?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'v3.0.0-rc1'
[qemu/ar7.git] / target / i386 / kvm.c
blob191bf141f7081e015724f6ad00c2d331ccbbca7b
1 /*
2 * QEMU KVM support
3 *
4 * Copyright (C) 2006-2008 Qumranet Technologies
5 * Copyright IBM, Corp. 2008
6 *
7 * Authors:
8 * Anthony Liguori <aliguori@us.ibm.com>
9 *
10 * This work is licensed under the terms of the GNU GPL, version 2 or later.
11 * See the COPYING file in the top-level directory.
12 *
13 */
14
15 #include "qemu/osdep.h"
16 #include "qapi/error.h"
17 #include <sys/ioctl.h>
18 #include <sys/utsname.h>
19
20 #include <linux/kvm.h>
21 #include "standard-headers/asm-x86/kvm_para.h"
22
23 #include "qemu-common.h"
24 #include "cpu.h"
25 #include "sysemu/sysemu.h"
26 #include "sysemu/hw_accel.h"
27 #include "sysemu/kvm_int.h"
28 #include "kvm_i386.h"
29 #include "hyperv.h"
30 #include "hyperv-proto.h"
31
32 #include "exec/gdbstub.h"
33 #include "qemu/host-utils.h"
34 #include "qemu/config-file.h"
35 #include "qemu/error-report.h"
36 #include "hw/i386/pc.h"
37 #include "hw/i386/apic.h"
38 #include "hw/i386/apic_internal.h"
39 #include "hw/i386/apic-msidef.h"
40 #include "hw/i386/intel_iommu.h"
41 #include "hw/i386/x86-iommu.h"
42
43 #include "hw/pci/pci.h"
44 #include "hw/pci/msi.h"
45 #include "hw/pci/msix.h"
46 #include "migration/blocker.h"
47 #include "exec/memattrs.h"
48 #include "trace.h"
49
50 //#define DEBUG_KVM
51
52 #ifdef DEBUG_KVM
53 #define DPRINTF(fmt, ...) \
54 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
55 #else
56 #define DPRINTF(fmt, ...) \
57 do { } while (0)
58 #endif
59
60 #define MSR_KVM_WALL_CLOCK 0x11
61 #define MSR_KVM_SYSTEM_TIME 0x12
62
63 /* A 4096-byte buffer can hold the 8-byte kvm_msrs header, plus
64 * 255 kvm_msr_entry structs */
65 #define MSR_BUF_SIZE 4096
66
67 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
68 KVM_CAP_INFO(SET_TSS_ADDR),
69 KVM_CAP_INFO(EXT_CPUID),
70 KVM_CAP_INFO(MP_STATE),
71 KVM_CAP_LAST_INFO
72 };
73
74 static bool has_msr_star;
75 static bool has_msr_hsave_pa;
76 static bool has_msr_tsc_aux;
77 static bool has_msr_tsc_adjust;
78 static bool has_msr_tsc_deadline;
79 static bool has_msr_feature_control;
80 static bool has_msr_misc_enable;
81 static bool has_msr_smbase;
82 static bool has_msr_bndcfgs;
83 static int lm_capable_kernel;
84 static bool has_msr_hv_hypercall;
85 static bool has_msr_hv_crash;
86 static bool has_msr_hv_reset;
87 static bool has_msr_hv_vpindex;
88 static bool hv_vpindex_settable;
89 static bool has_msr_hv_runtime;
90 static bool has_msr_hv_synic;
91 static bool has_msr_hv_stimer;
92 static bool has_msr_hv_frequencies;
93 static bool has_msr_hv_reenlightenment;
94 static bool has_msr_xss;
95 static bool has_msr_spec_ctrl;
96 static bool has_msr_virt_ssbd;
97 static bool has_msr_smi_count;
98
99 static uint32_t has_architectural_pmu_version;
100 static uint32_t num_architectural_pmu_gp_counters;
101 static uint32_t num_architectural_pmu_fixed_counters;
102
103 static int has_xsave;
104 static int has_xcrs;
105 static int has_pit_state2;
106
107 static bool has_msr_mcg_ext_ctl;
108
109 static struct kvm_cpuid2 *cpuid_cache;
110
111 int kvm_has_pit_state2(void)
112 {
113 return has_pit_state2;
114 }
115
116 bool kvm_has_smm(void)
117 {
118 return kvm_check_extension(kvm_state, KVM_CAP_X86_SMM);
119 }
120
121 bool kvm_has_adjust_clock_stable(void)
122 {
123 int ret = kvm_check_extension(kvm_state, KVM_CAP_ADJUST_CLOCK);
124
125 return (ret == KVM_CLOCK_TSC_STABLE);
126 }
127
128 bool kvm_allows_irq0_override(void)
129 {
130 return !kvm_irqchip_in_kernel() || kvm_has_gsi_routing();
131 }
132
133 static bool kvm_x2apic_api_set_flags(uint64_t flags)
134 {
135 KVMState *s = KVM_STATE(current_machine->accelerator);
136
137 return !kvm_vm_enable_cap(s, KVM_CAP_X2APIC_API, 0, flags);
138 }
139
140 #define MEMORIZE(fn, _result) \
141 ({ \
142 static bool _memorized; \
143 \
144 if (_memorized) { \
145 return _result; \
146 } \
147 _memorized = true; \
148 _result = fn; \
149 })
150
151 static bool has_x2apic_api;
152
153 bool kvm_has_x2apic_api(void)
154 {
155 return has_x2apic_api;
156 }
157
158 bool kvm_enable_x2apic(void)
159 {
160 return MEMORIZE(
161 kvm_x2apic_api_set_flags(KVM_X2APIC_API_USE_32BIT_IDS |
162 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK),
163 has_x2apic_api);
164 }
165
166 bool kvm_hv_vpindex_settable(void)
167 {
168 return hv_vpindex_settable;
169 }
170
171 static int kvm_get_tsc(CPUState *cs)
172 {
173 X86CPU *cpu = X86_CPU(cs);
174 CPUX86State *env = &cpu->env;
175 struct {
176 struct kvm_msrs info;
177 struct kvm_msr_entry entries[1];
178 } msr_data;
179 int ret;
180
181 if (env->tsc_valid) {
182 return 0;
183 }
184
185 msr_data.info.nmsrs = 1;
186 msr_data.entries[0].index = MSR_IA32_TSC;
187 env->tsc_valid = !runstate_is_running();
188
189 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MSRS, &msr_data);
190 if (ret < 0) {
191 return ret;
192 }
193
194 assert(ret == 1);
195 env->tsc = msr_data.entries[0].data;
196 return 0;
197 }
198
199 static inline void do_kvm_synchronize_tsc(CPUState *cpu, run_on_cpu_data arg)
200 {
201 kvm_get_tsc(cpu);
202 }
203
204 void kvm_synchronize_all_tsc(void)
205 {
206 CPUState *cpu;
207
208 if (kvm_enabled()) {
209 CPU_FOREACH(cpu) {
210 run_on_cpu(cpu, do_kvm_synchronize_tsc, RUN_ON_CPU_NULL);
211 }
212 }
213 }
214
215 static struct kvm_cpuid2 *try_get_cpuid(KVMState *s, int max)
216 {
217 struct kvm_cpuid2 *cpuid;
218 int r, size;
219
220 size = sizeof(*cpuid) + max * sizeof(*cpuid->entries);
221 cpuid = g_malloc0(size);
222 cpuid->nent = max;
223 r = kvm_ioctl(s, KVM_GET_SUPPORTED_CPUID, cpuid);
224 if (r == 0 && cpuid->nent >= max) {
225 r = -E2BIG;
226 }
227 if (r < 0) {
228 if (r == -E2BIG) {
229 g_free(cpuid);
230 return NULL;
231 } else {
232 fprintf(stderr, "KVM_GET_SUPPORTED_CPUID failed: %s\n",
233 strerror(-r));
234 exit(1);
235 }
236 }
237 return cpuid;
238 }
239
240 /* Run KVM_GET_SUPPORTED_CPUID ioctl(), allocating a buffer large enough
241 * for all entries.
242 */
243 static struct kvm_cpuid2 *get_supported_cpuid(KVMState *s)
244 {
245 struct kvm_cpuid2 *cpuid;
246 int max = 1;
247
248 if (cpuid_cache != NULL) {
249 return cpuid_cache;
250 }
251 while ((cpuid = try_get_cpuid(s, max)) == NULL) {
252 max *= 2;
253 }
254 cpuid_cache = cpuid;
255 return cpuid;
256 }
257
258 static const struct kvm_para_features {
259 int cap;
260 int feature;
261 } para_features[] = {
262 { KVM_CAP_CLOCKSOURCE, KVM_FEATURE_CLOCKSOURCE },
263 { KVM_CAP_NOP_IO_DELAY, KVM_FEATURE_NOP_IO_DELAY },
264 { KVM_CAP_PV_MMU, KVM_FEATURE_MMU_OP },
265 { KVM_CAP_ASYNC_PF, KVM_FEATURE_ASYNC_PF },
266 };
267
268 static int get_para_features(KVMState *s)
269 {
270 int i, features = 0;
271
272 for (i = 0; i < ARRAY_SIZE(para_features); i++) {
273 if (kvm_check_extension(s, para_features[i].cap)) {
274 features |= (1 << para_features[i].feature);
275 }
276 }
277
278 return features;
279 }
280
281 static bool host_tsx_blacklisted(void)
282 {
283 int family, model, stepping;\
284 char vendor[CPUID_VENDOR_SZ + 1];
285
286 host_vendor_fms(vendor, &family, &model, &stepping);
287
288 /* Check if we are running on a Haswell host known to have broken TSX */
289 return !strcmp(vendor, CPUID_VENDOR_INTEL) &&
290 (family == 6) &&
291 ((model == 63 && stepping < 4) ||
292 model == 60 || model == 69 || model == 70);
293 }
294
295 /* Returns the value for a specific register on the cpuid entry
296 */
297 static uint32_t cpuid_entry_get_reg(struct kvm_cpuid_entry2 *entry, int reg)
298 {
299 uint32_t ret = 0;
300 switch (reg) {
301 case R_EAX:
302 ret = entry->eax;
303 break;
304 case R_EBX:
305 ret = entry->ebx;
306 break;
307 case R_ECX:
308 ret = entry->ecx;
309 break;
310 case R_EDX:
311 ret = entry->edx;
312 break;
313 }
314 return ret;
315 }
316
317 /* Find matching entry for function/index on kvm_cpuid2 struct
318 */
319 static struct kvm_cpuid_entry2 *cpuid_find_entry(struct kvm_cpuid2 *cpuid,
320 uint32_t function,
321 uint32_t index)
322 {
323 int i;
324 for (i = 0; i < cpuid->nent; ++i) {
325 if (cpuid->entries[i].function == function &&
326 cpuid->entries[i].index == index) {
327 return &cpuid->entries[i];
328 }
329 }
330 /* not found: */
331 return NULL;
332 }
333
334 uint32_t kvm_arch_get_supported_cpuid(KVMState *s, uint32_t function,
335 uint32_t index, int reg)
336 {
337 struct kvm_cpuid2 *cpuid;
338 uint32_t ret = 0;
339 uint32_t cpuid_1_edx;
340 bool found = false;
341
342 cpuid = get_supported_cpuid(s);
343
344 struct kvm_cpuid_entry2 *entry = cpuid_find_entry(cpuid, function, index);
345 if (entry) {
346 found = true;
347 ret = cpuid_entry_get_reg(entry, reg);
348 }
349
350 /* Fixups for the data returned by KVM, below */
351
352 if (function == 1 && reg == R_EDX) {
353 /* KVM before 2.6.30 misreports the following features */
354 ret |= CPUID_MTRR | CPUID_PAT | CPUID_MCE | CPUID_MCA;
355 } else if (function == 1 && reg == R_ECX) {
356 /* We can set the hypervisor flag, even if KVM does not return it on
357 * GET_SUPPORTED_CPUID
358 */
359 ret |= CPUID_EXT_HYPERVISOR;
360 /* tsc-deadline flag is not returned by GET_SUPPORTED_CPUID, but it
361 * can be enabled if the kernel has KVM_CAP_TSC_DEADLINE_TIMER,
362 * and the irqchip is in the kernel.
363 */
364 if (kvm_irqchip_in_kernel() &&
365 kvm_check_extension(s, KVM_CAP_TSC_DEADLINE_TIMER)) {
366 ret |= CPUID_EXT_TSC_DEADLINE_TIMER;
367 }
368
369 /* x2apic is reported by GET_SUPPORTED_CPUID, but it can't be enabled
370 * without the in-kernel irqchip
371 */
372 if (!kvm_irqchip_in_kernel()) {
373 ret &= ~CPUID_EXT_X2APIC;
374 }
375
376 if (enable_cpu_pm) {
377 int disable_exits = kvm_check_extension(s,
378 KVM_CAP_X86_DISABLE_EXITS);
379
380 if (disable_exits & KVM_X86_DISABLE_EXITS_MWAIT) {
381 ret |= CPUID_EXT_MONITOR;
382 }
383 }
384 } else if (function == 6 && reg == R_EAX) {
385 ret |= CPUID_6_EAX_ARAT; /* safe to allow because of emulated APIC */
386 } else if (function == 7 && index == 0 && reg == R_EBX) {
387 if (host_tsx_blacklisted()) {
388 ret &= ~(CPUID_7_0_EBX_RTM | CPUID_7_0_EBX_HLE);
389 }
390 } else if (function == 0x80000001 && reg == R_ECX) {
391 /*
392 * It's safe to enable TOPOEXT even if it's not returned by
393 * GET_SUPPORTED_CPUID. Unconditionally enabling TOPOEXT here allows
394 * us to keep CPU models including TOPOEXT runnable on older kernels.
395 */
396 ret |= CPUID_EXT3_TOPOEXT;
397 } else if (function == 0x80000001 && reg == R_EDX) {
398 /* On Intel, kvm returns cpuid according to the Intel spec,
399 * so add missing bits according to the AMD spec:
400 */
401 cpuid_1_edx = kvm_arch_get_supported_cpuid(s, 1, 0, R_EDX);
402 ret |= cpuid_1_edx & CPUID_EXT2_AMD_ALIASES;
403 } else if (function == KVM_CPUID_FEATURES && reg == R_EAX) {
404 /* kvm_pv_unhalt is reported by GET_SUPPORTED_CPUID, but it can't
405 * be enabled without the in-kernel irqchip
406 */
407 if (!kvm_irqchip_in_kernel()) {
408 ret &= ~(1U << KVM_FEATURE_PV_UNHALT);
409 }
410 } else if (function == KVM_CPUID_FEATURES && reg == R_EDX) {
411 ret |= 1U << KVM_HINTS_REALTIME;
412 found = 1;
413 }
414
415 /* fallback for older kernels */
416 if ((function == KVM_CPUID_FEATURES) && !found) {
417 ret = get_para_features(s);
418 }
419
420 return ret;
421 }
422
423 typedef struct HWPoisonPage {
424 ram_addr_t ram_addr;
425 QLIST_ENTRY(HWPoisonPage) list;
426 } HWPoisonPage;
427
428 static QLIST_HEAD(, HWPoisonPage) hwpoison_page_list =
429 QLIST_HEAD_INITIALIZER(hwpoison_page_list);
430
431 static void kvm_unpoison_all(void *param)
432 {
433 HWPoisonPage *page, *next_page;
434
435 QLIST_FOREACH_SAFE(page, &hwpoison_page_list, list, next_page) {
436 QLIST_REMOVE(page, list);
437 qemu_ram_remap(page->ram_addr, TARGET_PAGE_SIZE);
438 g_free(page);
439 }
440 }
441
442 static void kvm_hwpoison_page_add(ram_addr_t ram_addr)
443 {
444 HWPoisonPage *page;
445
446 QLIST_FOREACH(page, &hwpoison_page_list, list) {
447 if (page->ram_addr == ram_addr) {
448 return;
449 }
450 }
451 page = g_new(HWPoisonPage, 1);
452 page->ram_addr = ram_addr;
453 QLIST_INSERT_HEAD(&hwpoison_page_list, page, list);
454 }
455
456 static int kvm_get_mce_cap_supported(KVMState *s, uint64_t *mce_cap,
457 int *max_banks)
458 {
459 int r;
460
461 r = kvm_check_extension(s, KVM_CAP_MCE);
462 if (r > 0) {
463 *max_banks = r;
464 return kvm_ioctl(s, KVM_X86_GET_MCE_CAP_SUPPORTED, mce_cap);
465 }
466 return -ENOSYS;
467 }
468
469 static void kvm_mce_inject(X86CPU *cpu, hwaddr paddr, int code)
470 {
471 CPUState *cs = CPU(cpu);
472 CPUX86State *env = &cpu->env;
473 uint64_t status = MCI_STATUS_VAL | MCI_STATUS_UC | MCI_STATUS_EN |
474 MCI_STATUS_MISCV | MCI_STATUS_ADDRV | MCI_STATUS_S;
475 uint64_t mcg_status = MCG_STATUS_MCIP;
476 int flags = 0;
477
478 if (code == BUS_MCEERR_AR) {
479 status |= MCI_STATUS_AR | 0x134;
480 mcg_status |= MCG_STATUS_EIPV;
481 } else {
482 status |= 0xc0;
483 mcg_status |= MCG_STATUS_RIPV;
484 }
485
486 flags = cpu_x86_support_mca_broadcast(env) ? MCE_INJECT_BROADCAST : 0;
487 /* We need to read back the value of MSR_EXT_MCG_CTL that was set by the
488 * guest kernel back into env->mcg_ext_ctl.
489 */
490 cpu_synchronize_state(cs);
491 if (env->mcg_ext_ctl & MCG_EXT_CTL_LMCE_EN) {
492 mcg_status |= MCG_STATUS_LMCE;
493 flags = 0;
494 }
495
496 cpu_x86_inject_mce(NULL, cpu, 9, status, mcg_status, paddr,
497 (MCM_ADDR_PHYS << 6) | 0xc, flags);
498 }
499
500 static void hardware_memory_error(void)
501 {
502 fprintf(stderr, "Hardware memory error!\n");
503 exit(1);
504 }
505
506 void kvm_arch_on_sigbus_vcpu(CPUState *c, int code, void *addr)
507 {
508 X86CPU *cpu = X86_CPU(c);
509 CPUX86State *env = &cpu->env;
510 ram_addr_t ram_addr;
511 hwaddr paddr;
512
513 /* If we get an action required MCE, it has been injected by KVM
514 * while the VM was running. An action optional MCE instead should
515 * be coming from the main thread, which qemu_init_sigbus identifies
516 * as the "early kill" thread.
517 */
518 assert(code == BUS_MCEERR_AR || code == BUS_MCEERR_AO);
519
520 if ((env->mcg_cap & MCG_SER_P) && addr) {
521 ram_addr = qemu_ram_addr_from_host(addr);
522 if (ram_addr != RAM_ADDR_INVALID &&
523 kvm_physical_memory_addr_from_host(c->kvm_state, addr, &paddr)) {
524 kvm_hwpoison_page_add(ram_addr);
525 kvm_mce_inject(cpu, paddr, code);
526 return;
527 }
528
529 fprintf(stderr, "Hardware memory error for memory used by "
530 "QEMU itself instead of guest system!\n");
531 }
532
533 if (code == BUS_MCEERR_AR) {
534 hardware_memory_error();
535 }
536
537 /* Hope we are lucky for AO MCE */
538 }
539
540 static int kvm_inject_mce_oldstyle(X86CPU *cpu)
541 {
542 CPUX86State *env = &cpu->env;
543
544 if (!kvm_has_vcpu_events() && env->exception_injected == EXCP12_MCHK) {
545 unsigned int bank, bank_num = env->mcg_cap & 0xff;
546 struct kvm_x86_mce mce;
547
548 env->exception_injected = -1;
549
550 /*
551 * There must be at least one bank in use if an MCE is pending.
552 * Find it and use its values for the event injection.
553 */
554 for (bank = 0; bank < bank_num; bank++) {
555 if (env->mce_banks[bank * 4 + 1] & MCI_STATUS_VAL) {
556 break;
557 }
558 }
559 assert(bank < bank_num);
560
561 mce.bank = bank;
562 mce.status = env->mce_banks[bank * 4 + 1];
563 mce.mcg_status = env->mcg_status;
564 mce.addr = env->mce_banks[bank * 4 + 2];
565 mce.misc = env->mce_banks[bank * 4 + 3];
566
567 return kvm_vcpu_ioctl(CPU(cpu), KVM_X86_SET_MCE, &mce);
568 }
569 return 0;
570 }
571
572 static void cpu_update_state(void *opaque, int running, RunState state)
573 {
574 CPUX86State *env = opaque;
575
576 if (running) {
577 env->tsc_valid = false;
578 }
579 }
580
581 unsigned long kvm_arch_vcpu_id(CPUState *cs)
582 {
583 X86CPU *cpu = X86_CPU(cs);
584 return cpu->apic_id;
585 }
586
587 #ifndef KVM_CPUID_SIGNATURE_NEXT
588 #define KVM_CPUID_SIGNATURE_NEXT 0x40000100
589 #endif
590
591 static bool hyperv_hypercall_available(X86CPU *cpu)
592 {
593 return cpu->hyperv_vapic ||
594 (cpu->hyperv_spinlock_attempts != HYPERV_SPINLOCK_NEVER_RETRY);
595 }
596
597 static bool hyperv_enabled(X86CPU *cpu)
598 {
599 CPUState *cs = CPU(cpu);
600 return kvm_check_extension(cs->kvm_state, KVM_CAP_HYPERV) > 0 &&
601 (hyperv_hypercall_available(cpu) ||
602 cpu->hyperv_time ||
603 cpu->hyperv_relaxed_timing ||
604 cpu->hyperv_crash ||
605 cpu->hyperv_reset ||
606 cpu->hyperv_vpindex ||
607 cpu->hyperv_runtime ||
608 cpu->hyperv_synic ||
609 cpu->hyperv_stimer ||
610 cpu->hyperv_reenlightenment ||
611 cpu->hyperv_tlbflush);
612 }
613
614 static int kvm_arch_set_tsc_khz(CPUState *cs)
615 {
616 X86CPU *cpu = X86_CPU(cs);
617 CPUX86State *env = &cpu->env;
618 int r;
619
620 if (!env->tsc_khz) {
621 return 0;
622 }
623
624 r = kvm_check_extension(cs->kvm_state, KVM_CAP_TSC_CONTROL) ?
625 kvm_vcpu_ioctl(cs, KVM_SET_TSC_KHZ, env->tsc_khz) :
626 -ENOTSUP;
627 if (r < 0) {
628 /* When KVM_SET_TSC_KHZ fails, it's an error only if the current
629 * TSC frequency doesn't match the one we want.
630 */
631 int cur_freq = kvm_check_extension(cs->kvm_state, KVM_CAP_GET_TSC_KHZ) ?
632 kvm_vcpu_ioctl(cs, KVM_GET_TSC_KHZ) :
633 -ENOTSUP;
634 if (cur_freq <= 0 || cur_freq != env->tsc_khz) {
635 warn_report("TSC frequency mismatch between "
636 "VM (%" PRId64 " kHz) and host (%d kHz), "
637 "and TSC scaling unavailable",
638 env->tsc_khz, cur_freq);
639 return r;
640 }
641 }
642
643 return 0;
644 }
645
646 static bool tsc_is_stable_and_known(CPUX86State *env)
647 {
648 if (!env->tsc_khz) {
649 return false;
650 }
651 return (env->features[FEAT_8000_0007_EDX] & CPUID_APM_INVTSC)
652 || env->user_tsc_khz;
653 }
654
655 static int hyperv_handle_properties(CPUState *cs)
656 {
657 X86CPU *cpu = X86_CPU(cs);
658 CPUX86State *env = &cpu->env;
659
660 if (cpu->hyperv_relaxed_timing) {
661 env->features[FEAT_HYPERV_EAX] |= HV_HYPERCALL_AVAILABLE;
662 }
663 if (cpu->hyperv_vapic) {
664 env->features[FEAT_HYPERV_EAX] |= HV_HYPERCALL_AVAILABLE;
665 env->features[FEAT_HYPERV_EAX] |= HV_APIC_ACCESS_AVAILABLE;
666 }
667 if (cpu->hyperv_time) {
668 if (kvm_check_extension(cs->kvm_state, KVM_CAP_HYPERV_TIME) <= 0) {
669 fprintf(stderr, "Hyper-V clocksources "
670 "(requested by 'hv-time' cpu flag) "
671 "are not supported by kernel\n");
672 return -ENOSYS;
673 }
674 env->features[FEAT_HYPERV_EAX] |= HV_HYPERCALL_AVAILABLE;
675 env->features[FEAT_HYPERV_EAX] |= HV_TIME_REF_COUNT_AVAILABLE;
676 env->features[FEAT_HYPERV_EAX] |= HV_REFERENCE_TSC_AVAILABLE;
677 }
678 if (cpu->hyperv_frequencies) {
679 if (!has_msr_hv_frequencies) {
680 fprintf(stderr, "Hyper-V frequency MSRs "
681 "(requested by 'hv-frequencies' cpu flag) "
682 "are not supported by kernel\n");
683 return -ENOSYS;
684 }
685 env->features[FEAT_HYPERV_EAX] |= HV_ACCESS_FREQUENCY_MSRS;
686 env->features[FEAT_HYPERV_EDX] |= HV_FREQUENCY_MSRS_AVAILABLE;
687 }
688 if (cpu->hyperv_crash) {
689 if (!has_msr_hv_crash) {
690 fprintf(stderr, "Hyper-V crash MSRs "
691 "(requested by 'hv-crash' cpu flag) "
692 "are not supported by kernel\n");
693 return -ENOSYS;
694 }
695 env->features[FEAT_HYPERV_EDX] |= HV_GUEST_CRASH_MSR_AVAILABLE;
696 }
697 if (cpu->hyperv_reenlightenment) {
698 if (!has_msr_hv_reenlightenment) {
699 fprintf(stderr,
700 "Hyper-V Reenlightenment MSRs "
701 "(requested by 'hv-reenlightenment' cpu flag) "
702 "are not supported by kernel\n");
703 return -ENOSYS;
704 }
705 env->features[FEAT_HYPERV_EAX] |= HV_ACCESS_REENLIGHTENMENTS_CONTROL;
706 }
707 env->features[FEAT_HYPERV_EDX] |= HV_CPU_DYNAMIC_PARTITIONING_AVAILABLE;
708 if (cpu->hyperv_reset) {
709 if (!has_msr_hv_reset) {
710 fprintf(stderr, "Hyper-V reset MSR "
711 "(requested by 'hv-reset' cpu flag) "
712 "is not supported by kernel\n");
713 return -ENOSYS;
714 }
715 env->features[FEAT_HYPERV_EAX] |= HV_RESET_AVAILABLE;
716 }
717 if (cpu->hyperv_vpindex) {
718 if (!has_msr_hv_vpindex) {
719 fprintf(stderr, "Hyper-V VP_INDEX MSR "
720 "(requested by 'hv-vpindex' cpu flag) "
721 "is not supported by kernel\n");
722 return -ENOSYS;
723 }
724 env->features[FEAT_HYPERV_EAX] |= HV_VP_INDEX_AVAILABLE;
725 }
726 if (cpu->hyperv_runtime) {
727 if (!has_msr_hv_runtime) {
728 fprintf(stderr, "Hyper-V VP_RUNTIME MSR "
729 "(requested by 'hv-runtime' cpu flag) "
730 "is not supported by kernel\n");
731 return -ENOSYS;
732 }
733 env->features[FEAT_HYPERV_EAX] |= HV_VP_RUNTIME_AVAILABLE;
734 }
735 if (cpu->hyperv_synic) {
736 if (!has_msr_hv_synic ||
737 kvm_vcpu_enable_cap(cs, KVM_CAP_HYPERV_SYNIC, 0)) {
738 fprintf(stderr, "Hyper-V SynIC is not supported by kernel\n");
739 return -ENOSYS;
740 }
741
742 env->features[FEAT_HYPERV_EAX] |= HV_SYNIC_AVAILABLE;
743 }
744 if (cpu->hyperv_stimer) {
745 if (!has_msr_hv_stimer) {
746 fprintf(stderr, "Hyper-V timers aren't supported by kernel\n");
747 return -ENOSYS;
748 }
749 env->features[FEAT_HYPERV_EAX] |= HV_SYNTIMERS_AVAILABLE;
750 }
751 return 0;
752 }
753
754 static int hyperv_init_vcpu(X86CPU *cpu)
755 {
756 if (cpu->hyperv_vpindex && !hv_vpindex_settable) {
757 /*
758 * the kernel doesn't support setting vp_index; assert that its value
759 * is in sync
760 */
761 int ret;
762 struct {
763 struct kvm_msrs info;
764 struct kvm_msr_entry entries[1];
765 } msr_data = {
766 .info.nmsrs = 1,
767 .entries[0].index = HV_X64_MSR_VP_INDEX,
768 };
769
770 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MSRS, &msr_data);
771 if (ret < 0) {
772 return ret;
773 }
774 assert(ret == 1);
775
776 if (msr_data.entries[0].data != hyperv_vp_index(cpu)) {
777 error_report("kernel's vp_index != QEMU's vp_index");
778 return -ENXIO;
779 }
780 }
781
782 return 0;
783 }
784
785 static Error *invtsc_mig_blocker;
786
787 #define KVM_MAX_CPUID_ENTRIES 100
788
789 int kvm_arch_init_vcpu(CPUState *cs)
790 {
791 struct {
792 struct kvm_cpuid2 cpuid;
793 struct kvm_cpuid_entry2 entries[KVM_MAX_CPUID_ENTRIES];
794 } QEMU_PACKED cpuid_data;
795 X86CPU *cpu = X86_CPU(cs);
796 CPUX86State *env = &cpu->env;
797 uint32_t limit, i, j, cpuid_i;
798 uint32_t unused;
799 struct kvm_cpuid_entry2 *c;
800 uint32_t signature[3];
801 int kvm_base = KVM_CPUID_SIGNATURE;
802 int r;
803 Error *local_err = NULL;
804
805 memset(&cpuid_data, 0, sizeof(cpuid_data));
806
807 cpuid_i = 0;
808
809 r = kvm_arch_set_tsc_khz(cs);
810 if (r < 0) {
811 goto fail;
812 }
813
814 /* vcpu's TSC frequency is either specified by user, or following
815 * the value used by KVM if the former is not present. In the
816 * latter case, we query it from KVM and record in env->tsc_khz,
817 * so that vcpu's TSC frequency can be migrated later via this field.
818 */
819 if (!env->tsc_khz) {
820 r = kvm_check_extension(cs->kvm_state, KVM_CAP_GET_TSC_KHZ) ?
821 kvm_vcpu_ioctl(cs, KVM_GET_TSC_KHZ) :
822 -ENOTSUP;
823 if (r > 0) {
824 env->tsc_khz = r;
825 }
826 }
827
828 /* Paravirtualization CPUIDs */
829 if (hyperv_enabled(cpu)) {
830 c = &cpuid_data.entries[cpuid_i++];
831 c->function = HV_CPUID_VENDOR_AND_MAX_FUNCTIONS;
832 if (!cpu->hyperv_vendor_id) {
833 memcpy(signature, "Microsoft Hv", 12);
834 } else {
835 size_t len = strlen(cpu->hyperv_vendor_id);
836
837 if (len > 12) {
838 error_report("hv-vendor-id truncated to 12 characters");
839 len = 12;
840 }
841 memset(signature, 0, 12);
842 memcpy(signature, cpu->hyperv_vendor_id, len);
843 }
844 c->eax = HV_CPUID_MIN;
845 c->ebx = signature[0];
846 c->ecx = signature[1];
847 c->edx = signature[2];
848
849 c = &cpuid_data.entries[cpuid_i++];
850 c->function = HV_CPUID_INTERFACE;
851 memcpy(signature, "Hv#1\0\0\0\0\0\0\0\0", 12);
852 c->eax = signature[0];
853 c->ebx = 0;
854 c->ecx = 0;
855 c->edx = 0;
856
857 c = &cpuid_data.entries[cpuid_i++];
858 c->function = HV_CPUID_VERSION;
859 c->eax = 0x00001bbc;
860 c->ebx = 0x00060001;
861
862 c = &cpuid_data.entries[cpuid_i++];
863 c->function = HV_CPUID_FEATURES;
864 r = hyperv_handle_properties(cs);
865 if (r) {
866 return r;
867 }
868 c->eax = env->features[FEAT_HYPERV_EAX];
869 c->ebx = env->features[FEAT_HYPERV_EBX];
870 c->edx = env->features[FEAT_HYPERV_EDX];
871
872 c = &cpuid_data.entries[cpuid_i++];
873 c->function = HV_CPUID_ENLIGHTMENT_INFO;
874 if (cpu->hyperv_relaxed_timing) {
875 c->eax |= HV_RELAXED_TIMING_RECOMMENDED;
876 }
877 if (cpu->hyperv_vapic) {
878 c->eax |= HV_APIC_ACCESS_RECOMMENDED;
879 }
880 if (cpu->hyperv_tlbflush) {
881 if (kvm_check_extension(cs->kvm_state,
882 KVM_CAP_HYPERV_TLBFLUSH) <= 0) {
883 fprintf(stderr, "Hyper-V TLB flush support "
884 "(requested by 'hv-tlbflush' cpu flag) "
885 " is not supported by kernel\n");
886 return -ENOSYS;
887 }
888 c->eax |= HV_REMOTE_TLB_FLUSH_RECOMMENDED;
889 c->eax |= HV_EX_PROCESSOR_MASKS_RECOMMENDED;
890 }
891
892 c->ebx = cpu->hyperv_spinlock_attempts;
893
894 c = &cpuid_data.entries[cpuid_i++];
895 c->function = HV_CPUID_IMPLEMENT_LIMITS;
896
897 c->eax = cpu->hv_max_vps;
898 c->ebx = 0x40;
899
900 kvm_base = KVM_CPUID_SIGNATURE_NEXT;
901 has_msr_hv_hypercall = true;
902 }
903
904 if (cpu->expose_kvm) {
905 memcpy(signature, "KVMKVMKVM\0\0\0", 12);
906 c = &cpuid_data.entries[cpuid_i++];
907 c->function = KVM_CPUID_SIGNATURE | kvm_base;
908 c->eax = KVM_CPUID_FEATURES | kvm_base;
909 c->ebx = signature[0];
910 c->ecx = signature[1];
911 c->edx = signature[2];
912
913 c = &cpuid_data.entries[cpuid_i++];
914 c->function = KVM_CPUID_FEATURES | kvm_base;
915 c->eax = env->features[FEAT_KVM];
916 c->edx = env->features[FEAT_KVM_HINTS];
917 }
918
919 cpu_x86_cpuid(env, 0, 0, &limit, &unused, &unused, &unused);
920
921 for (i = 0; i <= limit; i++) {
922 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
923 fprintf(stderr, "unsupported level value: 0x%x\n", limit);
924 abort();
925 }
926 c = &cpuid_data.entries[cpuid_i++];
927 assert(cpuid_i < 100);
928
929 switch (i) {
930 case 2: {
931 /* Keep reading function 2 till all the input is received */
932 int times;
933
934 c->function = i;
935 c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC |
936 KVM_CPUID_FLAG_STATE_READ_NEXT;
937 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
938 times = c->eax & 0xff;
939
940 for (j = 1; j < times; ++j) {
941 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
942 fprintf(stderr, "cpuid_data is full, no space for "
943 "cpuid(eax:2):eax & 0xf = 0x%x\n", times);
944 abort();
945 }
946 c = &cpuid_data.entries[cpuid_i++];
947 c->function = i;
948 c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC;
949 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
950 }
951 break;
952 }
953 case 4:
954 case 0xb:
955 case 0xd:
956 for (j = 0; ; j++) {
957 if (i == 0xd && j == 64) {
958 break;
959 }
960 c->function = i;
961 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
962 c->index = j;
963 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
964
965 if (i == 4 && c->eax == 0) {
966 break;
967 }
968 if (i == 0xb && !(c->ecx & 0xff00)) {
969 break;
970 }
971 if (i == 0xd && c->eax == 0) {
972 continue;
973 }
974 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
975 fprintf(stderr, "cpuid_data is full, no space for "
976 "cpuid(eax:0x%x,ecx:0x%x)\n", i, j);
977 abort();
978 }
979 c = &cpuid_data.entries[cpuid_i++];
980 }
981 break;
982 case 0x14: {
983 uint32_t times;
984
985 c->function = i;
986 c->index = 0;
987 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
988 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
989 times = c->eax;
990
991 for (j = 1; j <= times; ++j) {
992 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
993 fprintf(stderr, "cpuid_data is full, no space for "
994 "cpuid(eax:0x14,ecx:0x%x)\n", j);
995 abort();
996 }
997 c = &cpuid_data.entries[cpuid_i++];
998 c->function = i;
999 c->index = j;
1000 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1001 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
1002 }
1003 break;
1004 }
1005 default:
1006 c->function = i;
1007 c->flags = 0;
1008 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1009 break;
1010 }
1011 }
1012
1013 if (limit >= 0x0a) {
1014 uint32_t eax, edx;
1015
1016 cpu_x86_cpuid(env, 0x0a, 0, &eax, &unused, &unused, &edx);
1017
1018 has_architectural_pmu_version = eax & 0xff;
1019 if (has_architectural_pmu_version > 0) {
1020 num_architectural_pmu_gp_counters = (eax & 0xff00) >> 8;
1021
1022 /* Shouldn't be more than 32, since that's the number of bits
1023 * available in EBX to tell us _which_ counters are available.
1024 * Play it safe.
1025 */
1026 if (num_architectural_pmu_gp_counters > MAX_GP_COUNTERS) {
1027 num_architectural_pmu_gp_counters = MAX_GP_COUNTERS;
1028 }
1029
1030 if (has_architectural_pmu_version > 1) {
1031 num_architectural_pmu_fixed_counters = edx & 0x1f;
1032
1033 if (num_architectural_pmu_fixed_counters > MAX_FIXED_COUNTERS) {
1034 num_architectural_pmu_fixed_counters = MAX_FIXED_COUNTERS;
1035 }
1036 }
1037 }
1038 }
1039
1040 cpu_x86_cpuid(env, 0x80000000, 0, &limit, &unused, &unused, &unused);
1041
1042 for (i = 0x80000000; i <= limit; i++) {
1043 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1044 fprintf(stderr, "unsupported xlevel value: 0x%x\n", limit);
1045 abort();
1046 }
1047 c = &cpuid_data.entries[cpuid_i++];
1048 assert(cpuid_i < 100);
1049
1050 switch (i) {
1051 case 0x8000001d:
1052 /* Query for all AMD cache information leaves */
1053 for (j = 0; ; j++) {
1054 c->function = i;
1055 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1056 c->index = j;
1057 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
1058
1059 if (c->eax == 0) {
1060 break;
1061 }
1062 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1063 fprintf(stderr, "cpuid_data is full, no space for "
1064 "cpuid(eax:0x%x,ecx:0x%x)\n", i, j);
1065 abort();
1066 }
1067 c = &cpuid_data.entries[cpuid_i++];
1068 }
1069 break;
1070 default:
1071 c->function = i;
1072 c->flags = 0;
1073 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1074 break;
1075 }
1076 }
1077
1078 /* Call Centaur's CPUID instructions they are supported. */
1079 if (env->cpuid_xlevel2 > 0) {
1080 cpu_x86_cpuid(env, 0xC0000000, 0, &limit, &unused, &unused, &unused);
1081
1082 for (i = 0xC0000000; i <= limit; i++) {
1083 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1084 fprintf(stderr, "unsupported xlevel2 value: 0x%x\n", limit);
1085 abort();
1086 }
1087 c = &cpuid_data.entries[cpuid_i++];
1088
1089 c->function = i;
1090 c->flags = 0;
1091 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1092 }
1093 }
1094
1095 cpuid_data.cpuid.nent = cpuid_i;
1096
1097 if (((env->cpuid_version >> 8)&0xF) >= 6
1098 && (env->features[FEAT_1_EDX] & (CPUID_MCE | CPUID_MCA)) ==
1099 (CPUID_MCE | CPUID_MCA)
1100 && kvm_check_extension(cs->kvm_state, KVM_CAP_MCE) > 0) {
1101 uint64_t mcg_cap, unsupported_caps;
1102 int banks;
1103 int ret;
1104
1105 ret = kvm_get_mce_cap_supported(cs->kvm_state, &mcg_cap, &banks);
1106 if (ret < 0) {
1107 fprintf(stderr, "kvm_get_mce_cap_supported: %s", strerror(-ret));
1108 return ret;
1109 }
1110
1111 if (banks < (env->mcg_cap & MCG_CAP_BANKS_MASK)) {
1112 error_report("kvm: Unsupported MCE bank count (QEMU = %d, KVM = %d)",
1113 (int)(env->mcg_cap & MCG_CAP_BANKS_MASK), banks);
1114 return -ENOTSUP;
1115 }
1116
1117 unsupported_caps = env->mcg_cap & ~(mcg_cap | MCG_CAP_BANKS_MASK);
1118 if (unsupported_caps) {
1119 if (unsupported_caps & MCG_LMCE_P) {
1120 error_report("kvm: LMCE not supported");
1121 return -ENOTSUP;
1122 }
1123 warn_report("Unsupported MCG_CAP bits: 0x%" PRIx64,
1124 unsupported_caps);
1125 }
1126
1127 env->mcg_cap &= mcg_cap | MCG_CAP_BANKS_MASK;
1128 ret = kvm_vcpu_ioctl(cs, KVM_X86_SETUP_MCE, &env->mcg_cap);
1129 if (ret < 0) {
1130 fprintf(stderr, "KVM_X86_SETUP_MCE: %s", strerror(-ret));
1131 return ret;
1132 }
1133 }
1134
1135 qemu_add_vm_change_state_handler(cpu_update_state, env);
1136
1137 c = cpuid_find_entry(&cpuid_data.cpuid, 1, 0);
1138 if (c) {
1139 has_msr_feature_control = !!(c->ecx & CPUID_EXT_VMX) ||
1140 !!(c->ecx & CPUID_EXT_SMX);
1141 }
1142
1143 if (env->mcg_cap & MCG_LMCE_P) {
1144 has_msr_mcg_ext_ctl = has_msr_feature_control = true;
1145 }
1146
1147 if (!env->user_tsc_khz) {
1148 if ((env->features[FEAT_8000_0007_EDX] & CPUID_APM_INVTSC) &&
1149 invtsc_mig_blocker == NULL) {
1150 /* for migration */
1151 error_setg(&invtsc_mig_blocker,
1152 "State blocked by non-migratable CPU device"
1153 " (invtsc flag)");
1154 r = migrate_add_blocker(invtsc_mig_blocker, &local_err);
1155 if (local_err) {
1156 error_report_err(local_err);
1157 error_free(invtsc_mig_blocker);
1158 goto fail;
1159 }
1160 /* for savevm */
1161 vmstate_x86_cpu.unmigratable = 1;
1162 }
1163 }
1164
1165 if (cpu->vmware_cpuid_freq
1166 /* Guests depend on 0x40000000 to detect this feature, so only expose
1167 * it if KVM exposes leaf 0x40000000. (Conflicts with Hyper-V) */
1168 && cpu->expose_kvm
1169 && kvm_base == KVM_CPUID_SIGNATURE
1170 /* TSC clock must be stable and known for this feature. */
1171 && tsc_is_stable_and_known(env)) {
1172
1173 c = &cpuid_data.entries[cpuid_i++];
1174 c->function = KVM_CPUID_SIGNATURE | 0x10;
1175 c->eax = env->tsc_khz;
1176 /* LAPIC resolution of 1ns (freq: 1GHz) is hardcoded in KVM's
1177 * APIC_BUS_CYCLE_NS */
1178 c->ebx = 1000000;
1179 c->ecx = c->edx = 0;
1180
1181 c = cpuid_find_entry(&cpuid_data.cpuid, kvm_base, 0);
1182 c->eax = MAX(c->eax, KVM_CPUID_SIGNATURE | 0x10);
1183 }
1184
1185 cpuid_data.cpuid.nent = cpuid_i;
1186
1187 cpuid_data.cpuid.padding = 0;
1188 r = kvm_vcpu_ioctl(cs, KVM_SET_CPUID2, &cpuid_data);
1189 if (r) {
1190 goto fail;
1191 }
1192
1193 if (has_xsave) {
1194 env->kvm_xsave_buf = qemu_memalign(4096, sizeof(struct kvm_xsave));
1195 }
1196 cpu->kvm_msr_buf = g_malloc0(MSR_BUF_SIZE);
1197
1198 if (!(env->features[FEAT_8000_0001_EDX] & CPUID_EXT2_RDTSCP)) {
1199 has_msr_tsc_aux = false;
1200 }
1201
1202 r = hyperv_init_vcpu(cpu);
1203 if (r) {
1204 goto fail;
1205 }
1206
1207 return 0;
1208
1209 fail:
1210 migrate_del_blocker(invtsc_mig_blocker);
1211 return r;
1212 }
1213
1214 void kvm_arch_reset_vcpu(X86CPU *cpu)
1215 {
1216 CPUX86State *env = &cpu->env;
1217
1218 env->xcr0 = 1;
1219 if (kvm_irqchip_in_kernel()) {
1220 env->mp_state = cpu_is_bsp(cpu) ? KVM_MP_STATE_RUNNABLE :
1221 KVM_MP_STATE_UNINITIALIZED;
1222 } else {
1223 env->mp_state = KVM_MP_STATE_RUNNABLE;
1224 }
1225
1226 if (cpu->hyperv_synic) {
1227 int i;
1228 for (i = 0; i < ARRAY_SIZE(env->msr_hv_synic_sint); i++) {
1229 env->msr_hv_synic_sint[i] = HV_SINT_MASKED;
1230 }
1231 }
1232 }
1233
1234 void kvm_arch_do_init_vcpu(X86CPU *cpu)
1235 {
1236 CPUX86State *env = &cpu->env;
1237
1238 /* APs get directly into wait-for-SIPI state. */
1239 if (env->mp_state == KVM_MP_STATE_UNINITIALIZED) {
1240 env->mp_state = KVM_MP_STATE_INIT_RECEIVED;
1241 }
1242 }
1243
1244 static int kvm_get_supported_msrs(KVMState *s)
1245 {
1246 static int kvm_supported_msrs;
1247 int ret = 0;
1248
1249 /* first time */
1250 if (kvm_supported_msrs == 0) {
1251 struct kvm_msr_list msr_list, *kvm_msr_list;
1252
1253 kvm_supported_msrs = -1;
1254
1255 /* Obtain MSR list from KVM. These are the MSRs that we must
1256 * save/restore */
1257 msr_list.nmsrs = 0;
1258 ret = kvm_ioctl(s, KVM_GET_MSR_INDEX_LIST, &msr_list);
1259 if (ret < 0 && ret != -E2BIG) {
1260 return ret;
1261 }
1262 /* Old kernel modules had a bug and could write beyond the provided
1263 memory. Allocate at least a safe amount of 1K. */
1264 kvm_msr_list = g_malloc0(MAX(1024, sizeof(msr_list) +
1265 msr_list.nmsrs *
1266 sizeof(msr_list.indices[0])));
1267
1268 kvm_msr_list->nmsrs = msr_list.nmsrs;
1269 ret = kvm_ioctl(s, KVM_GET_MSR_INDEX_LIST, kvm_msr_list);
1270 if (ret >= 0) {
1271 int i;
1272
1273 for (i = 0; i < kvm_msr_list->nmsrs; i++) {
1274 switch (kvm_msr_list->indices[i]) {
1275 case MSR_STAR:
1276 has_msr_star = true;
1277 break;
1278 case MSR_VM_HSAVE_PA:
1279 has_msr_hsave_pa = true;
1280 break;
1281 case MSR_TSC_AUX:
1282 has_msr_tsc_aux = true;
1283 break;
1284 case MSR_TSC_ADJUST:
1285 has_msr_tsc_adjust = true;
1286 break;
1287 case MSR_IA32_TSCDEADLINE:
1288 has_msr_tsc_deadline = true;
1289 break;
1290 case MSR_IA32_SMBASE:
1291 has_msr_smbase = true;
1292 break;
1293 case MSR_SMI_COUNT:
1294 has_msr_smi_count = true;
1295 break;
1296 case MSR_IA32_MISC_ENABLE:
1297 has_msr_misc_enable = true;
1298 break;
1299 case MSR_IA32_BNDCFGS:
1300 has_msr_bndcfgs = true;
1301 break;
1302 case MSR_IA32_XSS:
1303 has_msr_xss = true;
1304 break;
1305 case HV_X64_MSR_CRASH_CTL:
1306 has_msr_hv_crash = true;
1307 break;
1308 case HV_X64_MSR_RESET:
1309 has_msr_hv_reset = true;
1310 break;
1311 case HV_X64_MSR_VP_INDEX:
1312 has_msr_hv_vpindex = true;
1313 break;
1314 case HV_X64_MSR_VP_RUNTIME:
1315 has_msr_hv_runtime = true;
1316 break;
1317 case HV_X64_MSR_SCONTROL:
1318 has_msr_hv_synic = true;
1319 break;
1320 case HV_X64_MSR_STIMER0_CONFIG:
1321 has_msr_hv_stimer = true;
1322 break;
1323 case HV_X64_MSR_TSC_FREQUENCY:
1324 has_msr_hv_frequencies = true;
1325 break;
1326 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1327 has_msr_hv_reenlightenment = true;
1328 break;
1329 case MSR_IA32_SPEC_CTRL:
1330 has_msr_spec_ctrl = true;
1331 break;
1332 case MSR_VIRT_SSBD:
1333 has_msr_virt_ssbd = true;
1334 break;
1335 }
1336 }
1337 }
1338
1339 g_free(kvm_msr_list);
1340 }
1341
1342 return ret;
1343 }
1344
1345 static Notifier smram_machine_done;
1346 static KVMMemoryListener smram_listener;
1347 static AddressSpace smram_address_space;
1348 static MemoryRegion smram_as_root;
1349 static MemoryRegion smram_as_mem;
1350
1351 static void register_smram_listener(Notifier *n, void *unused)
1352 {
1353 MemoryRegion *smram =
1354 (MemoryRegion *) object_resolve_path("/machine/smram", NULL);
1355
1356 /* Outer container... */
1357 memory_region_init(&smram_as_root, OBJECT(kvm_state), "mem-container-smram", ~0ull);
1358 memory_region_set_enabled(&smram_as_root, true);
1359
1360 /* ... with two regions inside: normal system memory with low
1361 * priority, and...
1362 */
1363 memory_region_init_alias(&smram_as_mem, OBJECT(kvm_state), "mem-smram",
1364 get_system_memory(), 0, ~0ull);
1365 memory_region_add_subregion_overlap(&smram_as_root, 0, &smram_as_mem, 0);
1366 memory_region_set_enabled(&smram_as_mem, true);
1367
1368 if (smram) {
1369 /* ... SMRAM with higher priority */
1370 memory_region_add_subregion_overlap(&smram_as_root, 0, smram, 10);
1371 memory_region_set_enabled(smram, true);
1372 }
1373
1374 address_space_init(&smram_address_space, &smram_as_root, "KVM-SMRAM");
1375 kvm_memory_listener_register(kvm_state, &smram_listener,
1376 &smram_address_space, 1);
1377 }
1378
1379 int kvm_arch_init(MachineState *ms, KVMState *s)
1380 {
1381 uint64_t identity_base = 0xfffbc000;
1382 uint64_t shadow_mem;
1383 int ret;
1384 struct utsname utsname;
1385
1386 #ifdef KVM_CAP_XSAVE
1387 has_xsave = kvm_check_extension(s, KVM_CAP_XSAVE);
1388 #endif
1389
1390 #ifdef KVM_CAP_XCRS
1391 has_xcrs = kvm_check_extension(s, KVM_CAP_XCRS);
1392 #endif
1393
1394 #ifdef KVM_CAP_PIT_STATE2
1395 has_pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2);
1396 #endif
1397
1398 hv_vpindex_settable = kvm_check_extension(s, KVM_CAP_HYPERV_VP_INDEX);
1399
1400 ret = kvm_get_supported_msrs(s);
1401 if (ret < 0) {
1402 return ret;
1403 }
1404
1405 uname(&utsname);
1406 lm_capable_kernel = strcmp(utsname.machine, "x86_64") == 0;
1407
1408 /*
1409 * On older Intel CPUs, KVM uses vm86 mode to emulate 16-bit code directly.
1410 * In order to use vm86 mode, an EPT identity map and a TSS are needed.
1411 * Since these must be part of guest physical memory, we need to allocate
1412 * them, both by setting their start addresses in the kernel and by
1413 * creating a corresponding e820 entry. We need 4 pages before the BIOS.
1414 *
1415 * Older KVM versions may not support setting the identity map base. In
1416 * that case we need to stick with the default, i.e. a 256K maximum BIOS
1417 * size.
1418 */
1419 if (kvm_check_extension(s, KVM_CAP_SET_IDENTITY_MAP_ADDR)) {
1420 /* Allows up to 16M BIOSes. */
1421 identity_base = 0xfeffc000;
1422
1423 ret = kvm_vm_ioctl(s, KVM_SET_IDENTITY_MAP_ADDR, &identity_base);
1424 if (ret < 0) {
1425 return ret;
1426 }
1427 }
1428
1429 /* Set TSS base one page after EPT identity map. */
1430 ret = kvm_vm_ioctl(s, KVM_SET_TSS_ADDR, identity_base + 0x1000);
1431 if (ret < 0) {
1432 return ret;
1433 }
1434
1435 /* Tell fw_cfg to notify the BIOS to reserve the range. */
1436 ret = e820_add_entry(identity_base, 0x4000, E820_RESERVED);
1437 if (ret < 0) {
1438 fprintf(stderr, "e820_add_entry() table is full\n");
1439 return ret;
1440 }
1441 qemu_register_reset(kvm_unpoison_all, NULL);
1442
1443 shadow_mem = machine_kvm_shadow_mem(ms);
1444 if (shadow_mem != -1) {
1445 shadow_mem /= 4096;
1446 ret = kvm_vm_ioctl(s, KVM_SET_NR_MMU_PAGES, shadow_mem);
1447 if (ret < 0) {
1448 return ret;
1449 }
1450 }
1451
1452 if (kvm_check_extension(s, KVM_CAP_X86_SMM) &&
1453 object_dynamic_cast(OBJECT(ms), TYPE_PC_MACHINE) &&
1454 pc_machine_is_smm_enabled(PC_MACHINE(ms))) {
1455 smram_machine_done.notify = register_smram_listener;
1456 qemu_add_machine_init_done_notifier(&smram_machine_done);
1457 }
1458
1459 if (enable_cpu_pm) {
1460 int disable_exits = kvm_check_extension(s, KVM_CAP_X86_DISABLE_EXITS);
1461 int ret;
1462
1463 /* Work around for kernel header with a typo. TODO: fix header and drop. */
1464 #if defined(KVM_X86_DISABLE_EXITS_HTL) && !defined(KVM_X86_DISABLE_EXITS_HLT)
1465 #define KVM_X86_DISABLE_EXITS_HLT KVM_X86_DISABLE_EXITS_HTL
1466 #endif
1467 if (disable_exits) {
1468 disable_exits &= (KVM_X86_DISABLE_EXITS_MWAIT |
1469 KVM_X86_DISABLE_EXITS_HLT |
1470 KVM_X86_DISABLE_EXITS_PAUSE);
1471 }
1472
1473 ret = kvm_vm_enable_cap(s, KVM_CAP_X86_DISABLE_EXITS, 0,
1474 disable_exits);
1475 if (ret < 0) {
1476 error_report("kvm: guest stopping CPU not supported: %s",
1477 strerror(-ret));
1478 }
1479 }
1480
1481 return 0;
1482 }
1483
1484 static void set_v8086_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
1485 {
1486 lhs->selector = rhs->selector;
1487 lhs->base = rhs->base;
1488 lhs->limit = rhs->limit;
1489 lhs->type = 3;
1490 lhs->present = 1;
1491 lhs->dpl = 3;
1492 lhs->db = 0;
1493 lhs->s = 1;
1494 lhs->l = 0;
1495 lhs->g = 0;
1496 lhs->avl = 0;
1497 lhs->unusable = 0;
1498 }
1499
1500 static void set_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
1501 {
1502 unsigned flags = rhs->flags;
1503 lhs->selector = rhs->selector;
1504 lhs->base = rhs->base;
1505 lhs->limit = rhs->limit;
1506 lhs->type = (flags >> DESC_TYPE_SHIFT) & 15;
1507 lhs->present = (flags & DESC_P_MASK) != 0;
1508 lhs->dpl = (flags >> DESC_DPL_SHIFT) & 3;
1509 lhs->db = (flags >> DESC_B_SHIFT) & 1;
1510 lhs->s = (flags & DESC_S_MASK) != 0;
1511 lhs->l = (flags >> DESC_L_SHIFT) & 1;
1512 lhs->g = (flags & DESC_G_MASK) != 0;
1513 lhs->avl = (flags & DESC_AVL_MASK) != 0;
1514 lhs->unusable = !lhs->present;
1515 lhs->padding = 0;
1516 }
1517
1518 static void get_seg(SegmentCache *lhs, const struct kvm_segment *rhs)
1519 {
1520 lhs->selector = rhs->selector;
1521 lhs->base = rhs->base;
1522 lhs->limit = rhs->limit;
1523 lhs->flags = (rhs->type << DESC_TYPE_SHIFT) |
1524 ((rhs->present && !rhs->unusable) * DESC_P_MASK) |
1525 (rhs->dpl << DESC_DPL_SHIFT) |
1526 (rhs->db << DESC_B_SHIFT) |
1527 (rhs->s * DESC_S_MASK) |
1528 (rhs->l << DESC_L_SHIFT) |
1529 (rhs->g * DESC_G_MASK) |
1530 (rhs->avl * DESC_AVL_MASK);
1531 }
1532
1533 static void kvm_getput_reg(__u64 *kvm_reg, target_ulong *qemu_reg, int set)
1534 {
1535 if (set) {
1536 *kvm_reg = *qemu_reg;
1537 } else {
1538 *qemu_reg = *kvm_reg;
1539 }
1540 }
1541
1542 static int kvm_getput_regs(X86CPU *cpu, int set)
1543 {
1544 CPUX86State *env = &cpu->env;
1545 struct kvm_regs regs;
1546 int ret = 0;
1547
1548 if (!set) {
1549 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_REGS, &regs);
1550 if (ret < 0) {
1551 return ret;
1552 }
1553 }
1554
1555 kvm_getput_reg(&regs.rax, &env->regs[R_EAX], set);
1556 kvm_getput_reg(&regs.rbx, &env->regs[R_EBX], set);
1557 kvm_getput_reg(&regs.rcx, &env->regs[R_ECX], set);
1558 kvm_getput_reg(&regs.rdx, &env->regs[R_EDX], set);
1559 kvm_getput_reg(&regs.rsi, &env->regs[R_ESI], set);
1560 kvm_getput_reg(&regs.rdi, &env->regs[R_EDI], set);
1561 kvm_getput_reg(&regs.rsp, &env->regs[R_ESP], set);
1562 kvm_getput_reg(&regs.rbp, &env->regs[R_EBP], set);
1563 #ifdef TARGET_X86_64
1564 kvm_getput_reg(&regs.r8, &env->regs[8], set);
1565 kvm_getput_reg(&regs.r9, &env->regs[9], set);
1566 kvm_getput_reg(&regs.r10, &env->regs[10], set);
1567 kvm_getput_reg(&regs.r11, &env->regs[11], set);
1568 kvm_getput_reg(&regs.r12, &env->regs[12], set);
1569 kvm_getput_reg(&regs.r13, &env->regs[13], set);
1570 kvm_getput_reg(&regs.r14, &env->regs[14], set);
1571 kvm_getput_reg(&regs.r15, &env->regs[15], set);
1572 #endif
1573
1574 kvm_getput_reg(&regs.rflags, &env->eflags, set);
1575 kvm_getput_reg(&regs.rip, &env->eip, set);
1576
1577 if (set) {
1578 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_REGS, &regs);
1579 }
1580
1581 return ret;
1582 }
1583
1584 static int kvm_put_fpu(X86CPU *cpu)
1585 {
1586 CPUX86State *env = &cpu->env;
1587 struct kvm_fpu fpu;
1588 int i;
1589
1590 memset(&fpu, 0, sizeof fpu);
1591 fpu.fsw = env->fpus & ~(7 << 11);
1592 fpu.fsw |= (env->fpstt & 7) << 11;
1593 fpu.fcw = env->fpuc;
1594 fpu.last_opcode = env->fpop;
1595 fpu.last_ip = env->fpip;
1596 fpu.last_dp = env->fpdp;
1597 for (i = 0; i < 8; ++i) {
1598 fpu.ftwx |= (!env->fptags[i]) << i;
1599 }
1600 memcpy(fpu.fpr, env->fpregs, sizeof env->fpregs);
1601 for (i = 0; i < CPU_NB_REGS; i++) {
1602 stq_p(&fpu.xmm[i][0], env->xmm_regs[i].ZMM_Q(0));
1603 stq_p(&fpu.xmm[i][8], env->xmm_regs[i].ZMM_Q(1));
1604 }
1605 fpu.mxcsr = env->mxcsr;
1606
1607 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_FPU, &fpu);
1608 }
1609
1610 #define XSAVE_FCW_FSW 0
1611 #define XSAVE_FTW_FOP 1
1612 #define XSAVE_CWD_RIP 2
1613 #define XSAVE_CWD_RDP 4
1614 #define XSAVE_MXCSR 6
1615 #define XSAVE_ST_SPACE 8
1616 #define XSAVE_XMM_SPACE 40
1617 #define XSAVE_XSTATE_BV 128
1618 #define XSAVE_YMMH_SPACE 144
1619 #define XSAVE_BNDREGS 240
1620 #define XSAVE_BNDCSR 256
1621 #define XSAVE_OPMASK 272
1622 #define XSAVE_ZMM_Hi256 288
1623 #define XSAVE_Hi16_ZMM 416
1624 #define XSAVE_PKRU 672
1625
1626 #define XSAVE_BYTE_OFFSET(word_offset) \
1627 ((word_offset) * sizeof_field(struct kvm_xsave, region[0]))
1628
1629 #define ASSERT_OFFSET(word_offset, field) \
1630 QEMU_BUILD_BUG_ON(XSAVE_BYTE_OFFSET(word_offset) != \
1631 offsetof(X86XSaveArea, field))
1632
1633 ASSERT_OFFSET(XSAVE_FCW_FSW, legacy.fcw);
1634 ASSERT_OFFSET(XSAVE_FTW_FOP, legacy.ftw);
1635 ASSERT_OFFSET(XSAVE_CWD_RIP, legacy.fpip);
1636 ASSERT_OFFSET(XSAVE_CWD_RDP, legacy.fpdp);
1637 ASSERT_OFFSET(XSAVE_MXCSR, legacy.mxcsr);
1638 ASSERT_OFFSET(XSAVE_ST_SPACE, legacy.fpregs);
1639 ASSERT_OFFSET(XSAVE_XMM_SPACE, legacy.xmm_regs);
1640 ASSERT_OFFSET(XSAVE_XSTATE_BV, header.xstate_bv);
1641 ASSERT_OFFSET(XSAVE_YMMH_SPACE, avx_state);
1642 ASSERT_OFFSET(XSAVE_BNDREGS, bndreg_state);
1643 ASSERT_OFFSET(XSAVE_BNDCSR, bndcsr_state);
1644 ASSERT_OFFSET(XSAVE_OPMASK, opmask_state);
1645 ASSERT_OFFSET(XSAVE_ZMM_Hi256, zmm_hi256_state);
1646 ASSERT_OFFSET(XSAVE_Hi16_ZMM, hi16_zmm_state);
1647 ASSERT_OFFSET(XSAVE_PKRU, pkru_state);
1648
1649 static int kvm_put_xsave(X86CPU *cpu)
1650 {
1651 CPUX86State *env = &cpu->env;
1652 X86XSaveArea *xsave = env->kvm_xsave_buf;
1653
1654 if (!has_xsave) {
1655 return kvm_put_fpu(cpu);
1656 }
1657 x86_cpu_xsave_all_areas(cpu, xsave);
1658
1659 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_XSAVE, xsave);
1660 }
1661
1662 static int kvm_put_xcrs(X86CPU *cpu)
1663 {
1664 CPUX86State *env = &cpu->env;
1665 struct kvm_xcrs xcrs = {};
1666
1667 if (!has_xcrs) {
1668 return 0;
1669 }
1670
1671 xcrs.nr_xcrs = 1;
1672 xcrs.flags = 0;
1673 xcrs.xcrs[0].xcr = 0;
1674 xcrs.xcrs[0].value = env->xcr0;
1675 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_XCRS, &xcrs);
1676 }
1677
1678 static int kvm_put_sregs(X86CPU *cpu)
1679 {
1680 CPUX86State *env = &cpu->env;
1681 struct kvm_sregs sregs;
1682
1683 memset(sregs.interrupt_bitmap, 0, sizeof(sregs.interrupt_bitmap));
1684 if (env->interrupt_injected >= 0) {
1685 sregs.interrupt_bitmap[env->interrupt_injected / 64] |=
1686 (uint64_t)1 << (env->interrupt_injected % 64);
1687 }
1688
1689 if ((env->eflags & VM_MASK)) {
1690 set_v8086_seg(&sregs.cs, &env->segs[R_CS]);
1691 set_v8086_seg(&sregs.ds, &env->segs[R_DS]);
1692 set_v8086_seg(&sregs.es, &env->segs[R_ES]);
1693 set_v8086_seg(&sregs.fs, &env->segs[R_FS]);
1694 set_v8086_seg(&sregs.gs, &env->segs[R_GS]);
1695 set_v8086_seg(&sregs.ss, &env->segs[R_SS]);
1696 } else {
1697 set_seg(&sregs.cs, &env->segs[R_CS]);
1698 set_seg(&sregs.ds, &env->segs[R_DS]);
1699 set_seg(&sregs.es, &env->segs[R_ES]);
1700 set_seg(&sregs.fs, &env->segs[R_FS]);
1701 set_seg(&sregs.gs, &env->segs[R_GS]);
1702 set_seg(&sregs.ss, &env->segs[R_SS]);
1703 }
1704
1705 set_seg(&sregs.tr, &env->tr);
1706 set_seg(&sregs.ldt, &env->ldt);
1707
1708 sregs.idt.limit = env->idt.limit;
1709 sregs.idt.base = env->idt.base;
1710 memset(sregs.idt.padding, 0, sizeof sregs.idt.padding);
1711 sregs.gdt.limit = env->gdt.limit;
1712 sregs.gdt.base = env->gdt.base;
1713 memset(sregs.gdt.padding, 0, sizeof sregs.gdt.padding);
1714
1715 sregs.cr0 = env->cr[0];
1716 sregs.cr2 = env->cr[2];
1717 sregs.cr3 = env->cr[3];
1718 sregs.cr4 = env->cr[4];
1719
1720 sregs.cr8 = cpu_get_apic_tpr(cpu->apic_state);
1721 sregs.apic_base = cpu_get_apic_base(cpu->apic_state);
1722
1723 sregs.efer = env->efer;
1724
1725 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_SREGS, &sregs);
1726 }
1727
1728 static void kvm_msr_buf_reset(X86CPU *cpu)
1729 {
1730 memset(cpu->kvm_msr_buf, 0, MSR_BUF_SIZE);
1731 }
1732
1733 static void kvm_msr_entry_add(X86CPU *cpu, uint32_t index, uint64_t value)
1734 {
1735 struct kvm_msrs *msrs = cpu->kvm_msr_buf;
1736 void *limit = ((void *)msrs) + MSR_BUF_SIZE;
1737 struct kvm_msr_entry *entry = &msrs->entries[msrs->nmsrs];
1738
1739 assert((void *)(entry + 1) <= limit);
1740
1741 entry->index = index;
1742 entry->reserved = 0;
1743 entry->data = value;
1744 msrs->nmsrs++;
1745 }
1746
1747 static int kvm_put_one_msr(X86CPU *cpu, int index, uint64_t value)
1748 {
1749 kvm_msr_buf_reset(cpu);
1750 kvm_msr_entry_add(cpu, index, value);
1751
1752 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MSRS, cpu->kvm_msr_buf);
1753 }
1754
1755 void kvm_put_apicbase(X86CPU *cpu, uint64_t value)
1756 {
1757 int ret;
1758
1759 ret = kvm_put_one_msr(cpu, MSR_IA32_APICBASE, value);
1760 assert(ret == 1);
1761 }
1762
1763 static int kvm_put_tscdeadline_msr(X86CPU *cpu)
1764 {
1765 CPUX86State *env = &cpu->env;
1766 int ret;
1767
1768 if (!has_msr_tsc_deadline) {
1769 return 0;
1770 }
1771
1772 ret = kvm_put_one_msr(cpu, MSR_IA32_TSCDEADLINE, env->tsc_deadline);
1773 if (ret < 0) {
1774 return ret;
1775 }
1776
1777 assert(ret == 1);
1778 return 0;
1779 }
1780
1781 /*
1782 * Provide a separate write service for the feature control MSR in order to
1783 * kick the VCPU out of VMXON or even guest mode on reset. This has to be done
1784 * before writing any other state because forcibly leaving nested mode
1785 * invalidates the VCPU state.
1786 */
1787 static int kvm_put_msr_feature_control(X86CPU *cpu)
1788 {
1789 int ret;
1790
1791 if (!has_msr_feature_control) {
1792 return 0;
1793 }
1794
1795 ret = kvm_put_one_msr(cpu, MSR_IA32_FEATURE_CONTROL,
1796 cpu->env.msr_ia32_feature_control);
1797 if (ret < 0) {
1798 return ret;
1799 }
1800
1801 assert(ret == 1);
1802 return 0;
1803 }
1804
1805 static int kvm_put_msrs(X86CPU *cpu, int level)
1806 {
1807 CPUX86State *env = &cpu->env;
1808 int i;
1809 int ret;
1810
1811 kvm_msr_buf_reset(cpu);
1812
1813 kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_CS, env->sysenter_cs);
1814 kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_ESP, env->sysenter_esp);
1815 kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_EIP, env->sysenter_eip);
1816 kvm_msr_entry_add(cpu, MSR_PAT, env->pat);
1817 if (has_msr_star) {
1818 kvm_msr_entry_add(cpu, MSR_STAR, env->star);
1819 }
1820 if (has_msr_hsave_pa) {
1821 kvm_msr_entry_add(cpu, MSR_VM_HSAVE_PA, env->vm_hsave);
1822 }
1823 if (has_msr_tsc_aux) {
1824 kvm_msr_entry_add(cpu, MSR_TSC_AUX, env->tsc_aux);
1825 }
1826 if (has_msr_tsc_adjust) {
1827 kvm_msr_entry_add(cpu, MSR_TSC_ADJUST, env->tsc_adjust);
1828 }
1829 if (has_msr_misc_enable) {
1830 kvm_msr_entry_add(cpu, MSR_IA32_MISC_ENABLE,
1831 env->msr_ia32_misc_enable);
1832 }
1833 if (has_msr_smbase) {
1834 kvm_msr_entry_add(cpu, MSR_IA32_SMBASE, env->smbase);
1835 }
1836 if (has_msr_smi_count) {
1837 kvm_msr_entry_add(cpu, MSR_SMI_COUNT, env->msr_smi_count);
1838 }
1839 if (has_msr_bndcfgs) {
1840 kvm_msr_entry_add(cpu, MSR_IA32_BNDCFGS, env->msr_bndcfgs);
1841 }
1842 if (has_msr_xss) {
1843 kvm_msr_entry_add(cpu, MSR_IA32_XSS, env->xss);
1844 }
1845 if (has_msr_spec_ctrl) {
1846 kvm_msr_entry_add(cpu, MSR_IA32_SPEC_CTRL, env->spec_ctrl);
1847 }
1848 if (has_msr_virt_ssbd) {
1849 kvm_msr_entry_add(cpu, MSR_VIRT_SSBD, env->virt_ssbd);
1850 }
1851
1852 #ifdef TARGET_X86_64
1853 if (lm_capable_kernel) {
1854 kvm_msr_entry_add(cpu, MSR_CSTAR, env->cstar);
1855 kvm_msr_entry_add(cpu, MSR_KERNELGSBASE, env->kernelgsbase);
1856 kvm_msr_entry_add(cpu, MSR_FMASK, env->fmask);
1857 kvm_msr_entry_add(cpu, MSR_LSTAR, env->lstar);
1858 }
1859 #endif
1860
1861 /*
1862 * The following MSRs have side effects on the guest or are too heavy
1863 * for normal writeback. Limit them to reset or full state updates.
1864 */
1865 if (level >= KVM_PUT_RESET_STATE) {
1866 kvm_msr_entry_add(cpu, MSR_IA32_TSC, env->tsc);
1867 kvm_msr_entry_add(cpu, MSR_KVM_SYSTEM_TIME, env->system_time_msr);
1868 kvm_msr_entry_add(cpu, MSR_KVM_WALL_CLOCK, env->wall_clock_msr);
1869 if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_ASYNC_PF)) {
1870 kvm_msr_entry_add(cpu, MSR_KVM_ASYNC_PF_EN, env->async_pf_en_msr);
1871 }
1872 if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_PV_EOI)) {
1873 kvm_msr_entry_add(cpu, MSR_KVM_PV_EOI_EN, env->pv_eoi_en_msr);
1874 }
1875 if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_STEAL_TIME)) {
1876 kvm_msr_entry_add(cpu, MSR_KVM_STEAL_TIME, env->steal_time_msr);
1877 }
1878 if (has_architectural_pmu_version > 0) {
1879 if (has_architectural_pmu_version > 1) {
1880 /* Stop the counter. */
1881 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR_CTRL, 0);
1882 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_CTRL, 0);
1883 }
1884
1885 /* Set the counter values. */
1886 for (i = 0; i < num_architectural_pmu_fixed_counters; i++) {
1887 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR0 + i,
1888 env->msr_fixed_counters[i]);
1889 }
1890 for (i = 0; i < num_architectural_pmu_gp_counters; i++) {
1891 kvm_msr_entry_add(cpu, MSR_P6_PERFCTR0 + i,
1892 env->msr_gp_counters[i]);
1893 kvm_msr_entry_add(cpu, MSR_P6_EVNTSEL0 + i,
1894 env->msr_gp_evtsel[i]);
1895 }
1896 if (has_architectural_pmu_version > 1) {
1897 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_STATUS,
1898 env->msr_global_status);
1899 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1900 env->msr_global_ovf_ctrl);
1901
1902 /* Now start the PMU. */
1903 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR_CTRL,
1904 env->msr_fixed_ctr_ctrl);
1905 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_CTRL,
1906 env->msr_global_ctrl);
1907 }
1908 }
1909 /*
1910 * Hyper-V partition-wide MSRs: to avoid clearing them on cpu hot-add,
1911 * only sync them to KVM on the first cpu
1912 */
1913 if (current_cpu == first_cpu) {
1914 if (has_msr_hv_hypercall) {
1915 kvm_msr_entry_add(cpu, HV_X64_MSR_GUEST_OS_ID,
1916 env->msr_hv_guest_os_id);
1917 kvm_msr_entry_add(cpu, HV_X64_MSR_HYPERCALL,
1918 env->msr_hv_hypercall);
1919 }
1920 if (cpu->hyperv_time) {
1921 kvm_msr_entry_add(cpu, HV_X64_MSR_REFERENCE_TSC,
1922 env->msr_hv_tsc);
1923 }
1924 if (cpu->hyperv_reenlightenment) {
1925 kvm_msr_entry_add(cpu, HV_X64_MSR_REENLIGHTENMENT_CONTROL,
1926 env->msr_hv_reenlightenment_control);
1927 kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_CONTROL,
1928 env->msr_hv_tsc_emulation_control);
1929 kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_STATUS,
1930 env->msr_hv_tsc_emulation_status);
1931 }
1932 }
1933 if (cpu->hyperv_vapic) {
1934 kvm_msr_entry_add(cpu, HV_X64_MSR_APIC_ASSIST_PAGE,
1935 env->msr_hv_vapic);
1936 }
1937 if (has_msr_hv_crash) {
1938 int j;
1939
1940 for (j = 0; j < HV_CRASH_PARAMS; j++)
1941 kvm_msr_entry_add(cpu, HV_X64_MSR_CRASH_P0 + j,
1942 env->msr_hv_crash_params[j]);
1943
1944 kvm_msr_entry_add(cpu, HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_NOTIFY);
1945 }
1946 if (has_msr_hv_runtime) {
1947 kvm_msr_entry_add(cpu, HV_X64_MSR_VP_RUNTIME, env->msr_hv_runtime);
1948 }
1949 if (cpu->hyperv_vpindex && hv_vpindex_settable) {
1950 kvm_msr_entry_add(cpu, HV_X64_MSR_VP_INDEX, hyperv_vp_index(cpu));
1951 }
1952 if (cpu->hyperv_synic) {
1953 int j;
1954
1955 kvm_msr_entry_add(cpu, HV_X64_MSR_SVERSION, HV_SYNIC_VERSION);
1956
1957 kvm_msr_entry_add(cpu, HV_X64_MSR_SCONTROL,
1958 env->msr_hv_synic_control);
1959 kvm_msr_entry_add(cpu, HV_X64_MSR_SIEFP,
1960 env->msr_hv_synic_evt_page);
1961 kvm_msr_entry_add(cpu, HV_X64_MSR_SIMP,
1962 env->msr_hv_synic_msg_page);
1963
1964 for (j = 0; j < ARRAY_SIZE(env->msr_hv_synic_sint); j++) {
1965 kvm_msr_entry_add(cpu, HV_X64_MSR_SINT0 + j,
1966 env->msr_hv_synic_sint[j]);
1967 }
1968 }
1969 if (has_msr_hv_stimer) {
1970 int j;
1971
1972 for (j = 0; j < ARRAY_SIZE(env->msr_hv_stimer_config); j++) {
1973 kvm_msr_entry_add(cpu, HV_X64_MSR_STIMER0_CONFIG + j * 2,
1974 env->msr_hv_stimer_config[j]);
1975 }
1976
1977 for (j = 0; j < ARRAY_SIZE(env->msr_hv_stimer_count); j++) {
1978 kvm_msr_entry_add(cpu, HV_X64_MSR_STIMER0_COUNT + j * 2,
1979 env->msr_hv_stimer_count[j]);
1980 }
1981 }
1982 if (env->features[FEAT_1_EDX] & CPUID_MTRR) {
1983 uint64_t phys_mask = MAKE_64BIT_MASK(0, cpu->phys_bits);
1984
1985 kvm_msr_entry_add(cpu, MSR_MTRRdefType, env->mtrr_deftype);
1986 kvm_msr_entry_add(cpu, MSR_MTRRfix64K_00000, env->mtrr_fixed[0]);
1987 kvm_msr_entry_add(cpu, MSR_MTRRfix16K_80000, env->mtrr_fixed[1]);
1988 kvm_msr_entry_add(cpu, MSR_MTRRfix16K_A0000, env->mtrr_fixed[2]);
1989 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C0000, env->mtrr_fixed[3]);
1990 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C8000, env->mtrr_fixed[4]);
1991 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D0000, env->mtrr_fixed[5]);
1992 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D8000, env->mtrr_fixed[6]);
1993 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E0000, env->mtrr_fixed[7]);
1994 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E8000, env->mtrr_fixed[8]);
1995 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F0000, env->mtrr_fixed[9]);
1996 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F8000, env->mtrr_fixed[10]);
1997 for (i = 0; i < MSR_MTRRcap_VCNT; i++) {
1998 /* The CPU GPs if we write to a bit above the physical limit of
1999 * the host CPU (and KVM emulates that)
2000 */
2001 uint64_t mask = env->mtrr_var[i].mask;
2002 mask &= phys_mask;
2003
2004 kvm_msr_entry_add(cpu, MSR_MTRRphysBase(i),
2005 env->mtrr_var[i].base);
2006 kvm_msr_entry_add(cpu, MSR_MTRRphysMask(i), mask);
2007 }
2008 }
2009 if (env->features[FEAT_7_0_EBX] & CPUID_7_0_EBX_INTEL_PT) {
2010 int addr_num = kvm_arch_get_supported_cpuid(kvm_state,
2011 0x14, 1, R_EAX) & 0x7;
2012
2013 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CTL,
2014 env->msr_rtit_ctrl);
2015 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_STATUS,
2016 env->msr_rtit_status);
2017 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_BASE,
2018 env->msr_rtit_output_base);
2019 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_MASK,
2020 env->msr_rtit_output_mask);
2021 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CR3_MATCH,
2022 env->msr_rtit_cr3_match);
2023 for (i = 0; i < addr_num; i++) {
2024 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_ADDR0_A + i,
2025 env->msr_rtit_addrs[i]);
2026 }
2027 }
2028
2029 /* Note: MSR_IA32_FEATURE_CONTROL is written separately, see
2030 * kvm_put_msr_feature_control. */
2031 }
2032 if (env->mcg_cap) {
2033 int i;
2034
2035 kvm_msr_entry_add(cpu, MSR_MCG_STATUS, env->mcg_status);
2036 kvm_msr_entry_add(cpu, MSR_MCG_CTL, env->mcg_ctl);
2037 if (has_msr_mcg_ext_ctl) {
2038 kvm_msr_entry_add(cpu, MSR_MCG_EXT_CTL, env->mcg_ext_ctl);
2039 }
2040 for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++) {
2041 kvm_msr_entry_add(cpu, MSR_MC0_CTL + i, env->mce_banks[i]);
2042 }
2043 }
2044
2045 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MSRS, cpu->kvm_msr_buf);
2046 if (ret < 0) {
2047 return ret;
2048 }
2049
2050 if (ret < cpu->kvm_msr_buf->nmsrs) {
2051 struct kvm_msr_entry *e = &cpu->kvm_msr_buf->entries[ret];
2052 error_report("error: failed to set MSR 0x%" PRIx32 " to 0x%" PRIx64,
2053 (uint32_t)e->index, (uint64_t)e->data);
2054 }
2055
2056 assert(ret == cpu->kvm_msr_buf->nmsrs);
2057 return 0;
2058 }
2059
2060
2061 static int kvm_get_fpu(X86CPU *cpu)
2062 {
2063 CPUX86State *env = &cpu->env;
2064 struct kvm_fpu fpu;
2065 int i, ret;
2066
2067 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_FPU, &fpu);
2068 if (ret < 0) {
2069 return ret;
2070 }
2071
2072 env->fpstt = (fpu.fsw >> 11) & 7;
2073 env->fpus = fpu.fsw;
2074 env->fpuc = fpu.fcw;
2075 env->fpop = fpu.last_opcode;
2076 env->fpip = fpu.last_ip;
2077 env->fpdp = fpu.last_dp;
2078 for (i = 0; i < 8; ++i) {
2079 env->fptags[i] = !((fpu.ftwx >> i) & 1);
2080 }
2081 memcpy(env->fpregs, fpu.fpr, sizeof env->fpregs);
2082 for (i = 0; i < CPU_NB_REGS; i++) {
2083 env->xmm_regs[i].ZMM_Q(0) = ldq_p(&fpu.xmm[i][0]);
2084 env->xmm_regs[i].ZMM_Q(1) = ldq_p(&fpu.xmm[i][8]);
2085 }
2086 env->mxcsr = fpu.mxcsr;
2087
2088 return 0;
2089 }
2090
2091 static int kvm_get_xsave(X86CPU *cpu)
2092 {
2093 CPUX86State *env = &cpu->env;
2094 X86XSaveArea *xsave = env->kvm_xsave_buf;
2095 int ret;
2096
2097 if (!has_xsave) {
2098 return kvm_get_fpu(cpu);
2099 }
2100
2101 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_XSAVE, xsave);
2102 if (ret < 0) {
2103 return ret;
2104 }
2105 x86_cpu_xrstor_all_areas(cpu, xsave);
2106
2107 return 0;
2108 }
2109
2110 static int kvm_get_xcrs(X86CPU *cpu)
2111 {
2112 CPUX86State *env = &cpu->env;
2113 int i, ret;
2114 struct kvm_xcrs xcrs;
2115
2116 if (!has_xcrs) {
2117 return 0;
2118 }
2119
2120 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_XCRS, &xcrs);
2121 if (ret < 0) {
2122 return ret;
2123 }
2124
2125 for (i = 0; i < xcrs.nr_xcrs; i++) {
2126 /* Only support xcr0 now */
2127 if (xcrs.xcrs[i].xcr == 0) {
2128 env->xcr0 = xcrs.xcrs[i].value;
2129 break;
2130 }
2131 }
2132 return 0;
2133 }
2134
2135 static int kvm_get_sregs(X86CPU *cpu)
2136 {
2137 CPUX86State *env = &cpu->env;
2138 struct kvm_sregs sregs;
2139 int bit, i, ret;
2140
2141 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
2142 if (ret < 0) {
2143 return ret;
2144 }
2145
2146 /* There can only be one pending IRQ set in the bitmap at a time, so try
2147 to find it and save its number instead (-1 for none). */
2148 env->interrupt_injected = -1;
2149 for (i = 0; i < ARRAY_SIZE(sregs.interrupt_bitmap); i++) {
2150 if (sregs.interrupt_bitmap[i]) {
2151 bit = ctz64(sregs.interrupt_bitmap[i]);
2152 env->interrupt_injected = i * 64 + bit;
2153 break;
2154 }
2155 }
2156
2157 get_seg(&env->segs[R_CS], &sregs.cs);
2158 get_seg(&env->segs[R_DS], &sregs.ds);
2159 get_seg(&env->segs[R_ES], &sregs.es);
2160 get_seg(&env->segs[R_FS], &sregs.fs);
2161 get_seg(&env->segs[R_GS], &sregs.gs);
2162 get_seg(&env->segs[R_SS], &sregs.ss);
2163
2164 get_seg(&env->tr, &sregs.tr);
2165 get_seg(&env->ldt, &sregs.ldt);
2166
2167 env->idt.limit = sregs.idt.limit;
2168 env->idt.base = sregs.idt.base;
2169 env->gdt.limit = sregs.gdt.limit;
2170 env->gdt.base = sregs.gdt.base;
2171
2172 env->cr[0] = sregs.cr0;
2173 env->cr[2] = sregs.cr2;
2174 env->cr[3] = sregs.cr3;
2175 env->cr[4] = sregs.cr4;
2176
2177 env->efer = sregs.efer;
2178
2179 /* changes to apic base and cr8/tpr are read back via kvm_arch_post_run */
2180 x86_update_hflags(env);
2181
2182 return 0;
2183 }
2184
2185 static int kvm_get_msrs(X86CPU *cpu)
2186 {
2187 CPUX86State *env = &cpu->env;
2188 struct kvm_msr_entry *msrs = cpu->kvm_msr_buf->entries;
2189 int ret, i;
2190 uint64_t mtrr_top_bits;
2191
2192 kvm_msr_buf_reset(cpu);
2193
2194 kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_CS, 0);
2195 kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_ESP, 0);
2196 kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_EIP, 0);
2197 kvm_msr_entry_add(cpu, MSR_PAT, 0);
2198 if (has_msr_star) {
2199 kvm_msr_entry_add(cpu, MSR_STAR, 0);
2200 }
2201 if (has_msr_hsave_pa) {
2202 kvm_msr_entry_add(cpu, MSR_VM_HSAVE_PA, 0);
2203 }
2204 if (has_msr_tsc_aux) {
2205 kvm_msr_entry_add(cpu, MSR_TSC_AUX, 0);
2206 }
2207 if (has_msr_tsc_adjust) {
2208 kvm_msr_entry_add(cpu, MSR_TSC_ADJUST, 0);
2209 }
2210 if (has_msr_tsc_deadline) {
2211 kvm_msr_entry_add(cpu, MSR_IA32_TSCDEADLINE, 0);
2212 }
2213 if (has_msr_misc_enable) {
2214 kvm_msr_entry_add(cpu, MSR_IA32_MISC_ENABLE, 0);
2215 }
2216 if (has_msr_smbase) {
2217 kvm_msr_entry_add(cpu, MSR_IA32_SMBASE, 0);
2218 }
2219 if (has_msr_smi_count) {
2220 kvm_msr_entry_add(cpu, MSR_SMI_COUNT, 0);
2221 }
2222 if (has_msr_feature_control) {
2223 kvm_msr_entry_add(cpu, MSR_IA32_FEATURE_CONTROL, 0);
2224 }
2225 if (has_msr_bndcfgs) {
2226 kvm_msr_entry_add(cpu, MSR_IA32_BNDCFGS, 0);
2227 }
2228 if (has_msr_xss) {
2229 kvm_msr_entry_add(cpu, MSR_IA32_XSS, 0);
2230 }
2231 if (has_msr_spec_ctrl) {
2232 kvm_msr_entry_add(cpu, MSR_IA32_SPEC_CTRL, 0);
2233 }
2234 if (has_msr_virt_ssbd) {
2235 kvm_msr_entry_add(cpu, MSR_VIRT_SSBD, 0);
2236 }
2237 if (!env->tsc_valid) {
2238 kvm_msr_entry_add(cpu, MSR_IA32_TSC, 0);
2239 env->tsc_valid = !runstate_is_running();
2240 }
2241
2242 #ifdef TARGET_X86_64
2243 if (lm_capable_kernel) {
2244 kvm_msr_entry_add(cpu, MSR_CSTAR, 0);
2245 kvm_msr_entry_add(cpu, MSR_KERNELGSBASE, 0);
2246 kvm_msr_entry_add(cpu, MSR_FMASK, 0);
2247 kvm_msr_entry_add(cpu, MSR_LSTAR, 0);
2248 }
2249 #endif
2250 kvm_msr_entry_add(cpu, MSR_KVM_SYSTEM_TIME, 0);
2251 kvm_msr_entry_add(cpu, MSR_KVM_WALL_CLOCK, 0);
2252 if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_ASYNC_PF)) {
2253 kvm_msr_entry_add(cpu, MSR_KVM_ASYNC_PF_EN, 0);
2254 }
2255 if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_PV_EOI)) {
2256 kvm_msr_entry_add(cpu, MSR_KVM_PV_EOI_EN, 0);
2257 }
2258 if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_STEAL_TIME)) {
2259 kvm_msr_entry_add(cpu, MSR_KVM_STEAL_TIME, 0);
2260 }
2261 if (has_architectural_pmu_version > 0) {
2262 if (has_architectural_pmu_version > 1) {
2263 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR_CTRL, 0);
2264 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_CTRL, 0);
2265 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_STATUS, 0);
2266 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_OVF_CTRL, 0);
2267 }
2268 for (i = 0; i < num_architectural_pmu_fixed_counters; i++) {
2269 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR0 + i, 0);
2270 }
2271 for (i = 0; i < num_architectural_pmu_gp_counters; i++) {
2272 kvm_msr_entry_add(cpu, MSR_P6_PERFCTR0 + i, 0);
2273 kvm_msr_entry_add(cpu, MSR_P6_EVNTSEL0 + i, 0);
2274 }
2275 }
2276
2277 if (env->mcg_cap) {
2278 kvm_msr_entry_add(cpu, MSR_MCG_STATUS, 0);
2279 kvm_msr_entry_add(cpu, MSR_MCG_CTL, 0);
2280 if (has_msr_mcg_ext_ctl) {
2281 kvm_msr_entry_add(cpu, MSR_MCG_EXT_CTL, 0);
2282 }
2283 for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++) {
2284 kvm_msr_entry_add(cpu, MSR_MC0_CTL + i, 0);
2285 }
2286 }
2287
2288 if (has_msr_hv_hypercall) {
2289 kvm_msr_entry_add(cpu, HV_X64_MSR_HYPERCALL, 0);
2290 kvm_msr_entry_add(cpu, HV_X64_MSR_GUEST_OS_ID, 0);
2291 }
2292 if (cpu->hyperv_vapic) {
2293 kvm_msr_entry_add(cpu, HV_X64_MSR_APIC_ASSIST_PAGE, 0);
2294 }
2295 if (cpu->hyperv_time) {
2296 kvm_msr_entry_add(cpu, HV_X64_MSR_REFERENCE_TSC, 0);
2297 }
2298 if (cpu->hyperv_reenlightenment) {
2299 kvm_msr_entry_add(cpu, HV_X64_MSR_REENLIGHTENMENT_CONTROL, 0);
2300 kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_CONTROL, 0);
2301 kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_STATUS, 0);
2302 }
2303 if (has_msr_hv_crash) {
2304 int j;
2305
2306 for (j = 0; j < HV_CRASH_PARAMS; j++) {
2307 kvm_msr_entry_add(cpu, HV_X64_MSR_CRASH_P0 + j, 0);
2308 }
2309 }
2310 if (has_msr_hv_runtime) {
2311 kvm_msr_entry_add(cpu, HV_X64_MSR_VP_RUNTIME, 0);
2312 }
2313 if (cpu->hyperv_synic) {
2314 uint32_t msr;
2315
2316 kvm_msr_entry_add(cpu, HV_X64_MSR_SCONTROL, 0);
2317 kvm_msr_entry_add(cpu, HV_X64_MSR_SIEFP, 0);
2318 kvm_msr_entry_add(cpu, HV_X64_MSR_SIMP, 0);
2319 for (msr = HV_X64_MSR_SINT0; msr <= HV_X64_MSR_SINT15; msr++) {
2320 kvm_msr_entry_add(cpu, msr, 0);
2321 }
2322 }
2323 if (has_msr_hv_stimer) {
2324 uint32_t msr;
2325
2326 for (msr = HV_X64_MSR_STIMER0_CONFIG; msr <= HV_X64_MSR_STIMER3_COUNT;
2327 msr++) {
2328 kvm_msr_entry_add(cpu, msr, 0);
2329 }
2330 }
2331 if (env->features[FEAT_1_EDX] & CPUID_MTRR) {
2332 kvm_msr_entry_add(cpu, MSR_MTRRdefType, 0);
2333 kvm_msr_entry_add(cpu, MSR_MTRRfix64K_00000, 0);
2334 kvm_msr_entry_add(cpu, MSR_MTRRfix16K_80000, 0);
2335 kvm_msr_entry_add(cpu, MSR_MTRRfix16K_A0000, 0);
2336 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C0000, 0);
2337 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C8000, 0);
2338 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D0000, 0);
2339 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D8000, 0);
2340 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E0000, 0);
2341 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E8000, 0);
2342 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F0000, 0);
2343 kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F8000, 0);
2344 for (i = 0; i < MSR_MTRRcap_VCNT; i++) {
2345 kvm_msr_entry_add(cpu, MSR_MTRRphysBase(i), 0);
2346 kvm_msr_entry_add(cpu, MSR_MTRRphysMask(i), 0);
2347 }
2348 }
2349
2350 if (env->features[FEAT_7_0_EBX] & CPUID_7_0_EBX_INTEL_PT) {
2351 int addr_num =
2352 kvm_arch_get_supported_cpuid(kvm_state, 0x14, 1, R_EAX) & 0x7;
2353
2354 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CTL, 0);
2355 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_STATUS, 0);
2356 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_BASE, 0);
2357 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_MASK, 0);
2358 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CR3_MATCH, 0);
2359 for (i = 0; i < addr_num; i++) {
2360 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_ADDR0_A + i, 0);
2361 }
2362 }
2363
2364 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MSRS, cpu->kvm_msr_buf);
2365 if (ret < 0) {
2366 return ret;
2367 }
2368
2369 if (ret < cpu->kvm_msr_buf->nmsrs) {
2370 struct kvm_msr_entry *e = &cpu->kvm_msr_buf->entries[ret];
2371 error_report("error: failed to get MSR 0x%" PRIx32,
2372 (uint32_t)e->index);
2373 }
2374
2375 assert(ret == cpu->kvm_msr_buf->nmsrs);
2376 /*
2377 * MTRR masks: Each mask consists of 5 parts
2378 * a 10..0: must be zero
2379 * b 11 : valid bit
2380 * c n-1.12: actual mask bits
2381 * d 51..n: reserved must be zero
2382 * e 63.52: reserved must be zero
2383 *
2384 * 'n' is the number of physical bits supported by the CPU and is
2385 * apparently always <= 52. We know our 'n' but don't know what
2386 * the destinations 'n' is; it might be smaller, in which case
2387 * it masks (c) on loading. It might be larger, in which case
2388 * we fill 'd' so that d..c is consistent irrespetive of the 'n'
2389 * we're migrating to.
2390 */
2391
2392 if (cpu->fill_mtrr_mask) {
2393 QEMU_BUILD_BUG_ON(TARGET_PHYS_ADDR_SPACE_BITS > 52);
2394 assert(cpu->phys_bits <= TARGET_PHYS_ADDR_SPACE_BITS);
2395 mtrr_top_bits = MAKE_64BIT_MASK(cpu->phys_bits, 52 - cpu->phys_bits);
2396 } else {
2397 mtrr_top_bits = 0;
2398 }
2399
2400 for (i = 0; i < ret; i++) {
2401 uint32_t index = msrs[i].index;
2402 switch (index) {
2403 case MSR_IA32_SYSENTER_CS:
2404 env->sysenter_cs = msrs[i].data;
2405 break;
2406 case MSR_IA32_SYSENTER_ESP:
2407 env->sysenter_esp = msrs[i].data;
2408 break;
2409 case MSR_IA32_SYSENTER_EIP:
2410 env->sysenter_eip = msrs[i].data;
2411 break;
2412 case MSR_PAT:
2413 env->pat = msrs[i].data;
2414 break;
2415 case MSR_STAR:
2416 env->star = msrs[i].data;
2417 break;
2418 #ifdef TARGET_X86_64
2419 case MSR_CSTAR:
2420 env->cstar = msrs[i].data;
2421 break;
2422 case MSR_KERNELGSBASE:
2423 env->kernelgsbase = msrs[i].data;
2424 break;
2425 case MSR_FMASK:
2426 env->fmask = msrs[i].data;
2427 break;
2428 case MSR_LSTAR:
2429 env->lstar = msrs[i].data;
2430 break;
2431 #endif
2432 case MSR_IA32_TSC:
2433 env->tsc = msrs[i].data;
2434 break;
2435 case MSR_TSC_AUX:
2436 env->tsc_aux = msrs[i].data;
2437 break;
2438 case MSR_TSC_ADJUST:
2439 env->tsc_adjust = msrs[i].data;
2440 break;
2441 case MSR_IA32_TSCDEADLINE:
2442 env->tsc_deadline = msrs[i].data;
2443 break;
2444 case MSR_VM_HSAVE_PA:
2445 env->vm_hsave = msrs[i].data;
2446 break;
2447 case MSR_KVM_SYSTEM_TIME:
2448 env->system_time_msr = msrs[i].data;
2449 break;
2450 case MSR_KVM_WALL_CLOCK:
2451 env->wall_clock_msr = msrs[i].data;
2452 break;
2453 case MSR_MCG_STATUS:
2454 env->mcg_status = msrs[i].data;
2455 break;
2456 case MSR_MCG_CTL:
2457 env->mcg_ctl = msrs[i].data;
2458 break;
2459 case MSR_MCG_EXT_CTL:
2460 env->mcg_ext_ctl = msrs[i].data;
2461 break;
2462 case MSR_IA32_MISC_ENABLE:
2463 env->msr_ia32_misc_enable = msrs[i].data;
2464 break;
2465 case MSR_IA32_SMBASE:
2466 env->smbase = msrs[i].data;
2467 break;
2468 case MSR_SMI_COUNT:
2469 env->msr_smi_count = msrs[i].data;
2470 break;
2471 case MSR_IA32_FEATURE_CONTROL:
2472 env->msr_ia32_feature_control = msrs[i].data;
2473 break;
2474 case MSR_IA32_BNDCFGS:
2475 env->msr_bndcfgs = msrs[i].data;
2476 break;
2477 case MSR_IA32_XSS:
2478 env->xss = msrs[i].data;
2479 break;
2480 default:
2481 if (msrs[i].index >= MSR_MC0_CTL &&
2482 msrs[i].index < MSR_MC0_CTL + (env->mcg_cap & 0xff) * 4) {
2483 env->mce_banks[msrs[i].index - MSR_MC0_CTL] = msrs[i].data;
2484 }
2485 break;
2486 case MSR_KVM_ASYNC_PF_EN:
2487 env->async_pf_en_msr = msrs[i].data;
2488 break;
2489 case MSR_KVM_PV_EOI_EN:
2490 env->pv_eoi_en_msr = msrs[i].data;
2491 break;
2492 case MSR_KVM_STEAL_TIME:
2493 env->steal_time_msr = msrs[i].data;
2494 break;
2495 case MSR_CORE_PERF_FIXED_CTR_CTRL:
2496 env->msr_fixed_ctr_ctrl = msrs[i].data;
2497 break;
2498 case MSR_CORE_PERF_GLOBAL_CTRL:
2499 env->msr_global_ctrl = msrs[i].data;
2500 break;
2501 case MSR_CORE_PERF_GLOBAL_STATUS:
2502 env->msr_global_status = msrs[i].data;
2503 break;
2504 case MSR_CORE_PERF_GLOBAL_OVF_CTRL:
2505 env->msr_global_ovf_ctrl = msrs[i].data;
2506 break;
2507 case MSR_CORE_PERF_FIXED_CTR0 ... MSR_CORE_PERF_FIXED_CTR0 + MAX_FIXED_COUNTERS - 1:
2508 env->msr_fixed_counters[index - MSR_CORE_PERF_FIXED_CTR0] = msrs[i].data;
2509 break;
2510 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR0 + MAX_GP_COUNTERS - 1:
2511 env->msr_gp_counters[index - MSR_P6_PERFCTR0] = msrs[i].data;
2512 break;
2513 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL0 + MAX_GP_COUNTERS - 1:
2514 env->msr_gp_evtsel[index - MSR_P6_EVNTSEL0] = msrs[i].data;
2515 break;
2516 case HV_X64_MSR_HYPERCALL:
2517 env->msr_hv_hypercall = msrs[i].data;
2518 break;
2519 case HV_X64_MSR_GUEST_OS_ID:
2520 env->msr_hv_guest_os_id = msrs[i].data;
2521 break;
2522 case HV_X64_MSR_APIC_ASSIST_PAGE:
2523 env->msr_hv_vapic = msrs[i].data;
2524 break;
2525 case HV_X64_MSR_REFERENCE_TSC:
2526 env->msr_hv_tsc = msrs[i].data;
2527 break;
2528 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2529 env->msr_hv_crash_params[index - HV_X64_MSR_CRASH_P0] = msrs[i].data;
2530 break;
2531 case HV_X64_MSR_VP_RUNTIME:
2532 env->msr_hv_runtime = msrs[i].data;
2533 break;
2534 case HV_X64_MSR_SCONTROL:
2535 env->msr_hv_synic_control = msrs[i].data;
2536 break;
2537 case HV_X64_MSR_SIEFP:
2538 env->msr_hv_synic_evt_page = msrs[i].data;
2539 break;
2540 case HV_X64_MSR_SIMP:
2541 env->msr_hv_synic_msg_page = msrs[i].data;
2542 break;
2543 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
2544 env->msr_hv_synic_sint[index - HV_X64_MSR_SINT0] = msrs[i].data;
2545 break;
2546 case HV_X64_MSR_STIMER0_CONFIG:
2547 case HV_X64_MSR_STIMER1_CONFIG:
2548 case HV_X64_MSR_STIMER2_CONFIG:
2549 case HV_X64_MSR_STIMER3_CONFIG:
2550 env->msr_hv_stimer_config[(index - HV_X64_MSR_STIMER0_CONFIG)/2] =
2551 msrs[i].data;
2552 break;
2553 case HV_X64_MSR_STIMER0_COUNT:
2554 case HV_X64_MSR_STIMER1_COUNT:
2555 case HV_X64_MSR_STIMER2_COUNT:
2556 case HV_X64_MSR_STIMER3_COUNT:
2557 env->msr_hv_stimer_count[(index - HV_X64_MSR_STIMER0_COUNT)/2] =
2558 msrs[i].data;
2559 break;
2560 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2561 env->msr_hv_reenlightenment_control = msrs[i].data;
2562 break;
2563 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2564 env->msr_hv_tsc_emulation_control = msrs[i].data;
2565 break;
2566 case HV_X64_MSR_TSC_EMULATION_STATUS:
2567 env->msr_hv_tsc_emulation_status = msrs[i].data;
2568 break;
2569 case MSR_MTRRdefType:
2570 env->mtrr_deftype = msrs[i].data;
2571 break;
2572 case MSR_MTRRfix64K_00000:
2573 env->mtrr_fixed[0] = msrs[i].data;
2574 break;
2575 case MSR_MTRRfix16K_80000:
2576 env->mtrr_fixed[1] = msrs[i].data;
2577 break;
2578 case MSR_MTRRfix16K_A0000:
2579 env->mtrr_fixed[2] = msrs[i].data;
2580 break;
2581 case MSR_MTRRfix4K_C0000:
2582 env->mtrr_fixed[3] = msrs[i].data;
2583 break;
2584 case MSR_MTRRfix4K_C8000:
2585 env->mtrr_fixed[4] = msrs[i].data;
2586 break;
2587 case MSR_MTRRfix4K_D0000:
2588 env->mtrr_fixed[5] = msrs[i].data;
2589 break;
2590 case MSR_MTRRfix4K_D8000:
2591 env->mtrr_fixed[6] = msrs[i].data;
2592 break;
2593 case MSR_MTRRfix4K_E0000:
2594 env->mtrr_fixed[7] = msrs[i].data;
2595 break;
2596 case MSR_MTRRfix4K_E8000:
2597 env->mtrr_fixed[8] = msrs[i].data;
2598 break;
2599 case MSR_MTRRfix4K_F0000:
2600 env->mtrr_fixed[9] = msrs[i].data;
2601 break;
2602 case MSR_MTRRfix4K_F8000:
2603 env->mtrr_fixed[10] = msrs[i].data;
2604 break;
2605 case MSR_MTRRphysBase(0) ... MSR_MTRRphysMask(MSR_MTRRcap_VCNT - 1):
2606 if (index & 1) {
2607 env->mtrr_var[MSR_MTRRphysIndex(index)].mask = msrs[i].data |
2608 mtrr_top_bits;
2609 } else {
2610 env->mtrr_var[MSR_MTRRphysIndex(index)].base = msrs[i].data;
2611 }
2612 break;
2613 case MSR_IA32_SPEC_CTRL:
2614 env->spec_ctrl = msrs[i].data;
2615 break;
2616 case MSR_VIRT_SSBD:
2617 env->virt_ssbd = msrs[i].data;
2618 break;
2619 case MSR_IA32_RTIT_CTL:
2620 env->msr_rtit_ctrl = msrs[i].data;
2621 break;
2622 case MSR_IA32_RTIT_STATUS:
2623 env->msr_rtit_status = msrs[i].data;
2624 break;
2625 case MSR_IA32_RTIT_OUTPUT_BASE:
2626 env->msr_rtit_output_base = msrs[i].data;
2627 break;
2628 case MSR_IA32_RTIT_OUTPUT_MASK:
2629 env->msr_rtit_output_mask = msrs[i].data;
2630 break;
2631 case MSR_IA32_RTIT_CR3_MATCH:
2632 env->msr_rtit_cr3_match = msrs[i].data;
2633 break;
2634 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2635 env->msr_rtit_addrs[index - MSR_IA32_RTIT_ADDR0_A] = msrs[i].data;
2636 break;
2637 }
2638 }
2639
2640 return 0;
2641 }
2642
2643 static int kvm_put_mp_state(X86CPU *cpu)
2644 {
2645 struct kvm_mp_state mp_state = { .mp_state = cpu->env.mp_state };
2646
2647 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
2648 }
2649
2650 static int kvm_get_mp_state(X86CPU *cpu)
2651 {
2652 CPUState *cs = CPU(cpu);
2653 CPUX86State *env = &cpu->env;
2654 struct kvm_mp_state mp_state;
2655 int ret;
2656
2657 ret = kvm_vcpu_ioctl(cs, KVM_GET_MP_STATE, &mp_state);
2658 if (ret < 0) {
2659 return ret;
2660 }
2661 env->mp_state = mp_state.mp_state;
2662 if (kvm_irqchip_in_kernel()) {
2663 cs->halted = (mp_state.mp_state == KVM_MP_STATE_HALTED);
2664 }
2665 return 0;
2666 }
2667
2668 static int kvm_get_apic(X86CPU *cpu)
2669 {
2670 DeviceState *apic = cpu->apic_state;
2671 struct kvm_lapic_state kapic;
2672 int ret;
2673
2674 if (apic && kvm_irqchip_in_kernel()) {
2675 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_LAPIC, &kapic);
2676 if (ret < 0) {
2677 return ret;
2678 }
2679
2680 kvm_get_apic_state(apic, &kapic);
2681 }
2682 return 0;
2683 }
2684
2685 static int kvm_put_vcpu_events(X86CPU *cpu, int level)
2686 {
2687 CPUState *cs = CPU(cpu);
2688 CPUX86State *env = &cpu->env;
2689 struct kvm_vcpu_events events = {};
2690
2691 if (!kvm_has_vcpu_events()) {
2692 return 0;
2693 }
2694
2695 events.exception.injected = (env->exception_injected >= 0);
2696 events.exception.nr = env->exception_injected;
2697 events.exception.has_error_code = env->has_error_code;
2698 events.exception.error_code = env->error_code;
2699 events.exception.pad = 0;
2700
2701 events.interrupt.injected = (env->interrupt_injected >= 0);
2702 events.interrupt.nr = env->interrupt_injected;
2703 events.interrupt.soft = env->soft_interrupt;
2704
2705 events.nmi.injected = env->nmi_injected;
2706 events.nmi.pending = env->nmi_pending;
2707 events.nmi.masked = !!(env->hflags2 & HF2_NMI_MASK);
2708 events.nmi.pad = 0;
2709
2710 events.sipi_vector = env->sipi_vector;
2711 events.flags = 0;
2712
2713 if (has_msr_smbase) {
2714 events.smi.smm = !!(env->hflags & HF_SMM_MASK);
2715 events.smi.smm_inside_nmi = !!(env->hflags2 & HF2_SMM_INSIDE_NMI_MASK);
2716 if (kvm_irqchip_in_kernel()) {
2717 /* As soon as these are moved to the kernel, remove them
2718 * from cs->interrupt_request.
2719 */
2720 events.smi.pending = cs->interrupt_request & CPU_INTERRUPT_SMI;
2721 events.smi.latched_init = cs->interrupt_request & CPU_INTERRUPT_INIT;
2722 cs->interrupt_request &= ~(CPU_INTERRUPT_INIT | CPU_INTERRUPT_SMI);
2723 } else {
2724 /* Keep these in cs->interrupt_request. */
2725 events.smi.pending = 0;
2726 events.smi.latched_init = 0;
2727 }
2728 /* Stop SMI delivery on old machine types to avoid a reboot
2729 * on an inward migration of an old VM.
2730 */
2731 if (!cpu->kvm_no_smi_migration) {
2732 events.flags |= KVM_VCPUEVENT_VALID_SMM;
2733 }
2734 }
2735
2736 if (level >= KVM_PUT_RESET_STATE) {
2737 events.flags |= KVM_VCPUEVENT_VALID_NMI_PENDING;
2738 if (env->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
2739 events.flags |= KVM_VCPUEVENT_VALID_SIPI_VECTOR;
2740 }
2741 }
2742
2743 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_VCPU_EVENTS, &events);
2744 }
2745
2746 static int kvm_get_vcpu_events(X86CPU *cpu)
2747 {
2748 CPUX86State *env = &cpu->env;
2749 struct kvm_vcpu_events events;
2750 int ret;
2751
2752 if (!kvm_has_vcpu_events()) {
2753 return 0;
2754 }
2755
2756 memset(&events, 0, sizeof(events));
2757 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_VCPU_EVENTS, &events);
2758 if (ret < 0) {
2759 return ret;
2760 }
2761 env->exception_injected =
2762 events.exception.injected ? events.exception.nr : -1;
2763 env->has_error_code = events.exception.has_error_code;
2764 env->error_code = events.exception.error_code;
2765
2766 env->interrupt_injected =
2767 events.interrupt.injected ? events.interrupt.nr : -1;
2768 env->soft_interrupt = events.interrupt.soft;
2769
2770 env->nmi_injected = events.nmi.injected;
2771 env->nmi_pending = events.nmi.pending;
2772 if (events.nmi.masked) {
2773 env->hflags2 |= HF2_NMI_MASK;
2774 } else {
2775 env->hflags2 &= ~HF2_NMI_MASK;
2776 }
2777
2778 if (events.flags & KVM_VCPUEVENT_VALID_SMM) {
2779 if (events.smi.smm) {
2780 env->hflags |= HF_SMM_MASK;
2781 } else {
2782 env->hflags &= ~HF_SMM_MASK;
2783 }
2784 if (events.smi.pending) {
2785 cpu_interrupt(CPU(cpu), CPU_INTERRUPT_SMI);
2786 } else {
2787 cpu_reset_interrupt(CPU(cpu), CPU_INTERRUPT_SMI);
2788 }
2789 if (events.smi.smm_inside_nmi) {
2790 env->hflags2 |= HF2_SMM_INSIDE_NMI_MASK;
2791 } else {
2792 env->hflags2 &= ~HF2_SMM_INSIDE_NMI_MASK;
2793 }
2794 if (events.smi.latched_init) {
2795 cpu_interrupt(CPU(cpu), CPU_INTERRUPT_INIT);
2796 } else {
2797 cpu_reset_interrupt(CPU(cpu), CPU_INTERRUPT_INIT);
2798 }
2799 }
2800
2801 env->sipi_vector = events.sipi_vector;
2802
2803 return 0;
2804 }
2805
2806 static int kvm_guest_debug_workarounds(X86CPU *cpu)
2807 {
2808 CPUState *cs = CPU(cpu);
2809 CPUX86State *env = &cpu->env;
2810 int ret = 0;
2811 unsigned long reinject_trap = 0;
2812
2813 if (!kvm_has_vcpu_events()) {
2814 if (env->exception_injected == 1) {
2815 reinject_trap = KVM_GUESTDBG_INJECT_DB;
2816 } else if (env->exception_injected == 3) {
2817 reinject_trap = KVM_GUESTDBG_INJECT_BP;
2818 }
2819 env->exception_injected = -1;
2820 }
2821
2822 /*
2823 * Kernels before KVM_CAP_X86_ROBUST_SINGLESTEP overwrote flags.TF
2824 * injected via SET_GUEST_DEBUG while updating GP regs. Work around this
2825 * by updating the debug state once again if single-stepping is on.
2826 * Another reason to call kvm_update_guest_debug here is a pending debug
2827 * trap raise by the guest. On kernels without SET_VCPU_EVENTS we have to
2828 * reinject them via SET_GUEST_DEBUG.
2829 */
2830 if (reinject_trap ||
2831 (!kvm_has_robust_singlestep() && cs->singlestep_enabled)) {
2832 ret = kvm_update_guest_debug(cs, reinject_trap);
2833 }
2834 return ret;
2835 }
2836
2837 static int kvm_put_debugregs(X86CPU *cpu)
2838 {
2839 CPUX86State *env = &cpu->env;
2840 struct kvm_debugregs dbgregs;
2841 int i;
2842
2843 if (!kvm_has_debugregs()) {
2844 return 0;
2845 }
2846
2847 for (i = 0; i < 4; i++) {
2848 dbgregs.db[i] = env->dr[i];
2849 }
2850 dbgregs.dr6 = env->dr[6];
2851 dbgregs.dr7 = env->dr[7];
2852 dbgregs.flags = 0;
2853
2854 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_DEBUGREGS, &dbgregs);
2855 }
2856
2857 static int kvm_get_debugregs(X86CPU *cpu)
2858 {
2859 CPUX86State *env = &cpu->env;
2860 struct kvm_debugregs dbgregs;
2861 int i, ret;
2862
2863 if (!kvm_has_debugregs()) {
2864 return 0;
2865 }
2866
2867 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_DEBUGREGS, &dbgregs);
2868 if (ret < 0) {
2869 return ret;
2870 }
2871 for (i = 0; i < 4; i++) {
2872 env->dr[i] = dbgregs.db[i];
2873 }
2874 env->dr[4] = env->dr[6] = dbgregs.dr6;
2875 env->dr[5] = env->dr[7] = dbgregs.dr7;
2876
2877 return 0;
2878 }
2879
2880 int kvm_arch_put_registers(CPUState *cpu, int level)
2881 {
2882 X86CPU *x86_cpu = X86_CPU(cpu);
2883 int ret;
2884
2885 assert(cpu_is_stopped(cpu) || qemu_cpu_is_self(cpu));
2886
2887 if (level >= KVM_PUT_RESET_STATE) {
2888 ret = kvm_put_msr_feature_control(x86_cpu);
2889 if (ret < 0) {
2890 return ret;
2891 }
2892 }
2893
2894 if (level == KVM_PUT_FULL_STATE) {
2895 /* We don't check for kvm_arch_set_tsc_khz() errors here,
2896 * because TSC frequency mismatch shouldn't abort migration,
2897 * unless the user explicitly asked for a more strict TSC
2898 * setting (e.g. using an explicit "tsc-freq" option).
2899 */
2900 kvm_arch_set_tsc_khz(cpu);
2901 }
2902
2903 ret = kvm_getput_regs(x86_cpu, 1);
2904 if (ret < 0) {
2905 return ret;
2906 }
2907 ret = kvm_put_xsave(x86_cpu);
2908 if (ret < 0) {
2909 return ret;
2910 }
2911 ret = kvm_put_xcrs(x86_cpu);
2912 if (ret < 0) {
2913 return ret;
2914 }
2915 ret = kvm_put_sregs(x86_cpu);
2916 if (ret < 0) {
2917 return ret;
2918 }
2919 /* must be before kvm_put_msrs */
2920 ret = kvm_inject_mce_oldstyle(x86_cpu);
2921 if (ret < 0) {
2922 return ret;
2923 }
2924 ret = kvm_put_msrs(x86_cpu, level);
2925 if (ret < 0) {
2926 return ret;
2927 }
2928 ret = kvm_put_vcpu_events(x86_cpu, level);
2929 if (ret < 0) {
2930 return ret;
2931 }
2932 if (level >= KVM_PUT_RESET_STATE) {
2933 ret = kvm_put_mp_state(x86_cpu);
2934 if (ret < 0) {
2935 return ret;
2936 }
2937 }
2938
2939 ret = kvm_put_tscdeadline_msr(x86_cpu);
2940 if (ret < 0) {
2941 return ret;
2942 }
2943 ret = kvm_put_debugregs(x86_cpu);
2944 if (ret < 0) {
2945 return ret;
2946 }
2947 /* must be last */
2948 ret = kvm_guest_debug_workarounds(x86_cpu);
2949 if (ret < 0) {
2950 return ret;
2951 }
2952 return 0;
2953 }
2954
2955 int kvm_arch_get_registers(CPUState *cs)
2956 {
2957 X86CPU *cpu = X86_CPU(cs);
2958 int ret;
2959
2960 assert(cpu_is_stopped(cs) || qemu_cpu_is_self(cs));
2961
2962 ret = kvm_get_vcpu_events(cpu);
2963 if (ret < 0) {
2964 goto out;
2965 }
2966 /*
2967 * KVM_GET_MPSTATE can modify CS and RIP, call it before
2968 * KVM_GET_REGS and KVM_GET_SREGS.
2969 */
2970 ret = kvm_get_mp_state(cpu);
2971 if (ret < 0) {
2972 goto out;
2973 }
2974 ret = kvm_getput_regs(cpu, 0);
2975 if (ret < 0) {
2976 goto out;
2977 }
2978 ret = kvm_get_xsave(cpu);
2979 if (ret < 0) {
2980 goto out;
2981 }
2982 ret = kvm_get_xcrs(cpu);
2983 if (ret < 0) {
2984 goto out;
2985 }
2986 ret = kvm_get_sregs(cpu);
2987 if (ret < 0) {
2988 goto out;
2989 }
2990 ret = kvm_get_msrs(cpu);
2991 if (ret < 0) {
2992 goto out;
2993 }
2994 ret = kvm_get_apic(cpu);
2995 if (ret < 0) {
2996 goto out;
2997 }
2998 ret = kvm_get_debugregs(cpu);
2999 if (ret < 0) {
3000 goto out;
3001 }
3002 ret = 0;
3003 out:
3004 cpu_sync_bndcs_hflags(&cpu->env);
3005 return ret;
3006 }
3007
3008 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
3009 {
3010 X86CPU *x86_cpu = X86_CPU(cpu);
3011 CPUX86State *env = &x86_cpu->env;
3012 int ret;
3013
3014 /* Inject NMI */
3015 if (cpu->interrupt_request & (CPU_INTERRUPT_NMI | CPU_INTERRUPT_SMI)) {
3016 if (cpu->interrupt_request & CPU_INTERRUPT_NMI) {
3017 qemu_mutex_lock_iothread();
3018 cpu->interrupt_request &= ~CPU_INTERRUPT_NMI;
3019 qemu_mutex_unlock_iothread();
3020 DPRINTF("injected NMI\n");
3021 ret = kvm_vcpu_ioctl(cpu, KVM_NMI);
3022 if (ret < 0) {
3023 fprintf(stderr, "KVM: injection failed, NMI lost (%s)\n",
3024 strerror(-ret));
3025 }
3026 }
3027 if (cpu->interrupt_request & CPU_INTERRUPT_SMI) {
3028 qemu_mutex_lock_iothread();
3029 cpu->interrupt_request &= ~CPU_INTERRUPT_SMI;
3030 qemu_mutex_unlock_iothread();
3031 DPRINTF("injected SMI\n");
3032 ret = kvm_vcpu_ioctl(cpu, KVM_SMI);
3033 if (ret < 0) {
3034 fprintf(stderr, "KVM: injection failed, SMI lost (%s)\n",
3035 strerror(-ret));
3036 }
3037 }
3038 }
3039
3040 if (!kvm_pic_in_kernel()) {
3041 qemu_mutex_lock_iothread();
3042 }
3043
3044 /* Force the VCPU out of its inner loop to process any INIT requests
3045 * or (for userspace APIC, but it is cheap to combine the checks here)
3046 * pending TPR access reports.
3047 */
3048 if (cpu->interrupt_request & (CPU_INTERRUPT_INIT | CPU_INTERRUPT_TPR)) {
3049 if ((cpu->interrupt_request & CPU_INTERRUPT_INIT) &&
3050 !(env->hflags & HF_SMM_MASK)) {
3051 cpu->exit_request = 1;
3052 }
3053 if (cpu->interrupt_request & CPU_INTERRUPT_TPR) {
3054 cpu->exit_request = 1;
3055 }
3056 }
3057
3058 if (!kvm_pic_in_kernel()) {
3059 /* Try to inject an interrupt if the guest can accept it */
3060 if (run->ready_for_interrupt_injection &&
3061 (cpu->interrupt_request & CPU_INTERRUPT_HARD) &&
3062 (env->eflags & IF_MASK)) {
3063 int irq;
3064
3065 cpu->interrupt_request &= ~CPU_INTERRUPT_HARD;
3066 irq = cpu_get_pic_interrupt(env);
3067 if (irq >= 0) {
3068 struct kvm_interrupt intr;
3069
3070 intr.irq = irq;
3071 DPRINTF("injected interrupt %d\n", irq);
3072 ret = kvm_vcpu_ioctl(cpu, KVM_INTERRUPT, &intr);
3073 if (ret < 0) {
3074 fprintf(stderr,
3075 "KVM: injection failed, interrupt lost (%s)\n",
3076 strerror(-ret));
3077 }
3078 }
3079 }
3080
3081 /* If we have an interrupt but the guest is not ready to receive an
3082 * interrupt, request an interrupt window exit. This will
3083 * cause a return to userspace as soon as the guest is ready to
3084 * receive interrupts. */
3085 if ((cpu->interrupt_request & CPU_INTERRUPT_HARD)) {
3086 run->request_interrupt_window = 1;
3087 } else {
3088 run->request_interrupt_window = 0;
3089 }
3090
3091 DPRINTF("setting tpr\n");
3092 run->cr8 = cpu_get_apic_tpr(x86_cpu->apic_state);
3093
3094 qemu_mutex_unlock_iothread();
3095 }
3096 }
3097
3098 MemTxAttrs kvm_arch_post_run(CPUState *cpu, struct kvm_run *run)
3099 {
3100 X86CPU *x86_cpu = X86_CPU(cpu);
3101 CPUX86State *env = &x86_cpu->env;
3102
3103 if (run->flags & KVM_RUN_X86_SMM) {
3104 env->hflags |= HF_SMM_MASK;
3105 } else {
3106 env->hflags &= ~HF_SMM_MASK;
3107 }
3108 if (run->if_flag) {
3109 env->eflags |= IF_MASK;
3110 } else {
3111 env->eflags &= ~IF_MASK;
3112 }
3113
3114 /* We need to protect the apic state against concurrent accesses from
3115 * different threads in case the userspace irqchip is used. */
3116 if (!kvm_irqchip_in_kernel()) {
3117 qemu_mutex_lock_iothread();
3118 }
3119 cpu_set_apic_tpr(x86_cpu->apic_state, run->cr8);
3120 cpu_set_apic_base(x86_cpu->apic_state, run->apic_base);
3121 if (!kvm_irqchip_in_kernel()) {
3122 qemu_mutex_unlock_iothread();
3123 }
3124 return cpu_get_mem_attrs(env);
3125 }
3126
3127 int kvm_arch_process_async_events(CPUState *cs)
3128 {
3129 X86CPU *cpu = X86_CPU(cs);
3130 CPUX86State *env = &cpu->env;
3131
3132 if (cs->interrupt_request & CPU_INTERRUPT_MCE) {
3133 /* We must not raise CPU_INTERRUPT_MCE if it's not supported. */
3134 assert(env->mcg_cap);
3135
3136 cs->interrupt_request &= ~CPU_INTERRUPT_MCE;
3137
3138 kvm_cpu_synchronize_state(cs);
3139
3140 if (env->exception_injected == EXCP08_DBLE) {
3141 /* this means triple fault */
3142 qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET);
3143 cs->exit_request = 1;
3144 return 0;
3145 }
3146 env->exception_injected = EXCP12_MCHK;
3147 env->has_error_code = 0;
3148
3149 cs->halted = 0;
3150 if (kvm_irqchip_in_kernel() && env->mp_state == KVM_MP_STATE_HALTED) {
3151 env->mp_state = KVM_MP_STATE_RUNNABLE;
3152 }
3153 }
3154
3155 if ((cs->interrupt_request & CPU_INTERRUPT_INIT) &&
3156 !(env->hflags & HF_SMM_MASK)) {
3157 kvm_cpu_synchronize_state(cs);
3158 do_cpu_init(cpu);
3159 }
3160
3161 if (kvm_irqchip_in_kernel()) {
3162 return 0;
3163 }
3164
3165 if (cs->interrupt_request & CPU_INTERRUPT_POLL) {
3166 cs->interrupt_request &= ~CPU_INTERRUPT_POLL;
3167 apic_poll_irq(cpu->apic_state);
3168 }
3169 if (((cs->interrupt_request & CPU_INTERRUPT_HARD) &&
3170 (env->eflags & IF_MASK)) ||
3171 (cs->interrupt_request & CPU_INTERRUPT_NMI)) {
3172 cs->halted = 0;
3173 }
3174 if (cs->interrupt_request & CPU_INTERRUPT_SIPI) {
3175 kvm_cpu_synchronize_state(cs);
3176 do_cpu_sipi(cpu);
3177 }
3178 if (cs->interrupt_request & CPU_INTERRUPT_TPR) {
3179 cs->interrupt_request &= ~CPU_INTERRUPT_TPR;
3180 kvm_cpu_synchronize_state(cs);
3181 apic_handle_tpr_access_report(cpu->apic_state, env->eip,
3182 env->tpr_access_type);
3183 }
3184
3185 return cs->halted;
3186 }
3187
3188 static int kvm_handle_halt(X86CPU *cpu)
3189 {
3190 CPUState *cs = CPU(cpu);
3191 CPUX86State *env = &cpu->env;
3192
3193 if (!((cs->interrupt_request & CPU_INTERRUPT_HARD) &&
3194 (env->eflags & IF_MASK)) &&
3195 !(cs->interrupt_request & CPU_INTERRUPT_NMI)) {
3196 cs->halted = 1;
3197 return EXCP_HLT;
3198 }
3199
3200 return 0;
3201 }
3202
3203 static int kvm_handle_tpr_access(X86CPU *cpu)
3204 {
3205 CPUState *cs = CPU(cpu);
3206 struct kvm_run *run = cs->kvm_run;
3207
3208 apic_handle_tpr_access_report(cpu->apic_state, run->tpr_access.rip,
3209 run->tpr_access.is_write ? TPR_ACCESS_WRITE
3210 : TPR_ACCESS_READ);
3211 return 1;
3212 }
3213
3214 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
3215 {
3216 static const uint8_t int3 = 0xcc;
3217
3218 if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 1, 0) ||
3219 cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&int3, 1, 1)) {
3220 return -EINVAL;
3221 }
3222 return 0;
3223 }
3224
3225 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
3226 {
3227 uint8_t int3;
3228
3229 if (cpu_memory_rw_debug(cs, bp->pc, &int3, 1, 0) || int3 != 0xcc ||
3230 cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 1, 1)) {
3231 return -EINVAL;
3232 }
3233 return 0;
3234 }
3235
3236 static struct {
3237 target_ulong addr;
3238 int len;
3239 int type;
3240 } hw_breakpoint[4];
3241
3242 static int nb_hw_breakpoint;
3243
3244 static int find_hw_breakpoint(target_ulong addr, int len, int type)
3245 {
3246 int n;
3247
3248 for (n = 0; n < nb_hw_breakpoint; n++) {
3249 if (hw_breakpoint[n].addr == addr && hw_breakpoint[n].type == type &&
3250 (hw_breakpoint[n].len == len || len == -1)) {
3251 return n;
3252 }
3253 }
3254 return -1;
3255 }
3256
3257 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
3258 target_ulong len, int type)
3259 {
3260 switch (type) {
3261 case GDB_BREAKPOINT_HW:
3262 len = 1;
3263 break;
3264 case GDB_WATCHPOINT_WRITE:
3265 case GDB_WATCHPOINT_ACCESS:
3266 switch (len) {
3267 case 1:
3268 break;
3269 case 2:
3270 case 4:
3271 case 8:
3272 if (addr & (len - 1)) {
3273 return -EINVAL;
3274 }
3275 break;
3276 default:
3277 return -EINVAL;
3278 }
3279 break;
3280 default:
3281 return -ENOSYS;
3282 }
3283
3284 if (nb_hw_breakpoint == 4) {
3285 return -ENOBUFS;
3286 }
3287 if (find_hw_breakpoint(addr, len, type) >= 0) {
3288 return -EEXIST;
3289 }
3290 hw_breakpoint[nb_hw_breakpoint].addr = addr;
3291 hw_breakpoint[nb_hw_breakpoint].len = len;
3292 hw_breakpoint[nb_hw_breakpoint].type = type;
3293 nb_hw_breakpoint++;
3294
3295 return 0;
3296 }
3297
3298 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
3299 target_ulong len, int type)
3300 {
3301 int n;
3302
3303 n = find_hw_breakpoint(addr, (type == GDB_BREAKPOINT_HW) ? 1 : len, type);
3304 if (n < 0) {
3305 return -ENOENT;
3306 }
3307 nb_hw_breakpoint--;
3308 hw_breakpoint[n] = hw_breakpoint[nb_hw_breakpoint];
3309
3310 return 0;
3311 }
3312
3313 void kvm_arch_remove_all_hw_breakpoints(void)
3314 {
3315 nb_hw_breakpoint = 0;
3316 }
3317
3318 static CPUWatchpoint hw_watchpoint;
3319
3320 static int kvm_handle_debug(X86CPU *cpu,
3321 struct kvm_debug_exit_arch *arch_info)
3322 {
3323 CPUState *cs = CPU(cpu);
3324 CPUX86State *env = &cpu->env;
3325 int ret = 0;
3326 int n;
3327
3328 if (arch_info->exception == 1) {
3329 if (arch_info->dr6 & (1 << 14)) {
3330 if (cs->singlestep_enabled) {
3331 ret = EXCP_DEBUG;
3332 }
3333 } else {
3334 for (n = 0; n < 4; n++) {
3335 if (arch_info->dr6 & (1 << n)) {
3336 switch ((arch_info->dr7 >> (16 + n*4)) & 0x3) {
3337 case 0x0:
3338 ret = EXCP_DEBUG;
3339 break;
3340 case 0x1:
3341 ret = EXCP_DEBUG;
3342 cs->watchpoint_hit = &hw_watchpoint;
3343 hw_watchpoint.vaddr = hw_breakpoint[n].addr;
3344 hw_watchpoint.flags = BP_MEM_WRITE;
3345 break;
3346 case 0x3:
3347 ret = EXCP_DEBUG;
3348 cs->watchpoint_hit = &hw_watchpoint;
3349 hw_watchpoint.vaddr = hw_breakpoint[n].addr;
3350 hw_watchpoint.flags = BP_MEM_ACCESS;
3351 break;
3352 }
3353 }
3354 }
3355 }
3356 } else if (kvm_find_sw_breakpoint(cs, arch_info->pc)) {
3357 ret = EXCP_DEBUG;
3358 }
3359 if (ret == 0) {
3360 cpu_synchronize_state(cs);
3361 assert(env->exception_injected == -1);
3362
3363 /* pass to guest */
3364 env->exception_injected = arch_info->exception;
3365 env->has_error_code = 0;
3366 }
3367
3368 return ret;
3369 }
3370
3371 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
3372 {
3373 const uint8_t type_code[] = {
3374 [GDB_BREAKPOINT_HW] = 0x0,
3375 [GDB_WATCHPOINT_WRITE] = 0x1,
3376 [GDB_WATCHPOINT_ACCESS] = 0x3
3377 };
3378 const uint8_t len_code[] = {
3379 [1] = 0x0, [2] = 0x1, [4] = 0x3, [8] = 0x2
3380 };
3381 int n;
3382
3383 if (kvm_sw_breakpoints_active(cpu)) {
3384 dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
3385 }
3386 if (nb_hw_breakpoint > 0) {
3387 dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
3388 dbg->arch.debugreg[7] = 0x0600;
3389 for (n = 0; n < nb_hw_breakpoint; n++) {
3390 dbg->arch.debugreg[n] = hw_breakpoint[n].addr;
3391 dbg->arch.debugreg[7] |= (2 << (n * 2)) |
3392 (type_code[hw_breakpoint[n].type] << (16 + n*4)) |
3393 ((uint32_t)len_code[hw_breakpoint[n].len] << (18 + n*4));
3394 }
3395 }
3396 }
3397
3398 static bool host_supports_vmx(void)
3399 {
3400 uint32_t ecx, unused;
3401
3402 host_cpuid(1, 0, &unused, &unused, &ecx, &unused);
3403 return ecx & CPUID_EXT_VMX;
3404 }
3405
3406 #define VMX_INVALID_GUEST_STATE 0x80000021
3407
3408 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
3409 {
3410 X86CPU *cpu = X86_CPU(cs);
3411 uint64_t code;
3412 int ret;
3413
3414 switch (run->exit_reason) {
3415 case KVM_EXIT_HLT:
3416 DPRINTF("handle_hlt\n");
3417 qemu_mutex_lock_iothread();
3418 ret = kvm_handle_halt(cpu);
3419 qemu_mutex_unlock_iothread();
3420 break;
3421 case KVM_EXIT_SET_TPR:
3422 ret = 0;
3423 break;
3424 case KVM_EXIT_TPR_ACCESS:
3425 qemu_mutex_lock_iothread();
3426 ret = kvm_handle_tpr_access(cpu);
3427 qemu_mutex_unlock_iothread();
3428 break;
3429 case KVM_EXIT_FAIL_ENTRY:
3430 code = run->fail_entry.hardware_entry_failure_reason;
3431 fprintf(stderr, "KVM: entry failed, hardware error 0x%" PRIx64 "\n",
3432 code);
3433 if (host_supports_vmx() && code == VMX_INVALID_GUEST_STATE) {
3434 fprintf(stderr,
3435 "\nIf you're running a guest on an Intel machine without "
3436 "unrestricted mode\n"
3437 "support, the failure can be most likely due to the guest "
3438 "entering an invalid\n"
3439 "state for Intel VT. For example, the guest maybe running "
3440 "in big real mode\n"
3441 "which is not supported on less recent Intel processors."
3442 "\n\n");
3443 }
3444 ret = -1;
3445 break;
3446 case KVM_EXIT_EXCEPTION:
3447 fprintf(stderr, "KVM: exception %d exit (error code 0x%x)\n",
3448 run->ex.exception, run->ex.error_code);
3449 ret = -1;
3450 break;
3451 case KVM_EXIT_DEBUG:
3452 DPRINTF("kvm_exit_debug\n");
3453 qemu_mutex_lock_iothread();
3454 ret = kvm_handle_debug(cpu, &run->debug.arch);
3455 qemu_mutex_unlock_iothread();
3456 break;
3457 case KVM_EXIT_HYPERV:
3458 ret = kvm_hv_handle_exit(cpu, &run->hyperv);
3459 break;
3460 case KVM_EXIT_IOAPIC_EOI:
3461 ioapic_eoi_broadcast(run->eoi.vector);
3462 ret = 0;
3463 break;
3464 default:
3465 fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
3466 ret = -1;
3467 break;
3468 }
3469
3470 return ret;
3471 }
3472
3473 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
3474 {
3475 X86CPU *cpu = X86_CPU(cs);
3476 CPUX86State *env = &cpu->env;
3477
3478 kvm_cpu_synchronize_state(cs);
3479 return !(env->cr[0] & CR0_PE_MASK) ||
3480 ((env->segs[R_CS].selector & 3) != 3);
3481 }
3482
3483 void kvm_arch_init_irq_routing(KVMState *s)
3484 {
3485 if (!kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
3486 /* If kernel can't do irq routing, interrupt source
3487 * override 0->2 cannot be set up as required by HPET.
3488 * So we have to disable it.
3489 */
3490 no_hpet = 1;
3491 }
3492 /* We know at this point that we're using the in-kernel
3493 * irqchip, so we can use irqfds, and on x86 we know
3494 * we can use msi via irqfd and GSI routing.
3495 */
3496 kvm_msi_via_irqfd_allowed = true;
3497 kvm_gsi_routing_allowed = true;
3498
3499 if (kvm_irqchip_is_split()) {
3500 int i;
3501
3502 /* If the ioapic is in QEMU and the lapics are in KVM, reserve
3503 MSI routes for signaling interrupts to the local apics. */
3504 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
3505 if (kvm_irqchip_add_msi_route(s, 0, NULL) < 0) {
3506 error_report("Could not enable split IRQ mode.");
3507 exit(1);
3508 }
3509 }
3510 }
3511 }
3512
3513 int kvm_arch_irqchip_create(MachineState *ms, KVMState *s)
3514 {
3515 int ret;
3516 if (machine_kernel_irqchip_split(ms)) {
3517 ret = kvm_vm_enable_cap(s, KVM_CAP_SPLIT_IRQCHIP, 0, 24);
3518 if (ret) {
3519 error_report("Could not enable split irqchip mode: %s",
3520 strerror(-ret));
3521 exit(1);
3522 } else {
3523 DPRINTF("Enabled KVM_CAP_SPLIT_IRQCHIP\n");
3524 kvm_split_irqchip = true;
3525 return 1;
3526 }
3527 } else {
3528 return 0;
3529 }
3530 }
3531
3532 /* Classic KVM device assignment interface. Will remain x86 only. */
3533 int kvm_device_pci_assign(KVMState *s, PCIHostDeviceAddress *dev_addr,
3534 uint32_t flags, uint32_t *dev_id)
3535 {
3536 struct kvm_assigned_pci_dev dev_data = {
3537 .segnr = dev_addr->domain,
3538 .busnr = dev_addr->bus,
3539 .devfn = PCI_DEVFN(dev_addr->slot, dev_addr->function),
3540 .flags = flags,
3541 };
3542 int ret;
3543
3544 dev_data.assigned_dev_id =
3545 (dev_addr->domain << 16) | (dev_addr->bus << 8) | dev_data.devfn;
3546
3547 ret = kvm_vm_ioctl(s, KVM_ASSIGN_PCI_DEVICE, &dev_data);
3548 if (ret < 0) {
3549 return ret;
3550 }
3551
3552 *dev_id = dev_data.assigned_dev_id;
3553
3554 return 0;
3555 }
3556
3557 int kvm_device_pci_deassign(KVMState *s, uint32_t dev_id)
3558 {
3559 struct kvm_assigned_pci_dev dev_data = {
3560 .assigned_dev_id = dev_id,
3561 };
3562
3563 return kvm_vm_ioctl(s, KVM_DEASSIGN_PCI_DEVICE, &dev_data);
3564 }
3565
3566 static int kvm_assign_irq_internal(KVMState *s, uint32_t dev_id,
3567 uint32_t irq_type, uint32_t guest_irq)
3568 {
3569 struct kvm_assigned_irq assigned_irq = {
3570 .assigned_dev_id = dev_id,
3571 .guest_irq = guest_irq,
3572 .flags = irq_type,
3573 };
3574
3575 if (kvm_check_extension(s, KVM_CAP_ASSIGN_DEV_IRQ)) {
3576 return kvm_vm_ioctl(s, KVM_ASSIGN_DEV_IRQ, &assigned_irq);
3577 } else {
3578 return kvm_vm_ioctl(s, KVM_ASSIGN_IRQ, &assigned_irq);
3579 }
3580 }
3581
3582 int kvm_device_intx_assign(KVMState *s, uint32_t dev_id, bool use_host_msi,
3583 uint32_t guest_irq)
3584 {
3585 uint32_t irq_type = KVM_DEV_IRQ_GUEST_INTX |
3586 (use_host_msi ? KVM_DEV_IRQ_HOST_MSI : KVM_DEV_IRQ_HOST_INTX);
3587
3588 return kvm_assign_irq_internal(s, dev_id, irq_type, guest_irq);
3589 }
3590
3591 int kvm_device_intx_set_mask(KVMState *s, uint32_t dev_id, bool masked)
3592 {
3593 struct kvm_assigned_pci_dev dev_data = {
3594 .assigned_dev_id = dev_id,
3595 .flags = masked ? KVM_DEV_ASSIGN_MASK_INTX : 0,
3596 };
3597
3598 return kvm_vm_ioctl(s, KVM_ASSIGN_SET_INTX_MASK, &dev_data);
3599 }
3600
3601 static int kvm_deassign_irq_internal(KVMState *s, uint32_t dev_id,
3602 uint32_t type)
3603 {
3604 struct kvm_assigned_irq assigned_irq = {
3605 .assigned_dev_id = dev_id,
3606 .flags = type,
3607 };
3608
3609 return kvm_vm_ioctl(s, KVM_DEASSIGN_DEV_IRQ, &assigned_irq);
3610 }
3611
3612 int kvm_device_intx_deassign(KVMState *s, uint32_t dev_id, bool use_host_msi)
3613 {
3614 return kvm_deassign_irq_internal(s, dev_id, KVM_DEV_IRQ_GUEST_INTX |
3615 (use_host_msi ? KVM_DEV_IRQ_HOST_MSI : KVM_DEV_IRQ_HOST_INTX));
3616 }
3617
3618 int kvm_device_msi_assign(KVMState *s, uint32_t dev_id, int virq)
3619 {
3620 return kvm_assign_irq_internal(s, dev_id, KVM_DEV_IRQ_HOST_MSI |
3621 KVM_DEV_IRQ_GUEST_MSI, virq);
3622 }
3623
3624 int kvm_device_msi_deassign(KVMState *s, uint32_t dev_id)
3625 {
3626 return kvm_deassign_irq_internal(s, dev_id, KVM_DEV_IRQ_GUEST_MSI |
3627 KVM_DEV_IRQ_HOST_MSI);
3628 }
3629
3630 bool kvm_device_msix_supported(KVMState *s)
3631 {
3632 /* The kernel lacks a corresponding KVM_CAP, so we probe by calling
3633 * KVM_ASSIGN_SET_MSIX_NR with an invalid parameter. */
3634 return kvm_vm_ioctl(s, KVM_ASSIGN_SET_MSIX_NR, NULL) == -EFAULT;
3635 }
3636
3637 int kvm_device_msix_init_vectors(KVMState *s, uint32_t dev_id,
3638 uint32_t nr_vectors)
3639 {
3640 struct kvm_assigned_msix_nr msix_nr = {
3641 .assigned_dev_id = dev_id,
3642 .entry_nr = nr_vectors,
3643 };
3644
3645 return kvm_vm_ioctl(s, KVM_ASSIGN_SET_MSIX_NR, &msix_nr);
3646 }
3647
3648 int kvm_device_msix_set_vector(KVMState *s, uint32_t dev_id, uint32_t vector,
3649 int virq)
3650 {
3651 struct kvm_assigned_msix_entry msix_entry = {
3652 .assigned_dev_id = dev_id,
3653 .gsi = virq,
3654 .entry = vector,
3655 };
3656
3657 return kvm_vm_ioctl(s, KVM_ASSIGN_SET_MSIX_ENTRY, &msix_entry);
3658 }
3659
3660 int kvm_device_msix_assign(KVMState *s, uint32_t dev_id)
3661 {
3662 return kvm_assign_irq_internal(s, dev_id, KVM_DEV_IRQ_HOST_MSIX |
3663 KVM_DEV_IRQ_GUEST_MSIX, 0);
3664 }
3665
3666 int kvm_device_msix_deassign(KVMState *s, uint32_t dev_id)
3667 {
3668 return kvm_deassign_irq_internal(s, dev_id, KVM_DEV_IRQ_GUEST_MSIX |
3669 KVM_DEV_IRQ_HOST_MSIX);
3670 }
3671
3672 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
3673 uint64_t address, uint32_t data, PCIDevice *dev)
3674 {
3675 X86IOMMUState *iommu = x86_iommu_get_default();
3676
3677 if (iommu) {
3678 int ret;
3679 MSIMessage src, dst;
3680 X86IOMMUClass *class = X86_IOMMU_GET_CLASS(iommu);
3681
3682 src.address = route->u.msi.address_hi;
3683 src.address <<= VTD_MSI_ADDR_HI_SHIFT;
3684 src.address |= route->u.msi.address_lo;
3685 src.data = route->u.msi.data;
3686
3687 ret = class->int_remap(iommu, &src, &dst, dev ? \
3688 pci_requester_id(dev) : \
3689 X86_IOMMU_SID_INVALID);
3690 if (ret) {
3691 trace_kvm_x86_fixup_msi_error(route->gsi);
3692 return 1;
3693 }
3694
3695 route->u.msi.address_hi = dst.address >> VTD_MSI_ADDR_HI_SHIFT;
3696 route->u.msi.address_lo = dst.address & VTD_MSI_ADDR_LO_MASK;
3697 route->u.msi.data = dst.data;
3698 }
3699
3700 return 0;
3701 }
3702
3703 typedef struct MSIRouteEntry MSIRouteEntry;
3704
3705 struct MSIRouteEntry {
3706 PCIDevice *dev; /* Device pointer */
3707 int vector; /* MSI/MSIX vector index */
3708 int virq; /* Virtual IRQ index */
3709 QLIST_ENTRY(MSIRouteEntry) list;
3710 };
3711
3712 /* List of used GSI routes */
3713 static QLIST_HEAD(, MSIRouteEntry) msi_route_list = \
3714 QLIST_HEAD_INITIALIZER(msi_route_list);
3715
3716 static void kvm_update_msi_routes_all(void *private, bool global,
3717 uint32_t index, uint32_t mask)
3718 {
3719 int cnt = 0;
3720 MSIRouteEntry *entry;
3721 MSIMessage msg;
3722 PCIDevice *dev;
3723
3724 /* TODO: explicit route update */
3725 QLIST_FOREACH(entry, &msi_route_list, list) {
3726 cnt++;
3727 dev = entry->dev;
3728 if (!msix_enabled(dev) && !msi_enabled(dev)) {
3729 continue;
3730 }
3731 msg = pci_get_msi_message(dev, entry->vector);
3732 kvm_irqchip_update_msi_route(kvm_state, entry->virq, msg, dev);
3733 }
3734 kvm_irqchip_commit_routes(kvm_state);
3735 trace_kvm_x86_update_msi_routes(cnt);
3736 }
3737
3738 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
3739 int vector, PCIDevice *dev)
3740 {
3741 static bool notify_list_inited = false;
3742 MSIRouteEntry *entry;
3743
3744 if (!dev) {
3745 /* These are (possibly) IOAPIC routes only used for split
3746 * kernel irqchip mode, while what we are housekeeping are
3747 * PCI devices only. */
3748 return 0;
3749 }
3750
3751 entry = g_new0(MSIRouteEntry, 1);
3752 entry->dev = dev;
3753 entry->vector = vector;
3754 entry->virq = route->gsi;
3755 QLIST_INSERT_HEAD(&msi_route_list, entry, list);
3756
3757 trace_kvm_x86_add_msi_route(route->gsi);
3758
3759 if (!notify_list_inited) {
3760 /* For the first time we do add route, add ourselves into
3761 * IOMMU's IEC notify list if needed. */
3762 X86IOMMUState *iommu = x86_iommu_get_default();
3763 if (iommu) {
3764 x86_iommu_iec_register_notifier(iommu,
3765 kvm_update_msi_routes_all,
3766 NULL);
3767 }
3768 notify_list_inited = true;
3769 }
3770 return 0;
3771 }
3772
3773 int kvm_arch_release_virq_post(int virq)
3774 {
3775 MSIRouteEntry *entry, *next;
3776 QLIST_FOREACH_SAFE(entry, &msi_route_list, list, next) {
3777 if (entry->virq == virq) {
3778 trace_kvm_x86_remove_msi_route(virq);
3779 QLIST_REMOVE(entry, list);
3780 g_free(entry);
3781 break;
3782 }
3783 }
3784 return 0;
3785 }
3786
3787 int kvm_arch_msi_data_to_gsi(uint32_t data)
3788 {
3789 abort();
3790 }
AR7 emulation for QEMU