2 * Kernel-based Virtual Machine driver for Linux
6 * Copyright (C) 2006 Qumranet, Inc.
9 * Yaniv Kamay <yaniv@qumranet.com>
10 * Avi Kivity <avi@qumranet.com>
12 * This work is licensed under the terms of the GNU GPL, version 2. See
13 * the COPYING file in the top-level directory.
18 #include "x86_emulate.h"
20 #include <linux/module.h>
21 #include <linux/kernel.h>
22 #include <linux/vmalloc.h>
23 #include <linux/highmem.h>
24 #include <linux/profile.h>
25 #include <linux/sched.h>
29 MODULE_AUTHOR("Qumranet");
30 MODULE_LICENSE("GPL");
32 #define IOPM_ALLOC_ORDER 2
33 #define MSRPM_ALLOC_ORDER 1
39 #define DR7_GD_MASK (1 << 13)
40 #define DR6_BD_MASK (1 << 13)
42 #define SEG_TYPE_LDT 2
43 #define SEG_TYPE_BUSY_TSS16 3
45 #define KVM_EFER_LMA (1 << 10)
46 #define KVM_EFER_LME (1 << 8)
48 #define SVM_FEATURE_NPT (1 << 0)
49 #define SVM_FEATURE_LBRV (1 << 1)
50 #define SVM_DEATURE_SVML (1 << 2)
52 unsigned long iopm_base
;
53 unsigned long msrpm_base
;
55 struct kvm_ldttss_desc
{
58 unsigned base1
: 8, type
: 5, dpl
: 2, p
: 1;
59 unsigned limit1
: 4, zero0
: 3, g
: 1, base2
: 8;
62 } __attribute__((packed
));
70 struct kvm_ldttss_desc
*tss_desc
;
72 struct page
*save_area
;
75 static DEFINE_PER_CPU(struct svm_cpu_data
*, svm_data
);
76 static uint32_t svm_features
;
78 struct svm_init_data
{
83 static u32 msrpm_ranges
[] = {0, 0xc0000000, 0xc0010000};
85 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
86 #define MSRS_RANGE_SIZE 2048
87 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
89 #define MAX_INST_SIZE 15
91 static inline u32
svm_has(u32 feat
)
93 return svm_features
& feat
;
96 static unsigned get_addr_size(struct kvm_vcpu
*vcpu
)
98 struct vmcb_save_area
*sa
= &vcpu
->svm
->vmcb
->save
;
101 if (!(sa
->cr0
& X86_CR0_PE
) || (sa
->rflags
& X86_EFLAGS_VM
))
104 cs_attrib
= sa
->cs
.attrib
;
106 return (cs_attrib
& SVM_SELECTOR_L_MASK
) ? 8 :
107 (cs_attrib
& SVM_SELECTOR_DB_MASK
) ? 4 : 2;
110 static inline u8
pop_irq(struct kvm_vcpu
*vcpu
)
112 int word_index
= __ffs(vcpu
->irq_summary
);
113 int bit_index
= __ffs(vcpu
->irq_pending
[word_index
]);
114 int irq
= word_index
* BITS_PER_LONG
+ bit_index
;
116 clear_bit(bit_index
, &vcpu
->irq_pending
[word_index
]);
117 if (!vcpu
->irq_pending
[word_index
])
118 clear_bit(word_index
, &vcpu
->irq_summary
);
122 static inline void push_irq(struct kvm_vcpu
*vcpu
, u8 irq
)
124 set_bit(irq
, vcpu
->irq_pending
);
125 set_bit(irq
/ BITS_PER_LONG
, &vcpu
->irq_summary
);
128 static inline void clgi(void)
130 asm volatile (SVM_CLGI
);
133 static inline void stgi(void)
135 asm volatile (SVM_STGI
);
138 static inline void invlpga(unsigned long addr
, u32 asid
)
140 asm volatile (SVM_INVLPGA :: "a"(addr
), "c"(asid
));
143 static inline unsigned long kvm_read_cr2(void)
147 asm volatile ("mov %%cr2, %0" : "=r" (cr2
));
151 static inline void kvm_write_cr2(unsigned long val
)
153 asm volatile ("mov %0, %%cr2" :: "r" (val
));
156 static inline unsigned long read_dr6(void)
160 asm volatile ("mov %%dr6, %0" : "=r" (dr6
));
164 static inline void write_dr6(unsigned long val
)
166 asm volatile ("mov %0, %%dr6" :: "r" (val
));
169 static inline unsigned long read_dr7(void)
173 asm volatile ("mov %%dr7, %0" : "=r" (dr7
));
177 static inline void write_dr7(unsigned long val
)
179 asm volatile ("mov %0, %%dr7" :: "r" (val
));
182 static inline void force_new_asid(struct kvm_vcpu
*vcpu
)
184 vcpu
->svm
->asid_generation
--;
187 static inline void flush_guest_tlb(struct kvm_vcpu
*vcpu
)
189 force_new_asid(vcpu
);
192 static void svm_set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
194 if (!(efer
& KVM_EFER_LMA
))
195 efer
&= ~KVM_EFER_LME
;
197 vcpu
->svm
->vmcb
->save
.efer
= efer
| MSR_EFER_SVME_MASK
;
198 vcpu
->shadow_efer
= efer
;
201 static void svm_inject_gp(struct kvm_vcpu
*vcpu
, unsigned error_code
)
203 vcpu
->svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
|
204 SVM_EVTINJ_VALID_ERR
|
205 SVM_EVTINJ_TYPE_EXEPT
|
207 vcpu
->svm
->vmcb
->control
.event_inj_err
= error_code
;
210 static void inject_ud(struct kvm_vcpu
*vcpu
)
212 vcpu
->svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
|
213 SVM_EVTINJ_TYPE_EXEPT
|
217 static int is_page_fault(uint32_t info
)
219 info
&= SVM_EVTINJ_VEC_MASK
| SVM_EVTINJ_TYPE_MASK
| SVM_EVTINJ_VALID
;
220 return info
== (PF_VECTOR
| SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_EXEPT
);
223 static int is_external_interrupt(u32 info
)
225 info
&= SVM_EVTINJ_TYPE_MASK
| SVM_EVTINJ_VALID
;
226 return info
== (SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
);
229 static void skip_emulated_instruction(struct kvm_vcpu
*vcpu
)
231 if (!vcpu
->svm
->next_rip
) {
232 printk(KERN_DEBUG
"%s: NOP\n", __FUNCTION__
);
235 if (vcpu
->svm
->next_rip
- vcpu
->svm
->vmcb
->save
.rip
> 15) {
236 printk(KERN_ERR
"%s: ip 0x%llx next 0x%llx\n",
238 vcpu
->svm
->vmcb
->save
.rip
,
239 vcpu
->svm
->next_rip
);
242 vcpu
->rip
= vcpu
->svm
->vmcb
->save
.rip
= vcpu
->svm
->next_rip
;
243 vcpu
->svm
->vmcb
->control
.int_state
&= ~SVM_INTERRUPT_SHADOW_MASK
;
245 vcpu
->interrupt_window_open
= 1;
248 static int has_svm(void)
250 uint32_t eax
, ebx
, ecx
, edx
;
252 if (boot_cpu_data
.x86_vendor
!= X86_VENDOR_AMD
) {
253 printk(KERN_INFO
"has_svm: not amd\n");
257 cpuid(0x80000000, &eax
, &ebx
, &ecx
, &edx
);
258 if (eax
< SVM_CPUID_FUNC
) {
259 printk(KERN_INFO
"has_svm: can't execute cpuid_8000000a\n");
263 cpuid(0x80000001, &eax
, &ebx
, &ecx
, &edx
);
264 if (!(ecx
& (1 << SVM_CPUID_FEATURE_SHIFT
))) {
265 printk(KERN_DEBUG
"has_svm: svm not available\n");
271 static void svm_hardware_disable(void *garbage
)
273 struct svm_cpu_data
*svm_data
274 = per_cpu(svm_data
, raw_smp_processor_id());
279 wrmsrl(MSR_VM_HSAVE_PA
, 0);
280 rdmsrl(MSR_EFER
, efer
);
281 wrmsrl(MSR_EFER
, efer
& ~MSR_EFER_SVME_MASK
);
282 per_cpu(svm_data
, raw_smp_processor_id()) = NULL
;
283 __free_page(svm_data
->save_area
);
288 static void svm_hardware_enable(void *garbage
)
291 struct svm_cpu_data
*svm_data
;
294 struct desc_ptr gdt_descr
;
296 struct Xgt_desc_struct gdt_descr
;
298 struct desc_struct
*gdt
;
299 int me
= raw_smp_processor_id();
302 printk(KERN_ERR
"svm_cpu_init: err EOPNOTSUPP on %d\n", me
);
305 svm_data
= per_cpu(svm_data
, me
);
308 printk(KERN_ERR
"svm_cpu_init: svm_data is NULL on %d\n",
313 svm_data
->asid_generation
= 1;
314 svm_data
->max_asid
= cpuid_ebx(SVM_CPUID_FUNC
) - 1;
315 svm_data
->next_asid
= svm_data
->max_asid
+ 1;
316 svm_features
= cpuid_edx(SVM_CPUID_FUNC
);
318 asm volatile ( "sgdt %0" : "=m"(gdt_descr
) );
319 gdt
= (struct desc_struct
*)gdt_descr
.address
;
320 svm_data
->tss_desc
= (struct kvm_ldttss_desc
*)(gdt
+ GDT_ENTRY_TSS
);
322 rdmsrl(MSR_EFER
, efer
);
323 wrmsrl(MSR_EFER
, efer
| MSR_EFER_SVME_MASK
);
325 wrmsrl(MSR_VM_HSAVE_PA
,
326 page_to_pfn(svm_data
->save_area
) << PAGE_SHIFT
);
329 static int svm_cpu_init(int cpu
)
331 struct svm_cpu_data
*svm_data
;
334 svm_data
= kzalloc(sizeof(struct svm_cpu_data
), GFP_KERNEL
);
338 svm_data
->save_area
= alloc_page(GFP_KERNEL
);
340 if (!svm_data
->save_area
)
343 per_cpu(svm_data
, cpu
) = svm_data
;
353 static int set_msr_interception(u32
*msrpm
, unsigned msr
,
358 for (i
= 0; i
< NUM_MSR_MAPS
; i
++) {
359 if (msr
>= msrpm_ranges
[i
] &&
360 msr
< msrpm_ranges
[i
] + MSRS_IN_RANGE
) {
361 u32 msr_offset
= (i
* MSRS_IN_RANGE
+ msr
-
362 msrpm_ranges
[i
]) * 2;
364 u32
*base
= msrpm
+ (msr_offset
/ 32);
365 u32 msr_shift
= msr_offset
% 32;
366 u32 mask
= ((write
) ? 0 : 2) | ((read
) ? 0 : 1);
367 *base
= (*base
& ~(0x3 << msr_shift
)) |
372 printk(KERN_DEBUG
"%s: not found 0x%x\n", __FUNCTION__
, msr
);
376 static __init
int svm_hardware_setup(void)
379 struct page
*iopm_pages
;
380 struct page
*msrpm_pages
;
381 void *iopm_va
, *msrpm_va
;
384 kvm_emulator_want_group7_invlpg();
386 iopm_pages
= alloc_pages(GFP_KERNEL
, IOPM_ALLOC_ORDER
);
391 iopm_va
= page_address(iopm_pages
);
392 memset(iopm_va
, 0xff, PAGE_SIZE
* (1 << IOPM_ALLOC_ORDER
));
393 clear_bit(0x80, iopm_va
); /* allow direct access to PC debug port */
394 iopm_base
= page_to_pfn(iopm_pages
) << PAGE_SHIFT
;
397 msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
403 msrpm_va
= page_address(msrpm_pages
);
404 memset(msrpm_va
, 0xff, PAGE_SIZE
* (1 << MSRPM_ALLOC_ORDER
));
405 msrpm_base
= page_to_pfn(msrpm_pages
) << PAGE_SHIFT
;
408 set_msr_interception(msrpm_va
, MSR_GS_BASE
, 1, 1);
409 set_msr_interception(msrpm_va
, MSR_FS_BASE
, 1, 1);
410 set_msr_interception(msrpm_va
, MSR_KERNEL_GS_BASE
, 1, 1);
411 set_msr_interception(msrpm_va
, MSR_LSTAR
, 1, 1);
412 set_msr_interception(msrpm_va
, MSR_CSTAR
, 1, 1);
413 set_msr_interception(msrpm_va
, MSR_SYSCALL_MASK
, 1, 1);
415 set_msr_interception(msrpm_va
, MSR_K6_STAR
, 1, 1);
416 set_msr_interception(msrpm_va
, MSR_IA32_SYSENTER_CS
, 1, 1);
417 set_msr_interception(msrpm_va
, MSR_IA32_SYSENTER_ESP
, 1, 1);
418 set_msr_interception(msrpm_va
, MSR_IA32_SYSENTER_EIP
, 1, 1);
420 for_each_online_cpu(cpu
) {
421 r
= svm_cpu_init(cpu
);
428 __free_pages(msrpm_pages
, MSRPM_ALLOC_ORDER
);
431 __free_pages(iopm_pages
, IOPM_ALLOC_ORDER
);
436 static __exit
void svm_hardware_unsetup(void)
438 __free_pages(pfn_to_page(msrpm_base
>> PAGE_SHIFT
), MSRPM_ALLOC_ORDER
);
439 __free_pages(pfn_to_page(iopm_base
>> PAGE_SHIFT
), IOPM_ALLOC_ORDER
);
440 iopm_base
= msrpm_base
= 0;
443 static void init_seg(struct vmcb_seg
*seg
)
446 seg
->attrib
= SVM_SELECTOR_P_MASK
| SVM_SELECTOR_S_MASK
|
447 SVM_SELECTOR_WRITE_MASK
; /* Read/Write Data Segment */
452 static void init_sys_seg(struct vmcb_seg
*seg
, uint32_t type
)
455 seg
->attrib
= SVM_SELECTOR_P_MASK
| type
;
460 static int svm_vcpu_setup(struct kvm_vcpu
*vcpu
)
465 static void init_vmcb(struct vmcb
*vmcb
)
467 struct vmcb_control_area
*control
= &vmcb
->control
;
468 struct vmcb_save_area
*save
= &vmcb
->save
;
470 control
->intercept_cr_read
= INTERCEPT_CR0_MASK
|
474 control
->intercept_cr_write
= INTERCEPT_CR0_MASK
|
478 control
->intercept_dr_read
= INTERCEPT_DR0_MASK
|
483 control
->intercept_dr_write
= INTERCEPT_DR0_MASK
|
490 control
->intercept_exceptions
= 1 << PF_VECTOR
;
493 control
->intercept
= (1ULL << INTERCEPT_INTR
) |
494 (1ULL << INTERCEPT_NMI
) |
495 (1ULL << INTERCEPT_SMI
) |
497 * selective cr0 intercept bug?
498 * 0: 0f 22 d8 mov %eax,%cr3
499 * 3: 0f 20 c0 mov %cr0,%eax
500 * 6: 0d 00 00 00 80 or $0x80000000,%eax
501 * b: 0f 22 c0 mov %eax,%cr0
502 * set cr3 ->interception
503 * get cr0 ->interception
504 * set cr0 -> no interception
506 /* (1ULL << INTERCEPT_SELECTIVE_CR0) | */
507 (1ULL << INTERCEPT_CPUID
) |
508 (1ULL << INTERCEPT_HLT
) |
509 (1ULL << INTERCEPT_INVLPGA
) |
510 (1ULL << INTERCEPT_IOIO_PROT
) |
511 (1ULL << INTERCEPT_MSR_PROT
) |
512 (1ULL << INTERCEPT_TASK_SWITCH
) |
513 (1ULL << INTERCEPT_SHUTDOWN
) |
514 (1ULL << INTERCEPT_VMRUN
) |
515 (1ULL << INTERCEPT_VMMCALL
) |
516 (1ULL << INTERCEPT_VMLOAD
) |
517 (1ULL << INTERCEPT_VMSAVE
) |
518 (1ULL << INTERCEPT_STGI
) |
519 (1ULL << INTERCEPT_CLGI
) |
520 (1ULL << INTERCEPT_SKINIT
) |
521 (1ULL << INTERCEPT_MONITOR
) |
522 (1ULL << INTERCEPT_MWAIT
);
524 control
->iopm_base_pa
= iopm_base
;
525 control
->msrpm_base_pa
= msrpm_base
;
526 control
->tsc_offset
= 0;
527 control
->int_ctl
= V_INTR_MASKING_MASK
;
535 save
->cs
.selector
= 0xf000;
536 /* Executable/Readable Code Segment */
537 save
->cs
.attrib
= SVM_SELECTOR_READ_MASK
| SVM_SELECTOR_P_MASK
|
538 SVM_SELECTOR_S_MASK
| SVM_SELECTOR_CODE_MASK
;
539 save
->cs
.limit
= 0xffff;
541 * cs.base should really be 0xffff0000, but vmx can't handle that, so
542 * be consistent with it.
544 * Replace when we have real mode working for vmx.
546 save
->cs
.base
= 0xf0000;
548 save
->gdtr
.limit
= 0xffff;
549 save
->idtr
.limit
= 0xffff;
551 init_sys_seg(&save
->ldtr
, SEG_TYPE_LDT
);
552 init_sys_seg(&save
->tr
, SEG_TYPE_BUSY_TSS16
);
554 save
->efer
= MSR_EFER_SVME_MASK
;
556 save
->dr6
= 0xffff0ff0;
559 save
->rip
= 0x0000fff0;
562 * cr0 val on cpu init should be 0x60000010, we enable cpu
563 * cache by default. the orderly way is to enable cache in bios.
565 save
->cr0
= 0x00000010 | X86_CR0_PG
| X86_CR0_WP
;
566 save
->cr4
= X86_CR4_PAE
;
570 static int svm_create_vcpu(struct kvm_vcpu
*vcpu
)
576 vcpu
->svm
= kzalloc(sizeof *vcpu
->svm
, GFP_KERNEL
);
579 page
= alloc_page(GFP_KERNEL
);
583 vcpu
->svm
->vmcb
= page_address(page
);
584 clear_page(vcpu
->svm
->vmcb
);
585 vcpu
->svm
->vmcb_pa
= page_to_pfn(page
) << PAGE_SHIFT
;
586 vcpu
->svm
->asid_generation
= 0;
587 memset(vcpu
->svm
->db_regs
, 0, sizeof(vcpu
->svm
->db_regs
));
588 init_vmcb(vcpu
->svm
->vmcb
);
591 vcpu
->fpu_active
= 1;
592 vcpu
->apic_base
= 0xfee00000 | MSR_IA32_APICBASE_ENABLE
;
593 if (vcpu
->vcpu_id
== 0)
594 vcpu
->apic_base
|= MSR_IA32_APICBASE_BSP
;
604 static void svm_free_vcpu(struct kvm_vcpu
*vcpu
)
609 __free_page(pfn_to_page(vcpu
->svm
->vmcb_pa
>> PAGE_SHIFT
));
613 static void svm_vcpu_load(struct kvm_vcpu
*vcpu
)
618 if (unlikely(cpu
!= vcpu
->cpu
)) {
622 * Make sure that the guest sees a monotonically
626 delta
= vcpu
->host_tsc
- tsc_this
;
627 vcpu
->svm
->vmcb
->control
.tsc_offset
+= delta
;
631 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
632 rdmsrl(host_save_user_msrs
[i
], vcpu
->svm
->host_user_msrs
[i
]);
635 static void svm_vcpu_put(struct kvm_vcpu
*vcpu
)
639 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
640 wrmsrl(host_save_user_msrs
[i
], vcpu
->svm
->host_user_msrs
[i
]);
642 rdtscll(vcpu
->host_tsc
);
646 static void svm_vcpu_decache(struct kvm_vcpu
*vcpu
)
650 static void svm_cache_regs(struct kvm_vcpu
*vcpu
)
652 vcpu
->regs
[VCPU_REGS_RAX
] = vcpu
->svm
->vmcb
->save
.rax
;
653 vcpu
->regs
[VCPU_REGS_RSP
] = vcpu
->svm
->vmcb
->save
.rsp
;
654 vcpu
->rip
= vcpu
->svm
->vmcb
->save
.rip
;
657 static void svm_decache_regs(struct kvm_vcpu
*vcpu
)
659 vcpu
->svm
->vmcb
->save
.rax
= vcpu
->regs
[VCPU_REGS_RAX
];
660 vcpu
->svm
->vmcb
->save
.rsp
= vcpu
->regs
[VCPU_REGS_RSP
];
661 vcpu
->svm
->vmcb
->save
.rip
= vcpu
->rip
;
664 static unsigned long svm_get_rflags(struct kvm_vcpu
*vcpu
)
666 return vcpu
->svm
->vmcb
->save
.rflags
;
669 static void svm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
671 vcpu
->svm
->vmcb
->save
.rflags
= rflags
;
674 static struct vmcb_seg
*svm_seg(struct kvm_vcpu
*vcpu
, int seg
)
676 struct vmcb_save_area
*save
= &vcpu
->svm
->vmcb
->save
;
679 case VCPU_SREG_CS
: return &save
->cs
;
680 case VCPU_SREG_DS
: return &save
->ds
;
681 case VCPU_SREG_ES
: return &save
->es
;
682 case VCPU_SREG_FS
: return &save
->fs
;
683 case VCPU_SREG_GS
: return &save
->gs
;
684 case VCPU_SREG_SS
: return &save
->ss
;
685 case VCPU_SREG_TR
: return &save
->tr
;
686 case VCPU_SREG_LDTR
: return &save
->ldtr
;
692 static u64
svm_get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
694 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
699 static void svm_get_segment(struct kvm_vcpu
*vcpu
,
700 struct kvm_segment
*var
, int seg
)
702 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
705 var
->limit
= s
->limit
;
706 var
->selector
= s
->selector
;
707 var
->type
= s
->attrib
& SVM_SELECTOR_TYPE_MASK
;
708 var
->s
= (s
->attrib
>> SVM_SELECTOR_S_SHIFT
) & 1;
709 var
->dpl
= (s
->attrib
>> SVM_SELECTOR_DPL_SHIFT
) & 3;
710 var
->present
= (s
->attrib
>> SVM_SELECTOR_P_SHIFT
) & 1;
711 var
->avl
= (s
->attrib
>> SVM_SELECTOR_AVL_SHIFT
) & 1;
712 var
->l
= (s
->attrib
>> SVM_SELECTOR_L_SHIFT
) & 1;
713 var
->db
= (s
->attrib
>> SVM_SELECTOR_DB_SHIFT
) & 1;
714 var
->g
= (s
->attrib
>> SVM_SELECTOR_G_SHIFT
) & 1;
715 var
->unusable
= !var
->present
;
718 static void svm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
720 struct vmcb_seg
*s
= svm_seg(vcpu
, VCPU_SREG_CS
);
722 *db
= (s
->attrib
>> SVM_SELECTOR_DB_SHIFT
) & 1;
723 *l
= (s
->attrib
>> SVM_SELECTOR_L_SHIFT
) & 1;
726 static void svm_get_idt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
728 dt
->limit
= vcpu
->svm
->vmcb
->save
.idtr
.limit
;
729 dt
->base
= vcpu
->svm
->vmcb
->save
.idtr
.base
;
732 static void svm_set_idt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
734 vcpu
->svm
->vmcb
->save
.idtr
.limit
= dt
->limit
;
735 vcpu
->svm
->vmcb
->save
.idtr
.base
= dt
->base
;
738 static void svm_get_gdt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
740 dt
->limit
= vcpu
->svm
->vmcb
->save
.gdtr
.limit
;
741 dt
->base
= vcpu
->svm
->vmcb
->save
.gdtr
.base
;
744 static void svm_set_gdt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
746 vcpu
->svm
->vmcb
->save
.gdtr
.limit
= dt
->limit
;
747 vcpu
->svm
->vmcb
->save
.gdtr
.base
= dt
->base
;
750 static void svm_decache_cr4_guest_bits(struct kvm_vcpu
*vcpu
)
754 static void svm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
757 if (vcpu
->shadow_efer
& KVM_EFER_LME
) {
758 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
759 vcpu
->shadow_efer
|= KVM_EFER_LMA
;
760 vcpu
->svm
->vmcb
->save
.efer
|= KVM_EFER_LMA
| KVM_EFER_LME
;
763 if (is_paging(vcpu
) && !(cr0
& X86_CR0_PG
) ) {
764 vcpu
->shadow_efer
&= ~KVM_EFER_LMA
;
765 vcpu
->svm
->vmcb
->save
.efer
&= ~(KVM_EFER_LMA
| KVM_EFER_LME
);
769 if ((vcpu
->cr0
& X86_CR0_TS
) && !(cr0
& X86_CR0_TS
)) {
770 vcpu
->svm
->vmcb
->control
.intercept_exceptions
&= ~(1 << NM_VECTOR
);
771 vcpu
->fpu_active
= 1;
775 cr0
|= X86_CR0_PG
| X86_CR0_WP
;
776 cr0
&= ~(X86_CR0_CD
| X86_CR0_NW
);
777 vcpu
->svm
->vmcb
->save
.cr0
= cr0
;
780 static void svm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
783 vcpu
->svm
->vmcb
->save
.cr4
= cr4
| X86_CR4_PAE
;
786 static void svm_set_segment(struct kvm_vcpu
*vcpu
,
787 struct kvm_segment
*var
, int seg
)
789 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
792 s
->limit
= var
->limit
;
793 s
->selector
= var
->selector
;
797 s
->attrib
= (var
->type
& SVM_SELECTOR_TYPE_MASK
);
798 s
->attrib
|= (var
->s
& 1) << SVM_SELECTOR_S_SHIFT
;
799 s
->attrib
|= (var
->dpl
& 3) << SVM_SELECTOR_DPL_SHIFT
;
800 s
->attrib
|= (var
->present
& 1) << SVM_SELECTOR_P_SHIFT
;
801 s
->attrib
|= (var
->avl
& 1) << SVM_SELECTOR_AVL_SHIFT
;
802 s
->attrib
|= (var
->l
& 1) << SVM_SELECTOR_L_SHIFT
;
803 s
->attrib
|= (var
->db
& 1) << SVM_SELECTOR_DB_SHIFT
;
804 s
->attrib
|= (var
->g
& 1) << SVM_SELECTOR_G_SHIFT
;
806 if (seg
== VCPU_SREG_CS
)
807 vcpu
->svm
->vmcb
->save
.cpl
808 = (vcpu
->svm
->vmcb
->save
.cs
.attrib
809 >> SVM_SELECTOR_DPL_SHIFT
) & 3;
815 vcpu->svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
816 vcpu->svm->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
820 static int svm_guest_debug(struct kvm_vcpu
*vcpu
, struct kvm_debug_guest
*dbg
)
825 static void load_host_msrs(struct kvm_vcpu
*vcpu
)
828 wrmsrl(MSR_GS_BASE
, vcpu
->svm
->host_gs_base
);
832 static void save_host_msrs(struct kvm_vcpu
*vcpu
)
835 rdmsrl(MSR_GS_BASE
, vcpu
->svm
->host_gs_base
);
839 static void new_asid(struct kvm_vcpu
*vcpu
, struct svm_cpu_data
*svm_data
)
841 if (svm_data
->next_asid
> svm_data
->max_asid
) {
842 ++svm_data
->asid_generation
;
843 svm_data
->next_asid
= 1;
844 vcpu
->svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ALL_ASID
;
847 vcpu
->cpu
= svm_data
->cpu
;
848 vcpu
->svm
->asid_generation
= svm_data
->asid_generation
;
849 vcpu
->svm
->vmcb
->control
.asid
= svm_data
->next_asid
++;
852 static void svm_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
854 invlpga(address
, vcpu
->svm
->vmcb
->control
.asid
); // is needed?
857 static unsigned long svm_get_dr(struct kvm_vcpu
*vcpu
, int dr
)
859 return vcpu
->svm
->db_regs
[dr
];
862 static void svm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long value
,
867 if (vcpu
->svm
->vmcb
->save
.dr7
& DR7_GD_MASK
) {
868 vcpu
->svm
->vmcb
->save
.dr7
&= ~DR7_GD_MASK
;
869 vcpu
->svm
->vmcb
->save
.dr6
|= DR6_BD_MASK
;
870 *exception
= DB_VECTOR
;
876 vcpu
->svm
->db_regs
[dr
] = value
;
879 if (vcpu
->cr4
& X86_CR4_DE
) {
880 *exception
= UD_VECTOR
;
884 if (value
& ~((1ULL << 32) - 1)) {
885 *exception
= GP_VECTOR
;
888 vcpu
->svm
->vmcb
->save
.dr7
= value
;
892 printk(KERN_DEBUG
"%s: unexpected dr %u\n",
894 *exception
= UD_VECTOR
;
899 static int pf_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
901 u32 exit_int_info
= vcpu
->svm
->vmcb
->control
.exit_int_info
;
904 enum emulation_result er
;
907 if (is_external_interrupt(exit_int_info
))
908 push_irq(vcpu
, exit_int_info
& SVM_EVTINJ_VEC_MASK
);
910 spin_lock(&vcpu
->kvm
->lock
);
912 fault_address
= vcpu
->svm
->vmcb
->control
.exit_info_2
;
913 error_code
= vcpu
->svm
->vmcb
->control
.exit_info_1
;
914 r
= kvm_mmu_page_fault(vcpu
, fault_address
, error_code
);
916 spin_unlock(&vcpu
->kvm
->lock
);
920 spin_unlock(&vcpu
->kvm
->lock
);
923 er
= emulate_instruction(vcpu
, kvm_run
, fault_address
, error_code
);
924 spin_unlock(&vcpu
->kvm
->lock
);
929 case EMULATE_DO_MMIO
:
930 ++vcpu
->stat
.mmio_exits
;
933 vcpu_printf(vcpu
, "%s: emulate fail\n", __FUNCTION__
);
939 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
943 static int nm_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
945 vcpu
->svm
->vmcb
->control
.intercept_exceptions
&= ~(1 << NM_VECTOR
);
946 if (!(vcpu
->cr0
& X86_CR0_TS
))
947 vcpu
->svm
->vmcb
->save
.cr0
&= ~X86_CR0_TS
;
948 vcpu
->fpu_active
= 1;
953 static int shutdown_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
956 * VMCB is undefined after a SHUTDOWN intercept
957 * so reinitialize it.
959 clear_page(vcpu
->svm
->vmcb
);
960 init_vmcb(vcpu
->svm
->vmcb
);
962 kvm_run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
966 static int io_get_override(struct kvm_vcpu
*vcpu
,
967 struct vmcb_seg
**seg
,
970 u8 inst
[MAX_INST_SIZE
];
975 rip
= vcpu
->svm
->vmcb
->save
.rip
;
976 ins_length
= vcpu
->svm
->next_rip
- rip
;
977 rip
+= vcpu
->svm
->vmcb
->save
.cs
.base
;
979 if (ins_length
> MAX_INST_SIZE
)
981 "%s: inst length err, cs base 0x%llx rip 0x%llx "
982 "next rip 0x%llx ins_length %u\n",
984 vcpu
->svm
->vmcb
->save
.cs
.base
,
985 vcpu
->svm
->vmcb
->save
.rip
,
986 vcpu
->svm
->vmcb
->control
.exit_info_2
,
989 if (kvm_read_guest(vcpu
, rip
, ins_length
, inst
) != ins_length
)
995 for (i
= 0; i
< ins_length
; i
++)
1006 *seg
= &vcpu
->svm
->vmcb
->save
.cs
;
1009 *seg
= &vcpu
->svm
->vmcb
->save
.ss
;
1012 *seg
= &vcpu
->svm
->vmcb
->save
.ds
;
1015 *seg
= &vcpu
->svm
->vmcb
->save
.es
;
1018 *seg
= &vcpu
->svm
->vmcb
->save
.fs
;
1021 *seg
= &vcpu
->svm
->vmcb
->save
.gs
;
1026 printk(KERN_DEBUG
"%s: unexpected\n", __FUNCTION__
);
1030 static unsigned long io_adress(struct kvm_vcpu
*vcpu
, int ins
, gva_t
*address
)
1032 unsigned long addr_mask
;
1034 struct vmcb_seg
*seg
;
1036 struct vmcb_save_area
*save_area
= &vcpu
->svm
->vmcb
->save
;
1037 u16 cs_attrib
= save_area
->cs
.attrib
;
1038 unsigned addr_size
= get_addr_size(vcpu
);
1040 if (!io_get_override(vcpu
, &seg
, &addr_override
))
1044 addr_size
= (addr_size
== 2) ? 4: (addr_size
>> 1);
1047 reg
= &vcpu
->regs
[VCPU_REGS_RDI
];
1048 seg
= &vcpu
->svm
->vmcb
->save
.es
;
1050 reg
= &vcpu
->regs
[VCPU_REGS_RSI
];
1051 seg
= (seg
) ? seg
: &vcpu
->svm
->vmcb
->save
.ds
;
1054 addr_mask
= ~0ULL >> (64 - (addr_size
* 8));
1056 if ((cs_attrib
& SVM_SELECTOR_L_MASK
) &&
1057 !(vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_VM
)) {
1058 *address
= (*reg
& addr_mask
);
1062 if (!(seg
->attrib
& SVM_SELECTOR_P_SHIFT
)) {
1063 svm_inject_gp(vcpu
, 0);
1067 *address
= (*reg
& addr_mask
) + seg
->base
;
1071 static int io_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1073 u32 io_info
= vcpu
->svm
->vmcb
->control
.exit_info_1
; //address size bug?
1074 int size
, down
, in
, string
, rep
;
1076 unsigned long count
;
1079 ++vcpu
->stat
.io_exits
;
1081 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->control
.exit_info_2
;
1083 in
= (io_info
& SVM_IOIO_TYPE_MASK
) != 0;
1084 port
= io_info
>> 16;
1085 size
= (io_info
& SVM_IOIO_SIZE_MASK
) >> SVM_IOIO_SIZE_SHIFT
;
1086 string
= (io_info
& SVM_IOIO_STR_MASK
) != 0;
1087 rep
= (io_info
& SVM_IOIO_REP_MASK
) != 0;
1089 down
= (vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_DF
) != 0;
1094 addr_mask
= io_adress(vcpu
, in
, &address
);
1096 printk(KERN_DEBUG
"%s: get io address failed\n",
1102 count
= vcpu
->regs
[VCPU_REGS_RCX
] & addr_mask
;
1104 return kvm_setup_pio(vcpu
, kvm_run
, in
, size
, count
, string
, down
,
1105 address
, rep
, port
);
1108 static int nop_on_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1113 static int halt_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1115 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 1;
1116 skip_emulated_instruction(vcpu
);
1117 return kvm_emulate_halt(vcpu
);
1120 static int vmmcall_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1122 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 3;
1123 skip_emulated_instruction(vcpu
);
1124 return kvm_hypercall(vcpu
, kvm_run
);
1127 static int invalid_op_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1133 static int task_switch_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1135 printk(KERN_DEBUG
"%s: task swiche is unsupported\n", __FUNCTION__
);
1136 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
1140 static int cpuid_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1142 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 2;
1143 kvm_emulate_cpuid(vcpu
);
1147 static int emulate_on_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1149 if (emulate_instruction(vcpu
, NULL
, 0, 0) != EMULATE_DONE
)
1150 printk(KERN_ERR
"%s: failed\n", __FUNCTION__
);
1154 static int svm_get_msr(struct kvm_vcpu
*vcpu
, unsigned ecx
, u64
*data
)
1157 case MSR_IA32_TIME_STAMP_COUNTER
: {
1161 *data
= vcpu
->svm
->vmcb
->control
.tsc_offset
+ tsc
;
1165 *data
= vcpu
->svm
->vmcb
->save
.star
;
1167 #ifdef CONFIG_X86_64
1169 *data
= vcpu
->svm
->vmcb
->save
.lstar
;
1172 *data
= vcpu
->svm
->vmcb
->save
.cstar
;
1174 case MSR_KERNEL_GS_BASE
:
1175 *data
= vcpu
->svm
->vmcb
->save
.kernel_gs_base
;
1177 case MSR_SYSCALL_MASK
:
1178 *data
= vcpu
->svm
->vmcb
->save
.sfmask
;
1181 case MSR_IA32_SYSENTER_CS
:
1182 *data
= vcpu
->svm
->vmcb
->save
.sysenter_cs
;
1184 case MSR_IA32_SYSENTER_EIP
:
1185 *data
= vcpu
->svm
->vmcb
->save
.sysenter_eip
;
1187 case MSR_IA32_SYSENTER_ESP
:
1188 *data
= vcpu
->svm
->vmcb
->save
.sysenter_esp
;
1191 return kvm_get_msr_common(vcpu
, ecx
, data
);
1196 static int rdmsr_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1198 u32 ecx
= vcpu
->regs
[VCPU_REGS_RCX
];
1201 if (svm_get_msr(vcpu
, ecx
, &data
))
1202 svm_inject_gp(vcpu
, 0);
1204 vcpu
->svm
->vmcb
->save
.rax
= data
& 0xffffffff;
1205 vcpu
->regs
[VCPU_REGS_RDX
] = data
>> 32;
1206 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 2;
1207 skip_emulated_instruction(vcpu
);
1212 static int svm_set_msr(struct kvm_vcpu
*vcpu
, unsigned ecx
, u64 data
)
1215 case MSR_IA32_TIME_STAMP_COUNTER
: {
1219 vcpu
->svm
->vmcb
->control
.tsc_offset
= data
- tsc
;
1223 vcpu
->svm
->vmcb
->save
.star
= data
;
1225 #ifdef CONFIG_X86_64
1227 vcpu
->svm
->vmcb
->save
.lstar
= data
;
1230 vcpu
->svm
->vmcb
->save
.cstar
= data
;
1232 case MSR_KERNEL_GS_BASE
:
1233 vcpu
->svm
->vmcb
->save
.kernel_gs_base
= data
;
1235 case MSR_SYSCALL_MASK
:
1236 vcpu
->svm
->vmcb
->save
.sfmask
= data
;
1239 case MSR_IA32_SYSENTER_CS
:
1240 vcpu
->svm
->vmcb
->save
.sysenter_cs
= data
;
1242 case MSR_IA32_SYSENTER_EIP
:
1243 vcpu
->svm
->vmcb
->save
.sysenter_eip
= data
;
1245 case MSR_IA32_SYSENTER_ESP
:
1246 vcpu
->svm
->vmcb
->save
.sysenter_esp
= data
;
1249 return kvm_set_msr_common(vcpu
, ecx
, data
);
1254 static int wrmsr_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1256 u32 ecx
= vcpu
->regs
[VCPU_REGS_RCX
];
1257 u64 data
= (vcpu
->svm
->vmcb
->save
.rax
& -1u)
1258 | ((u64
)(vcpu
->regs
[VCPU_REGS_RDX
] & -1u) << 32);
1259 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 2;
1260 if (svm_set_msr(vcpu
, ecx
, data
))
1261 svm_inject_gp(vcpu
, 0);
1263 skip_emulated_instruction(vcpu
);
1267 static int msr_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1269 if (vcpu
->svm
->vmcb
->control
.exit_info_1
)
1270 return wrmsr_interception(vcpu
, kvm_run
);
1272 return rdmsr_interception(vcpu
, kvm_run
);
1275 static int interrupt_window_interception(struct kvm_vcpu
*vcpu
,
1276 struct kvm_run
*kvm_run
)
1279 * If the user space waits to inject interrupts, exit as soon as
1282 if (kvm_run
->request_interrupt_window
&&
1283 !vcpu
->irq_summary
) {
1284 ++vcpu
->stat
.irq_window_exits
;
1285 kvm_run
->exit_reason
= KVM_EXIT_IRQ_WINDOW_OPEN
;
1292 static int (*svm_exit_handlers
[])(struct kvm_vcpu
*vcpu
,
1293 struct kvm_run
*kvm_run
) = {
1294 [SVM_EXIT_READ_CR0
] = emulate_on_interception
,
1295 [SVM_EXIT_READ_CR3
] = emulate_on_interception
,
1296 [SVM_EXIT_READ_CR4
] = emulate_on_interception
,
1298 [SVM_EXIT_WRITE_CR0
] = emulate_on_interception
,
1299 [SVM_EXIT_WRITE_CR3
] = emulate_on_interception
,
1300 [SVM_EXIT_WRITE_CR4
] = emulate_on_interception
,
1301 [SVM_EXIT_READ_DR0
] = emulate_on_interception
,
1302 [SVM_EXIT_READ_DR1
] = emulate_on_interception
,
1303 [SVM_EXIT_READ_DR2
] = emulate_on_interception
,
1304 [SVM_EXIT_READ_DR3
] = emulate_on_interception
,
1305 [SVM_EXIT_WRITE_DR0
] = emulate_on_interception
,
1306 [SVM_EXIT_WRITE_DR1
] = emulate_on_interception
,
1307 [SVM_EXIT_WRITE_DR2
] = emulate_on_interception
,
1308 [SVM_EXIT_WRITE_DR3
] = emulate_on_interception
,
1309 [SVM_EXIT_WRITE_DR5
] = emulate_on_interception
,
1310 [SVM_EXIT_WRITE_DR7
] = emulate_on_interception
,
1311 [SVM_EXIT_EXCP_BASE
+ PF_VECTOR
] = pf_interception
,
1312 [SVM_EXIT_EXCP_BASE
+ NM_VECTOR
] = nm_interception
,
1313 [SVM_EXIT_INTR
] = nop_on_interception
,
1314 [SVM_EXIT_NMI
] = nop_on_interception
,
1315 [SVM_EXIT_SMI
] = nop_on_interception
,
1316 [SVM_EXIT_INIT
] = nop_on_interception
,
1317 [SVM_EXIT_VINTR
] = interrupt_window_interception
,
1318 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
1319 [SVM_EXIT_CPUID
] = cpuid_interception
,
1320 [SVM_EXIT_HLT
] = halt_interception
,
1321 [SVM_EXIT_INVLPG
] = emulate_on_interception
,
1322 [SVM_EXIT_INVLPGA
] = invalid_op_interception
,
1323 [SVM_EXIT_IOIO
] = io_interception
,
1324 [SVM_EXIT_MSR
] = msr_interception
,
1325 [SVM_EXIT_TASK_SWITCH
] = task_switch_interception
,
1326 [SVM_EXIT_SHUTDOWN
] = shutdown_interception
,
1327 [SVM_EXIT_VMRUN
] = invalid_op_interception
,
1328 [SVM_EXIT_VMMCALL
] = vmmcall_interception
,
1329 [SVM_EXIT_VMLOAD
] = invalid_op_interception
,
1330 [SVM_EXIT_VMSAVE
] = invalid_op_interception
,
1331 [SVM_EXIT_STGI
] = invalid_op_interception
,
1332 [SVM_EXIT_CLGI
] = invalid_op_interception
,
1333 [SVM_EXIT_SKINIT
] = invalid_op_interception
,
1334 [SVM_EXIT_MONITOR
] = invalid_op_interception
,
1335 [SVM_EXIT_MWAIT
] = invalid_op_interception
,
1339 static int handle_exit(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1341 u32 exit_code
= vcpu
->svm
->vmcb
->control
.exit_code
;
1343 if (is_external_interrupt(vcpu
->svm
->vmcb
->control
.exit_int_info
) &&
1344 exit_code
!= SVM_EXIT_EXCP_BASE
+ PF_VECTOR
)
1345 printk(KERN_ERR
"%s: unexpected exit_ini_info 0x%x "
1347 __FUNCTION__
, vcpu
->svm
->vmcb
->control
.exit_int_info
,
1350 if (exit_code
>= ARRAY_SIZE(svm_exit_handlers
)
1351 || svm_exit_handlers
[exit_code
] == 0) {
1352 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
1353 kvm_run
->hw
.hardware_exit_reason
= exit_code
;
1357 return svm_exit_handlers
[exit_code
](vcpu
, kvm_run
);
1360 static void reload_tss(struct kvm_vcpu
*vcpu
)
1362 int cpu
= raw_smp_processor_id();
1364 struct svm_cpu_data
*svm_data
= per_cpu(svm_data
, cpu
);
1365 svm_data
->tss_desc
->type
= 9; //available 32/64-bit TSS
1369 static void pre_svm_run(struct kvm_vcpu
*vcpu
)
1371 int cpu
= raw_smp_processor_id();
1373 struct svm_cpu_data
*svm_data
= per_cpu(svm_data
, cpu
);
1375 vcpu
->svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_DO_NOTHING
;
1376 if (vcpu
->cpu
!= cpu
||
1377 vcpu
->svm
->asid_generation
!= svm_data
->asid_generation
)
1378 new_asid(vcpu
, svm_data
);
1382 static inline void kvm_do_inject_irq(struct kvm_vcpu
*vcpu
)
1384 struct vmcb_control_area
*control
;
1386 control
= &vcpu
->svm
->vmcb
->control
;
1387 control
->int_vector
= pop_irq(vcpu
);
1388 control
->int_ctl
&= ~V_INTR_PRIO_MASK
;
1389 control
->int_ctl
|= V_IRQ_MASK
|
1390 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT
);
1393 static void kvm_reput_irq(struct kvm_vcpu
*vcpu
)
1395 struct vmcb_control_area
*control
= &vcpu
->svm
->vmcb
->control
;
1397 if (control
->int_ctl
& V_IRQ_MASK
) {
1398 control
->int_ctl
&= ~V_IRQ_MASK
;
1399 push_irq(vcpu
, control
->int_vector
);
1402 vcpu
->interrupt_window_open
=
1403 !(control
->int_state
& SVM_INTERRUPT_SHADOW_MASK
);
1406 static void do_interrupt_requests(struct kvm_vcpu
*vcpu
,
1407 struct kvm_run
*kvm_run
)
1409 struct vmcb_control_area
*control
= &vcpu
->svm
->vmcb
->control
;
1411 vcpu
->interrupt_window_open
=
1412 (!(control
->int_state
& SVM_INTERRUPT_SHADOW_MASK
) &&
1413 (vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_IF
));
1415 if (vcpu
->interrupt_window_open
&& vcpu
->irq_summary
)
1417 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1419 kvm_do_inject_irq(vcpu
);
1422 * Interrupts blocked. Wait for unblock.
1424 if (!vcpu
->interrupt_window_open
&&
1425 (vcpu
->irq_summary
|| kvm_run
->request_interrupt_window
)) {
1426 control
->intercept
|= 1ULL << INTERCEPT_VINTR
;
1428 control
->intercept
&= ~(1ULL << INTERCEPT_VINTR
);
1431 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
,
1432 struct kvm_run
*kvm_run
)
1434 kvm_run
->ready_for_interrupt_injection
= (vcpu
->interrupt_window_open
&&
1435 vcpu
->irq_summary
== 0);
1436 kvm_run
->if_flag
= (vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_IF
) != 0;
1437 kvm_run
->cr8
= vcpu
->cr8
;
1438 kvm_run
->apic_base
= vcpu
->apic_base
;
1442 * Check if userspace requested an interrupt window, and that the
1443 * interrupt window is open.
1445 * No need to exit to userspace if we already have an interrupt queued.
1447 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
,
1448 struct kvm_run
*kvm_run
)
1450 return (!vcpu
->irq_summary
&&
1451 kvm_run
->request_interrupt_window
&&
1452 vcpu
->interrupt_window_open
&&
1453 (vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_IF
));
1456 static void save_db_regs(unsigned long *db_regs
)
1458 asm volatile ("mov %%dr0, %0" : "=r"(db_regs
[0]));
1459 asm volatile ("mov %%dr1, %0" : "=r"(db_regs
[1]));
1460 asm volatile ("mov %%dr2, %0" : "=r"(db_regs
[2]));
1461 asm volatile ("mov %%dr3, %0" : "=r"(db_regs
[3]));
1464 static void load_db_regs(unsigned long *db_regs
)
1466 asm volatile ("mov %0, %%dr0" : : "r"(db_regs
[0]));
1467 asm volatile ("mov %0, %%dr1" : : "r"(db_regs
[1]));
1468 asm volatile ("mov %0, %%dr2" : : "r"(db_regs
[2]));
1469 asm volatile ("mov %0, %%dr3" : : "r"(db_regs
[3]));
1472 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
)
1474 force_new_asid(vcpu
);
1477 static int svm_vcpu_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1485 r
= kvm_mmu_reload(vcpu
);
1489 if (!vcpu
->mmio_read_completed
)
1490 do_interrupt_requests(vcpu
, kvm_run
);
1494 vcpu
->guest_mode
= 1;
1496 if (test_and_clear_bit(KVM_TLB_FLUSH
, &vcpu
->requests
))
1497 svm_flush_tlb(vcpu
);
1501 save_host_msrs(vcpu
);
1502 fs_selector
= read_fs();
1503 gs_selector
= read_gs();
1504 ldt_selector
= read_ldt();
1505 vcpu
->svm
->host_cr2
= kvm_read_cr2();
1506 vcpu
->svm
->host_dr6
= read_dr6();
1507 vcpu
->svm
->host_dr7
= read_dr7();
1508 vcpu
->svm
->vmcb
->save
.cr2
= vcpu
->cr2
;
1510 if (vcpu
->svm
->vmcb
->save
.dr7
& 0xff) {
1512 save_db_regs(vcpu
->svm
->host_db_regs
);
1513 load_db_regs(vcpu
->svm
->db_regs
);
1516 if (vcpu
->fpu_active
) {
1517 fx_save(vcpu
->host_fx_image
);
1518 fx_restore(vcpu
->guest_fx_image
);
1522 #ifdef CONFIG_X86_64
1523 "push %%rbx; push %%rcx; push %%rdx;"
1524 "push %%rsi; push %%rdi; push %%rbp;"
1525 "push %%r8; push %%r9; push %%r10; push %%r11;"
1526 "push %%r12; push %%r13; push %%r14; push %%r15;"
1528 "push %%ebx; push %%ecx; push %%edx;"
1529 "push %%esi; push %%edi; push %%ebp;"
1532 #ifdef CONFIG_X86_64
1533 "mov %c[rbx](%[vcpu]), %%rbx \n\t"
1534 "mov %c[rcx](%[vcpu]), %%rcx \n\t"
1535 "mov %c[rdx](%[vcpu]), %%rdx \n\t"
1536 "mov %c[rsi](%[vcpu]), %%rsi \n\t"
1537 "mov %c[rdi](%[vcpu]), %%rdi \n\t"
1538 "mov %c[rbp](%[vcpu]), %%rbp \n\t"
1539 "mov %c[r8](%[vcpu]), %%r8 \n\t"
1540 "mov %c[r9](%[vcpu]), %%r9 \n\t"
1541 "mov %c[r10](%[vcpu]), %%r10 \n\t"
1542 "mov %c[r11](%[vcpu]), %%r11 \n\t"
1543 "mov %c[r12](%[vcpu]), %%r12 \n\t"
1544 "mov %c[r13](%[vcpu]), %%r13 \n\t"
1545 "mov %c[r14](%[vcpu]), %%r14 \n\t"
1546 "mov %c[r15](%[vcpu]), %%r15 \n\t"
1548 "mov %c[rbx](%[vcpu]), %%ebx \n\t"
1549 "mov %c[rcx](%[vcpu]), %%ecx \n\t"
1550 "mov %c[rdx](%[vcpu]), %%edx \n\t"
1551 "mov %c[rsi](%[vcpu]), %%esi \n\t"
1552 "mov %c[rdi](%[vcpu]), %%edi \n\t"
1553 "mov %c[rbp](%[vcpu]), %%ebp \n\t"
1556 #ifdef CONFIG_X86_64
1557 /* Enter guest mode */
1559 "mov %c[svm](%[vcpu]), %%rax \n\t"
1560 "mov %c[vmcb](%%rax), %%rax \n\t"
1566 /* Enter guest mode */
1568 "mov %c[svm](%[vcpu]), %%eax \n\t"
1569 "mov %c[vmcb](%%eax), %%eax \n\t"
1576 /* Save guest registers, load host registers */
1577 #ifdef CONFIG_X86_64
1578 "mov %%rbx, %c[rbx](%[vcpu]) \n\t"
1579 "mov %%rcx, %c[rcx](%[vcpu]) \n\t"
1580 "mov %%rdx, %c[rdx](%[vcpu]) \n\t"
1581 "mov %%rsi, %c[rsi](%[vcpu]) \n\t"
1582 "mov %%rdi, %c[rdi](%[vcpu]) \n\t"
1583 "mov %%rbp, %c[rbp](%[vcpu]) \n\t"
1584 "mov %%r8, %c[r8](%[vcpu]) \n\t"
1585 "mov %%r9, %c[r9](%[vcpu]) \n\t"
1586 "mov %%r10, %c[r10](%[vcpu]) \n\t"
1587 "mov %%r11, %c[r11](%[vcpu]) \n\t"
1588 "mov %%r12, %c[r12](%[vcpu]) \n\t"
1589 "mov %%r13, %c[r13](%[vcpu]) \n\t"
1590 "mov %%r14, %c[r14](%[vcpu]) \n\t"
1591 "mov %%r15, %c[r15](%[vcpu]) \n\t"
1593 "pop %%r15; pop %%r14; pop %%r13; pop %%r12;"
1594 "pop %%r11; pop %%r10; pop %%r9; pop %%r8;"
1595 "pop %%rbp; pop %%rdi; pop %%rsi;"
1596 "pop %%rdx; pop %%rcx; pop %%rbx; \n\t"
1598 "mov %%ebx, %c[rbx](%[vcpu]) \n\t"
1599 "mov %%ecx, %c[rcx](%[vcpu]) \n\t"
1600 "mov %%edx, %c[rdx](%[vcpu]) \n\t"
1601 "mov %%esi, %c[rsi](%[vcpu]) \n\t"
1602 "mov %%edi, %c[rdi](%[vcpu]) \n\t"
1603 "mov %%ebp, %c[rbp](%[vcpu]) \n\t"
1605 "pop %%ebp; pop %%edi; pop %%esi;"
1606 "pop %%edx; pop %%ecx; pop %%ebx; \n\t"
1610 [svm
]"i"(offsetof(struct kvm_vcpu
, svm
)),
1611 [vmcb
]"i"(offsetof(struct vcpu_svm
, vmcb_pa
)),
1612 [rbx
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RBX
])),
1613 [rcx
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RCX
])),
1614 [rdx
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RDX
])),
1615 [rsi
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RSI
])),
1616 [rdi
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RDI
])),
1617 [rbp
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RBP
]))
1618 #ifdef CONFIG_X86_64
1619 ,[r8
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R8
])),
1620 [r9
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R9
])),
1621 [r10
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R10
])),
1622 [r11
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R11
])),
1623 [r12
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R12
])),
1624 [r13
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R13
])),
1625 [r14
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R14
])),
1626 [r15
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R15
]))
1630 vcpu
->guest_mode
= 0;
1632 if (vcpu
->fpu_active
) {
1633 fx_save(vcpu
->guest_fx_image
);
1634 fx_restore(vcpu
->host_fx_image
);
1637 if ((vcpu
->svm
->vmcb
->save
.dr7
& 0xff))
1638 load_db_regs(vcpu
->svm
->host_db_regs
);
1640 vcpu
->cr2
= vcpu
->svm
->vmcb
->save
.cr2
;
1642 write_dr6(vcpu
->svm
->host_dr6
);
1643 write_dr7(vcpu
->svm
->host_dr7
);
1644 kvm_write_cr2(vcpu
->svm
->host_cr2
);
1646 load_fs(fs_selector
);
1647 load_gs(gs_selector
);
1648 load_ldt(ldt_selector
);
1649 load_host_msrs(vcpu
);
1654 * Profile KVM exit RIPs:
1656 if (unlikely(prof_on
== KVM_PROFILING
))
1657 profile_hit(KVM_PROFILING
,
1658 (void *)(unsigned long)vcpu
->svm
->vmcb
->save
.rip
);
1662 kvm_reput_irq(vcpu
);
1664 vcpu
->svm
->next_rip
= 0;
1666 if (vcpu
->svm
->vmcb
->control
.exit_code
== SVM_EXIT_ERR
) {
1667 kvm_run
->exit_reason
= KVM_EXIT_FAIL_ENTRY
;
1668 kvm_run
->fail_entry
.hardware_entry_failure_reason
1669 = vcpu
->svm
->vmcb
->control
.exit_code
;
1670 post_kvm_run_save(vcpu
, kvm_run
);
1674 r
= handle_exit(vcpu
, kvm_run
);
1676 if (signal_pending(current
)) {
1677 ++vcpu
->stat
.signal_exits
;
1678 post_kvm_run_save(vcpu
, kvm_run
);
1679 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
1683 if (dm_request_for_irq_injection(vcpu
, kvm_run
)) {
1684 ++vcpu
->stat
.request_irq_exits
;
1685 post_kvm_run_save(vcpu
, kvm_run
);
1686 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
1692 post_kvm_run_save(vcpu
, kvm_run
);
1696 static void svm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
1698 vcpu
->svm
->vmcb
->save
.cr3
= root
;
1699 force_new_asid(vcpu
);
1701 if (vcpu
->fpu_active
) {
1702 vcpu
->svm
->vmcb
->control
.intercept_exceptions
|= (1 << NM_VECTOR
);
1703 vcpu
->svm
->vmcb
->save
.cr0
|= X86_CR0_TS
;
1704 vcpu
->fpu_active
= 0;
1708 static void svm_inject_page_fault(struct kvm_vcpu
*vcpu
,
1712 uint32_t exit_int_info
= vcpu
->svm
->vmcb
->control
.exit_int_info
;
1714 ++vcpu
->stat
.pf_guest
;
1716 if (is_page_fault(exit_int_info
)) {
1718 vcpu
->svm
->vmcb
->control
.event_inj_err
= 0;
1719 vcpu
->svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
|
1720 SVM_EVTINJ_VALID_ERR
|
1721 SVM_EVTINJ_TYPE_EXEPT
|
1726 vcpu
->svm
->vmcb
->save
.cr2
= addr
;
1727 vcpu
->svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
|
1728 SVM_EVTINJ_VALID_ERR
|
1729 SVM_EVTINJ_TYPE_EXEPT
|
1731 vcpu
->svm
->vmcb
->control
.event_inj_err
= err_code
;
1735 static int is_disabled(void)
1739 rdmsrl(MSR_VM_CR
, vm_cr
);
1740 if (vm_cr
& (1 << SVM_VM_CR_SVM_DISABLE
))
1747 svm_patch_hypercall(struct kvm_vcpu
*vcpu
, unsigned char *hypercall
)
1750 * Patch in the VMMCALL instruction:
1752 hypercall
[0] = 0x0f;
1753 hypercall
[1] = 0x01;
1754 hypercall
[2] = 0xd9;
1755 hypercall
[3] = 0xc3;
1758 static struct kvm_arch_ops svm_arch_ops
= {
1759 .cpu_has_kvm_support
= has_svm
,
1760 .disabled_by_bios
= is_disabled
,
1761 .hardware_setup
= svm_hardware_setup
,
1762 .hardware_unsetup
= svm_hardware_unsetup
,
1763 .hardware_enable
= svm_hardware_enable
,
1764 .hardware_disable
= svm_hardware_disable
,
1766 .vcpu_create
= svm_create_vcpu
,
1767 .vcpu_free
= svm_free_vcpu
,
1769 .vcpu_load
= svm_vcpu_load
,
1770 .vcpu_put
= svm_vcpu_put
,
1771 .vcpu_decache
= svm_vcpu_decache
,
1773 .set_guest_debug
= svm_guest_debug
,
1774 .get_msr
= svm_get_msr
,
1775 .set_msr
= svm_set_msr
,
1776 .get_segment_base
= svm_get_segment_base
,
1777 .get_segment
= svm_get_segment
,
1778 .set_segment
= svm_set_segment
,
1779 .get_cs_db_l_bits
= svm_get_cs_db_l_bits
,
1780 .decache_cr4_guest_bits
= svm_decache_cr4_guest_bits
,
1781 .set_cr0
= svm_set_cr0
,
1782 .set_cr3
= svm_set_cr3
,
1783 .set_cr4
= svm_set_cr4
,
1784 .set_efer
= svm_set_efer
,
1785 .get_idt
= svm_get_idt
,
1786 .set_idt
= svm_set_idt
,
1787 .get_gdt
= svm_get_gdt
,
1788 .set_gdt
= svm_set_gdt
,
1789 .get_dr
= svm_get_dr
,
1790 .set_dr
= svm_set_dr
,
1791 .cache_regs
= svm_cache_regs
,
1792 .decache_regs
= svm_decache_regs
,
1793 .get_rflags
= svm_get_rflags
,
1794 .set_rflags
= svm_set_rflags
,
1796 .invlpg
= svm_invlpg
,
1797 .tlb_flush
= svm_flush_tlb
,
1798 .inject_page_fault
= svm_inject_page_fault
,
1800 .inject_gp
= svm_inject_gp
,
1802 .run
= svm_vcpu_run
,
1803 .skip_emulated_instruction
= skip_emulated_instruction
,
1804 .vcpu_setup
= svm_vcpu_setup
,
1805 .patch_hypercall
= svm_patch_hypercall
,
1808 static int __init
svm_init(void)
1810 return kvm_init_arch(&svm_arch_ops
, THIS_MODULE
);
1813 static void __exit
svm_exit(void)
1818 module_init(svm_init
)
1819 module_exit(svm_exit
)