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>
19 #include <linux/kvm.h>
21 #include "qemu-common.h"
26 #include "host-utils.h"
28 #ifdef CONFIG_KVM_PARA
29 #include <linux/kvm_para.h>
35 #define dprintf(fmt, ...) \
36 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
38 #define dprintf(fmt, ...) \
42 #define MSR_KVM_WALL_CLOCK 0x11
43 #define MSR_KVM_SYSTEM_TIME 0x12
45 #ifdef KVM_CAP_EXT_CPUID
47 static struct kvm_cpuid2
*try_get_cpuid(KVMState
*s
, int max
)
49 struct kvm_cpuid2
*cpuid
;
52 size
= sizeof(*cpuid
) + max
* sizeof(*cpuid
->entries
);
53 cpuid
= (struct kvm_cpuid2
*)qemu_mallocz(size
);
55 r
= kvm_ioctl(s
, KVM_GET_SUPPORTED_CPUID
, cpuid
);
56 if (r
== 0 && cpuid
->nent
>= max
) {
64 fprintf(stderr
, "KVM_GET_SUPPORTED_CPUID failed: %s\n",
72 uint32_t kvm_arch_get_supported_cpuid(CPUState
*env
, uint32_t function
, int reg
)
74 struct kvm_cpuid2
*cpuid
;
79 if (!kvm_check_extension(env
->kvm_state
, KVM_CAP_EXT_CPUID
)) {
84 while ((cpuid
= try_get_cpuid(env
->kvm_state
, max
)) == NULL
) {
88 for (i
= 0; i
< cpuid
->nent
; ++i
) {
89 if (cpuid
->entries
[i
].function
== function
) {
92 ret
= cpuid
->entries
[i
].eax
;
95 ret
= cpuid
->entries
[i
].ebx
;
98 ret
= cpuid
->entries
[i
].ecx
;
101 ret
= cpuid
->entries
[i
].edx
;
104 /* KVM before 2.6.30 misreports the following features */
105 ret
|= CPUID_MTRR
| CPUID_PAT
| CPUID_MCE
| CPUID_MCA
;
108 /* On Intel, kvm returns cpuid according to the Intel spec,
109 * so add missing bits according to the AMD spec:
111 cpuid_1_edx
= kvm_arch_get_supported_cpuid(env
, 1, R_EDX
);
112 ret
|= cpuid_1_edx
& 0xdfeff7ff;
127 uint32_t kvm_arch_get_supported_cpuid(CPUState
*env
, uint32_t function
, int reg
)
134 static void kvm_trim_features(uint32_t *features
, uint32_t supported
)
139 for (i
= 0; i
< 32; ++i
) {
141 if ((*features
& mask
) && !(supported
& mask
)) {
147 #ifdef CONFIG_KVM_PARA
148 struct kvm_para_features
{
151 } para_features
[] = {
152 #ifdef KVM_CAP_CLOCKSOURCE
153 { KVM_CAP_CLOCKSOURCE
, KVM_FEATURE_CLOCKSOURCE
},
155 #ifdef KVM_CAP_NOP_IO_DELAY
156 { KVM_CAP_NOP_IO_DELAY
, KVM_FEATURE_NOP_IO_DELAY
},
158 #ifdef KVM_CAP_PV_MMU
159 { KVM_CAP_PV_MMU
, KVM_FEATURE_MMU_OP
},
161 #ifdef KVM_CAP_CR3_CACHE
162 { KVM_CAP_CR3_CACHE
, KVM_FEATURE_CR3_CACHE
},
167 static int get_para_features(CPUState
*env
)
171 for (i
= 0; i
< ARRAY_SIZE(para_features
) - 1; i
++) {
172 if (kvm_check_extension(env
->kvm_state
, para_features
[i
].cap
))
173 features
|= (1 << para_features
[i
].feature
);
180 int kvm_arch_init_vcpu(CPUState
*env
)
183 struct kvm_cpuid2 cpuid
;
184 struct kvm_cpuid_entry2 entries
[100];
185 } __attribute__((packed
)) cpuid_data
;
186 uint32_t limit
, i
, j
, cpuid_i
;
188 struct kvm_cpuid_entry2
*c
;
189 #ifdef KVM_CPUID_SIGNATURE
190 uint32_t signature
[3];
193 env
->mp_state
= KVM_MP_STATE_RUNNABLE
;
195 kvm_trim_features(&env
->cpuid_features
,
196 kvm_arch_get_supported_cpuid(env
, 1, R_EDX
));
198 i
= env
->cpuid_ext_features
& CPUID_EXT_HYPERVISOR
;
199 kvm_trim_features(&env
->cpuid_ext_features
,
200 kvm_arch_get_supported_cpuid(env
, 1, R_ECX
));
201 env
->cpuid_ext_features
|= i
;
203 kvm_trim_features(&env
->cpuid_ext2_features
,
204 kvm_arch_get_supported_cpuid(env
, 0x80000001, R_EDX
));
205 kvm_trim_features(&env
->cpuid_ext3_features
,
206 kvm_arch_get_supported_cpuid(env
, 0x80000001, R_ECX
));
210 #ifdef CONFIG_KVM_PARA
211 /* Paravirtualization CPUIDs */
212 memcpy(signature
, "KVMKVMKVM\0\0\0", 12);
213 c
= &cpuid_data
.entries
[cpuid_i
++];
214 memset(c
, 0, sizeof(*c
));
215 c
->function
= KVM_CPUID_SIGNATURE
;
217 c
->ebx
= signature
[0];
218 c
->ecx
= signature
[1];
219 c
->edx
= signature
[2];
221 c
= &cpuid_data
.entries
[cpuid_i
++];
222 memset(c
, 0, sizeof(*c
));
223 c
->function
= KVM_CPUID_FEATURES
;
224 c
->eax
= env
->cpuid_kvm_features
& get_para_features(env
);
227 cpu_x86_cpuid(env
, 0, 0, &limit
, &unused
, &unused
, &unused
);
229 for (i
= 0; i
<= limit
; i
++) {
230 c
= &cpuid_data
.entries
[cpuid_i
++];
234 /* Keep reading function 2 till all the input is received */
238 c
->flags
= KVM_CPUID_FLAG_STATEFUL_FUNC
|
239 KVM_CPUID_FLAG_STATE_READ_NEXT
;
240 cpu_x86_cpuid(env
, i
, 0, &c
->eax
, &c
->ebx
, &c
->ecx
, &c
->edx
);
241 times
= c
->eax
& 0xff;
243 for (j
= 1; j
< times
; ++j
) {
244 c
= &cpuid_data
.entries
[cpuid_i
++];
246 c
->flags
= KVM_CPUID_FLAG_STATEFUL_FUNC
;
247 cpu_x86_cpuid(env
, i
, 0, &c
->eax
, &c
->ebx
, &c
->ecx
, &c
->edx
);
256 c
->flags
= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
258 cpu_x86_cpuid(env
, i
, j
, &c
->eax
, &c
->ebx
, &c
->ecx
, &c
->edx
);
260 if (i
== 4 && c
->eax
== 0)
262 if (i
== 0xb && !(c
->ecx
& 0xff00))
264 if (i
== 0xd && c
->eax
== 0)
267 c
= &cpuid_data
.entries
[cpuid_i
++];
273 cpu_x86_cpuid(env
, i
, 0, &c
->eax
, &c
->ebx
, &c
->ecx
, &c
->edx
);
277 cpu_x86_cpuid(env
, 0x80000000, 0, &limit
, &unused
, &unused
, &unused
);
279 for (i
= 0x80000000; i
<= limit
; i
++) {
280 c
= &cpuid_data
.entries
[cpuid_i
++];
284 cpu_x86_cpuid(env
, i
, 0, &c
->eax
, &c
->ebx
, &c
->ecx
, &c
->edx
);
287 cpuid_data
.cpuid
.nent
= cpuid_i
;
289 return kvm_vcpu_ioctl(env
, KVM_SET_CPUID2
, &cpuid_data
);
292 void kvm_arch_reset_vcpu(CPUState
*env
)
294 env
->exception_injected
= -1;
295 env
->interrupt_injected
= -1;
296 env
->nmi_injected
= 0;
297 env
->nmi_pending
= 0;
300 static int kvm_has_msr_star(CPUState
*env
)
302 static int has_msr_star
;
306 if (has_msr_star
== 0) {
307 struct kvm_msr_list msr_list
, *kvm_msr_list
;
311 /* Obtain MSR list from KVM. These are the MSRs that we must
314 ret
= kvm_ioctl(env
->kvm_state
, KVM_GET_MSR_INDEX_LIST
, &msr_list
);
315 if (ret
< 0 && ret
!= -E2BIG
) {
318 /* Old kernel modules had a bug and could write beyond the provided
319 memory. Allocate at least a safe amount of 1K. */
320 kvm_msr_list
= qemu_mallocz(MAX(1024, sizeof(msr_list
) +
322 sizeof(msr_list
.indices
[0])));
324 kvm_msr_list
->nmsrs
= msr_list
.nmsrs
;
325 ret
= kvm_ioctl(env
->kvm_state
, KVM_GET_MSR_INDEX_LIST
, kvm_msr_list
);
329 for (i
= 0; i
< kvm_msr_list
->nmsrs
; i
++) {
330 if (kvm_msr_list
->indices
[i
] == MSR_STAR
) {
340 if (has_msr_star
== 1)
345 int kvm_arch_init(KVMState
*s
, int smp_cpus
)
349 /* create vm86 tss. KVM uses vm86 mode to emulate 16-bit code
350 * directly. In order to use vm86 mode, a TSS is needed. Since this
351 * must be part of guest physical memory, we need to allocate it. Older
352 * versions of KVM just assumed that it would be at the end of physical
353 * memory but that doesn't work with more than 4GB of memory. We simply
354 * refuse to work with those older versions of KVM. */
355 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, KVM_CAP_SET_TSS_ADDR
);
357 fprintf(stderr
, "kvm does not support KVM_CAP_SET_TSS_ADDR\n");
361 /* this address is 3 pages before the bios, and the bios should present
362 * as unavaible memory. FIXME, need to ensure the e820 map deals with
365 return kvm_vm_ioctl(s
, KVM_SET_TSS_ADDR
, 0xfffbd000);
368 static void set_v8086_seg(struct kvm_segment
*lhs
, const SegmentCache
*rhs
)
370 lhs
->selector
= rhs
->selector
;
371 lhs
->base
= rhs
->base
;
372 lhs
->limit
= rhs
->limit
;
384 static void set_seg(struct kvm_segment
*lhs
, const SegmentCache
*rhs
)
386 unsigned flags
= rhs
->flags
;
387 lhs
->selector
= rhs
->selector
;
388 lhs
->base
= rhs
->base
;
389 lhs
->limit
= rhs
->limit
;
390 lhs
->type
= (flags
>> DESC_TYPE_SHIFT
) & 15;
391 lhs
->present
= (flags
& DESC_P_MASK
) != 0;
392 lhs
->dpl
= rhs
->selector
& 3;
393 lhs
->db
= (flags
>> DESC_B_SHIFT
) & 1;
394 lhs
->s
= (flags
& DESC_S_MASK
) != 0;
395 lhs
->l
= (flags
>> DESC_L_SHIFT
) & 1;
396 lhs
->g
= (flags
& DESC_G_MASK
) != 0;
397 lhs
->avl
= (flags
& DESC_AVL_MASK
) != 0;
401 static void get_seg(SegmentCache
*lhs
, const struct kvm_segment
*rhs
)
403 lhs
->selector
= rhs
->selector
;
404 lhs
->base
= rhs
->base
;
405 lhs
->limit
= rhs
->limit
;
407 (rhs
->type
<< DESC_TYPE_SHIFT
)
408 | (rhs
->present
* DESC_P_MASK
)
409 | (rhs
->dpl
<< DESC_DPL_SHIFT
)
410 | (rhs
->db
<< DESC_B_SHIFT
)
411 | (rhs
->s
* DESC_S_MASK
)
412 | (rhs
->l
<< DESC_L_SHIFT
)
413 | (rhs
->g
* DESC_G_MASK
)
414 | (rhs
->avl
* DESC_AVL_MASK
);
417 static void kvm_getput_reg(__u64
*kvm_reg
, target_ulong
*qemu_reg
, int set
)
420 *kvm_reg
= *qemu_reg
;
422 *qemu_reg
= *kvm_reg
;
425 static int kvm_getput_regs(CPUState
*env
, int set
)
427 struct kvm_regs regs
;
431 ret
= kvm_vcpu_ioctl(env
, KVM_GET_REGS
, ®s
);
436 kvm_getput_reg(®s
.rax
, &env
->regs
[R_EAX
], set
);
437 kvm_getput_reg(®s
.rbx
, &env
->regs
[R_EBX
], set
);
438 kvm_getput_reg(®s
.rcx
, &env
->regs
[R_ECX
], set
);
439 kvm_getput_reg(®s
.rdx
, &env
->regs
[R_EDX
], set
);
440 kvm_getput_reg(®s
.rsi
, &env
->regs
[R_ESI
], set
);
441 kvm_getput_reg(®s
.rdi
, &env
->regs
[R_EDI
], set
);
442 kvm_getput_reg(®s
.rsp
, &env
->regs
[R_ESP
], set
);
443 kvm_getput_reg(®s
.rbp
, &env
->regs
[R_EBP
], set
);
445 kvm_getput_reg(®s
.r8
, &env
->regs
[8], set
);
446 kvm_getput_reg(®s
.r9
, &env
->regs
[9], set
);
447 kvm_getput_reg(®s
.r10
, &env
->regs
[10], set
);
448 kvm_getput_reg(®s
.r11
, &env
->regs
[11], set
);
449 kvm_getput_reg(®s
.r12
, &env
->regs
[12], set
);
450 kvm_getput_reg(®s
.r13
, &env
->regs
[13], set
);
451 kvm_getput_reg(®s
.r14
, &env
->regs
[14], set
);
452 kvm_getput_reg(®s
.r15
, &env
->regs
[15], set
);
455 kvm_getput_reg(®s
.rflags
, &env
->eflags
, set
);
456 kvm_getput_reg(®s
.rip
, &env
->eip
, set
);
459 ret
= kvm_vcpu_ioctl(env
, KVM_SET_REGS
, ®s
);
464 static int kvm_put_fpu(CPUState
*env
)
469 memset(&fpu
, 0, sizeof fpu
);
470 fpu
.fsw
= env
->fpus
& ~(7 << 11);
471 fpu
.fsw
|= (env
->fpstt
& 7) << 11;
473 for (i
= 0; i
< 8; ++i
)
474 fpu
.ftwx
|= (!env
->fptags
[i
]) << i
;
475 memcpy(fpu
.fpr
, env
->fpregs
, sizeof env
->fpregs
);
476 memcpy(fpu
.xmm
, env
->xmm_regs
, sizeof env
->xmm_regs
);
477 fpu
.mxcsr
= env
->mxcsr
;
479 return kvm_vcpu_ioctl(env
, KVM_SET_FPU
, &fpu
);
482 static int kvm_put_sregs(CPUState
*env
)
484 struct kvm_sregs sregs
;
486 memset(sregs
.interrupt_bitmap
, 0, sizeof(sregs
.interrupt_bitmap
));
487 if (env
->interrupt_injected
>= 0) {
488 sregs
.interrupt_bitmap
[env
->interrupt_injected
/ 64] |=
489 (uint64_t)1 << (env
->interrupt_injected
% 64);
492 if ((env
->eflags
& VM_MASK
)) {
493 set_v8086_seg(&sregs
.cs
, &env
->segs
[R_CS
]);
494 set_v8086_seg(&sregs
.ds
, &env
->segs
[R_DS
]);
495 set_v8086_seg(&sregs
.es
, &env
->segs
[R_ES
]);
496 set_v8086_seg(&sregs
.fs
, &env
->segs
[R_FS
]);
497 set_v8086_seg(&sregs
.gs
, &env
->segs
[R_GS
]);
498 set_v8086_seg(&sregs
.ss
, &env
->segs
[R_SS
]);
500 set_seg(&sregs
.cs
, &env
->segs
[R_CS
]);
501 set_seg(&sregs
.ds
, &env
->segs
[R_DS
]);
502 set_seg(&sregs
.es
, &env
->segs
[R_ES
]);
503 set_seg(&sregs
.fs
, &env
->segs
[R_FS
]);
504 set_seg(&sregs
.gs
, &env
->segs
[R_GS
]);
505 set_seg(&sregs
.ss
, &env
->segs
[R_SS
]);
507 if (env
->cr
[0] & CR0_PE_MASK
) {
508 /* force ss cpl to cs cpl */
509 sregs
.ss
.selector
= (sregs
.ss
.selector
& ~3) |
510 (sregs
.cs
.selector
& 3);
511 sregs
.ss
.dpl
= sregs
.ss
.selector
& 3;
515 set_seg(&sregs
.tr
, &env
->tr
);
516 set_seg(&sregs
.ldt
, &env
->ldt
);
518 sregs
.idt
.limit
= env
->idt
.limit
;
519 sregs
.idt
.base
= env
->idt
.base
;
520 sregs
.gdt
.limit
= env
->gdt
.limit
;
521 sregs
.gdt
.base
= env
->gdt
.base
;
523 sregs
.cr0
= env
->cr
[0];
524 sregs
.cr2
= env
->cr
[2];
525 sregs
.cr3
= env
->cr
[3];
526 sregs
.cr4
= env
->cr
[4];
528 sregs
.cr8
= cpu_get_apic_tpr(env
);
529 sregs
.apic_base
= cpu_get_apic_base(env
);
531 sregs
.efer
= env
->efer
;
533 return kvm_vcpu_ioctl(env
, KVM_SET_SREGS
, &sregs
);
536 static void kvm_msr_entry_set(struct kvm_msr_entry
*entry
,
537 uint32_t index
, uint64_t value
)
539 entry
->index
= index
;
543 static int kvm_put_msrs(CPUState
*env
)
546 struct kvm_msrs info
;
547 struct kvm_msr_entry entries
[100];
549 struct kvm_msr_entry
*msrs
= msr_data
.entries
;
552 kvm_msr_entry_set(&msrs
[n
++], MSR_IA32_SYSENTER_CS
, env
->sysenter_cs
);
553 kvm_msr_entry_set(&msrs
[n
++], MSR_IA32_SYSENTER_ESP
, env
->sysenter_esp
);
554 kvm_msr_entry_set(&msrs
[n
++], MSR_IA32_SYSENTER_EIP
, env
->sysenter_eip
);
555 if (kvm_has_msr_star(env
))
556 kvm_msr_entry_set(&msrs
[n
++], MSR_STAR
, env
->star
);
557 kvm_msr_entry_set(&msrs
[n
++], MSR_IA32_TSC
, env
->tsc
);
559 /* FIXME if lm capable */
560 kvm_msr_entry_set(&msrs
[n
++], MSR_CSTAR
, env
->cstar
);
561 kvm_msr_entry_set(&msrs
[n
++], MSR_KERNELGSBASE
, env
->kernelgsbase
);
562 kvm_msr_entry_set(&msrs
[n
++], MSR_FMASK
, env
->fmask
);
563 kvm_msr_entry_set(&msrs
[n
++], MSR_LSTAR
, env
->lstar
);
565 kvm_msr_entry_set(&msrs
[n
++], MSR_KVM_SYSTEM_TIME
, env
->system_time_msr
);
566 kvm_msr_entry_set(&msrs
[n
++], MSR_KVM_WALL_CLOCK
, env
->wall_clock_msr
);
568 msr_data
.info
.nmsrs
= n
;
570 return kvm_vcpu_ioctl(env
, KVM_SET_MSRS
, &msr_data
);
575 static int kvm_get_fpu(CPUState
*env
)
580 ret
= kvm_vcpu_ioctl(env
, KVM_GET_FPU
, &fpu
);
584 env
->fpstt
= (fpu
.fsw
>> 11) & 7;
587 for (i
= 0; i
< 8; ++i
)
588 env
->fptags
[i
] = !((fpu
.ftwx
>> i
) & 1);
589 memcpy(env
->fpregs
, fpu
.fpr
, sizeof env
->fpregs
);
590 memcpy(env
->xmm_regs
, fpu
.xmm
, sizeof env
->xmm_regs
);
591 env
->mxcsr
= fpu
.mxcsr
;
596 static int kvm_get_sregs(CPUState
*env
)
598 struct kvm_sregs sregs
;
602 ret
= kvm_vcpu_ioctl(env
, KVM_GET_SREGS
, &sregs
);
606 /* There can only be one pending IRQ set in the bitmap at a time, so try
607 to find it and save its number instead (-1 for none). */
608 env
->interrupt_injected
= -1;
609 for (i
= 0; i
< ARRAY_SIZE(sregs
.interrupt_bitmap
); i
++) {
610 if (sregs
.interrupt_bitmap
[i
]) {
611 bit
= ctz64(sregs
.interrupt_bitmap
[i
]);
612 env
->interrupt_injected
= i
* 64 + bit
;
617 get_seg(&env
->segs
[R_CS
], &sregs
.cs
);
618 get_seg(&env
->segs
[R_DS
], &sregs
.ds
);
619 get_seg(&env
->segs
[R_ES
], &sregs
.es
);
620 get_seg(&env
->segs
[R_FS
], &sregs
.fs
);
621 get_seg(&env
->segs
[R_GS
], &sregs
.gs
);
622 get_seg(&env
->segs
[R_SS
], &sregs
.ss
);
624 get_seg(&env
->tr
, &sregs
.tr
);
625 get_seg(&env
->ldt
, &sregs
.ldt
);
627 env
->idt
.limit
= sregs
.idt
.limit
;
628 env
->idt
.base
= sregs
.idt
.base
;
629 env
->gdt
.limit
= sregs
.gdt
.limit
;
630 env
->gdt
.base
= sregs
.gdt
.base
;
632 env
->cr
[0] = sregs
.cr0
;
633 env
->cr
[2] = sregs
.cr2
;
634 env
->cr
[3] = sregs
.cr3
;
635 env
->cr
[4] = sregs
.cr4
;
637 cpu_set_apic_base(env
, sregs
.apic_base
);
639 env
->efer
= sregs
.efer
;
640 //cpu_set_apic_tpr(env, sregs.cr8);
642 #define HFLAG_COPY_MASK ~( \
643 HF_CPL_MASK | HF_PE_MASK | HF_MP_MASK | HF_EM_MASK | \
644 HF_TS_MASK | HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK | \
645 HF_OSFXSR_MASK | HF_LMA_MASK | HF_CS32_MASK | \
646 HF_SS32_MASK | HF_CS64_MASK | HF_ADDSEG_MASK)
650 hflags
= (env
->segs
[R_CS
].flags
>> DESC_DPL_SHIFT
) & HF_CPL_MASK
;
651 hflags
|= (env
->cr
[0] & CR0_PE_MASK
) << (HF_PE_SHIFT
- CR0_PE_SHIFT
);
652 hflags
|= (env
->cr
[0] << (HF_MP_SHIFT
- CR0_MP_SHIFT
)) &
653 (HF_MP_MASK
| HF_EM_MASK
| HF_TS_MASK
);
654 hflags
|= (env
->eflags
& (HF_TF_MASK
| HF_VM_MASK
| HF_IOPL_MASK
));
655 hflags
|= (env
->cr
[4] & CR4_OSFXSR_MASK
) <<
656 (HF_OSFXSR_SHIFT
- CR4_OSFXSR_SHIFT
);
658 if (env
->efer
& MSR_EFER_LMA
) {
659 hflags
|= HF_LMA_MASK
;
662 if ((hflags
& HF_LMA_MASK
) && (env
->segs
[R_CS
].flags
& DESC_L_MASK
)) {
663 hflags
|= HF_CS32_MASK
| HF_SS32_MASK
| HF_CS64_MASK
;
665 hflags
|= (env
->segs
[R_CS
].flags
& DESC_B_MASK
) >>
666 (DESC_B_SHIFT
- HF_CS32_SHIFT
);
667 hflags
|= (env
->segs
[R_SS
].flags
& DESC_B_MASK
) >>
668 (DESC_B_SHIFT
- HF_SS32_SHIFT
);
669 if (!(env
->cr
[0] & CR0_PE_MASK
) ||
670 (env
->eflags
& VM_MASK
) ||
671 !(hflags
& HF_CS32_MASK
)) {
672 hflags
|= HF_ADDSEG_MASK
;
674 hflags
|= ((env
->segs
[R_DS
].base
|
675 env
->segs
[R_ES
].base
|
676 env
->segs
[R_SS
].base
) != 0) <<
680 env
->hflags
= (env
->hflags
& HFLAG_COPY_MASK
) | hflags
;
685 static int kvm_get_msrs(CPUState
*env
)
688 struct kvm_msrs info
;
689 struct kvm_msr_entry entries
[100];
691 struct kvm_msr_entry
*msrs
= msr_data
.entries
;
695 msrs
[n
++].index
= MSR_IA32_SYSENTER_CS
;
696 msrs
[n
++].index
= MSR_IA32_SYSENTER_ESP
;
697 msrs
[n
++].index
= MSR_IA32_SYSENTER_EIP
;
698 if (kvm_has_msr_star(env
))
699 msrs
[n
++].index
= MSR_STAR
;
700 msrs
[n
++].index
= MSR_IA32_TSC
;
702 /* FIXME lm_capable_kernel */
703 msrs
[n
++].index
= MSR_CSTAR
;
704 msrs
[n
++].index
= MSR_KERNELGSBASE
;
705 msrs
[n
++].index
= MSR_FMASK
;
706 msrs
[n
++].index
= MSR_LSTAR
;
708 msrs
[n
++].index
= MSR_KVM_SYSTEM_TIME
;
709 msrs
[n
++].index
= MSR_KVM_WALL_CLOCK
;
711 msr_data
.info
.nmsrs
= n
;
712 ret
= kvm_vcpu_ioctl(env
, KVM_GET_MSRS
, &msr_data
);
716 for (i
= 0; i
< ret
; i
++) {
717 switch (msrs
[i
].index
) {
718 case MSR_IA32_SYSENTER_CS
:
719 env
->sysenter_cs
= msrs
[i
].data
;
721 case MSR_IA32_SYSENTER_ESP
:
722 env
->sysenter_esp
= msrs
[i
].data
;
724 case MSR_IA32_SYSENTER_EIP
:
725 env
->sysenter_eip
= msrs
[i
].data
;
728 env
->star
= msrs
[i
].data
;
732 env
->cstar
= msrs
[i
].data
;
734 case MSR_KERNELGSBASE
:
735 env
->kernelgsbase
= msrs
[i
].data
;
738 env
->fmask
= msrs
[i
].data
;
741 env
->lstar
= msrs
[i
].data
;
745 env
->tsc
= msrs
[i
].data
;
747 case MSR_KVM_SYSTEM_TIME
:
748 env
->system_time_msr
= msrs
[i
].data
;
750 case MSR_KVM_WALL_CLOCK
:
751 env
->wall_clock_msr
= msrs
[i
].data
;
759 static int kvm_put_mp_state(CPUState
*env
)
761 struct kvm_mp_state mp_state
= { .mp_state
= env
->mp_state
};
763 return kvm_vcpu_ioctl(env
, KVM_SET_MP_STATE
, &mp_state
);
766 static int kvm_get_mp_state(CPUState
*env
)
768 struct kvm_mp_state mp_state
;
771 ret
= kvm_vcpu_ioctl(env
, KVM_GET_MP_STATE
, &mp_state
);
775 env
->mp_state
= mp_state
.mp_state
;
779 static int kvm_put_vcpu_events(CPUState
*env
)
781 #ifdef KVM_CAP_VCPU_EVENTS
782 struct kvm_vcpu_events events
;
784 if (!kvm_has_vcpu_events()) {
788 events
.exception
.injected
= (env
->exception_injected
>= 0);
789 events
.exception
.nr
= env
->exception_injected
;
790 events
.exception
.has_error_code
= env
->has_error_code
;
791 events
.exception
.error_code
= env
->error_code
;
793 events
.interrupt
.injected
= (env
->interrupt_injected
>= 0);
794 events
.interrupt
.nr
= env
->interrupt_injected
;
795 events
.interrupt
.soft
= env
->soft_interrupt
;
797 events
.nmi
.injected
= env
->nmi_injected
;
798 events
.nmi
.pending
= env
->nmi_pending
;
799 events
.nmi
.masked
= !!(env
->hflags2
& HF2_NMI_MASK
);
801 events
.sipi_vector
= env
->sipi_vector
;
804 KVM_VCPUEVENT_VALID_NMI_PENDING
| KVM_VCPUEVENT_VALID_SIPI_VECTOR
;
806 return kvm_vcpu_ioctl(env
, KVM_SET_VCPU_EVENTS
, &events
);
812 static int kvm_get_vcpu_events(CPUState
*env
)
814 #ifdef KVM_CAP_VCPU_EVENTS
815 struct kvm_vcpu_events events
;
818 if (!kvm_has_vcpu_events()) {
822 ret
= kvm_vcpu_ioctl(env
, KVM_GET_VCPU_EVENTS
, &events
);
826 env
->exception_injected
=
827 events
.exception
.injected
? events
.exception
.nr
: -1;
828 env
->has_error_code
= events
.exception
.has_error_code
;
829 env
->error_code
= events
.exception
.error_code
;
831 env
->interrupt_injected
=
832 events
.interrupt
.injected
? events
.interrupt
.nr
: -1;
833 env
->soft_interrupt
= events
.interrupt
.soft
;
835 env
->nmi_injected
= events
.nmi
.injected
;
836 env
->nmi_pending
= events
.nmi
.pending
;
837 if (events
.nmi
.masked
) {
838 env
->hflags2
|= HF2_NMI_MASK
;
840 env
->hflags2
&= ~HF2_NMI_MASK
;
843 env
->sipi_vector
= events
.sipi_vector
;
849 int kvm_arch_put_registers(CPUState
*env
)
853 ret
= kvm_getput_regs(env
, 1);
857 ret
= kvm_put_fpu(env
);
861 ret
= kvm_put_sregs(env
);
865 ret
= kvm_put_msrs(env
);
869 ret
= kvm_put_mp_state(env
);
873 ret
= kvm_put_vcpu_events(env
);
880 int kvm_arch_get_registers(CPUState
*env
)
884 ret
= kvm_getput_regs(env
, 0);
888 ret
= kvm_get_fpu(env
);
892 ret
= kvm_get_sregs(env
);
896 ret
= kvm_get_msrs(env
);
900 ret
= kvm_get_mp_state(env
);
904 ret
= kvm_get_vcpu_events(env
);
911 int kvm_arch_pre_run(CPUState
*env
, struct kvm_run
*run
)
913 /* Try to inject an interrupt if the guest can accept it */
914 if (run
->ready_for_interrupt_injection
&&
915 (env
->interrupt_request
& CPU_INTERRUPT_HARD
) &&
916 (env
->eflags
& IF_MASK
)) {
919 env
->interrupt_request
&= ~CPU_INTERRUPT_HARD
;
920 irq
= cpu_get_pic_interrupt(env
);
922 struct kvm_interrupt intr
;
925 dprintf("injected interrupt %d\n", irq
);
926 kvm_vcpu_ioctl(env
, KVM_INTERRUPT
, &intr
);
930 /* If we have an interrupt but the guest is not ready to receive an
931 * interrupt, request an interrupt window exit. This will
932 * cause a return to userspace as soon as the guest is ready to
933 * receive interrupts. */
934 if ((env
->interrupt_request
& CPU_INTERRUPT_HARD
))
935 run
->request_interrupt_window
= 1;
937 run
->request_interrupt_window
= 0;
939 dprintf("setting tpr\n");
940 run
->cr8
= cpu_get_apic_tpr(env
);
945 int kvm_arch_post_run(CPUState
*env
, struct kvm_run
*run
)
948 env
->eflags
|= IF_MASK
;
950 env
->eflags
&= ~IF_MASK
;
952 cpu_set_apic_tpr(env
, run
->cr8
);
953 cpu_set_apic_base(env
, run
->apic_base
);
958 static int kvm_handle_halt(CPUState
*env
)
960 if (!((env
->interrupt_request
& CPU_INTERRUPT_HARD
) &&
961 (env
->eflags
& IF_MASK
)) &&
962 !(env
->interrupt_request
& CPU_INTERRUPT_NMI
)) {
964 env
->exception_index
= EXCP_HLT
;
971 int kvm_arch_handle_exit(CPUState
*env
, struct kvm_run
*run
)
975 switch (run
->exit_reason
) {
977 dprintf("handle_hlt\n");
978 ret
= kvm_handle_halt(env
);
985 #ifdef KVM_CAP_SET_GUEST_DEBUG
986 int kvm_arch_insert_sw_breakpoint(CPUState
*env
, struct kvm_sw_breakpoint
*bp
)
988 static const uint8_t int3
= 0xcc;
990 if (cpu_memory_rw_debug(env
, bp
->pc
, (uint8_t *)&bp
->saved_insn
, 1, 0) ||
991 cpu_memory_rw_debug(env
, bp
->pc
, (uint8_t *)&int3
, 1, 1))
996 int kvm_arch_remove_sw_breakpoint(CPUState
*env
, struct kvm_sw_breakpoint
*bp
)
1000 if (cpu_memory_rw_debug(env
, bp
->pc
, &int3
, 1, 0) || int3
!= 0xcc ||
1001 cpu_memory_rw_debug(env
, bp
->pc
, (uint8_t *)&bp
->saved_insn
, 1, 1))
1012 static int nb_hw_breakpoint
;
1014 static int find_hw_breakpoint(target_ulong addr
, int len
, int type
)
1018 for (n
= 0; n
< nb_hw_breakpoint
; n
++)
1019 if (hw_breakpoint
[n
].addr
== addr
&& hw_breakpoint
[n
].type
== type
&&
1020 (hw_breakpoint
[n
].len
== len
|| len
== -1))
1025 int kvm_arch_insert_hw_breakpoint(target_ulong addr
,
1026 target_ulong len
, int type
)
1029 case GDB_BREAKPOINT_HW
:
1032 case GDB_WATCHPOINT_WRITE
:
1033 case GDB_WATCHPOINT_ACCESS
:
1040 if (addr
& (len
- 1))
1051 if (nb_hw_breakpoint
== 4)
1054 if (find_hw_breakpoint(addr
, len
, type
) >= 0)
1057 hw_breakpoint
[nb_hw_breakpoint
].addr
= addr
;
1058 hw_breakpoint
[nb_hw_breakpoint
].len
= len
;
1059 hw_breakpoint
[nb_hw_breakpoint
].type
= type
;
1065 int kvm_arch_remove_hw_breakpoint(target_ulong addr
,
1066 target_ulong len
, int type
)
1070 n
= find_hw_breakpoint(addr
, (type
== GDB_BREAKPOINT_HW
) ? 1 : len
, type
);
1075 hw_breakpoint
[n
] = hw_breakpoint
[nb_hw_breakpoint
];
1080 void kvm_arch_remove_all_hw_breakpoints(void)
1082 nb_hw_breakpoint
= 0;
1085 static CPUWatchpoint hw_watchpoint
;
1087 int kvm_arch_debug(struct kvm_debug_exit_arch
*arch_info
)
1092 if (arch_info
->exception
== 1) {
1093 if (arch_info
->dr6
& (1 << 14)) {
1094 if (cpu_single_env
->singlestep_enabled
)
1097 for (n
= 0; n
< 4; n
++)
1098 if (arch_info
->dr6
& (1 << n
))
1099 switch ((arch_info
->dr7
>> (16 + n
*4)) & 0x3) {
1105 cpu_single_env
->watchpoint_hit
= &hw_watchpoint
;
1106 hw_watchpoint
.vaddr
= hw_breakpoint
[n
].addr
;
1107 hw_watchpoint
.flags
= BP_MEM_WRITE
;
1111 cpu_single_env
->watchpoint_hit
= &hw_watchpoint
;
1112 hw_watchpoint
.vaddr
= hw_breakpoint
[n
].addr
;
1113 hw_watchpoint
.flags
= BP_MEM_ACCESS
;
1117 } else if (kvm_find_sw_breakpoint(cpu_single_env
, arch_info
->pc
))
1121 kvm_update_guest_debug(cpu_single_env
,
1122 (arch_info
->exception
== 1) ?
1123 KVM_GUESTDBG_INJECT_DB
: KVM_GUESTDBG_INJECT_BP
);
1128 void kvm_arch_update_guest_debug(CPUState
*env
, struct kvm_guest_debug
*dbg
)
1130 const uint8_t type_code
[] = {
1131 [GDB_BREAKPOINT_HW
] = 0x0,
1132 [GDB_WATCHPOINT_WRITE
] = 0x1,
1133 [GDB_WATCHPOINT_ACCESS
] = 0x3
1135 const uint8_t len_code
[] = {
1136 [1] = 0x0, [2] = 0x1, [4] = 0x3, [8] = 0x2
1140 if (kvm_sw_breakpoints_active(env
))
1141 dbg
->control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_USE_SW_BP
;
1143 if (nb_hw_breakpoint
> 0) {
1144 dbg
->control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_USE_HW_BP
;
1145 dbg
->arch
.debugreg
[7] = 0x0600;
1146 for (n
= 0; n
< nb_hw_breakpoint
; n
++) {
1147 dbg
->arch
.debugreg
[n
] = hw_breakpoint
[n
].addr
;
1148 dbg
->arch
.debugreg
[7] |= (2 << (n
* 2)) |
1149 (type_code
[hw_breakpoint
[n
].type
] << (16 + n
*4)) |
1150 (len_code
[hw_breakpoint
[n
].len
] << (18 + n
*4));
1154 #endif /* KVM_CAP_SET_GUEST_DEBUG */