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"
29 #define dprintf(fmt, ...) \
30 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
32 #define dprintf(fmt, ...) \
36 int kvm_arch_init_vcpu(CPUState
*env
)
39 struct kvm_cpuid cpuid
;
40 struct kvm_cpuid_entry entries
[100];
41 } __attribute__((packed
)) cpuid_data
;
42 uint32_t limit
, i
, cpuid_i
;
43 uint32_t eax
, ebx
, ecx
, edx
;
47 cpu_x86_cpuid(env
, 0, &eax
, &ebx
, &ecx
, &edx
);
50 for (i
= 0; i
<= limit
; i
++) {
51 struct kvm_cpuid_entry
*c
= &cpuid_data
.entries
[cpuid_i
++];
53 cpu_x86_cpuid(env
, i
, &eax
, &ebx
, &ecx
, &edx
);
61 cpu_x86_cpuid(env
, 0x80000000, &eax
, &ebx
, &ecx
, &edx
);
64 for (i
= 0x80000000; i
<= limit
; i
++) {
65 struct kvm_cpuid_entry
*c
= &cpuid_data
.entries
[cpuid_i
++];
67 cpu_x86_cpuid(env
, i
, &eax
, &ebx
, &ecx
, &edx
);
75 cpuid_data
.cpuid
.nent
= cpuid_i
;
77 return kvm_vcpu_ioctl(env
, KVM_SET_CPUID
, &cpuid_data
);
80 static int kvm_has_msr_star(CPUState
*env
)
82 static int has_msr_star
;
86 if (has_msr_star
== 0) {
87 struct kvm_msr_list msr_list
, *kvm_msr_list
;
91 /* Obtain MSR list from KVM. These are the MSRs that we must
94 ret
= kvm_ioctl(env
->kvm_state
, KVM_GET_MSR_INDEX_LIST
, &msr_list
);
98 kvm_msr_list
= qemu_mallocz(sizeof(msr_list
) +
99 msr_list
.nmsrs
* sizeof(msr_list
.indices
[0]));
100 if (kvm_msr_list
== NULL
)
103 kvm_msr_list
->nmsrs
= msr_list
.nmsrs
;
104 ret
= kvm_ioctl(env
->kvm_state
, KVM_GET_MSR_INDEX_LIST
, kvm_msr_list
);
108 for (i
= 0; i
< kvm_msr_list
->nmsrs
; i
++) {
109 if (kvm_msr_list
->indices
[i
] == MSR_STAR
) {
119 if (has_msr_star
== 1)
124 int kvm_arch_init(KVMState
*s
, int smp_cpus
)
128 /* create vm86 tss. KVM uses vm86 mode to emulate 16-bit code
129 * directly. In order to use vm86 mode, a TSS is needed. Since this
130 * must be part of guest physical memory, we need to allocate it. Older
131 * versions of KVM just assumed that it would be at the end of physical
132 * memory but that doesn't work with more than 4GB of memory. We simply
133 * refuse to work with those older versions of KVM. */
134 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, KVM_CAP_SET_TSS_ADDR
);
136 fprintf(stderr
, "kvm does not support KVM_CAP_SET_TSS_ADDR\n");
140 /* this address is 3 pages before the bios, and the bios should present
141 * as unavaible memory. FIXME, need to ensure the e820 map deals with
144 return kvm_vm_ioctl(s
, KVM_SET_TSS_ADDR
, 0xfffbd000);
147 static void set_v8086_seg(struct kvm_segment
*lhs
, const SegmentCache
*rhs
)
149 lhs
->selector
= rhs
->selector
;
150 lhs
->base
= rhs
->base
;
151 lhs
->limit
= rhs
->limit
;
163 static void set_seg(struct kvm_segment
*lhs
, const SegmentCache
*rhs
)
165 unsigned flags
= rhs
->flags
;
166 lhs
->selector
= rhs
->selector
;
167 lhs
->base
= rhs
->base
;
168 lhs
->limit
= rhs
->limit
;
169 lhs
->type
= (flags
>> DESC_TYPE_SHIFT
) & 15;
170 lhs
->present
= (flags
& DESC_P_MASK
) != 0;
171 lhs
->dpl
= rhs
->selector
& 3;
172 lhs
->db
= (flags
>> DESC_B_SHIFT
) & 1;
173 lhs
->s
= (flags
& DESC_S_MASK
) != 0;
174 lhs
->l
= (flags
>> DESC_L_SHIFT
) & 1;
175 lhs
->g
= (flags
& DESC_G_MASK
) != 0;
176 lhs
->avl
= (flags
& DESC_AVL_MASK
) != 0;
180 static void get_seg(SegmentCache
*lhs
, const struct kvm_segment
*rhs
)
182 lhs
->selector
= rhs
->selector
;
183 lhs
->base
= rhs
->base
;
184 lhs
->limit
= rhs
->limit
;
186 (rhs
->type
<< DESC_TYPE_SHIFT
)
187 | (rhs
->present
* DESC_P_MASK
)
188 | (rhs
->dpl
<< DESC_DPL_SHIFT
)
189 | (rhs
->db
<< DESC_B_SHIFT
)
190 | (rhs
->s
* DESC_S_MASK
)
191 | (rhs
->l
<< DESC_L_SHIFT
)
192 | (rhs
->g
* DESC_G_MASK
)
193 | (rhs
->avl
* DESC_AVL_MASK
);
196 static void kvm_getput_reg(__u64
*kvm_reg
, target_ulong
*qemu_reg
, int set
)
199 *kvm_reg
= *qemu_reg
;
201 *qemu_reg
= *kvm_reg
;
204 static int kvm_getput_regs(CPUState
*env
, int set
)
206 struct kvm_regs regs
;
210 ret
= kvm_vcpu_ioctl(env
, KVM_GET_REGS
, ®s
);
215 kvm_getput_reg(®s
.rax
, &env
->regs
[R_EAX
], set
);
216 kvm_getput_reg(®s
.rbx
, &env
->regs
[R_EBX
], set
);
217 kvm_getput_reg(®s
.rcx
, &env
->regs
[R_ECX
], set
);
218 kvm_getput_reg(®s
.rdx
, &env
->regs
[R_EDX
], set
);
219 kvm_getput_reg(®s
.rsi
, &env
->regs
[R_ESI
], set
);
220 kvm_getput_reg(®s
.rdi
, &env
->regs
[R_EDI
], set
);
221 kvm_getput_reg(®s
.rsp
, &env
->regs
[R_ESP
], set
);
222 kvm_getput_reg(®s
.rbp
, &env
->regs
[R_EBP
], set
);
224 kvm_getput_reg(®s
.r8
, &env
->regs
[8], set
);
225 kvm_getput_reg(®s
.r9
, &env
->regs
[9], set
);
226 kvm_getput_reg(®s
.r10
, &env
->regs
[10], set
);
227 kvm_getput_reg(®s
.r11
, &env
->regs
[11], set
);
228 kvm_getput_reg(®s
.r12
, &env
->regs
[12], set
);
229 kvm_getput_reg(®s
.r13
, &env
->regs
[13], set
);
230 kvm_getput_reg(®s
.r14
, &env
->regs
[14], set
);
231 kvm_getput_reg(®s
.r15
, &env
->regs
[15], set
);
234 kvm_getput_reg(®s
.rflags
, &env
->eflags
, set
);
235 kvm_getput_reg(®s
.rip
, &env
->eip
, set
);
238 ret
= kvm_vcpu_ioctl(env
, KVM_SET_REGS
, ®s
);
243 static int kvm_put_fpu(CPUState
*env
)
248 memset(&fpu
, 0, sizeof fpu
);
249 fpu
.fsw
= env
->fpus
& ~(7 << 11);
250 fpu
.fsw
|= (env
->fpstt
& 7) << 11;
252 for (i
= 0; i
< 8; ++i
)
253 fpu
.ftwx
|= (!env
->fptags
[i
]) << i
;
254 memcpy(fpu
.fpr
, env
->fpregs
, sizeof env
->fpregs
);
255 memcpy(fpu
.xmm
, env
->xmm_regs
, sizeof env
->xmm_regs
);
256 fpu
.mxcsr
= env
->mxcsr
;
258 return kvm_vcpu_ioctl(env
, KVM_SET_FPU
, &fpu
);
261 static int kvm_put_sregs(CPUState
*env
)
263 struct kvm_sregs sregs
;
265 memcpy(sregs
.interrupt_bitmap
,
266 env
->interrupt_bitmap
,
267 sizeof(sregs
.interrupt_bitmap
));
269 if ((env
->eflags
& VM_MASK
)) {
270 set_v8086_seg(&sregs
.cs
, &env
->segs
[R_CS
]);
271 set_v8086_seg(&sregs
.ds
, &env
->segs
[R_DS
]);
272 set_v8086_seg(&sregs
.es
, &env
->segs
[R_ES
]);
273 set_v8086_seg(&sregs
.fs
, &env
->segs
[R_FS
]);
274 set_v8086_seg(&sregs
.gs
, &env
->segs
[R_GS
]);
275 set_v8086_seg(&sregs
.ss
, &env
->segs
[R_SS
]);
277 set_seg(&sregs
.cs
, &env
->segs
[R_CS
]);
278 set_seg(&sregs
.ds
, &env
->segs
[R_DS
]);
279 set_seg(&sregs
.es
, &env
->segs
[R_ES
]);
280 set_seg(&sregs
.fs
, &env
->segs
[R_FS
]);
281 set_seg(&sregs
.gs
, &env
->segs
[R_GS
]);
282 set_seg(&sregs
.ss
, &env
->segs
[R_SS
]);
284 if (env
->cr
[0] & CR0_PE_MASK
) {
285 /* force ss cpl to cs cpl */
286 sregs
.ss
.selector
= (sregs
.ss
.selector
& ~3) |
287 (sregs
.cs
.selector
& 3);
288 sregs
.ss
.dpl
= sregs
.ss
.selector
& 3;
292 set_seg(&sregs
.tr
, &env
->tr
);
293 set_seg(&sregs
.ldt
, &env
->ldt
);
295 sregs
.idt
.limit
= env
->idt
.limit
;
296 sregs
.idt
.base
= env
->idt
.base
;
297 sregs
.gdt
.limit
= env
->gdt
.limit
;
298 sregs
.gdt
.base
= env
->gdt
.base
;
300 sregs
.cr0
= env
->cr
[0];
301 sregs
.cr2
= env
->cr
[2];
302 sregs
.cr3
= env
->cr
[3];
303 sregs
.cr4
= env
->cr
[4];
305 sregs
.cr8
= cpu_get_apic_tpr(env
);
306 sregs
.apic_base
= cpu_get_apic_base(env
);
308 sregs
.efer
= env
->efer
;
310 return kvm_vcpu_ioctl(env
, KVM_SET_SREGS
, &sregs
);
313 static void kvm_msr_entry_set(struct kvm_msr_entry
*entry
,
314 uint32_t index
, uint64_t value
)
316 entry
->index
= index
;
320 static int kvm_put_msrs(CPUState
*env
)
323 struct kvm_msrs info
;
324 struct kvm_msr_entry entries
[100];
326 struct kvm_msr_entry
*msrs
= msr_data
.entries
;
329 kvm_msr_entry_set(&msrs
[n
++], MSR_IA32_SYSENTER_CS
, env
->sysenter_cs
);
330 kvm_msr_entry_set(&msrs
[n
++], MSR_IA32_SYSENTER_ESP
, env
->sysenter_esp
);
331 kvm_msr_entry_set(&msrs
[n
++], MSR_IA32_SYSENTER_EIP
, env
->sysenter_eip
);
332 if (kvm_has_msr_star(env
))
333 kvm_msr_entry_set(&msrs
[n
++], MSR_STAR
, env
->star
);
334 kvm_msr_entry_set(&msrs
[n
++], MSR_IA32_TSC
, env
->tsc
);
336 /* FIXME if lm capable */
337 kvm_msr_entry_set(&msrs
[n
++], MSR_CSTAR
, env
->cstar
);
338 kvm_msr_entry_set(&msrs
[n
++], MSR_KERNELGSBASE
, env
->kernelgsbase
);
339 kvm_msr_entry_set(&msrs
[n
++], MSR_FMASK
, env
->fmask
);
340 kvm_msr_entry_set(&msrs
[n
++], MSR_LSTAR
, env
->lstar
);
342 msr_data
.info
.nmsrs
= n
;
344 return kvm_vcpu_ioctl(env
, KVM_SET_MSRS
, &msr_data
);
349 static int kvm_get_fpu(CPUState
*env
)
354 ret
= kvm_vcpu_ioctl(env
, KVM_GET_FPU
, &fpu
);
358 env
->fpstt
= (fpu
.fsw
>> 11) & 7;
361 for (i
= 0; i
< 8; ++i
)
362 env
->fptags
[i
] = !((fpu
.ftwx
>> i
) & 1);
363 memcpy(env
->fpregs
, fpu
.fpr
, sizeof env
->fpregs
);
364 memcpy(env
->xmm_regs
, fpu
.xmm
, sizeof env
->xmm_regs
);
365 env
->mxcsr
= fpu
.mxcsr
;
370 static int kvm_get_sregs(CPUState
*env
)
372 struct kvm_sregs sregs
;
376 ret
= kvm_vcpu_ioctl(env
, KVM_GET_SREGS
, &sregs
);
380 memcpy(env
->interrupt_bitmap
,
381 sregs
.interrupt_bitmap
,
382 sizeof(sregs
.interrupt_bitmap
));
384 get_seg(&env
->segs
[R_CS
], &sregs
.cs
);
385 get_seg(&env
->segs
[R_DS
], &sregs
.ds
);
386 get_seg(&env
->segs
[R_ES
], &sregs
.es
);
387 get_seg(&env
->segs
[R_FS
], &sregs
.fs
);
388 get_seg(&env
->segs
[R_GS
], &sregs
.gs
);
389 get_seg(&env
->segs
[R_SS
], &sregs
.ss
);
391 get_seg(&env
->tr
, &sregs
.tr
);
392 get_seg(&env
->ldt
, &sregs
.ldt
);
394 env
->idt
.limit
= sregs
.idt
.limit
;
395 env
->idt
.base
= sregs
.idt
.base
;
396 env
->gdt
.limit
= sregs
.gdt
.limit
;
397 env
->gdt
.base
= sregs
.gdt
.base
;
399 env
->cr
[0] = sregs
.cr0
;
400 env
->cr
[2] = sregs
.cr2
;
401 env
->cr
[3] = sregs
.cr3
;
402 env
->cr
[4] = sregs
.cr4
;
404 cpu_set_apic_base(env
, sregs
.apic_base
);
406 env
->efer
= sregs
.efer
;
407 //cpu_set_apic_tpr(env, sregs.cr8);
409 #define HFLAG_COPY_MASK ~( \
410 HF_CPL_MASK | HF_PE_MASK | HF_MP_MASK | HF_EM_MASK | \
411 HF_TS_MASK | HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK | \
412 HF_OSFXSR_MASK | HF_LMA_MASK | HF_CS32_MASK | \
413 HF_SS32_MASK | HF_CS64_MASK | HF_ADDSEG_MASK)
417 hflags
= (env
->segs
[R_CS
].flags
>> DESC_DPL_SHIFT
) & HF_CPL_MASK
;
418 hflags
|= (env
->cr
[0] & CR0_PE_MASK
) << (HF_PE_SHIFT
- CR0_PE_SHIFT
);
419 hflags
|= (env
->cr
[0] << (HF_MP_SHIFT
- CR0_MP_SHIFT
)) &
420 (HF_MP_MASK
| HF_EM_MASK
| HF_TS_MASK
);
421 hflags
|= (env
->eflags
& (HF_TF_MASK
| HF_VM_MASK
| HF_IOPL_MASK
));
422 hflags
|= (env
->cr
[4] & CR4_OSFXSR_MASK
) <<
423 (HF_OSFXSR_SHIFT
- CR4_OSFXSR_SHIFT
);
425 if (env
->efer
& MSR_EFER_LMA
) {
426 hflags
|= HF_LMA_MASK
;
429 if ((hflags
& HF_LMA_MASK
) && (env
->segs
[R_CS
].flags
& DESC_L_MASK
)) {
430 hflags
|= HF_CS32_MASK
| HF_SS32_MASK
| HF_CS64_MASK
;
432 hflags
|= (env
->segs
[R_CS
].flags
& DESC_B_MASK
) >>
433 (DESC_B_SHIFT
- HF_CS32_SHIFT
);
434 hflags
|= (env
->segs
[R_SS
].flags
& DESC_B_MASK
) >>
435 (DESC_B_SHIFT
- HF_SS32_SHIFT
);
436 if (!(env
->cr
[0] & CR0_PE_MASK
) ||
437 (env
->eflags
& VM_MASK
) ||
438 !(hflags
& HF_CS32_MASK
)) {
439 hflags
|= HF_ADDSEG_MASK
;
441 hflags
|= ((env
->segs
[R_DS
].base
|
442 env
->segs
[R_ES
].base
|
443 env
->segs
[R_SS
].base
) != 0) <<
447 env
->hflags
= (env
->hflags
& HFLAG_COPY_MASK
) | hflags
;
448 env
->cc_src
= env
->eflags
& (CC_O
| CC_S
| CC_Z
| CC_A
| CC_P
| CC_C
);
449 env
->df
= 1 - (2 * ((env
->eflags
>> 10) & 1));
450 env
->cc_op
= CC_OP_EFLAGS
;
451 env
->eflags
&= ~(DF_MASK
| CC_O
| CC_S
| CC_Z
| CC_A
| CC_P
| CC_C
);
456 static int kvm_get_msrs(CPUState
*env
)
459 struct kvm_msrs info
;
460 struct kvm_msr_entry entries
[100];
462 struct kvm_msr_entry
*msrs
= msr_data
.entries
;
466 msrs
[n
++].index
= MSR_IA32_SYSENTER_CS
;
467 msrs
[n
++].index
= MSR_IA32_SYSENTER_ESP
;
468 msrs
[n
++].index
= MSR_IA32_SYSENTER_EIP
;
469 if (kvm_has_msr_star(env
))
470 msrs
[n
++].index
= MSR_STAR
;
471 msrs
[n
++].index
= MSR_IA32_TSC
;
473 /* FIXME lm_capable_kernel */
474 msrs
[n
++].index
= MSR_CSTAR
;
475 msrs
[n
++].index
= MSR_KERNELGSBASE
;
476 msrs
[n
++].index
= MSR_FMASK
;
477 msrs
[n
++].index
= MSR_LSTAR
;
479 msr_data
.info
.nmsrs
= n
;
480 ret
= kvm_vcpu_ioctl(env
, KVM_GET_MSRS
, &msr_data
);
484 for (i
= 0; i
< ret
; i
++) {
485 switch (msrs
[i
].index
) {
486 case MSR_IA32_SYSENTER_CS
:
487 env
->sysenter_cs
= msrs
[i
].data
;
489 case MSR_IA32_SYSENTER_ESP
:
490 env
->sysenter_esp
= msrs
[i
].data
;
492 case MSR_IA32_SYSENTER_EIP
:
493 env
->sysenter_eip
= msrs
[i
].data
;
496 env
->star
= msrs
[i
].data
;
500 env
->cstar
= msrs
[i
].data
;
502 case MSR_KERNELGSBASE
:
503 env
->kernelgsbase
= msrs
[i
].data
;
506 env
->fmask
= msrs
[i
].data
;
509 env
->lstar
= msrs
[i
].data
;
513 env
->tsc
= msrs
[i
].data
;
521 int kvm_arch_put_registers(CPUState
*env
)
525 ret
= kvm_getput_regs(env
, 1);
529 ret
= kvm_put_fpu(env
);
533 ret
= kvm_put_sregs(env
);
537 ret
= kvm_put_msrs(env
);
544 int kvm_arch_get_registers(CPUState
*env
)
548 ret
= kvm_getput_regs(env
, 0);
552 ret
= kvm_get_fpu(env
);
556 ret
= kvm_get_sregs(env
);
560 ret
= kvm_get_msrs(env
);
567 int kvm_arch_pre_run(CPUState
*env
, struct kvm_run
*run
)
569 /* Try to inject an interrupt if the guest can accept it */
570 if (run
->ready_for_interrupt_injection
&&
571 (env
->interrupt_request
& CPU_INTERRUPT_HARD
) &&
572 (env
->eflags
& IF_MASK
)) {
575 env
->interrupt_request
&= ~CPU_INTERRUPT_HARD
;
576 irq
= cpu_get_pic_interrupt(env
);
578 struct kvm_interrupt intr
;
581 dprintf("injected interrupt %d\n", irq
);
582 kvm_vcpu_ioctl(env
, KVM_INTERRUPT
, &intr
);
586 /* If we have an interrupt but the guest is not ready to receive an
587 * interrupt, request an interrupt window exit. This will
588 * cause a return to userspace as soon as the guest is ready to
589 * receive interrupts. */
590 if ((env
->interrupt_request
& CPU_INTERRUPT_HARD
))
591 run
->request_interrupt_window
= 1;
593 run
->request_interrupt_window
= 0;
595 dprintf("setting tpr\n");
596 run
->cr8
= cpu_get_apic_tpr(env
);
601 int kvm_arch_post_run(CPUState
*env
, struct kvm_run
*run
)
604 env
->eflags
|= IF_MASK
;
606 env
->eflags
&= ~IF_MASK
;
608 cpu_set_apic_tpr(env
, run
->cr8
);
609 cpu_set_apic_base(env
, run
->apic_base
);
614 static int kvm_handle_halt(CPUState
*env
)
616 if (!((env
->interrupt_request
& CPU_INTERRUPT_HARD
) &&
617 (env
->eflags
& IF_MASK
)) &&
618 !(env
->interrupt_request
& CPU_INTERRUPT_NMI
)) {
620 env
->exception_index
= EXCP_HLT
;
627 int kvm_arch_handle_exit(CPUState
*env
, struct kvm_run
*run
)
631 switch (run
->exit_reason
) {
633 dprintf("handle_hlt\n");
634 ret
= kvm_handle_halt(env
);