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>
22 #include "qemu-common.h"
27 #include "host-utils.h"
32 #ifdef CONFIG_KVM_PARA
33 #include <linux/kvm_para.h>
39 #define DPRINTF(fmt, ...) \
40 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
42 #define DPRINTF(fmt, ...) \
46 #define MSR_KVM_WALL_CLOCK 0x11
47 #define MSR_KVM_SYSTEM_TIME 0x12
50 #define BUS_MCEERR_AR 4
53 #define BUS_MCEERR_AO 5
56 const KVMCapabilityInfo kvm_arch_required_capabilities
[] = {
57 KVM_CAP_INFO(SET_TSS_ADDR
),
58 KVM_CAP_INFO(EXT_CPUID
),
59 KVM_CAP_INFO(MP_STATE
),
63 static bool has_msr_star
;
64 static bool has_msr_hsave_pa
;
65 #if defined(CONFIG_KVM_PARA) && defined(KVM_CAP_ASYNC_PF)
66 static bool has_msr_async_pf_en
;
68 static int lm_capable_kernel
;
70 static struct kvm_cpuid2
*try_get_cpuid(KVMState
*s
, int max
)
72 struct kvm_cpuid2
*cpuid
;
75 size
= sizeof(*cpuid
) + max
* sizeof(*cpuid
->entries
);
76 cpuid
= (struct kvm_cpuid2
*)qemu_mallocz(size
);
78 r
= kvm_ioctl(s
, KVM_GET_SUPPORTED_CPUID
, cpuid
);
79 if (r
== 0 && cpuid
->nent
>= max
) {
87 fprintf(stderr
, "KVM_GET_SUPPORTED_CPUID failed: %s\n",
95 uint32_t kvm_arch_get_supported_cpuid(CPUState
*env
, uint32_t function
,
96 uint32_t index
, int reg
)
98 struct kvm_cpuid2
*cpuid
;
101 uint32_t cpuid_1_edx
;
104 while ((cpuid
= try_get_cpuid(env
->kvm_state
, max
)) == NULL
) {
108 for (i
= 0; i
< cpuid
->nent
; ++i
) {
109 if (cpuid
->entries
[i
].function
== function
&&
110 cpuid
->entries
[i
].index
== index
) {
113 ret
= cpuid
->entries
[i
].eax
;
116 ret
= cpuid
->entries
[i
].ebx
;
119 ret
= cpuid
->entries
[i
].ecx
;
122 ret
= cpuid
->entries
[i
].edx
;
125 /* KVM before 2.6.30 misreports the following features */
126 ret
|= CPUID_MTRR
| CPUID_PAT
| CPUID_MCE
| CPUID_MCA
;
129 /* On Intel, kvm returns cpuid according to the Intel spec,
130 * so add missing bits according to the AMD spec:
132 cpuid_1_edx
= kvm_arch_get_supported_cpuid(env
, 1, 0, R_EDX
);
133 ret
|= cpuid_1_edx
& 0x183f7ff;
146 #ifdef CONFIG_KVM_PARA
147 struct kvm_para_features
{
150 } para_features
[] = {
151 { KVM_CAP_CLOCKSOURCE
, KVM_FEATURE_CLOCKSOURCE
},
152 { KVM_CAP_NOP_IO_DELAY
, KVM_FEATURE_NOP_IO_DELAY
},
153 { KVM_CAP_PV_MMU
, KVM_FEATURE_MMU_OP
},
154 #ifdef KVM_CAP_ASYNC_PF
155 { KVM_CAP_ASYNC_PF
, KVM_FEATURE_ASYNC_PF
},
160 static int get_para_features(CPUState
*env
)
164 for (i
= 0; i
< ARRAY_SIZE(para_features
) - 1; i
++) {
165 if (kvm_check_extension(env
->kvm_state
, para_features
[i
].cap
)) {
166 features
|= (1 << para_features
[i
].feature
);
169 #ifdef KVM_CAP_ASYNC_PF
170 has_msr_async_pf_en
= features
& (1 << KVM_FEATURE_ASYNC_PF
);
174 #endif /* CONFIG_KVM_PARA */
177 static int kvm_get_mce_cap_supported(KVMState
*s
, uint64_t *mce_cap
,
182 r
= kvm_check_extension(s
, KVM_CAP_MCE
);
185 return kvm_ioctl(s
, KVM_X86_GET_MCE_CAP_SUPPORTED
, mce_cap
);
190 static void kvm_mce_inject(CPUState
*env
, target_phys_addr_t paddr
, int code
)
192 uint64_t status
= MCI_STATUS_VAL
| MCI_STATUS_UC
| MCI_STATUS_EN
|
193 MCI_STATUS_MISCV
| MCI_STATUS_ADDRV
| MCI_STATUS_S
;
194 uint64_t mcg_status
= MCG_STATUS_MCIP
;
196 if (code
== BUS_MCEERR_AR
) {
197 status
|= MCI_STATUS_AR
| 0x134;
198 mcg_status
|= MCG_STATUS_EIPV
;
201 mcg_status
|= MCG_STATUS_RIPV
;
203 cpu_x86_inject_mce(NULL
, env
, 9, status
, mcg_status
, paddr
,
204 (MCM_ADDR_PHYS
<< 6) | 0xc,
205 cpu_x86_support_mca_broadcast(env
) ?
206 MCE_INJECT_BROADCAST
: 0);
208 #endif /* KVM_CAP_MCE */
210 static void hardware_memory_error(void)
212 fprintf(stderr
, "Hardware memory error!\n");
216 int kvm_arch_on_sigbus_vcpu(CPUState
*env
, int code
, void *addr
)
220 target_phys_addr_t paddr
;
222 if ((env
->mcg_cap
& MCG_SER_P
) && addr
223 && (code
== BUS_MCEERR_AR
|| code
== BUS_MCEERR_AO
)) {
224 if (qemu_ram_addr_from_host(addr
, &ram_addr
) ||
225 !kvm_physical_memory_addr_from_ram(env
->kvm_state
, ram_addr
,
227 fprintf(stderr
, "Hardware memory error for memory used by "
228 "QEMU itself instead of guest system!\n");
229 /* Hope we are lucky for AO MCE */
230 if (code
== BUS_MCEERR_AO
) {
233 hardware_memory_error();
236 kvm_mce_inject(env
, paddr
, code
);
238 #endif /* KVM_CAP_MCE */
240 if (code
== BUS_MCEERR_AO
) {
242 } else if (code
== BUS_MCEERR_AR
) {
243 hardware_memory_error();
251 int kvm_arch_on_sigbus(int code
, void *addr
)
254 if ((first_cpu
->mcg_cap
& MCG_SER_P
) && addr
&& code
== BUS_MCEERR_AO
) {
256 target_phys_addr_t paddr
;
258 /* Hope we are lucky for AO MCE */
259 if (qemu_ram_addr_from_host(addr
, &ram_addr
) ||
260 !kvm_physical_memory_addr_from_ram(first_cpu
->kvm_state
, ram_addr
,
262 fprintf(stderr
, "Hardware memory error for memory used by "
263 "QEMU itself instead of guest system!: %p\n", addr
);
266 kvm_mce_inject(first_cpu
, paddr
, code
);
268 #endif /* KVM_CAP_MCE */
270 if (code
== BUS_MCEERR_AO
) {
272 } else if (code
== BUS_MCEERR_AR
) {
273 hardware_memory_error();
281 static int kvm_inject_mce_oldstyle(CPUState
*env
)
284 if (!kvm_has_vcpu_events() && env
->exception_injected
== EXCP12_MCHK
) {
285 unsigned int bank
, bank_num
= env
->mcg_cap
& 0xff;
286 struct kvm_x86_mce mce
;
288 env
->exception_injected
= -1;
291 * There must be at least one bank in use if an MCE is pending.
292 * Find it and use its values for the event injection.
294 for (bank
= 0; bank
< bank_num
; bank
++) {
295 if (env
->mce_banks
[bank
* 4 + 1] & MCI_STATUS_VAL
) {
299 assert(bank
< bank_num
);
302 mce
.status
= env
->mce_banks
[bank
* 4 + 1];
303 mce
.mcg_status
= env
->mcg_status
;
304 mce
.addr
= env
->mce_banks
[bank
* 4 + 2];
305 mce
.misc
= env
->mce_banks
[bank
* 4 + 3];
307 return kvm_vcpu_ioctl(env
, KVM_X86_SET_MCE
, &mce
);
309 #endif /* KVM_CAP_MCE */
313 static void cpu_update_state(void *opaque
, int running
, int reason
)
315 CPUState
*env
= opaque
;
318 env
->tsc_valid
= false;
322 int kvm_arch_init_vcpu(CPUState
*env
)
325 struct kvm_cpuid2 cpuid
;
326 struct kvm_cpuid_entry2 entries
[100];
327 } __attribute__((packed
)) cpuid_data
;
328 uint32_t limit
, i
, j
, cpuid_i
;
330 struct kvm_cpuid_entry2
*c
;
331 #ifdef CONFIG_KVM_PARA
332 uint32_t signature
[3];
335 env
->cpuid_features
&= kvm_arch_get_supported_cpuid(env
, 1, 0, R_EDX
);
337 i
= env
->cpuid_ext_features
& CPUID_EXT_HYPERVISOR
;
338 env
->cpuid_ext_features
&= kvm_arch_get_supported_cpuid(env
, 1, 0, R_ECX
);
339 env
->cpuid_ext_features
|= i
;
341 env
->cpuid_ext2_features
&= kvm_arch_get_supported_cpuid(env
, 0x80000001,
343 env
->cpuid_ext3_features
&= kvm_arch_get_supported_cpuid(env
, 0x80000001,
345 env
->cpuid_svm_features
&= kvm_arch_get_supported_cpuid(env
, 0x8000000A,
351 #ifdef CONFIG_KVM_PARA
352 /* Paravirtualization CPUIDs */
353 memcpy(signature
, "KVMKVMKVM\0\0\0", 12);
354 c
= &cpuid_data
.entries
[cpuid_i
++];
355 memset(c
, 0, sizeof(*c
));
356 c
->function
= KVM_CPUID_SIGNATURE
;
358 c
->ebx
= signature
[0];
359 c
->ecx
= signature
[1];
360 c
->edx
= signature
[2];
362 c
= &cpuid_data
.entries
[cpuid_i
++];
363 memset(c
, 0, sizeof(*c
));
364 c
->function
= KVM_CPUID_FEATURES
;
365 c
->eax
= env
->cpuid_kvm_features
& get_para_features(env
);
368 cpu_x86_cpuid(env
, 0, 0, &limit
, &unused
, &unused
, &unused
);
370 for (i
= 0; i
<= limit
; i
++) {
371 c
= &cpuid_data
.entries
[cpuid_i
++];
375 /* Keep reading function 2 till all the input is received */
379 c
->flags
= KVM_CPUID_FLAG_STATEFUL_FUNC
|
380 KVM_CPUID_FLAG_STATE_READ_NEXT
;
381 cpu_x86_cpuid(env
, i
, 0, &c
->eax
, &c
->ebx
, &c
->ecx
, &c
->edx
);
382 times
= c
->eax
& 0xff;
384 for (j
= 1; j
< times
; ++j
) {
385 c
= &cpuid_data
.entries
[cpuid_i
++];
387 c
->flags
= KVM_CPUID_FLAG_STATEFUL_FUNC
;
388 cpu_x86_cpuid(env
, i
, 0, &c
->eax
, &c
->ebx
, &c
->ecx
, &c
->edx
);
397 c
->flags
= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
399 cpu_x86_cpuid(env
, i
, j
, &c
->eax
, &c
->ebx
, &c
->ecx
, &c
->edx
);
401 if (i
== 4 && c
->eax
== 0) {
404 if (i
== 0xb && !(c
->ecx
& 0xff00)) {
407 if (i
== 0xd && c
->eax
== 0) {
410 c
= &cpuid_data
.entries
[cpuid_i
++];
416 cpu_x86_cpuid(env
, i
, 0, &c
->eax
, &c
->ebx
, &c
->ecx
, &c
->edx
);
420 cpu_x86_cpuid(env
, 0x80000000, 0, &limit
, &unused
, &unused
, &unused
);
422 for (i
= 0x80000000; i
<= limit
; i
++) {
423 c
= &cpuid_data
.entries
[cpuid_i
++];
427 cpu_x86_cpuid(env
, i
, 0, &c
->eax
, &c
->ebx
, &c
->ecx
, &c
->edx
);
430 cpuid_data
.cpuid
.nent
= cpuid_i
;
433 if (((env
->cpuid_version
>> 8)&0xF) >= 6
434 && (env
->cpuid_features
&(CPUID_MCE
|CPUID_MCA
)) == (CPUID_MCE
|CPUID_MCA
)
435 && kvm_check_extension(env
->kvm_state
, KVM_CAP_MCE
) > 0) {
440 ret
= kvm_get_mce_cap_supported(env
->kvm_state
, &mcg_cap
, &banks
);
442 fprintf(stderr
, "kvm_get_mce_cap_supported: %s", strerror(-ret
));
446 if (banks
> MCE_BANKS_DEF
) {
447 banks
= MCE_BANKS_DEF
;
449 mcg_cap
&= MCE_CAP_DEF
;
451 ret
= kvm_vcpu_ioctl(env
, KVM_X86_SETUP_MCE
, &mcg_cap
);
453 fprintf(stderr
, "KVM_X86_SETUP_MCE: %s", strerror(-ret
));
457 env
->mcg_cap
= mcg_cap
;
461 qemu_add_vm_change_state_handler(cpu_update_state
, env
);
463 return kvm_vcpu_ioctl(env
, KVM_SET_CPUID2
, &cpuid_data
);
466 void kvm_arch_reset_vcpu(CPUState
*env
)
468 env
->exception_injected
= -1;
469 env
->interrupt_injected
= -1;
471 if (kvm_irqchip_in_kernel()) {
472 env
->mp_state
= cpu_is_bsp(env
) ? KVM_MP_STATE_RUNNABLE
:
473 KVM_MP_STATE_UNINITIALIZED
;
475 env
->mp_state
= KVM_MP_STATE_RUNNABLE
;
479 static int kvm_get_supported_msrs(KVMState
*s
)
481 static int kvm_supported_msrs
;
485 if (kvm_supported_msrs
== 0) {
486 struct kvm_msr_list msr_list
, *kvm_msr_list
;
488 kvm_supported_msrs
= -1;
490 /* Obtain MSR list from KVM. These are the MSRs that we must
493 ret
= kvm_ioctl(s
, KVM_GET_MSR_INDEX_LIST
, &msr_list
);
494 if (ret
< 0 && ret
!= -E2BIG
) {
497 /* Old kernel modules had a bug and could write beyond the provided
498 memory. Allocate at least a safe amount of 1K. */
499 kvm_msr_list
= qemu_mallocz(MAX(1024, sizeof(msr_list
) +
501 sizeof(msr_list
.indices
[0])));
503 kvm_msr_list
->nmsrs
= msr_list
.nmsrs
;
504 ret
= kvm_ioctl(s
, KVM_GET_MSR_INDEX_LIST
, kvm_msr_list
);
508 for (i
= 0; i
< kvm_msr_list
->nmsrs
; i
++) {
509 if (kvm_msr_list
->indices
[i
] == MSR_STAR
) {
513 if (kvm_msr_list
->indices
[i
] == MSR_VM_HSAVE_PA
) {
514 has_msr_hsave_pa
= true;
526 int kvm_arch_init(KVMState
*s
)
528 uint64_t identity_base
= 0xfffbc000;
530 struct utsname utsname
;
532 ret
= kvm_get_supported_msrs(s
);
538 lm_capable_kernel
= strcmp(utsname
.machine
, "x86_64") == 0;
541 * On older Intel CPUs, KVM uses vm86 mode to emulate 16-bit code directly.
542 * In order to use vm86 mode, an EPT identity map and a TSS are needed.
543 * Since these must be part of guest physical memory, we need to allocate
544 * them, both by setting their start addresses in the kernel and by
545 * creating a corresponding e820 entry. We need 4 pages before the BIOS.
547 * Older KVM versions may not support setting the identity map base. In
548 * that case we need to stick with the default, i.e. a 256K maximum BIOS
551 #ifdef KVM_CAP_SET_IDENTITY_MAP_ADDR
552 if (kvm_check_extension(s
, KVM_CAP_SET_IDENTITY_MAP_ADDR
)) {
553 /* Allows up to 16M BIOSes. */
554 identity_base
= 0xfeffc000;
556 ret
= kvm_vm_ioctl(s
, KVM_SET_IDENTITY_MAP_ADDR
, &identity_base
);
562 /* Set TSS base one page after EPT identity map. */
563 ret
= kvm_vm_ioctl(s
, KVM_SET_TSS_ADDR
, identity_base
+ 0x1000);
568 /* Tell fw_cfg to notify the BIOS to reserve the range. */
569 ret
= e820_add_entry(identity_base
, 0x4000, E820_RESERVED
);
571 fprintf(stderr
, "e820_add_entry() table is full\n");
578 static void set_v8086_seg(struct kvm_segment
*lhs
, const SegmentCache
*rhs
)
580 lhs
->selector
= rhs
->selector
;
581 lhs
->base
= rhs
->base
;
582 lhs
->limit
= rhs
->limit
;
594 static void set_seg(struct kvm_segment
*lhs
, const SegmentCache
*rhs
)
596 unsigned flags
= rhs
->flags
;
597 lhs
->selector
= rhs
->selector
;
598 lhs
->base
= rhs
->base
;
599 lhs
->limit
= rhs
->limit
;
600 lhs
->type
= (flags
>> DESC_TYPE_SHIFT
) & 15;
601 lhs
->present
= (flags
& DESC_P_MASK
) != 0;
602 lhs
->dpl
= (flags
>> DESC_DPL_SHIFT
) & 3;
603 lhs
->db
= (flags
>> DESC_B_SHIFT
) & 1;
604 lhs
->s
= (flags
& DESC_S_MASK
) != 0;
605 lhs
->l
= (flags
>> DESC_L_SHIFT
) & 1;
606 lhs
->g
= (flags
& DESC_G_MASK
) != 0;
607 lhs
->avl
= (flags
& DESC_AVL_MASK
) != 0;
611 static void get_seg(SegmentCache
*lhs
, const struct kvm_segment
*rhs
)
613 lhs
->selector
= rhs
->selector
;
614 lhs
->base
= rhs
->base
;
615 lhs
->limit
= rhs
->limit
;
616 lhs
->flags
= (rhs
->type
<< DESC_TYPE_SHIFT
) |
617 (rhs
->present
* DESC_P_MASK
) |
618 (rhs
->dpl
<< DESC_DPL_SHIFT
) |
619 (rhs
->db
<< DESC_B_SHIFT
) |
620 (rhs
->s
* DESC_S_MASK
) |
621 (rhs
->l
<< DESC_L_SHIFT
) |
622 (rhs
->g
* DESC_G_MASK
) |
623 (rhs
->avl
* DESC_AVL_MASK
);
626 static void kvm_getput_reg(__u64
*kvm_reg
, target_ulong
*qemu_reg
, int set
)
629 *kvm_reg
= *qemu_reg
;
631 *qemu_reg
= *kvm_reg
;
635 static int kvm_getput_regs(CPUState
*env
, int set
)
637 struct kvm_regs regs
;
641 ret
= kvm_vcpu_ioctl(env
, KVM_GET_REGS
, ®s
);
647 kvm_getput_reg(®s
.rax
, &env
->regs
[R_EAX
], set
);
648 kvm_getput_reg(®s
.rbx
, &env
->regs
[R_EBX
], set
);
649 kvm_getput_reg(®s
.rcx
, &env
->regs
[R_ECX
], set
);
650 kvm_getput_reg(®s
.rdx
, &env
->regs
[R_EDX
], set
);
651 kvm_getput_reg(®s
.rsi
, &env
->regs
[R_ESI
], set
);
652 kvm_getput_reg(®s
.rdi
, &env
->regs
[R_EDI
], set
);
653 kvm_getput_reg(®s
.rsp
, &env
->regs
[R_ESP
], set
);
654 kvm_getput_reg(®s
.rbp
, &env
->regs
[R_EBP
], set
);
656 kvm_getput_reg(®s
.r8
, &env
->regs
[8], set
);
657 kvm_getput_reg(®s
.r9
, &env
->regs
[9], set
);
658 kvm_getput_reg(®s
.r10
, &env
->regs
[10], set
);
659 kvm_getput_reg(®s
.r11
, &env
->regs
[11], set
);
660 kvm_getput_reg(®s
.r12
, &env
->regs
[12], set
);
661 kvm_getput_reg(®s
.r13
, &env
->regs
[13], set
);
662 kvm_getput_reg(®s
.r14
, &env
->regs
[14], set
);
663 kvm_getput_reg(®s
.r15
, &env
->regs
[15], set
);
666 kvm_getput_reg(®s
.rflags
, &env
->eflags
, set
);
667 kvm_getput_reg(®s
.rip
, &env
->eip
, set
);
670 ret
= kvm_vcpu_ioctl(env
, KVM_SET_REGS
, ®s
);
676 static int kvm_put_fpu(CPUState
*env
)
681 memset(&fpu
, 0, sizeof fpu
);
682 fpu
.fsw
= env
->fpus
& ~(7 << 11);
683 fpu
.fsw
|= (env
->fpstt
& 7) << 11;
685 for (i
= 0; i
< 8; ++i
) {
686 fpu
.ftwx
|= (!env
->fptags
[i
]) << i
;
688 memcpy(fpu
.fpr
, env
->fpregs
, sizeof env
->fpregs
);
689 memcpy(fpu
.xmm
, env
->xmm_regs
, sizeof env
->xmm_regs
);
690 fpu
.mxcsr
= env
->mxcsr
;
692 return kvm_vcpu_ioctl(env
, KVM_SET_FPU
, &fpu
);
696 #define XSAVE_CWD_RIP 2
697 #define XSAVE_CWD_RDP 4
698 #define XSAVE_MXCSR 6
699 #define XSAVE_ST_SPACE 8
700 #define XSAVE_XMM_SPACE 40
701 #define XSAVE_XSTATE_BV 128
702 #define XSAVE_YMMH_SPACE 144
705 static int kvm_put_xsave(CPUState
*env
)
709 struct kvm_xsave
* xsave
;
710 uint16_t cwd
, swd
, twd
, fop
;
712 if (!kvm_has_xsave()) {
713 return kvm_put_fpu(env
);
716 xsave
= qemu_memalign(4096, sizeof(struct kvm_xsave
));
717 memset(xsave
, 0, sizeof(struct kvm_xsave
));
718 cwd
= swd
= twd
= fop
= 0;
719 swd
= env
->fpus
& ~(7 << 11);
720 swd
|= (env
->fpstt
& 7) << 11;
722 for (i
= 0; i
< 8; ++i
) {
723 twd
|= (!env
->fptags
[i
]) << i
;
725 xsave
->region
[0] = (uint32_t)(swd
<< 16) + cwd
;
726 xsave
->region
[1] = (uint32_t)(fop
<< 16) + twd
;
727 memcpy(&xsave
->region
[XSAVE_ST_SPACE
], env
->fpregs
,
729 memcpy(&xsave
->region
[XSAVE_XMM_SPACE
], env
->xmm_regs
,
730 sizeof env
->xmm_regs
);
731 xsave
->region
[XSAVE_MXCSR
] = env
->mxcsr
;
732 *(uint64_t *)&xsave
->region
[XSAVE_XSTATE_BV
] = env
->xstate_bv
;
733 memcpy(&xsave
->region
[XSAVE_YMMH_SPACE
], env
->ymmh_regs
,
734 sizeof env
->ymmh_regs
);
735 r
= kvm_vcpu_ioctl(env
, KVM_SET_XSAVE
, xsave
);
739 return kvm_put_fpu(env
);
743 static int kvm_put_xcrs(CPUState
*env
)
746 struct kvm_xcrs xcrs
;
748 if (!kvm_has_xcrs()) {
754 xcrs
.xcrs
[0].xcr
= 0;
755 xcrs
.xcrs
[0].value
= env
->xcr0
;
756 return kvm_vcpu_ioctl(env
, KVM_SET_XCRS
, &xcrs
);
762 static int kvm_put_sregs(CPUState
*env
)
764 struct kvm_sregs sregs
;
766 memset(sregs
.interrupt_bitmap
, 0, sizeof(sregs
.interrupt_bitmap
));
767 if (env
->interrupt_injected
>= 0) {
768 sregs
.interrupt_bitmap
[env
->interrupt_injected
/ 64] |=
769 (uint64_t)1 << (env
->interrupt_injected
% 64);
772 if ((env
->eflags
& VM_MASK
)) {
773 set_v8086_seg(&sregs
.cs
, &env
->segs
[R_CS
]);
774 set_v8086_seg(&sregs
.ds
, &env
->segs
[R_DS
]);
775 set_v8086_seg(&sregs
.es
, &env
->segs
[R_ES
]);
776 set_v8086_seg(&sregs
.fs
, &env
->segs
[R_FS
]);
777 set_v8086_seg(&sregs
.gs
, &env
->segs
[R_GS
]);
778 set_v8086_seg(&sregs
.ss
, &env
->segs
[R_SS
]);
780 set_seg(&sregs
.cs
, &env
->segs
[R_CS
]);
781 set_seg(&sregs
.ds
, &env
->segs
[R_DS
]);
782 set_seg(&sregs
.es
, &env
->segs
[R_ES
]);
783 set_seg(&sregs
.fs
, &env
->segs
[R_FS
]);
784 set_seg(&sregs
.gs
, &env
->segs
[R_GS
]);
785 set_seg(&sregs
.ss
, &env
->segs
[R_SS
]);
788 set_seg(&sregs
.tr
, &env
->tr
);
789 set_seg(&sregs
.ldt
, &env
->ldt
);
791 sregs
.idt
.limit
= env
->idt
.limit
;
792 sregs
.idt
.base
= env
->idt
.base
;
793 sregs
.gdt
.limit
= env
->gdt
.limit
;
794 sregs
.gdt
.base
= env
->gdt
.base
;
796 sregs
.cr0
= env
->cr
[0];
797 sregs
.cr2
= env
->cr
[2];
798 sregs
.cr3
= env
->cr
[3];
799 sregs
.cr4
= env
->cr
[4];
801 sregs
.cr8
= cpu_get_apic_tpr(env
->apic_state
);
802 sregs
.apic_base
= cpu_get_apic_base(env
->apic_state
);
804 sregs
.efer
= env
->efer
;
806 return kvm_vcpu_ioctl(env
, KVM_SET_SREGS
, &sregs
);
809 static void kvm_msr_entry_set(struct kvm_msr_entry
*entry
,
810 uint32_t index
, uint64_t value
)
812 entry
->index
= index
;
816 static int kvm_put_msrs(CPUState
*env
, int level
)
819 struct kvm_msrs info
;
820 struct kvm_msr_entry entries
[100];
822 struct kvm_msr_entry
*msrs
= msr_data
.entries
;
825 kvm_msr_entry_set(&msrs
[n
++], MSR_IA32_SYSENTER_CS
, env
->sysenter_cs
);
826 kvm_msr_entry_set(&msrs
[n
++], MSR_IA32_SYSENTER_ESP
, env
->sysenter_esp
);
827 kvm_msr_entry_set(&msrs
[n
++], MSR_IA32_SYSENTER_EIP
, env
->sysenter_eip
);
829 kvm_msr_entry_set(&msrs
[n
++], MSR_STAR
, env
->star
);
831 if (has_msr_hsave_pa
) {
832 kvm_msr_entry_set(&msrs
[n
++], MSR_VM_HSAVE_PA
, env
->vm_hsave
);
835 if (lm_capable_kernel
) {
836 kvm_msr_entry_set(&msrs
[n
++], MSR_CSTAR
, env
->cstar
);
837 kvm_msr_entry_set(&msrs
[n
++], MSR_KERNELGSBASE
, env
->kernelgsbase
);
838 kvm_msr_entry_set(&msrs
[n
++], MSR_FMASK
, env
->fmask
);
839 kvm_msr_entry_set(&msrs
[n
++], MSR_LSTAR
, env
->lstar
);
842 if (level
== KVM_PUT_FULL_STATE
) {
844 * KVM is yet unable to synchronize TSC values of multiple VCPUs on
845 * writeback. Until this is fixed, we only write the offset to SMP
846 * guests after migration, desynchronizing the VCPUs, but avoiding
847 * huge jump-backs that would occur without any writeback at all.
849 if (smp_cpus
== 1 || env
->tsc
!= 0) {
850 kvm_msr_entry_set(&msrs
[n
++], MSR_IA32_TSC
, env
->tsc
);
854 * The following paravirtual MSRs have side effects on the guest or are
855 * too heavy for normal writeback. Limit them to reset or full state
858 if (level
>= KVM_PUT_RESET_STATE
) {
859 kvm_msr_entry_set(&msrs
[n
++], MSR_KVM_SYSTEM_TIME
,
860 env
->system_time_msr
);
861 kvm_msr_entry_set(&msrs
[n
++], MSR_KVM_WALL_CLOCK
, env
->wall_clock_msr
);
862 #if defined(CONFIG_KVM_PARA) && defined(KVM_CAP_ASYNC_PF)
863 if (has_msr_async_pf_en
) {
864 kvm_msr_entry_set(&msrs
[n
++], MSR_KVM_ASYNC_PF_EN
,
865 env
->async_pf_en_msr
);
873 kvm_msr_entry_set(&msrs
[n
++], MSR_MCG_STATUS
, env
->mcg_status
);
874 kvm_msr_entry_set(&msrs
[n
++], MSR_MCG_CTL
, env
->mcg_ctl
);
875 for (i
= 0; i
< (env
->mcg_cap
& 0xff) * 4; i
++) {
876 kvm_msr_entry_set(&msrs
[n
++], MSR_MC0_CTL
+ i
, env
->mce_banks
[i
]);
881 msr_data
.info
.nmsrs
= n
;
883 return kvm_vcpu_ioctl(env
, KVM_SET_MSRS
, &msr_data
);
888 static int kvm_get_fpu(CPUState
*env
)
893 ret
= kvm_vcpu_ioctl(env
, KVM_GET_FPU
, &fpu
);
898 env
->fpstt
= (fpu
.fsw
>> 11) & 7;
901 for (i
= 0; i
< 8; ++i
) {
902 env
->fptags
[i
] = !((fpu
.ftwx
>> i
) & 1);
904 memcpy(env
->fpregs
, fpu
.fpr
, sizeof env
->fpregs
);
905 memcpy(env
->xmm_regs
, fpu
.xmm
, sizeof env
->xmm_regs
);
906 env
->mxcsr
= fpu
.mxcsr
;
911 static int kvm_get_xsave(CPUState
*env
)
914 struct kvm_xsave
* xsave
;
916 uint16_t cwd
, swd
, twd
, fop
;
918 if (!kvm_has_xsave()) {
919 return kvm_get_fpu(env
);
922 xsave
= qemu_memalign(4096, sizeof(struct kvm_xsave
));
923 ret
= kvm_vcpu_ioctl(env
, KVM_GET_XSAVE
, xsave
);
929 cwd
= (uint16_t)xsave
->region
[0];
930 swd
= (uint16_t)(xsave
->region
[0] >> 16);
931 twd
= (uint16_t)xsave
->region
[1];
932 fop
= (uint16_t)(xsave
->region
[1] >> 16);
933 env
->fpstt
= (swd
>> 11) & 7;
936 for (i
= 0; i
< 8; ++i
) {
937 env
->fptags
[i
] = !((twd
>> i
) & 1);
939 env
->mxcsr
= xsave
->region
[XSAVE_MXCSR
];
940 memcpy(env
->fpregs
, &xsave
->region
[XSAVE_ST_SPACE
],
942 memcpy(env
->xmm_regs
, &xsave
->region
[XSAVE_XMM_SPACE
],
943 sizeof env
->xmm_regs
);
944 env
->xstate_bv
= *(uint64_t *)&xsave
->region
[XSAVE_XSTATE_BV
];
945 memcpy(env
->ymmh_regs
, &xsave
->region
[XSAVE_YMMH_SPACE
],
946 sizeof env
->ymmh_regs
);
950 return kvm_get_fpu(env
);
954 static int kvm_get_xcrs(CPUState
*env
)
958 struct kvm_xcrs xcrs
;
960 if (!kvm_has_xcrs()) {
964 ret
= kvm_vcpu_ioctl(env
, KVM_GET_XCRS
, &xcrs
);
969 for (i
= 0; i
< xcrs
.nr_xcrs
; i
++) {
970 /* Only support xcr0 now */
971 if (xcrs
.xcrs
[0].xcr
== 0) {
972 env
->xcr0
= xcrs
.xcrs
[0].value
;
982 static int kvm_get_sregs(CPUState
*env
)
984 struct kvm_sregs sregs
;
988 ret
= kvm_vcpu_ioctl(env
, KVM_GET_SREGS
, &sregs
);
993 /* There can only be one pending IRQ set in the bitmap at a time, so try
994 to find it and save its number instead (-1 for none). */
995 env
->interrupt_injected
= -1;
996 for (i
= 0; i
< ARRAY_SIZE(sregs
.interrupt_bitmap
); i
++) {
997 if (sregs
.interrupt_bitmap
[i
]) {
998 bit
= ctz64(sregs
.interrupt_bitmap
[i
]);
999 env
->interrupt_injected
= i
* 64 + bit
;
1004 get_seg(&env
->segs
[R_CS
], &sregs
.cs
);
1005 get_seg(&env
->segs
[R_DS
], &sregs
.ds
);
1006 get_seg(&env
->segs
[R_ES
], &sregs
.es
);
1007 get_seg(&env
->segs
[R_FS
], &sregs
.fs
);
1008 get_seg(&env
->segs
[R_GS
], &sregs
.gs
);
1009 get_seg(&env
->segs
[R_SS
], &sregs
.ss
);
1011 get_seg(&env
->tr
, &sregs
.tr
);
1012 get_seg(&env
->ldt
, &sregs
.ldt
);
1014 env
->idt
.limit
= sregs
.idt
.limit
;
1015 env
->idt
.base
= sregs
.idt
.base
;
1016 env
->gdt
.limit
= sregs
.gdt
.limit
;
1017 env
->gdt
.base
= sregs
.gdt
.base
;
1019 env
->cr
[0] = sregs
.cr0
;
1020 env
->cr
[2] = sregs
.cr2
;
1021 env
->cr
[3] = sregs
.cr3
;
1022 env
->cr
[4] = sregs
.cr4
;
1024 cpu_set_apic_base(env
->apic_state
, sregs
.apic_base
);
1026 env
->efer
= sregs
.efer
;
1027 //cpu_set_apic_tpr(env->apic_state, sregs.cr8);
1029 #define HFLAG_COPY_MASK \
1030 ~( HF_CPL_MASK | HF_PE_MASK | HF_MP_MASK | HF_EM_MASK | \
1031 HF_TS_MASK | HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK | \
1032 HF_OSFXSR_MASK | HF_LMA_MASK | HF_CS32_MASK | \
1033 HF_SS32_MASK | HF_CS64_MASK | HF_ADDSEG_MASK)
1035 hflags
= (env
->segs
[R_CS
].flags
>> DESC_DPL_SHIFT
) & HF_CPL_MASK
;
1036 hflags
|= (env
->cr
[0] & CR0_PE_MASK
) << (HF_PE_SHIFT
- CR0_PE_SHIFT
);
1037 hflags
|= (env
->cr
[0] << (HF_MP_SHIFT
- CR0_MP_SHIFT
)) &
1038 (HF_MP_MASK
| HF_EM_MASK
| HF_TS_MASK
);
1039 hflags
|= (env
->eflags
& (HF_TF_MASK
| HF_VM_MASK
| HF_IOPL_MASK
));
1040 hflags
|= (env
->cr
[4] & CR4_OSFXSR_MASK
) <<
1041 (HF_OSFXSR_SHIFT
- CR4_OSFXSR_SHIFT
);
1043 if (env
->efer
& MSR_EFER_LMA
) {
1044 hflags
|= HF_LMA_MASK
;
1047 if ((hflags
& HF_LMA_MASK
) && (env
->segs
[R_CS
].flags
& DESC_L_MASK
)) {
1048 hflags
|= HF_CS32_MASK
| HF_SS32_MASK
| HF_CS64_MASK
;
1050 hflags
|= (env
->segs
[R_CS
].flags
& DESC_B_MASK
) >>
1051 (DESC_B_SHIFT
- HF_CS32_SHIFT
);
1052 hflags
|= (env
->segs
[R_SS
].flags
& DESC_B_MASK
) >>
1053 (DESC_B_SHIFT
- HF_SS32_SHIFT
);
1054 if (!(env
->cr
[0] & CR0_PE_MASK
) || (env
->eflags
& VM_MASK
) ||
1055 !(hflags
& HF_CS32_MASK
)) {
1056 hflags
|= HF_ADDSEG_MASK
;
1058 hflags
|= ((env
->segs
[R_DS
].base
| env
->segs
[R_ES
].base
|
1059 env
->segs
[R_SS
].base
) != 0) << HF_ADDSEG_SHIFT
;
1062 env
->hflags
= (env
->hflags
& HFLAG_COPY_MASK
) | hflags
;
1067 static int kvm_get_msrs(CPUState
*env
)
1070 struct kvm_msrs info
;
1071 struct kvm_msr_entry entries
[100];
1073 struct kvm_msr_entry
*msrs
= msr_data
.entries
;
1077 msrs
[n
++].index
= MSR_IA32_SYSENTER_CS
;
1078 msrs
[n
++].index
= MSR_IA32_SYSENTER_ESP
;
1079 msrs
[n
++].index
= MSR_IA32_SYSENTER_EIP
;
1081 msrs
[n
++].index
= MSR_STAR
;
1083 if (has_msr_hsave_pa
) {
1084 msrs
[n
++].index
= MSR_VM_HSAVE_PA
;
1087 if (!env
->tsc_valid
) {
1088 msrs
[n
++].index
= MSR_IA32_TSC
;
1089 env
->tsc_valid
= !vm_running
;
1092 #ifdef TARGET_X86_64
1093 if (lm_capable_kernel
) {
1094 msrs
[n
++].index
= MSR_CSTAR
;
1095 msrs
[n
++].index
= MSR_KERNELGSBASE
;
1096 msrs
[n
++].index
= MSR_FMASK
;
1097 msrs
[n
++].index
= MSR_LSTAR
;
1100 msrs
[n
++].index
= MSR_KVM_SYSTEM_TIME
;
1101 msrs
[n
++].index
= MSR_KVM_WALL_CLOCK
;
1102 #if defined(CONFIG_KVM_PARA) && defined(KVM_CAP_ASYNC_PF)
1103 if (has_msr_async_pf_en
) {
1104 msrs
[n
++].index
= MSR_KVM_ASYNC_PF_EN
;
1110 msrs
[n
++].index
= MSR_MCG_STATUS
;
1111 msrs
[n
++].index
= MSR_MCG_CTL
;
1112 for (i
= 0; i
< (env
->mcg_cap
& 0xff) * 4; i
++) {
1113 msrs
[n
++].index
= MSR_MC0_CTL
+ i
;
1118 msr_data
.info
.nmsrs
= n
;
1119 ret
= kvm_vcpu_ioctl(env
, KVM_GET_MSRS
, &msr_data
);
1124 for (i
= 0; i
< ret
; i
++) {
1125 switch (msrs
[i
].index
) {
1126 case MSR_IA32_SYSENTER_CS
:
1127 env
->sysenter_cs
= msrs
[i
].data
;
1129 case MSR_IA32_SYSENTER_ESP
:
1130 env
->sysenter_esp
= msrs
[i
].data
;
1132 case MSR_IA32_SYSENTER_EIP
:
1133 env
->sysenter_eip
= msrs
[i
].data
;
1136 env
->star
= msrs
[i
].data
;
1138 #ifdef TARGET_X86_64
1140 env
->cstar
= msrs
[i
].data
;
1142 case MSR_KERNELGSBASE
:
1143 env
->kernelgsbase
= msrs
[i
].data
;
1146 env
->fmask
= msrs
[i
].data
;
1149 env
->lstar
= msrs
[i
].data
;
1153 env
->tsc
= msrs
[i
].data
;
1155 case MSR_VM_HSAVE_PA
:
1156 env
->vm_hsave
= msrs
[i
].data
;
1158 case MSR_KVM_SYSTEM_TIME
:
1159 env
->system_time_msr
= msrs
[i
].data
;
1161 case MSR_KVM_WALL_CLOCK
:
1162 env
->wall_clock_msr
= msrs
[i
].data
;
1165 case MSR_MCG_STATUS
:
1166 env
->mcg_status
= msrs
[i
].data
;
1169 env
->mcg_ctl
= msrs
[i
].data
;
1174 if (msrs
[i
].index
>= MSR_MC0_CTL
&&
1175 msrs
[i
].index
< MSR_MC0_CTL
+ (env
->mcg_cap
& 0xff) * 4) {
1176 env
->mce_banks
[msrs
[i
].index
- MSR_MC0_CTL
] = msrs
[i
].data
;
1180 #if defined(CONFIG_KVM_PARA) && defined(KVM_CAP_ASYNC_PF)
1181 case MSR_KVM_ASYNC_PF_EN
:
1182 env
->async_pf_en_msr
= msrs
[i
].data
;
1191 static int kvm_put_mp_state(CPUState
*env
)
1193 struct kvm_mp_state mp_state
= { .mp_state
= env
->mp_state
};
1195 return kvm_vcpu_ioctl(env
, KVM_SET_MP_STATE
, &mp_state
);
1198 static int kvm_get_mp_state(CPUState
*env
)
1200 struct kvm_mp_state mp_state
;
1203 ret
= kvm_vcpu_ioctl(env
, KVM_GET_MP_STATE
, &mp_state
);
1207 env
->mp_state
= mp_state
.mp_state
;
1208 if (kvm_irqchip_in_kernel()) {
1209 env
->halted
= (mp_state
.mp_state
== KVM_MP_STATE_HALTED
);
1214 static int kvm_put_vcpu_events(CPUState
*env
, int level
)
1216 #ifdef KVM_CAP_VCPU_EVENTS
1217 struct kvm_vcpu_events events
;
1219 if (!kvm_has_vcpu_events()) {
1223 events
.exception
.injected
= (env
->exception_injected
>= 0);
1224 events
.exception
.nr
= env
->exception_injected
;
1225 events
.exception
.has_error_code
= env
->has_error_code
;
1226 events
.exception
.error_code
= env
->error_code
;
1228 events
.interrupt
.injected
= (env
->interrupt_injected
>= 0);
1229 events
.interrupt
.nr
= env
->interrupt_injected
;
1230 events
.interrupt
.soft
= env
->soft_interrupt
;
1232 events
.nmi
.injected
= env
->nmi_injected
;
1233 events
.nmi
.pending
= env
->nmi_pending
;
1234 events
.nmi
.masked
= !!(env
->hflags2
& HF2_NMI_MASK
);
1236 events
.sipi_vector
= env
->sipi_vector
;
1239 if (level
>= KVM_PUT_RESET_STATE
) {
1241 KVM_VCPUEVENT_VALID_NMI_PENDING
| KVM_VCPUEVENT_VALID_SIPI_VECTOR
;
1244 return kvm_vcpu_ioctl(env
, KVM_SET_VCPU_EVENTS
, &events
);
1250 static int kvm_get_vcpu_events(CPUState
*env
)
1252 #ifdef KVM_CAP_VCPU_EVENTS
1253 struct kvm_vcpu_events events
;
1256 if (!kvm_has_vcpu_events()) {
1260 ret
= kvm_vcpu_ioctl(env
, KVM_GET_VCPU_EVENTS
, &events
);
1264 env
->exception_injected
=
1265 events
.exception
.injected
? events
.exception
.nr
: -1;
1266 env
->has_error_code
= events
.exception
.has_error_code
;
1267 env
->error_code
= events
.exception
.error_code
;
1269 env
->interrupt_injected
=
1270 events
.interrupt
.injected
? events
.interrupt
.nr
: -1;
1271 env
->soft_interrupt
= events
.interrupt
.soft
;
1273 env
->nmi_injected
= events
.nmi
.injected
;
1274 env
->nmi_pending
= events
.nmi
.pending
;
1275 if (events
.nmi
.masked
) {
1276 env
->hflags2
|= HF2_NMI_MASK
;
1278 env
->hflags2
&= ~HF2_NMI_MASK
;
1281 env
->sipi_vector
= events
.sipi_vector
;
1287 static int kvm_guest_debug_workarounds(CPUState
*env
)
1290 #ifdef KVM_CAP_SET_GUEST_DEBUG
1291 unsigned long reinject_trap
= 0;
1293 if (!kvm_has_vcpu_events()) {
1294 if (env
->exception_injected
== 1) {
1295 reinject_trap
= KVM_GUESTDBG_INJECT_DB
;
1296 } else if (env
->exception_injected
== 3) {
1297 reinject_trap
= KVM_GUESTDBG_INJECT_BP
;
1299 env
->exception_injected
= -1;
1303 * Kernels before KVM_CAP_X86_ROBUST_SINGLESTEP overwrote flags.TF
1304 * injected via SET_GUEST_DEBUG while updating GP regs. Work around this
1305 * by updating the debug state once again if single-stepping is on.
1306 * Another reason to call kvm_update_guest_debug here is a pending debug
1307 * trap raise by the guest. On kernels without SET_VCPU_EVENTS we have to
1308 * reinject them via SET_GUEST_DEBUG.
1310 if (reinject_trap
||
1311 (!kvm_has_robust_singlestep() && env
->singlestep_enabled
)) {
1312 ret
= kvm_update_guest_debug(env
, reinject_trap
);
1314 #endif /* KVM_CAP_SET_GUEST_DEBUG */
1318 static int kvm_put_debugregs(CPUState
*env
)
1320 #ifdef KVM_CAP_DEBUGREGS
1321 struct kvm_debugregs dbgregs
;
1324 if (!kvm_has_debugregs()) {
1328 for (i
= 0; i
< 4; i
++) {
1329 dbgregs
.db
[i
] = env
->dr
[i
];
1331 dbgregs
.dr6
= env
->dr
[6];
1332 dbgregs
.dr7
= env
->dr
[7];
1335 return kvm_vcpu_ioctl(env
, KVM_SET_DEBUGREGS
, &dbgregs
);
1341 static int kvm_get_debugregs(CPUState
*env
)
1343 #ifdef KVM_CAP_DEBUGREGS
1344 struct kvm_debugregs dbgregs
;
1347 if (!kvm_has_debugregs()) {
1351 ret
= kvm_vcpu_ioctl(env
, KVM_GET_DEBUGREGS
, &dbgregs
);
1355 for (i
= 0; i
< 4; i
++) {
1356 env
->dr
[i
] = dbgregs
.db
[i
];
1358 env
->dr
[4] = env
->dr
[6] = dbgregs
.dr6
;
1359 env
->dr
[5] = env
->dr
[7] = dbgregs
.dr7
;
1365 int kvm_arch_put_registers(CPUState
*env
, int level
)
1369 assert(cpu_is_stopped(env
) || qemu_cpu_is_self(env
));
1371 ret
= kvm_getput_regs(env
, 1);
1375 ret
= kvm_put_xsave(env
);
1379 ret
= kvm_put_xcrs(env
);
1383 ret
= kvm_put_sregs(env
);
1387 /* must be before kvm_put_msrs */
1388 ret
= kvm_inject_mce_oldstyle(env
);
1392 ret
= kvm_put_msrs(env
, level
);
1396 if (level
>= KVM_PUT_RESET_STATE
) {
1397 ret
= kvm_put_mp_state(env
);
1402 ret
= kvm_put_vcpu_events(env
, level
);
1406 ret
= kvm_put_debugregs(env
);
1411 ret
= kvm_guest_debug_workarounds(env
);
1418 int kvm_arch_get_registers(CPUState
*env
)
1422 assert(cpu_is_stopped(env
) || qemu_cpu_is_self(env
));
1424 ret
= kvm_getput_regs(env
, 0);
1428 ret
= kvm_get_xsave(env
);
1432 ret
= kvm_get_xcrs(env
);
1436 ret
= kvm_get_sregs(env
);
1440 ret
= kvm_get_msrs(env
);
1444 ret
= kvm_get_mp_state(env
);
1448 ret
= kvm_get_vcpu_events(env
);
1452 ret
= kvm_get_debugregs(env
);
1459 void kvm_arch_pre_run(CPUState
*env
, struct kvm_run
*run
)
1464 if (env
->interrupt_request
& CPU_INTERRUPT_NMI
) {
1465 env
->interrupt_request
&= ~CPU_INTERRUPT_NMI
;
1466 DPRINTF("injected NMI\n");
1467 ret
= kvm_vcpu_ioctl(env
, KVM_NMI
);
1469 fprintf(stderr
, "KVM: injection failed, NMI lost (%s)\n",
1474 if (!kvm_irqchip_in_kernel()) {
1475 /* Force the VCPU out of its inner loop to process the INIT request */
1476 if (env
->interrupt_request
& CPU_INTERRUPT_INIT
) {
1477 env
->exit_request
= 1;
1480 /* Try to inject an interrupt if the guest can accept it */
1481 if (run
->ready_for_interrupt_injection
&&
1482 (env
->interrupt_request
& CPU_INTERRUPT_HARD
) &&
1483 (env
->eflags
& IF_MASK
)) {
1486 env
->interrupt_request
&= ~CPU_INTERRUPT_HARD
;
1487 irq
= cpu_get_pic_interrupt(env
);
1489 struct kvm_interrupt intr
;
1492 DPRINTF("injected interrupt %d\n", irq
);
1493 ret
= kvm_vcpu_ioctl(env
, KVM_INTERRUPT
, &intr
);
1496 "KVM: injection failed, interrupt lost (%s)\n",
1502 /* If we have an interrupt but the guest is not ready to receive an
1503 * interrupt, request an interrupt window exit. This will
1504 * cause a return to userspace as soon as the guest is ready to
1505 * receive interrupts. */
1506 if ((env
->interrupt_request
& CPU_INTERRUPT_HARD
)) {
1507 run
->request_interrupt_window
= 1;
1509 run
->request_interrupt_window
= 0;
1512 DPRINTF("setting tpr\n");
1513 run
->cr8
= cpu_get_apic_tpr(env
->apic_state
);
1517 void kvm_arch_post_run(CPUState
*env
, struct kvm_run
*run
)
1520 env
->eflags
|= IF_MASK
;
1522 env
->eflags
&= ~IF_MASK
;
1524 cpu_set_apic_tpr(env
->apic_state
, run
->cr8
);
1525 cpu_set_apic_base(env
->apic_state
, run
->apic_base
);
1528 int kvm_arch_process_async_events(CPUState
*env
)
1530 if (env
->interrupt_request
& CPU_INTERRUPT_MCE
) {
1531 /* We must not raise CPU_INTERRUPT_MCE if it's not supported. */
1532 assert(env
->mcg_cap
);
1534 env
->interrupt_request
&= ~CPU_INTERRUPT_MCE
;
1536 kvm_cpu_synchronize_state(env
);
1538 if (env
->exception_injected
== EXCP08_DBLE
) {
1539 /* this means triple fault */
1540 qemu_system_reset_request();
1541 env
->exit_request
= 1;
1544 env
->exception_injected
= EXCP12_MCHK
;
1545 env
->has_error_code
= 0;
1548 if (kvm_irqchip_in_kernel() && env
->mp_state
== KVM_MP_STATE_HALTED
) {
1549 env
->mp_state
= KVM_MP_STATE_RUNNABLE
;
1553 if (kvm_irqchip_in_kernel()) {
1557 if (env
->interrupt_request
& (CPU_INTERRUPT_HARD
| CPU_INTERRUPT_NMI
)) {
1560 if (env
->interrupt_request
& CPU_INTERRUPT_INIT
) {
1561 kvm_cpu_synchronize_state(env
);
1564 if (env
->interrupt_request
& CPU_INTERRUPT_SIPI
) {
1565 kvm_cpu_synchronize_state(env
);
1572 static int kvm_handle_halt(CPUState
*env
)
1574 if (!((env
->interrupt_request
& CPU_INTERRUPT_HARD
) &&
1575 (env
->eflags
& IF_MASK
)) &&
1576 !(env
->interrupt_request
& CPU_INTERRUPT_NMI
)) {
1584 static bool host_supports_vmx(void)
1586 uint32_t ecx
, unused
;
1588 host_cpuid(1, 0, &unused
, &unused
, &ecx
, &unused
);
1589 return ecx
& CPUID_EXT_VMX
;
1592 #define VMX_INVALID_GUEST_STATE 0x80000021
1594 int kvm_arch_handle_exit(CPUState
*env
, struct kvm_run
*run
)
1599 switch (run
->exit_reason
) {
1601 DPRINTF("handle_hlt\n");
1602 ret
= kvm_handle_halt(env
);
1604 case KVM_EXIT_SET_TPR
:
1607 case KVM_EXIT_FAIL_ENTRY
:
1608 code
= run
->fail_entry
.hardware_entry_failure_reason
;
1609 fprintf(stderr
, "KVM: entry failed, hardware error 0x%" PRIx64
"\n",
1611 if (host_supports_vmx() && code
== VMX_INVALID_GUEST_STATE
) {
1613 "\nIf you're runnning a guest on an Intel machine without "
1614 "unrestricted mode\n"
1615 "support, the failure can be most likely due to the guest "
1616 "entering an invalid\n"
1617 "state for Intel VT. For example, the guest maybe running "
1618 "in big real mode\n"
1619 "which is not supported on less recent Intel processors."
1624 case KVM_EXIT_EXCEPTION
:
1625 fprintf(stderr
, "KVM: exception %d exit (error code 0x%x)\n",
1626 run
->ex
.exception
, run
->ex
.error_code
);
1630 fprintf(stderr
, "KVM: unknown exit reason %d\n", run
->exit_reason
);
1638 #ifdef KVM_CAP_SET_GUEST_DEBUG
1639 int kvm_arch_insert_sw_breakpoint(CPUState
*env
, struct kvm_sw_breakpoint
*bp
)
1641 static const uint8_t int3
= 0xcc;
1643 if (cpu_memory_rw_debug(env
, bp
->pc
, (uint8_t *)&bp
->saved_insn
, 1, 0) ||
1644 cpu_memory_rw_debug(env
, bp
->pc
, (uint8_t *)&int3
, 1, 1)) {
1650 int kvm_arch_remove_sw_breakpoint(CPUState
*env
, struct kvm_sw_breakpoint
*bp
)
1654 if (cpu_memory_rw_debug(env
, bp
->pc
, &int3
, 1, 0) || int3
!= 0xcc ||
1655 cpu_memory_rw_debug(env
, bp
->pc
, (uint8_t *)&bp
->saved_insn
, 1, 1)) {
1667 static int nb_hw_breakpoint
;
1669 static int find_hw_breakpoint(target_ulong addr
, int len
, int type
)
1673 for (n
= 0; n
< nb_hw_breakpoint
; n
++) {
1674 if (hw_breakpoint
[n
].addr
== addr
&& hw_breakpoint
[n
].type
== type
&&
1675 (hw_breakpoint
[n
].len
== len
|| len
== -1)) {
1682 int kvm_arch_insert_hw_breakpoint(target_ulong addr
,
1683 target_ulong len
, int type
)
1686 case GDB_BREAKPOINT_HW
:
1689 case GDB_WATCHPOINT_WRITE
:
1690 case GDB_WATCHPOINT_ACCESS
:
1697 if (addr
& (len
- 1)) {
1709 if (nb_hw_breakpoint
== 4) {
1712 if (find_hw_breakpoint(addr
, len
, type
) >= 0) {
1715 hw_breakpoint
[nb_hw_breakpoint
].addr
= addr
;
1716 hw_breakpoint
[nb_hw_breakpoint
].len
= len
;
1717 hw_breakpoint
[nb_hw_breakpoint
].type
= type
;
1723 int kvm_arch_remove_hw_breakpoint(target_ulong addr
,
1724 target_ulong len
, int type
)
1728 n
= find_hw_breakpoint(addr
, (type
== GDB_BREAKPOINT_HW
) ? 1 : len
, type
);
1733 hw_breakpoint
[n
] = hw_breakpoint
[nb_hw_breakpoint
];
1738 void kvm_arch_remove_all_hw_breakpoints(void)
1740 nb_hw_breakpoint
= 0;
1743 static CPUWatchpoint hw_watchpoint
;
1745 int kvm_arch_debug(struct kvm_debug_exit_arch
*arch_info
)
1750 if (arch_info
->exception
== 1) {
1751 if (arch_info
->dr6
& (1 << 14)) {
1752 if (cpu_single_env
->singlestep_enabled
) {
1756 for (n
= 0; n
< 4; n
++) {
1757 if (arch_info
->dr6
& (1 << n
)) {
1758 switch ((arch_info
->dr7
>> (16 + n
*4)) & 0x3) {
1764 cpu_single_env
->watchpoint_hit
= &hw_watchpoint
;
1765 hw_watchpoint
.vaddr
= hw_breakpoint
[n
].addr
;
1766 hw_watchpoint
.flags
= BP_MEM_WRITE
;
1770 cpu_single_env
->watchpoint_hit
= &hw_watchpoint
;
1771 hw_watchpoint
.vaddr
= hw_breakpoint
[n
].addr
;
1772 hw_watchpoint
.flags
= BP_MEM_ACCESS
;
1778 } else if (kvm_find_sw_breakpoint(cpu_single_env
, arch_info
->pc
)) {
1782 cpu_synchronize_state(cpu_single_env
);
1783 assert(cpu_single_env
->exception_injected
== -1);
1785 cpu_single_env
->exception_injected
= arch_info
->exception
;
1786 cpu_single_env
->has_error_code
= 0;
1792 void kvm_arch_update_guest_debug(CPUState
*env
, struct kvm_guest_debug
*dbg
)
1794 const uint8_t type_code
[] = {
1795 [GDB_BREAKPOINT_HW
] = 0x0,
1796 [GDB_WATCHPOINT_WRITE
] = 0x1,
1797 [GDB_WATCHPOINT_ACCESS
] = 0x3
1799 const uint8_t len_code
[] = {
1800 [1] = 0x0, [2] = 0x1, [4] = 0x3, [8] = 0x2
1804 if (kvm_sw_breakpoints_active(env
)) {
1805 dbg
->control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_USE_SW_BP
;
1807 if (nb_hw_breakpoint
> 0) {
1808 dbg
->control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_USE_HW_BP
;
1809 dbg
->arch
.debugreg
[7] = 0x0600;
1810 for (n
= 0; n
< nb_hw_breakpoint
; n
++) {
1811 dbg
->arch
.debugreg
[n
] = hw_breakpoint
[n
].addr
;
1812 dbg
->arch
.debugreg
[7] |= (2 << (n
* 2)) |
1813 (type_code
[hw_breakpoint
[n
].type
] << (16 + n
*4)) |
1814 ((uint32_t)len_code
[hw_breakpoint
[n
].len
] << (18 + n
*4));
1818 #endif /* KVM_CAP_SET_GUEST_DEBUG */
1820 bool kvm_arch_stop_on_emulation_error(CPUState
*env
)
1822 return !(env
->cr
[0] & CR0_PE_MASK
) ||
1823 ((env
->segs
[R_CS
].selector
& 3) != 3);