVM_DEV_IRQ_GUEST_MSI, virq);
3332 }
3333
3334 int kvm_device_msi_deassign(KVMState *s, uint32_t dev_id)
3335 {
3336 return kvm_deassign_irq_internal(s, dev_id, KVM_DEV_IRQ_GUEST_MSI |
3337 KVM_DEV_IRQ_HOST_MSI);
3338 }
3339
3340 bool kvm_device_msix_supported(KVMState *s)
3341 {
3342 /* The kernel lacks a corresponding KVM_CAP, so we probe by calling
3343 * KVM_ASSIGN_SET_MSIX_NR with an invalid parameter. */
3344 return kvm_vm_ioctl(s, KVM_ASSIGN_SET_MSIX_NR, NULL) == -EFAULT;
3345 }
3346
3347 int kvm_device_msix_init_vectors(KVMState *s, uint32_t dev_id,
3348 uint32_t nr_vectors)
3349 {
3350 struct kvm_assigned_msix_nr msix_nr = {
3351 .assigned_dev_id = dev_id,
3352 .entry_nr = nr_vectors,
3353 };
3354
3355 return kvm_vm_ioctl(s, KVM_ASSIGN_SET_MSIX_NR, &msix_nr);
3356 }
3357
3358 int kvm_device_msix_set_vector(KVMState *s, uint32_t dev_id, uint32_t vector,
3359 int virq)
3360 {
3361 struct kvm_assigned_msix_entry msix_entry = {
3362 .assigned_dev_id = dev_id,
3363 .gsi = virq,
3364 .entry = vector,
3365 };
3366
3367 return kvm_vm_ioctl(s, KVM_ASSIGN_SET_MSIX_ENTRY, &msix_entry);
3368 }
3369
3370 int kvm_device_msix_assign(KVMState *s, uint32_t dev_id)
3371 {
3372 return kvm_assign_irq_internal(s, dev_id, KVM_DEV_IRQ_HOST_MSIX |
3373 KVM_DEV_IRQ_GUEST_MSIX, 0);
3374 }
3375
3376 int kvm_device_msix_deassign(KVMState *s, uint32_t dev_id)
3377 {
3378 return kvm_deassign_irq_internal(s, dev_id, KVM_DEV_IRQ_GUEST_MSIX |
3379 KVM_DEV_IRQ_HOST_MSIX);
3380 }
3381
3382 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
3383 uint64_t address, uint32_t data, PCIDevice *dev)
3384 {
3385 X86IOMMUState *iommu = x86_iommu_get_default();
3386
3387 if (iommu) {
3388 int ret;
3389 MSIMessage src, dst;
3390 X86IOMMUClass *class = X86_IOMMU_GET_CLASS(iommu);
3391
3392 src.address = route->u.msi.address_hi;
3393 src.address <<= VTD_MSI_ADDR_HI_SHIFT;
3394 src.address |= route->u.msi.address_lo;
3395 src.data = route->u.msi.data;
3396
3397 ret = class->int_remap(iommu, &src, &dst, dev ? \
3398 pci_requester_id(dev) : \
3399 X86_IOMMU_SID_INVALID);
3400 if (ret) {
3401 trace_kvm_x86_fixup_msi_error(route->gsi);
3402 return 1;
3403 }
3404
3405 route->u.msi.address_hi = dst.address >> VTD_MSI_ADDR_HI_SHIFT;
3406 route->u.msi.address_lo = dst.address & VTD_MSI_ADDR_LO_MASK;
3407 route->u.msi.data = dst.data;
3408 }
3409
3410 return 0;
3411 }
3412
3413 typedef struct MSIRouteEntry MSIRouteEntry;
3414
3415 struct MSIRouteEntry {
3416 PCIDevice *dev; /* Device pointer */
3417 int vector; /* MSI/MSIX vector index */
3418 int virq; /* Virtual IRQ index */
3419 QLIST_ENTRY(MSIRouteEntry) list;
3420 };
3421
3422 /* List of used GSI routes */
3423 static QLIST_HEAD(, MSIRouteEntry) msi_route_list = \
3424 QLIST_HEAD_INITIALIZER(msi_route_list);
3425
3426 static void kvm_update_msi_routes_all(void *private, bool global,
3427 uint32_t index, uint32_t mask)
3428 {
3429 int cnt = 0;
3430 MSIRouteEntry *entry;
3431 MSIMessage msg;
3432 PCIDevice *dev;
3433
3434 /* TODO: explicit route update */
3435 QLIST_FOREACH(entry, &msi_route_list, list) {
3436 cnt++;
3437 dev = entry->dev;
3438 if (!msix_enabled(dev) && !msi_enabled(dev)) {
3439 continue;
3440 }
3441 msg = pci_get_msi_message(dev, entry->vector);
3442 kvm_irqchip_update_msi_route(kvm_state, entry->virq, msg, dev);
3443 }
3444 kvm_irqchip_commit_routes(kvm_state);
3445 trace_kvm_x86_update_msi_routes(cnt);
3446 }
3447
3448 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
3449 int vector, PCIDevice *dev)
3450 {
3451 static bool notify_list_inited = false;
3452 MSIRouteEntry *entry;
3453
3454 if (!dev) {
3455 /* These are (possibly) IOAPIC routes only used for split
3456 * kernel irqchip mode, while what we are housekeeping are
3457 * PCI devices only. */
3458 return 0;
3459 }
3460
3461 entry = g_new0(MSIRouteEntry, 1);
3462 entry->dev = dev;
3463 entry->vector = vector;
3464 entry->virq = route->gsi;
3465 QLIST_INSERT_HEAD(&msi_route_list, entry, list);
3466
3467 trace_kvm_x86_add_msi_route(route->gsi);
3468
3469 if (!notify_list_inited) {
3470 /* For the first time we do add route, add ourselves into
3471 * IOMMU's IEC notify list if needed. */
3472 X86IOMMUState *iommu = x86_iommu_get_default();
3473 if (iommu) {
3474 x86_iommu_iec_register_notifier(iommu,
3475 kvm_update_msi_routes_all,
3476 NULL);
3477 }
3478 notify_list_inited = true;
3479 }
3480 return 0;
3481 }
3482
3483 int kvm_arch_release_virq_post(int virq)
3484 {
3485 MSIRouteEntry *entry, *next;
3486 QLIST_FOREACH_SAFE(entry, &msi_route_list, list, next) {
3487 if (entry->virq == virq) {
3488 trace_kvm_x86_remove_msi_route(virq);
3489 QLIST_REMOVE(entry, list);
3490 break;
3491 }
3492 }
3493 return 0;
3494 }
3495
3496 int kvm_arch_msi_data_to_gsi(uint32_t data)
3497 {
3498 abort();
3499 }
armbru's QEMU hackery