4 * Copyright (C) 2006-2008 Qumranet Technologies
5 * Copyright IBM, Corp. 2008
8 * Anthony Liguori <aliguori@us.ibm.com>
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.
15 #include <sys/types.h>
16 #include <sys/ioctl.h>
18 #include <sys/utsname.h>
20 #include <linux/kvm.h>
21 #include <linux/kvm_para.h>
23 #include "qemu-common.h"
28 #include "host-utils.h"
36 #define DPRINTF(fmt, ...) \
37 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
39 #define DPRINTF(fmt, ...) \
43 #define MSR_KVM_WALL_CLOCK 0x11
44 #define MSR_KVM_SYSTEM_TIME 0x12
47 #define BUS_MCEERR_AR 4
50 #define BUS_MCEERR_AO 5
53 const KVMCapabilityInfo kvm_arch_required_capabilities
[] = {
54 KVM_CAP_INFO(SET_TSS_ADDR
),
55 KVM_CAP_INFO(EXT_CPUID
),
56 KVM_CAP_INFO(MP_STATE
),
60 static bool has_msr_star
;
61 static bool has_msr_hsave_pa
;
62 static bool has_msr_async_pf_en
;
63 static int lm_capable_kernel
;
65 static struct kvm_cpuid2
*try_get_cpuid(KVMState
*s
, int max
)
67 struct kvm_cpuid2
*cpuid
;
70 size
= sizeof(*cpuid
) + max
* sizeof(*cpuid
->entries
);
71 cpuid
= (struct kvm_cpuid2
*)g_malloc0(size
);
73 r
= kvm_ioctl(s
, KVM_GET_SUPPORTED_CPUID
, cpuid
);
74 if (r
== 0 && cpuid
->nent
>= max
) {
82 fprintf(stderr
, "KVM_GET_SUPPORTED_CPUID failed: %s\n",
90 struct kvm_para_features
{
94 { KVM_CAP_CLOCKSOURCE
, KVM_FEATURE_CLOCKSOURCE
},
95 { KVM_CAP_NOP_IO_DELAY
, KVM_FEATURE_NOP_IO_DELAY
},
96 { KVM_CAP_PV_MMU
, KVM_FEATURE_MMU_OP
},
97 { KVM_CAP_ASYNC_PF
, KVM_FEATURE_ASYNC_PF
},
101 static int get_para_features(KVMState
*s
)
105 for (i
= 0; i
< ARRAY_SIZE(para_features
) - 1; i
++) {
106 if (kvm_check_extension(s
, para_features
[i
].cap
)) {
107 features
|= (1 << para_features
[i
].feature
);
115 uint32_t kvm_arch_get_supported_cpuid(KVMState
*s
, uint32_t function
,
116 uint32_t index
, int reg
)
118 struct kvm_cpuid2
*cpuid
;
121 uint32_t cpuid_1_edx
;
122 int has_kvm_features
= 0;
125 while ((cpuid
= try_get_cpuid(s
, max
)) == NULL
) {
129 for (i
= 0; i
< cpuid
->nent
; ++i
) {
130 if (cpuid
->entries
[i
].function
== function
&&
131 cpuid
->entries
[i
].index
== index
) {
132 if (cpuid
->entries
[i
].function
== KVM_CPUID_FEATURES
) {
133 has_kvm_features
= 1;
137 ret
= cpuid
->entries
[i
].eax
;
140 ret
= cpuid
->entries
[i
].ebx
;
143 ret
= cpuid
->entries
[i
].ecx
;
146 ret
= cpuid
->entries
[i
].edx
;
149 /* KVM before 2.6.30 misreports the following features */
150 ret
|= CPUID_MTRR
| CPUID_PAT
| CPUID_MCE
| CPUID_MCA
;
153 /* On Intel, kvm returns cpuid according to the Intel spec,
154 * so add missing bits according to the AMD spec:
156 cpuid_1_edx
= kvm_arch_get_supported_cpuid(s
, 1, 0, R_EDX
);
157 ret
|= cpuid_1_edx
& 0x183f7ff;
167 /* fallback for older kernels */
168 if (!has_kvm_features
&& (function
== KVM_CPUID_FEATURES
)) {
169 ret
= get_para_features(s
);
175 typedef struct HWPoisonPage
{
177 QLIST_ENTRY(HWPoisonPage
) list
;
180 static QLIST_HEAD(, HWPoisonPage
) hwpoison_page_list
=
181 QLIST_HEAD_INITIALIZER(hwpoison_page_list
);
183 static void kvm_unpoison_all(void *param
)
185 HWPoisonPage
*page
, *next_page
;
187 QLIST_FOREACH_SAFE(page
, &hwpoison_page_list
, list
, next_page
) {
188 QLIST_REMOVE(page
, list
);
189 qemu_ram_remap(page
->ram_addr
, TARGET_PAGE_SIZE
);
194 static void kvm_hwpoison_page_add(ram_addr_t ram_addr
)
198 QLIST_FOREACH(page
, &hwpoison_page_list
, list
) {
199 if (page
->ram_addr
== ram_addr
) {
203 page
= g_malloc(sizeof(HWPoisonPage
));
204 page
->ram_addr
= ram_addr
;
205 QLIST_INSERT_HEAD(&hwpoison_page_list
, page
, list
);
208 static int kvm_get_mce_cap_supported(KVMState
*s
, uint64_t *mce_cap
,
213 r
= kvm_check_extension(s
, KVM_CAP_MCE
);
216 return kvm_ioctl(s
, KVM_X86_GET_MCE_CAP_SUPPORTED
, mce_cap
);
221 static void kvm_mce_inject(CPUState
*env
, target_phys_addr_t paddr
, int code
)
223 uint64_t status
= MCI_STATUS_VAL
| MCI_STATUS_UC
| MCI_STATUS_EN
|
224 MCI_STATUS_MISCV
| MCI_STATUS_ADDRV
| MCI_STATUS_S
;
225 uint64_t mcg_status
= MCG_STATUS_MCIP
;
227 if (code
== BUS_MCEERR_AR
) {
228 status
|= MCI_STATUS_AR
| 0x134;
229 mcg_status
|= MCG_STATUS_EIPV
;
232 mcg_status
|= MCG_STATUS_RIPV
;
234 cpu_x86_inject_mce(NULL
, env
, 9, status
, mcg_status
, paddr
,
235 (MCM_ADDR_PHYS
<< 6) | 0xc,
236 cpu_x86_support_mca_broadcast(env
) ?
237 MCE_INJECT_BROADCAST
: 0);
240 static void hardware_memory_error(void)
242 fprintf(stderr
, "Hardware memory error!\n");
246 int kvm_arch_on_sigbus_vcpu(CPUState
*env
, int code
, void *addr
)
249 target_phys_addr_t paddr
;
251 if ((env
->mcg_cap
& MCG_SER_P
) && addr
252 && (code
== BUS_MCEERR_AR
|| code
== BUS_MCEERR_AO
)) {
253 if (qemu_ram_addr_from_host(addr
, &ram_addr
) ||
254 !kvm_physical_memory_addr_from_ram(env
->kvm_state
, ram_addr
,
256 fprintf(stderr
, "Hardware memory error for memory used by "
257 "QEMU itself instead of guest system!\n");
258 /* Hope we are lucky for AO MCE */
259 if (code
== BUS_MCEERR_AO
) {
262 hardware_memory_error();
265 kvm_hwpoison_page_add(ram_addr
);
266 kvm_mce_inject(env
, paddr
, code
);
268 if (code
== BUS_MCEERR_AO
) {
270 } else if (code
== BUS_MCEERR_AR
) {
271 hardware_memory_error();
279 int kvm_arch_on_sigbus(int code
, void *addr
)
281 if ((first_cpu
->mcg_cap
& MCG_SER_P
) && addr
&& code
== BUS_MCEERR_AO
) {
283 target_phys_addr_t paddr
;
285 /* Hope we are lucky for AO MCE */
286 if (qemu_ram_addr_from_host(addr
, &ram_addr
) ||
287 !kvm_physical_memory_addr_from_ram(first_cpu
->kvm_state
, ram_addr
,
289 fprintf(stderr
, "Hardware memory error for memory used by "
290 "QEMU itself instead of guest system!: %p\n", addr
);
293 kvm_hwpoison_page_add(ram_addr
);
294 kvm_mce_inject(first_cpu
, paddr
, code
);
296 if (code
== BUS_MCEERR_AO
) {
298 } else if (code
== BUS_MCEERR_AR
) {
299 hardware_memory_error();
307 static int kvm_inject_mce_oldstyle(CPUState
*env
)
309 if (!kvm_has_vcpu_events() && env
->exception_injected
== EXCP12_MCHK
) {
310 unsigned int bank
, bank_num
= env
->mcg_cap
& 0xff;
311 struct kvm_x86_mce mce
;
313 env
->exception_injected
= -1;
316 * There must be at least one bank in use if an MCE is pending.
317 * Find it and use its values for the event injection.
319 for (bank
= 0; bank
< bank_num
; bank
++) {
320 if (env
->mce_banks
[bank
* 4 + 1] & MCI_STATUS_VAL
) {
324 assert(bank
< bank_num
);
327 mce
.status
= env
->mce_banks
[bank
* 4 + 1];
328 mce
.mcg_status
= env
->mcg_status
;
329 mce
.addr
= env
->mce_banks
[bank
* 4 + 2];
330 mce
.misc
= env
->mce_banks
[bank
* 4 + 3];
332 return kvm_vcpu_ioctl(env
, KVM_X86_SET_MCE
, &mce
);
337 static void cpu_update_state(void *opaque
, int running
, RunState state
)
339 CPUState
*env
= opaque
;
342 env
->tsc_valid
= false;
346 int kvm_arch_init_vcpu(CPUState
*env
)
349 struct kvm_cpuid2 cpuid
;
350 struct kvm_cpuid_entry2 entries
[100];
351 } QEMU_PACKED cpuid_data
;
352 KVMState
*s
= env
->kvm_state
;
353 uint32_t limit
, i
, j
, cpuid_i
;
355 struct kvm_cpuid_entry2
*c
;
356 uint32_t signature
[3];
359 env
->cpuid_features
&= kvm_arch_get_supported_cpuid(s
, 1, 0, R_EDX
);
361 i
= env
->cpuid_ext_features
& CPUID_EXT_HYPERVISOR
;
362 env
->cpuid_ext_features
&= kvm_arch_get_supported_cpuid(s
, 1, 0, R_ECX
);
363 env
->cpuid_ext_features
|= i
;
365 env
->cpuid_ext2_features
&= kvm_arch_get_supported_cpuid(s
, 0x80000001,
367 env
->cpuid_ext3_features
&= kvm_arch_get_supported_cpuid(s
, 0x80000001,
369 env
->cpuid_svm_features
&= kvm_arch_get_supported_cpuid(s
, 0x8000000A,
374 /* Paravirtualization CPUIDs */
375 memcpy(signature
, "KVMKVMKVM\0\0\0", 12);
376 c
= &cpuid_data
.entries
[cpuid_i
++];
377 memset(c
, 0, sizeof(*c
));
378 c
->function
= KVM_CPUID_SIGNATURE
;
380 c
->ebx
= signature
[0];
381 c
->ecx
= signature
[1];
382 c
->edx
= signature
[2];
384 c
= &cpuid_data
.entries
[cpuid_i
++];
385 memset(c
, 0, sizeof(*c
));
386 c
->function
= KVM_CPUID_FEATURES
;
387 c
->eax
= env
->cpuid_kvm_features
&
388 kvm_arch_get_supported_cpuid(s
, KVM_CPUID_FEATURES
, 0, R_EAX
);
390 has_msr_async_pf_en
= c
->eax
& (1 << KVM_FEATURE_ASYNC_PF
);
392 cpu_x86_cpuid(env
, 0, 0, &limit
, &unused
, &unused
, &unused
);
394 for (i
= 0; i
<= limit
; i
++) {
395 c
= &cpuid_data
.entries
[cpuid_i
++];
399 /* Keep reading function 2 till all the input is received */
403 c
->flags
= KVM_CPUID_FLAG_STATEFUL_FUNC
|
404 KVM_CPUID_FLAG_STATE_READ_NEXT
;
405 cpu_x86_cpuid(env
, i
, 0, &c
->eax
, &c
->ebx
, &c
->ecx
, &c
->edx
);
406 times
= c
->eax
& 0xff;
408 for (j
= 1; j
< times
; ++j
) {
409 c
= &cpuid_data
.entries
[cpuid_i
++];
411 c
->flags
= KVM_CPUID_FLAG_STATEFUL_FUNC
;
412 cpu_x86_cpuid(env
, i
, 0, &c
->eax
, &c
->ebx
, &c
->ecx
, &c
->edx
);
420 if (i
== 0xd && j
== 64) {
424 c
->flags
= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
426 cpu_x86_cpuid(env
, i
, j
, &c
->eax
, &c
->ebx
, &c
->ecx
, &c
->edx
);
428 if (i
== 4 && c
->eax
== 0) {
431 if (i
== 0xb && !(c
->ecx
& 0xff00)) {
434 if (i
== 0xd && c
->eax
== 0) {
437 c
= &cpuid_data
.entries
[cpuid_i
++];
443 cpu_x86_cpuid(env
, i
, 0, &c
->eax
, &c
->ebx
, &c
->ecx
, &c
->edx
);
447 cpu_x86_cpuid(env
, 0x80000000, 0, &limit
, &unused
, &unused
, &unused
);
449 for (i
= 0x80000000; i
<= limit
; i
++) {
450 c
= &cpuid_data
.entries
[cpuid_i
++];
454 cpu_x86_cpuid(env
, i
, 0, &c
->eax
, &c
->ebx
, &c
->ecx
, &c
->edx
);
457 /* Call Centaur's CPUID instructions they are supported. */
458 if (env
->cpuid_xlevel2
> 0) {
459 env
->cpuid_ext4_features
&=
460 kvm_arch_get_supported_cpuid(s
, 0xC0000001, 0, R_EDX
);
461 cpu_x86_cpuid(env
, 0xC0000000, 0, &limit
, &unused
, &unused
, &unused
);
463 for (i
= 0xC0000000; i
<= limit
; i
++) {
464 c
= &cpuid_data
.entries
[cpuid_i
++];
468 cpu_x86_cpuid(env
, i
, 0, &c
->eax
, &c
->ebx
, &c
->ecx
, &c
->edx
);
472 cpuid_data
.cpuid
.nent
= cpuid_i
;
474 if (((env
->cpuid_version
>> 8)&0xF) >= 6
475 && (env
->cpuid_features
&(CPUID_MCE
|CPUID_MCA
)) == (CPUID_MCE
|CPUID_MCA
)
476 && kvm_check_extension(env
->kvm_state
, KVM_CAP_MCE
) > 0) {
481 ret
= kvm_get_mce_cap_supported(env
->kvm_state
, &mcg_cap
, &banks
);
483 fprintf(stderr
, "kvm_get_mce_cap_supported: %s", strerror(-ret
));
487 if (banks
> MCE_BANKS_DEF
) {
488 banks
= MCE_BANKS_DEF
;
490 mcg_cap
&= MCE_CAP_DEF
;
492 ret
= kvm_vcpu_ioctl(env
, KVM_X86_SETUP_MCE
, &mcg_cap
);
494 fprintf(stderr
, "KVM_X86_SETUP_MCE: %s", strerror(-ret
));
498 env
->mcg_cap
= mcg_cap
;
501 qemu_add_vm_change_state_handler(cpu_update_state
, env
);
503 r
= kvm_vcpu_ioctl(env
, KVM_SET_CPUID2
, &cpuid_data
);
508 r
= kvm_check_extension(env
->kvm_state
, KVM_CAP_TSC_CONTROL
);
509 if (r
&& env
->tsc_khz
) {
510 r
= kvm_vcpu_ioctl(env
, KVM_SET_TSC_KHZ
, env
->tsc_khz
);
512 fprintf(stderr
, "KVM_SET_TSC_KHZ failed\n");
520 void kvm_arch_reset_vcpu(CPUState
*env
)
522 env
->exception_injected
= -1;
523 env
->interrupt_injected
= -1;
525 if (kvm_irqchip_in_kernel()) {
526 env
->mp_state
= cpu_is_bsp(env
) ? KVM_MP_STATE_RUNNABLE
:
527 KVM_MP_STATE_UNINITIALIZED
;
529 env
->mp_state
= KVM_MP_STATE_RUNNABLE
;
533 static int kvm_get_supported_msrs(KVMState
*s
)
535 static int kvm_supported_msrs
;
539 if (kvm_supported_msrs
== 0) {
540 struct kvm_msr_list msr_list
, *kvm_msr_list
;
542 kvm_supported_msrs
= -1;
544 /* Obtain MSR list from KVM. These are the MSRs that we must
547 ret
= kvm_ioctl(s
, KVM_GET_MSR_INDEX_LIST
, &msr_list
);
548 if (ret
< 0 && ret
!= -E2BIG
) {
551 /* Old kernel modules had a bug and could write beyond the provided
552 memory. Allocate at least a safe amount of 1K. */
553 kvm_msr_list
= g_malloc0(MAX(1024, sizeof(msr_list
) +
555 sizeof(msr_list
.indices
[0])));
557 kvm_msr_list
->nmsrs
= msr_list
.nmsrs
;
558 ret
= kvm_ioctl(s
, KVM_GET_MSR_INDEX_LIST
, kvm_msr_list
);
562 for (i
= 0; i
< kvm_msr_list
->nmsrs
; i
++) {
563 if (kvm_msr_list
->indices
[i
] == MSR_STAR
) {
567 if (kvm_msr_list
->indices
[i
] == MSR_VM_HSAVE_PA
) {
568 has_msr_hsave_pa
= true;
574 g_free(kvm_msr_list
);
580 int kvm_arch_init(KVMState
*s
)
582 uint64_t identity_base
= 0xfffbc000;
584 struct utsname utsname
;
586 ret
= kvm_get_supported_msrs(s
);
592 lm_capable_kernel
= strcmp(utsname
.machine
, "x86_64") == 0;
595 * On older Intel CPUs, KVM uses vm86 mode to emulate 16-bit code directly.
596 * In order to use vm86 mode, an EPT identity map and a TSS are needed.
597 * Since these must be part of guest physical memory, we need to allocate
598 * them, both by setting their start addresses in the kernel and by
599 * creating a corresponding e820 entry. We need 4 pages before the BIOS.
601 * Older KVM versions may not support setting the identity map base. In
602 * that case we need to stick with the default, i.e. a 256K maximum BIOS
605 if (kvm_check_extension(s
, KVM_CAP_SET_IDENTITY_MAP_ADDR
)) {
606 /* Allows up to 16M BIOSes. */
607 identity_base
= 0xfeffc000;
609 ret
= kvm_vm_ioctl(s
, KVM_SET_IDENTITY_MAP_ADDR
, &identity_base
);
615 /* Set TSS base one page after EPT identity map. */
616 ret
= kvm_vm_ioctl(s
, KVM_SET_TSS_ADDR
, identity_base
+ 0x1000);
621 /* Tell fw_cfg to notify the BIOS to reserve the range. */
622 ret
= e820_add_entry(identity_base
, 0x4000, E820_RESERVED
);
624 fprintf(stderr
, "e820_add_entry() table is full\n");
627 qemu_register_reset(kvm_unpoison_all
, NULL
);
632 static void set_v8086_seg(struct kvm_segment
*lhs
, const SegmentCache
*rhs
)
634 lhs
->selector
= rhs
->selector
;
635 lhs
->base
= rhs
->base
;
636 lhs
->limit
= rhs
->limit
;
648 static void set_seg(struct kvm_segment
*lhs
, const SegmentCache
*rhs
)
650 unsigned flags
= rhs
->flags
;
651 lhs
->selector
= rhs
->selector
;
652 lhs
->base
= rhs
->base
;
653 lhs
->limit
= rhs
->limit
;
654 lhs
->type
= (flags
>> DESC_TYPE_SHIFT
) & 15;
655 lhs
->present
= (flags
& DESC_P_MASK
) != 0;
656 lhs
->dpl
= (flags
>> DESC_DPL_SHIFT
) & 3;
657 lhs
->db
= (flags
>> DESC_B_SHIFT
) & 1;
658 lhs
->s
= (flags
& DESC_S_MASK
) != 0;
659 lhs
->l
= (flags
>> DESC_L_SHIFT
) & 1;
660 lhs
->g
= (flags
& DESC_G_MASK
) != 0;
661 lhs
->avl
= (flags
& DESC_AVL_MASK
) != 0;
665 static void get_seg(SegmentCache
*lhs
, const struct kvm_segment
*rhs
)
667 lhs
->selector
= rhs
->selector
;
668 lhs
->base
= rhs
->base
;
669 lhs
->limit
= rhs
->limit
;
670 lhs
->flags
= (rhs
->type
<< DESC_TYPE_SHIFT
) |
671 (rhs
->present
* DESC_P_MASK
) |
672 (rhs
->dpl
<< DESC_DPL_SHIFT
) |
673 (rhs
->db
<< DESC_B_SHIFT
) |
674 (rhs
->s
* DESC_S_MASK
) |
675 (rhs
->l
<< DESC_L_SHIFT
) |
676 (rhs
->g
* DESC_G_MASK
) |
677 (rhs
->avl
* DESC_AVL_MASK
);
680 static void kvm_getput_reg(__u64
*kvm_reg
, target_ulong
*qemu_reg
, int set
)
683 *kvm_reg
= *qemu_reg
;
685 *qemu_reg
= *kvm_reg
;
689 static int kvm_getput_regs(CPUState
*env
, int set
)
691 struct kvm_regs regs
;
695 ret
= kvm_vcpu_ioctl(env
, KVM_GET_REGS
, ®s
);
701 kvm_getput_reg(®s
.rax
, &env
->regs
[R_EAX
], set
);
702 kvm_getput_reg(®s
.rbx
, &env
->regs
[R_EBX
], set
);
703 kvm_getput_reg(®s
.rcx
, &env
->regs
[R_ECX
], set
);
704 kvm_getput_reg(®s
.rdx
, &env
->regs
[R_EDX
], set
);
705 kvm_getput_reg(®s
.rsi
, &env
->regs
[R_ESI
], set
);
706 kvm_getput_reg(®s
.rdi
, &env
->regs
[R_EDI
], set
);
707 kvm_getput_reg(®s
.rsp
, &env
->regs
[R_ESP
], set
);
708 kvm_getput_reg(®s
.rbp
, &env
->regs
[R_EBP
], set
);
710 kvm_getput_reg(®s
.r8
, &env
->regs
[8], set
);
711 kvm_getput_reg(®s
.r9
, &env
->regs
[9], set
);
712 kvm_getput_reg(®s
.r10
, &env
->regs
[10], set
);
713 kvm_getput_reg(®s
.r11
, &env
->regs
[11], set
);
714 kvm_getput_reg(®s
.r12
, &env
->regs
[12], set
);
715 kvm_getput_reg(®s
.r13
, &env
->regs
[13], set
);
716 kvm_getput_reg(®s
.r14
, &env
->regs
[14], set
);
717 kvm_getput_reg(®s
.r15
, &env
->regs
[15], set
);
720 kvm_getput_reg(®s
.rflags
, &env
->eflags
, set
);
721 kvm_getput_reg(®s
.rip
, &env
->eip
, set
);
724 ret
= kvm_vcpu_ioctl(env
, KVM_SET_REGS
, ®s
);
730 static int kvm_put_fpu(CPUState
*env
)
735 memset(&fpu
, 0, sizeof fpu
);
736 fpu
.fsw
= env
->fpus
& ~(7 << 11);
737 fpu
.fsw
|= (env
->fpstt
& 7) << 11;
739 fpu
.last_opcode
= env
->fpop
;
740 fpu
.last_ip
= env
->fpip
;
741 fpu
.last_dp
= env
->fpdp
;
742 for (i
= 0; i
< 8; ++i
) {
743 fpu
.ftwx
|= (!env
->fptags
[i
]) << i
;
745 memcpy(fpu
.fpr
, env
->fpregs
, sizeof env
->fpregs
);
746 memcpy(fpu
.xmm
, env
->xmm_regs
, sizeof env
->xmm_regs
);
747 fpu
.mxcsr
= env
->mxcsr
;
749 return kvm_vcpu_ioctl(env
, KVM_SET_FPU
, &fpu
);
752 #define XSAVE_CWD_RIP 2
753 #define XSAVE_CWD_RDP 4
754 #define XSAVE_MXCSR 6
755 #define XSAVE_ST_SPACE 8
756 #define XSAVE_XMM_SPACE 40
757 #define XSAVE_XSTATE_BV 128
758 #define XSAVE_YMMH_SPACE 144
760 static int kvm_put_xsave(CPUState
*env
)
763 struct kvm_xsave
* xsave
;
764 uint16_t cwd
, swd
, twd
;
766 if (!kvm_has_xsave()) {
767 return kvm_put_fpu(env
);
770 xsave
= qemu_memalign(4096, sizeof(struct kvm_xsave
));
771 memset(xsave
, 0, sizeof(struct kvm_xsave
));
773 swd
= env
->fpus
& ~(7 << 11);
774 swd
|= (env
->fpstt
& 7) << 11;
776 for (i
= 0; i
< 8; ++i
) {
777 twd
|= (!env
->fptags
[i
]) << i
;
779 xsave
->region
[0] = (uint32_t)(swd
<< 16) + cwd
;
780 xsave
->region
[1] = (uint32_t)(env
->fpop
<< 16) + twd
;
781 memcpy(&xsave
->region
[XSAVE_CWD_RIP
], &env
->fpip
, sizeof(env
->fpip
));
782 memcpy(&xsave
->region
[XSAVE_CWD_RDP
], &env
->fpdp
, sizeof(env
->fpdp
));
783 memcpy(&xsave
->region
[XSAVE_ST_SPACE
], env
->fpregs
,
785 memcpy(&xsave
->region
[XSAVE_XMM_SPACE
], env
->xmm_regs
,
786 sizeof env
->xmm_regs
);
787 xsave
->region
[XSAVE_MXCSR
] = env
->mxcsr
;
788 *(uint64_t *)&xsave
->region
[XSAVE_XSTATE_BV
] = env
->xstate_bv
;
789 memcpy(&xsave
->region
[XSAVE_YMMH_SPACE
], env
->ymmh_regs
,
790 sizeof env
->ymmh_regs
);
791 r
= kvm_vcpu_ioctl(env
, KVM_SET_XSAVE
, xsave
);
796 static int kvm_put_xcrs(CPUState
*env
)
798 struct kvm_xcrs xcrs
;
800 if (!kvm_has_xcrs()) {
806 xcrs
.xcrs
[0].xcr
= 0;
807 xcrs
.xcrs
[0].value
= env
->xcr0
;
808 return kvm_vcpu_ioctl(env
, KVM_SET_XCRS
, &xcrs
);
811 static int kvm_put_sregs(CPUState
*env
)
813 struct kvm_sregs sregs
;
815 memset(sregs
.interrupt_bitmap
, 0, sizeof(sregs
.interrupt_bitmap
));
816 if (env
->interrupt_injected
>= 0) {
817 sregs
.interrupt_bitmap
[env
->interrupt_injected
/ 64] |=
818 (uint64_t)1 << (env
->interrupt_injected
% 64);
821 if ((env
->eflags
& VM_MASK
)) {
822 set_v8086_seg(&sregs
.cs
, &env
->segs
[R_CS
]);
823 set_v8086_seg(&sregs
.ds
, &env
->segs
[R_DS
]);
824 set_v8086_seg(&sregs
.es
, &env
->segs
[R_ES
]);
825 set_v8086_seg(&sregs
.fs
, &env
->segs
[R_FS
]);
826 set_v8086_seg(&sregs
.gs
, &env
->segs
[R_GS
]);
827 set_v8086_seg(&sregs
.ss
, &env
->segs
[R_SS
]);
829 set_seg(&sregs
.cs
, &env
->segs
[R_CS
]);
830 set_seg(&sregs
.ds
, &env
->segs
[R_DS
]);
831 set_seg(&sregs
.es
, &env
->segs
[R_ES
]);
832 set_seg(&sregs
.fs
, &env
->segs
[R_FS
]);
833 set_seg(&sregs
.gs
, &env
->segs
[R_GS
]);
834 set_seg(&sregs
.ss
, &env
->segs
[R_SS
]);
837 set_seg(&sregs
.tr
, &env
->tr
);
838 set_seg(&sregs
.ldt
, &env
->ldt
);
840 sregs
.idt
.limit
= env
->idt
.limit
;
841 sregs
.idt
.base
= env
->idt
.base
;
842 sregs
.gdt
.limit
= env
->gdt
.limit
;
843 sregs
.gdt
.base
= env
->gdt
.base
;
845 sregs
.cr0
= env
->cr
[0];
846 sregs
.cr2
= env
->cr
[2];
847 sregs
.cr3
= env
->cr
[3];
848 sregs
.cr4
= env
->cr
[4];
850 sregs
.cr8
= cpu_get_apic_tpr(env
->apic_state
);
851 sregs
.apic_base
= cpu_get_apic_base(env
->apic_state
);
853 sregs
.efer
= env
->efer
;
855 return kvm_vcpu_ioctl(env
, KVM_SET_SREGS
, &sregs
);
858 static void kvm_msr_entry_set(struct kvm_msr_entry
*entry
,
859 uint32_t index
, uint64_t value
)
861 entry
->index
= index
;
865 static int kvm_put_msrs(CPUState
*env
, int level
)
868 struct kvm_msrs info
;
869 struct kvm_msr_entry entries
[100];
871 struct kvm_msr_entry
*msrs
= msr_data
.entries
;
874 kvm_msr_entry_set(&msrs
[n
++], MSR_IA32_SYSENTER_CS
, env
->sysenter_cs
);
875 kvm_msr_entry_set(&msrs
[n
++], MSR_IA32_SYSENTER_ESP
, env
->sysenter_esp
);
876 kvm_msr_entry_set(&msrs
[n
++], MSR_IA32_SYSENTER_EIP
, env
->sysenter_eip
);
877 kvm_msr_entry_set(&msrs
[n
++], MSR_PAT
, env
->pat
);
879 kvm_msr_entry_set(&msrs
[n
++], MSR_STAR
, env
->star
);
881 if (has_msr_hsave_pa
) {
882 kvm_msr_entry_set(&msrs
[n
++], MSR_VM_HSAVE_PA
, env
->vm_hsave
);
885 if (lm_capable_kernel
) {
886 kvm_msr_entry_set(&msrs
[n
++], MSR_CSTAR
, env
->cstar
);
887 kvm_msr_entry_set(&msrs
[n
++], MSR_KERNELGSBASE
, env
->kernelgsbase
);
888 kvm_msr_entry_set(&msrs
[n
++], MSR_FMASK
, env
->fmask
);
889 kvm_msr_entry_set(&msrs
[n
++], MSR_LSTAR
, env
->lstar
);
892 if (level
== KVM_PUT_FULL_STATE
) {
894 * KVM is yet unable to synchronize TSC values of multiple VCPUs on
895 * writeback. Until this is fixed, we only write the offset to SMP
896 * guests after migration, desynchronizing the VCPUs, but avoiding
897 * huge jump-backs that would occur without any writeback at all.
899 if (smp_cpus
== 1 || env
->tsc
!= 0) {
900 kvm_msr_entry_set(&msrs
[n
++], MSR_IA32_TSC
, env
->tsc
);
904 * The following paravirtual MSRs have side effects on the guest or are
905 * too heavy for normal writeback. Limit them to reset or full state
908 if (level
>= KVM_PUT_RESET_STATE
) {
909 kvm_msr_entry_set(&msrs
[n
++], MSR_KVM_SYSTEM_TIME
,
910 env
->system_time_msr
);
911 kvm_msr_entry_set(&msrs
[n
++], MSR_KVM_WALL_CLOCK
, env
->wall_clock_msr
);
912 if (has_msr_async_pf_en
) {
913 kvm_msr_entry_set(&msrs
[n
++], MSR_KVM_ASYNC_PF_EN
,
914 env
->async_pf_en_msr
);
920 kvm_msr_entry_set(&msrs
[n
++], MSR_MCG_STATUS
, env
->mcg_status
);
921 kvm_msr_entry_set(&msrs
[n
++], MSR_MCG_CTL
, env
->mcg_ctl
);
922 for (i
= 0; i
< (env
->mcg_cap
& 0xff) * 4; i
++) {
923 kvm_msr_entry_set(&msrs
[n
++], MSR_MC0_CTL
+ i
, env
->mce_banks
[i
]);
927 msr_data
.info
.nmsrs
= n
;
929 return kvm_vcpu_ioctl(env
, KVM_SET_MSRS
, &msr_data
);
934 static int kvm_get_fpu(CPUState
*env
)
939 ret
= kvm_vcpu_ioctl(env
, KVM_GET_FPU
, &fpu
);
944 env
->fpstt
= (fpu
.fsw
>> 11) & 7;
947 env
->fpop
= fpu
.last_opcode
;
948 env
->fpip
= fpu
.last_ip
;
949 env
->fpdp
= fpu
.last_dp
;
950 for (i
= 0; i
< 8; ++i
) {
951 env
->fptags
[i
] = !((fpu
.ftwx
>> i
) & 1);
953 memcpy(env
->fpregs
, fpu
.fpr
, sizeof env
->fpregs
);
954 memcpy(env
->xmm_regs
, fpu
.xmm
, sizeof env
->xmm_regs
);
955 env
->mxcsr
= fpu
.mxcsr
;
960 static int kvm_get_xsave(CPUState
*env
)
962 struct kvm_xsave
* xsave
;
964 uint16_t cwd
, swd
, twd
;
966 if (!kvm_has_xsave()) {
967 return kvm_get_fpu(env
);
970 xsave
= qemu_memalign(4096, sizeof(struct kvm_xsave
));
971 ret
= kvm_vcpu_ioctl(env
, KVM_GET_XSAVE
, xsave
);
977 cwd
= (uint16_t)xsave
->region
[0];
978 swd
= (uint16_t)(xsave
->region
[0] >> 16);
979 twd
= (uint16_t)xsave
->region
[1];
980 env
->fpop
= (uint16_t)(xsave
->region
[1] >> 16);
981 env
->fpstt
= (swd
>> 11) & 7;
984 for (i
= 0; i
< 8; ++i
) {
985 env
->fptags
[i
] = !((twd
>> i
) & 1);
987 memcpy(&env
->fpip
, &xsave
->region
[XSAVE_CWD_RIP
], sizeof(env
->fpip
));
988 memcpy(&env
->fpdp
, &xsave
->region
[XSAVE_CWD_RDP
], sizeof(env
->fpdp
));
989 env
->mxcsr
= xsave
->region
[XSAVE_MXCSR
];
990 memcpy(env
->fpregs
, &xsave
->region
[XSAVE_ST_SPACE
],
992 memcpy(env
->xmm_regs
, &xsave
->region
[XSAVE_XMM_SPACE
],
993 sizeof env
->xmm_regs
);
994 env
->xstate_bv
= *(uint64_t *)&xsave
->region
[XSAVE_XSTATE_BV
];
995 memcpy(env
->ymmh_regs
, &xsave
->region
[XSAVE_YMMH_SPACE
],
996 sizeof env
->ymmh_regs
);
1001 static int kvm_get_xcrs(CPUState
*env
)
1004 struct kvm_xcrs xcrs
;
1006 if (!kvm_has_xcrs()) {
1010 ret
= kvm_vcpu_ioctl(env
, KVM_GET_XCRS
, &xcrs
);
1015 for (i
= 0; i
< xcrs
.nr_xcrs
; i
++) {
1016 /* Only support xcr0 now */
1017 if (xcrs
.xcrs
[0].xcr
== 0) {
1018 env
->xcr0
= xcrs
.xcrs
[0].value
;
1025 static int kvm_get_sregs(CPUState
*env
)
1027 struct kvm_sregs sregs
;
1031 ret
= kvm_vcpu_ioctl(env
, KVM_GET_SREGS
, &sregs
);
1036 /* There can only be one pending IRQ set in the bitmap at a time, so try
1037 to find it and save its number instead (-1 for none). */
1038 env
->interrupt_injected
= -1;
1039 for (i
= 0; i
< ARRAY_SIZE(sregs
.interrupt_bitmap
); i
++) {
1040 if (sregs
.interrupt_bitmap
[i
]) {
1041 bit
= ctz64(sregs
.interrupt_bitmap
[i
]);
1042 env
->interrupt_injected
= i
* 64 + bit
;
1047 get_seg(&env
->segs
[R_CS
], &sregs
.cs
);
1048 get_seg(&env
->segs
[R_DS
], &sregs
.ds
);
1049 get_seg(&env
->segs
[R_ES
], &sregs
.es
);
1050 get_seg(&env
->segs
[R_FS
], &sregs
.fs
);
1051 get_seg(&env
->segs
[R_GS
], &sregs
.gs
);
1052 get_seg(&env
->segs
[R_SS
], &sregs
.ss
);
1054 get_seg(&env
->tr
, &sregs
.tr
);
1055 get_seg(&env
->ldt
, &sregs
.ldt
);
1057 env
->idt
.limit
= sregs
.idt
.limit
;
1058 env
->idt
.base
= sregs
.idt
.base
;
1059 env
->gdt
.limit
= sregs
.gdt
.limit
;
1060 env
->gdt
.base
= sregs
.gdt
.base
;
1062 env
->cr
[0] = sregs
.cr0
;
1063 env
->cr
[2] = sregs
.cr2
;
1064 env
->cr
[3] = sregs
.cr3
;
1065 env
->cr
[4] = sregs
.cr4
;
1067 cpu_set_apic_base(env
->apic_state
, sregs
.apic_base
);
1069 env
->efer
= sregs
.efer
;
1070 //cpu_set_apic_tpr(env->apic_state, sregs.cr8);
1072 #define HFLAG_COPY_MASK \
1073 ~( HF_CPL_MASK | HF_PE_MASK | HF_MP_MASK | HF_EM_MASK | \
1074 HF_TS_MASK | HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK | \
1075 HF_OSFXSR_MASK | HF_LMA_MASK | HF_CS32_MASK | \
1076 HF_SS32_MASK | HF_CS64_MASK | HF_ADDSEG_MASK)
1078 hflags
= (env
->segs
[R_CS
].flags
>> DESC_DPL_SHIFT
) & HF_CPL_MASK
;
1079 hflags
|= (env
->cr
[0] & CR0_PE_MASK
) << (HF_PE_SHIFT
- CR0_PE_SHIFT
);
1080 hflags
|= (env
->cr
[0] << (HF_MP_SHIFT
- CR0_MP_SHIFT
)) &
1081 (HF_MP_MASK
| HF_EM_MASK
| HF_TS_MASK
);
1082 hflags
|= (env
->eflags
& (HF_TF_MASK
| HF_VM_MASK
| HF_IOPL_MASK
));
1083 hflags
|= (env
->cr
[4] & CR4_OSFXSR_MASK
) <<
1084 (HF_OSFXSR_SHIFT
- CR4_OSFXSR_SHIFT
);
1086 if (env
->efer
& MSR_EFER_LMA
) {
1087 hflags
|= HF_LMA_MASK
;
1090 if ((hflags
& HF_LMA_MASK
) && (env
->segs
[R_CS
].flags
& DESC_L_MASK
)) {
1091 hflags
|= HF_CS32_MASK
| HF_SS32_MASK
| HF_CS64_MASK
;
1093 hflags
|= (env
->segs
[R_CS
].flags
& DESC_B_MASK
) >>
1094 (DESC_B_SHIFT
- HF_CS32_SHIFT
);
1095 hflags
|= (env
->segs
[R_SS
].flags
& DESC_B_MASK
) >>
1096 (DESC_B_SHIFT
- HF_SS32_SHIFT
);
1097 if (!(env
->cr
[0] & CR0_PE_MASK
) || (env
->eflags
& VM_MASK
) ||
1098 !(hflags
& HF_CS32_MASK
)) {
1099 hflags
|= HF_ADDSEG_MASK
;
1101 hflags
|= ((env
->segs
[R_DS
].base
| env
->segs
[R_ES
].base
|
1102 env
->segs
[R_SS
].base
) != 0) << HF_ADDSEG_SHIFT
;
1105 env
->hflags
= (env
->hflags
& HFLAG_COPY_MASK
) | hflags
;
1110 static int kvm_get_msrs(CPUState
*env
)
1113 struct kvm_msrs info
;
1114 struct kvm_msr_entry entries
[100];
1116 struct kvm_msr_entry
*msrs
= msr_data
.entries
;
1120 msrs
[n
++].index
= MSR_IA32_SYSENTER_CS
;
1121 msrs
[n
++].index
= MSR_IA32_SYSENTER_ESP
;
1122 msrs
[n
++].index
= MSR_IA32_SYSENTER_EIP
;
1123 msrs
[n
++].index
= MSR_PAT
;
1125 msrs
[n
++].index
= MSR_STAR
;
1127 if (has_msr_hsave_pa
) {
1128 msrs
[n
++].index
= MSR_VM_HSAVE_PA
;
1131 if (!env
->tsc_valid
) {
1132 msrs
[n
++].index
= MSR_IA32_TSC
;
1133 env
->tsc_valid
= !runstate_is_running();
1136 #ifdef TARGET_X86_64
1137 if (lm_capable_kernel
) {
1138 msrs
[n
++].index
= MSR_CSTAR
;
1139 msrs
[n
++].index
= MSR_KERNELGSBASE
;
1140 msrs
[n
++].index
= MSR_FMASK
;
1141 msrs
[n
++].index
= MSR_LSTAR
;
1144 msrs
[n
++].index
= MSR_KVM_SYSTEM_TIME
;
1145 msrs
[n
++].index
= MSR_KVM_WALL_CLOCK
;
1146 if (has_msr_async_pf_en
) {
1147 msrs
[n
++].index
= MSR_KVM_ASYNC_PF_EN
;
1151 msrs
[n
++].index
= MSR_MCG_STATUS
;
1152 msrs
[n
++].index
= MSR_MCG_CTL
;
1153 for (i
= 0; i
< (env
->mcg_cap
& 0xff) * 4; i
++) {
1154 msrs
[n
++].index
= MSR_MC0_CTL
+ i
;
1158 msr_data
.info
.nmsrs
= n
;
1159 ret
= kvm_vcpu_ioctl(env
, KVM_GET_MSRS
, &msr_data
);
1164 for (i
= 0; i
< ret
; i
++) {
1165 switch (msrs
[i
].index
) {
1166 case MSR_IA32_SYSENTER_CS
:
1167 env
->sysenter_cs
= msrs
[i
].data
;
1169 case MSR_IA32_SYSENTER_ESP
:
1170 env
->sysenter_esp
= msrs
[i
].data
;
1172 case MSR_IA32_SYSENTER_EIP
:
1173 env
->sysenter_eip
= msrs
[i
].data
;
1176 env
->pat
= msrs
[i
].data
;
1179 env
->star
= msrs
[i
].data
;
1181 #ifdef TARGET_X86_64
1183 env
->cstar
= msrs
[i
].data
;
1185 case MSR_KERNELGSBASE
:
1186 env
->kernelgsbase
= msrs
[i
].data
;
1189 env
->fmask
= msrs
[i
].data
;
1192 env
->lstar
= msrs
[i
].data
;
1196 env
->tsc
= msrs
[i
].data
;
1198 case MSR_VM_HSAVE_PA
:
1199 env
->vm_hsave
= msrs
[i
].data
;
1201 case MSR_KVM_SYSTEM_TIME
:
1202 env
->system_time_msr
= msrs
[i
].data
;
1204 case MSR_KVM_WALL_CLOCK
:
1205 env
->wall_clock_msr
= msrs
[i
].data
;
1207 case MSR_MCG_STATUS
:
1208 env
->mcg_status
= msrs
[i
].data
;
1211 env
->mcg_ctl
= msrs
[i
].data
;
1214 if (msrs
[i
].index
>= MSR_MC0_CTL
&&
1215 msrs
[i
].index
< MSR_MC0_CTL
+ (env
->mcg_cap
& 0xff) * 4) {
1216 env
->mce_banks
[msrs
[i
].index
- MSR_MC0_CTL
] = msrs
[i
].data
;
1219 case MSR_KVM_ASYNC_PF_EN
:
1220 env
->async_pf_en_msr
= msrs
[i
].data
;
1228 static int kvm_put_mp_state(CPUState
*env
)
1230 struct kvm_mp_state mp_state
= { .mp_state
= env
->mp_state
};
1232 return kvm_vcpu_ioctl(env
, KVM_SET_MP_STATE
, &mp_state
);
1235 static int kvm_get_mp_state(CPUState
*env
)
1237 struct kvm_mp_state mp_state
;
1240 ret
= kvm_vcpu_ioctl(env
, KVM_GET_MP_STATE
, &mp_state
);
1244 env
->mp_state
= mp_state
.mp_state
;
1245 if (kvm_irqchip_in_kernel()) {
1246 env
->halted
= (mp_state
.mp_state
== KVM_MP_STATE_HALTED
);
1251 static int kvm_put_vcpu_events(CPUState
*env
, int level
)
1253 struct kvm_vcpu_events events
;
1255 if (!kvm_has_vcpu_events()) {
1259 events
.exception
.injected
= (env
->exception_injected
>= 0);
1260 events
.exception
.nr
= env
->exception_injected
;
1261 events
.exception
.has_error_code
= env
->has_error_code
;
1262 events
.exception
.error_code
= env
->error_code
;
1264 events
.interrupt
.injected
= (env
->interrupt_injected
>= 0);
1265 events
.interrupt
.nr
= env
->interrupt_injected
;
1266 events
.interrupt
.soft
= env
->soft_interrupt
;
1268 events
.nmi
.injected
= env
->nmi_injected
;
1269 events
.nmi
.pending
= env
->nmi_pending
;
1270 events
.nmi
.masked
= !!(env
->hflags2
& HF2_NMI_MASK
);
1272 events
.sipi_vector
= env
->sipi_vector
;
1275 if (level
>= KVM_PUT_RESET_STATE
) {
1277 KVM_VCPUEVENT_VALID_NMI_PENDING
| KVM_VCPUEVENT_VALID_SIPI_VECTOR
;
1280 return kvm_vcpu_ioctl(env
, KVM_SET_VCPU_EVENTS
, &events
);
1283 static int kvm_get_vcpu_events(CPUState
*env
)
1285 struct kvm_vcpu_events events
;
1288 if (!kvm_has_vcpu_events()) {
1292 ret
= kvm_vcpu_ioctl(env
, KVM_GET_VCPU_EVENTS
, &events
);
1296 env
->exception_injected
=
1297 events
.exception
.injected
? events
.exception
.nr
: -1;
1298 env
->has_error_code
= events
.exception
.has_error_code
;
1299 env
->error_code
= events
.exception
.error_code
;
1301 env
->interrupt_injected
=
1302 events
.interrupt
.injected
? events
.interrupt
.nr
: -1;
1303 env
->soft_interrupt
= events
.interrupt
.soft
;
1305 env
->nmi_injected
= events
.nmi
.injected
;
1306 env
->nmi_pending
= events
.nmi
.pending
;
1307 if (events
.nmi
.masked
) {
1308 env
->hflags2
|= HF2_NMI_MASK
;
1310 env
->hflags2
&= ~HF2_NMI_MASK
;
1313 env
->sipi_vector
= events
.sipi_vector
;
1318 static int kvm_guest_debug_workarounds(CPUState
*env
)
1321 unsigned long reinject_trap
= 0;
1323 if (!kvm_has_vcpu_events()) {
1324 if (env
->exception_injected
== 1) {
1325 reinject_trap
= KVM_GUESTDBG_INJECT_DB
;
1326 } else if (env
->exception_injected
== 3) {
1327 reinject_trap
= KVM_GUESTDBG_INJECT_BP
;
1329 env
->exception_injected
= -1;
1333 * Kernels before KVM_CAP_X86_ROBUST_SINGLESTEP overwrote flags.TF
1334 * injected via SET_GUEST_DEBUG while updating GP regs. Work around this
1335 * by updating the debug state once again if single-stepping is on.
1336 * Another reason to call kvm_update_guest_debug here is a pending debug
1337 * trap raise by the guest. On kernels without SET_VCPU_EVENTS we have to
1338 * reinject them via SET_GUEST_DEBUG.
1340 if (reinject_trap
||
1341 (!kvm_has_robust_singlestep() && env
->singlestep_enabled
)) {
1342 ret
= kvm_update_guest_debug(env
, reinject_trap
);
1347 static int kvm_put_debugregs(CPUState
*env
)
1349 struct kvm_debugregs dbgregs
;
1352 if (!kvm_has_debugregs()) {
1356 for (i
= 0; i
< 4; i
++) {
1357 dbgregs
.db
[i
] = env
->dr
[i
];
1359 dbgregs
.dr6
= env
->dr
[6];
1360 dbgregs
.dr7
= env
->dr
[7];
1363 return kvm_vcpu_ioctl(env
, KVM_SET_DEBUGREGS
, &dbgregs
);
1366 static int kvm_get_debugregs(CPUState
*env
)
1368 struct kvm_debugregs dbgregs
;
1371 if (!kvm_has_debugregs()) {
1375 ret
= kvm_vcpu_ioctl(env
, KVM_GET_DEBUGREGS
, &dbgregs
);
1379 for (i
= 0; i
< 4; i
++) {
1380 env
->dr
[i
] = dbgregs
.db
[i
];
1382 env
->dr
[4] = env
->dr
[6] = dbgregs
.dr6
;
1383 env
->dr
[5] = env
->dr
[7] = dbgregs
.dr7
;
1388 int kvm_arch_put_registers(CPUState
*env
, int level
)
1392 assert(cpu_is_stopped(env
) || qemu_cpu_is_self(env
));
1394 ret
= kvm_getput_regs(env
, 1);
1398 ret
= kvm_put_xsave(env
);
1402 ret
= kvm_put_xcrs(env
);
1406 ret
= kvm_put_sregs(env
);
1410 /* must be before kvm_put_msrs */
1411 ret
= kvm_inject_mce_oldstyle(env
);
1415 ret
= kvm_put_msrs(env
, level
);
1419 if (level
>= KVM_PUT_RESET_STATE
) {
1420 ret
= kvm_put_mp_state(env
);
1425 ret
= kvm_put_vcpu_events(env
, level
);
1429 ret
= kvm_put_debugregs(env
);
1434 ret
= kvm_guest_debug_workarounds(env
);
1441 int kvm_arch_get_registers(CPUState
*env
)
1445 assert(cpu_is_stopped(env
) || qemu_cpu_is_self(env
));
1447 ret
= kvm_getput_regs(env
, 0);
1451 ret
= kvm_get_xsave(env
);
1455 ret
= kvm_get_xcrs(env
);
1459 ret
= kvm_get_sregs(env
);
1463 ret
= kvm_get_msrs(env
);
1467 ret
= kvm_get_mp_state(env
);
1471 ret
= kvm_get_vcpu_events(env
);
1475 ret
= kvm_get_debugregs(env
);
1482 void kvm_arch_pre_run(CPUState
*env
, struct kvm_run
*run
)
1487 if (env
->interrupt_request
& CPU_INTERRUPT_NMI
) {
1488 env
->interrupt_request
&= ~CPU_INTERRUPT_NMI
;
1489 DPRINTF("injected NMI\n");
1490 ret
= kvm_vcpu_ioctl(env
, KVM_NMI
);
1492 fprintf(stderr
, "KVM: injection failed, NMI lost (%s)\n",
1497 if (!kvm_irqchip_in_kernel()) {
1498 /* Force the VCPU out of its inner loop to process the INIT request */
1499 if (env
->interrupt_request
& CPU_INTERRUPT_INIT
) {
1500 env
->exit_request
= 1;
1503 /* Try to inject an interrupt if the guest can accept it */
1504 if (run
->ready_for_interrupt_injection
&&
1505 (env
->interrupt_request
& CPU_INTERRUPT_HARD
) &&
1506 (env
->eflags
& IF_MASK
)) {
1509 env
->interrupt_request
&= ~CPU_INTERRUPT_HARD
;
1510 irq
= cpu_get_pic_interrupt(env
);
1512 struct kvm_interrupt intr
;
1515 DPRINTF("injected interrupt %d\n", irq
);
1516 ret
= kvm_vcpu_ioctl(env
, KVM_INTERRUPT
, &intr
);
1519 "KVM: injection failed, interrupt lost (%s)\n",
1525 /* If we have an interrupt but the guest is not ready to receive an
1526 * interrupt, request an interrupt window exit. This will
1527 * cause a return to userspace as soon as the guest is ready to
1528 * receive interrupts. */
1529 if ((env
->interrupt_request
& CPU_INTERRUPT_HARD
)) {
1530 run
->request_interrupt_window
= 1;
1532 run
->request_interrupt_window
= 0;
1535 DPRINTF("setting tpr\n");
1536 run
->cr8
= cpu_get_apic_tpr(env
->apic_state
);
1540 void kvm_arch_post_run(CPUState
*env
, struct kvm_run
*run
)
1543 env
->eflags
|= IF_MASK
;
1545 env
->eflags
&= ~IF_MASK
;
1547 cpu_set_apic_tpr(env
->apic_state
, run
->cr8
);
1548 cpu_set_apic_base(env
->apic_state
, run
->apic_base
);
1551 int kvm_arch_process_async_events(CPUState
*env
)
1553 if (env
->interrupt_request
& CPU_INTERRUPT_MCE
) {
1554 /* We must not raise CPU_INTERRUPT_MCE if it's not supported. */
1555 assert(env
->mcg_cap
);
1557 env
->interrupt_request
&= ~CPU_INTERRUPT_MCE
;
1559 kvm_cpu_synchronize_state(env
);
1561 if (env
->exception_injected
== EXCP08_DBLE
) {
1562 /* this means triple fault */
1563 qemu_system_reset_request();
1564 env
->exit_request
= 1;
1567 env
->exception_injected
= EXCP12_MCHK
;
1568 env
->has_error_code
= 0;
1571 if (kvm_irqchip_in_kernel() && env
->mp_state
== KVM_MP_STATE_HALTED
) {
1572 env
->mp_state
= KVM_MP_STATE_RUNNABLE
;
1576 if (kvm_irqchip_in_kernel()) {
1580 if (((env
->interrupt_request
& CPU_INTERRUPT_HARD
) &&
1581 (env
->eflags
& IF_MASK
)) ||
1582 (env
->interrupt_request
& CPU_INTERRUPT_NMI
)) {
1585 if (env
->interrupt_request
& CPU_INTERRUPT_INIT
) {
1586 kvm_cpu_synchronize_state(env
);
1589 if (env
->interrupt_request
& CPU_INTERRUPT_SIPI
) {
1590 kvm_cpu_synchronize_state(env
);
1597 static int kvm_handle_halt(CPUState
*env
)
1599 if (!((env
->interrupt_request
& CPU_INTERRUPT_HARD
) &&
1600 (env
->eflags
& IF_MASK
)) &&
1601 !(env
->interrupt_request
& CPU_INTERRUPT_NMI
)) {
1609 int kvm_arch_insert_sw_breakpoint(CPUState
*env
, struct kvm_sw_breakpoint
*bp
)
1611 static const uint8_t int3
= 0xcc;
1613 if (cpu_memory_rw_debug(env
, bp
->pc
, (uint8_t *)&bp
->saved_insn
, 1, 0) ||
1614 cpu_memory_rw_debug(env
, bp
->pc
, (uint8_t *)&int3
, 1, 1)) {
1620 int kvm_arch_remove_sw_breakpoint(CPUState
*env
, struct kvm_sw_breakpoint
*bp
)
1624 if (cpu_memory_rw_debug(env
, bp
->pc
, &int3
, 1, 0) || int3
!= 0xcc ||
1625 cpu_memory_rw_debug(env
, bp
->pc
, (uint8_t *)&bp
->saved_insn
, 1, 1)) {
1637 static int nb_hw_breakpoint
;
1639 static int find_hw_breakpoint(target_ulong addr
, int len
, int type
)
1643 for (n
= 0; n
< nb_hw_breakpoint
; n
++) {
1644 if (hw_breakpoint
[n
].addr
== addr
&& hw_breakpoint
[n
].type
== type
&&
1645 (hw_breakpoint
[n
].len
== len
|| len
== -1)) {
1652 int kvm_arch_insert_hw_breakpoint(target_ulong addr
,
1653 target_ulong len
, int type
)
1656 case GDB_BREAKPOINT_HW
:
1659 case GDB_WATCHPOINT_WRITE
:
1660 case GDB_WATCHPOINT_ACCESS
:
1667 if (addr
& (len
- 1)) {
1679 if (nb_hw_breakpoint
== 4) {
1682 if (find_hw_breakpoint(addr
, len
, type
) >= 0) {
1685 hw_breakpoint
[nb_hw_breakpoint
].addr
= addr
;
1686 hw_breakpoint
[nb_hw_breakpoint
].len
= len
;
1687 hw_breakpoint
[nb_hw_breakpoint
].type
= type
;
1693 int kvm_arch_remove_hw_breakpoint(target_ulong addr
,
1694 target_ulong len
, int type
)
1698 n
= find_hw_breakpoint(addr
, (type
== GDB_BREAKPOINT_HW
) ? 1 : len
, type
);
1703 hw_breakpoint
[n
] = hw_breakpoint
[nb_hw_breakpoint
];
1708 void kvm_arch_remove_all_hw_breakpoints(void)
1710 nb_hw_breakpoint
= 0;
1713 static CPUWatchpoint hw_watchpoint
;
1715 static int kvm_handle_debug(struct kvm_debug_exit_arch
*arch_info
)
1720 if (arch_info
->exception
== 1) {
1721 if (arch_info
->dr6
& (1 << 14)) {
1722 if (cpu_single_env
->singlestep_enabled
) {
1726 for (n
= 0; n
< 4; n
++) {
1727 if (arch_info
->dr6
& (1 << n
)) {
1728 switch ((arch_info
->dr7
>> (16 + n
*4)) & 0x3) {
1734 cpu_single_env
->watchpoint_hit
= &hw_watchpoint
;
1735 hw_watchpoint
.vaddr
= hw_breakpoint
[n
].addr
;
1736 hw_watchpoint
.flags
= BP_MEM_WRITE
;
1740 cpu_single_env
->watchpoint_hit
= &hw_watchpoint
;
1741 hw_watchpoint
.vaddr
= hw_breakpoint
[n
].addr
;
1742 hw_watchpoint
.flags
= BP_MEM_ACCESS
;
1748 } else if (kvm_find_sw_breakpoint(cpu_single_env
, arch_info
->pc
)) {
1752 cpu_synchronize_state(cpu_single_env
);
1753 assert(cpu_single_env
->exception_injected
== -1);
1756 cpu_single_env
->exception_injected
= arch_info
->exception
;
1757 cpu_single_env
->has_error_code
= 0;
1763 void kvm_arch_update_guest_debug(CPUState
*env
, struct kvm_guest_debug
*dbg
)
1765 const uint8_t type_code
[] = {
1766 [GDB_BREAKPOINT_HW
] = 0x0,
1767 [GDB_WATCHPOINT_WRITE
] = 0x1,
1768 [GDB_WATCHPOINT_ACCESS
] = 0x3
1770 const uint8_t len_code
[] = {
1771 [1] = 0x0, [2] = 0x1, [4] = 0x3, [8] = 0x2
1775 if (kvm_sw_breakpoints_active(env
)) {
1776 dbg
->control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_USE_SW_BP
;
1778 if (nb_hw_breakpoint
> 0) {
1779 dbg
->control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_USE_HW_BP
;
1780 dbg
->arch
.debugreg
[7] = 0x0600;
1781 for (n
= 0; n
< nb_hw_breakpoint
; n
++) {
1782 dbg
->arch
.debugreg
[n
] = hw_breakpoint
[n
].addr
;
1783 dbg
->arch
.debugreg
[7] |= (2 << (n
* 2)) |
1784 (type_code
[hw_breakpoint
[n
].type
] << (16 + n
*4)) |
1785 ((uint32_t)len_code
[hw_breakpoint
[n
].len
] << (18 + n
*4));
1790 static bool host_supports_vmx(void)
1792 uint32_t ecx
, unused
;
1794 host_cpuid(1, 0, &unused
, &unused
, &ecx
, &unused
);
1795 return ecx
& CPUID_EXT_VMX
;
1798 #define VMX_INVALID_GUEST_STATE 0x80000021
1800 int kvm_arch_handle_exit(CPUState
*env
, struct kvm_run
*run
)
1805 switch (run
->exit_reason
) {
1807 DPRINTF("handle_hlt\n");
1808 ret
= kvm_handle_halt(env
);
1810 case KVM_EXIT_SET_TPR
:
1813 case KVM_EXIT_FAIL_ENTRY
:
1814 code
= run
->fail_entry
.hardware_entry_failure_reason
;
1815 fprintf(stderr
, "KVM: entry failed, hardware error 0x%" PRIx64
"\n",
1817 if (host_supports_vmx() && code
== VMX_INVALID_GUEST_STATE
) {
1819 "\nIf you're runnning a guest on an Intel machine without "
1820 "unrestricted mode\n"
1821 "support, the failure can be most likely due to the guest "
1822 "entering an invalid\n"
1823 "state for Intel VT. For example, the guest maybe running "
1824 "in big real mode\n"
1825 "which is not supported on less recent Intel processors."
1830 case KVM_EXIT_EXCEPTION
:
1831 fprintf(stderr
, "KVM: exception %d exit (error code 0x%x)\n",
1832 run
->ex
.exception
, run
->ex
.error_code
);
1835 case KVM_EXIT_DEBUG
:
1836 DPRINTF("kvm_exit_debug\n");
1837 ret
= kvm_handle_debug(&run
->debug
.arch
);
1840 fprintf(stderr
, "KVM: unknown exit reason %d\n", run
->exit_reason
);
1848 bool kvm_arch_stop_on_emulation_error(CPUState
*env
)
1850 return !(env
->cr
[0] & CR0_PE_MASK
) ||
1851 ((env
->segs
[R_CS
].selector
& 3) != 3);