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.
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/highmem.h>
23 #include "x86_emulate.h"
25 MODULE_AUTHOR("Qumranet");
26 MODULE_LICENSE("GPL");
28 #define IOPM_ALLOC_ORDER 2
29 #define MSRPM_ALLOC_ORDER 1
35 #define DR7_GD_MASK (1 << 13)
36 #define DR6_BD_MASK (1 << 13)
37 #define CR4_DE_MASK (1UL << 3)
39 #define SEG_TYPE_LDT 2
40 #define SEG_TYPE_BUSY_TSS16 3
42 #define KVM_EFER_LMA (1 << 10)
43 #define KVM_EFER_LME (1 << 8)
45 unsigned long iopm_base
;
46 unsigned long msrpm_base
;
48 struct kvm_ldttss_desc
{
51 unsigned base1
: 8, type
: 5, dpl
: 2, p
: 1;
52 unsigned limit1
: 4, zero0
: 3, g
: 1, base2
: 8;
55 } __attribute__((packed
));
60 uint64_t asid_generation
;
63 struct kvm_ldttss_desc
*tss_desc
;
65 struct page
*save_area
;
68 static DEFINE_PER_CPU(struct svm_cpu_data
*, svm_data
);
70 struct svm_init_data
{
75 static u32 msrpm_ranges
[] = {0, 0xc0000000, 0xc0010000};
77 #define NUM_MSR_MAPS (sizeof(msrpm_ranges) / sizeof(*msrpm_ranges))
78 #define MSRS_RANGE_SIZE 2048
79 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
81 #define MAX_INST_SIZE 15
83 static unsigned get_addr_size(struct kvm_vcpu
*vcpu
)
85 struct vmcb_save_area
*sa
= &vcpu
->svm
->vmcb
->save
;
88 if (!(sa
->cr0
& CR0_PE_MASK
) || (sa
->rflags
& X86_EFLAGS_VM
))
91 cs_attrib
= sa
->cs
.attrib
;
93 return (cs_attrib
& SVM_SELECTOR_L_MASK
) ? 8 :
94 (cs_attrib
& SVM_SELECTOR_DB_MASK
) ? 4 : 2;
97 static inline u8
pop_irq(struct kvm_vcpu
*vcpu
)
99 int word_index
= __ffs(vcpu
->irq_summary
);
100 int bit_index
= __ffs(vcpu
->irq_pending
[word_index
]);
101 int irq
= word_index
* BITS_PER_LONG
+ bit_index
;
103 clear_bit(bit_index
, &vcpu
->irq_pending
[word_index
]);
104 if (!vcpu
->irq_pending
[word_index
])
105 clear_bit(word_index
, &vcpu
->irq_summary
);
109 static inline void push_irq(struct kvm_vcpu
*vcpu
, u8 irq
)
111 set_bit(irq
, vcpu
->irq_pending
);
112 set_bit(irq
/ BITS_PER_LONG
, &vcpu
->irq_summary
);
115 static inline void clgi(void)
117 asm volatile (SVM_CLGI
);
120 static inline void stgi(void)
122 asm volatile (SVM_STGI
);
125 static inline void invlpga(unsigned long addr
, u32 asid
)
127 asm volatile (SVM_INVLPGA :: "a"(addr
), "c"(asid
));
130 static inline unsigned long kvm_read_cr2(void)
134 asm volatile ("mov %%cr2, %0" : "=r" (cr2
));
138 static inline void kvm_write_cr2(unsigned long val
)
140 asm volatile ("mov %0, %%cr2" :: "r" (val
));
143 static inline unsigned long read_dr6(void)
147 asm volatile ("mov %%dr6, %0" : "=r" (dr6
));
151 static inline void write_dr6(unsigned long val
)
153 asm volatile ("mov %0, %%dr6" :: "r" (val
));
156 static inline unsigned long read_dr7(void)
160 asm volatile ("mov %%dr7, %0" : "=r" (dr7
));
164 static inline void write_dr7(unsigned long val
)
166 asm volatile ("mov %0, %%dr7" :: "r" (val
));
169 static inline int svm_is_long_mode(struct kvm_vcpu
*vcpu
)
171 return vcpu
->svm
->vmcb
->save
.efer
& KVM_EFER_LMA
;
174 static inline void force_new_asid(struct kvm_vcpu
*vcpu
)
176 vcpu
->svm
->asid_generation
--;
179 static inline void flush_guest_tlb(struct kvm_vcpu
*vcpu
)
181 force_new_asid(vcpu
);
184 static void svm_set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
186 if (!(efer
& KVM_EFER_LMA
))
187 efer
&= ~KVM_EFER_LME
;
189 vcpu
->svm
->vmcb
->save
.efer
= efer
| MSR_EFER_SVME_MASK
;
190 vcpu
->shadow_efer
= efer
;
193 static void svm_inject_gp(struct kvm_vcpu
*vcpu
, unsigned error_code
)
195 vcpu
->svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
|
196 SVM_EVTINJ_VALID_ERR
|
197 SVM_EVTINJ_TYPE_EXEPT
|
199 vcpu
->svm
->vmcb
->control
.event_inj_err
= error_code
;
202 static void inject_ud(struct kvm_vcpu
*vcpu
)
204 vcpu
->svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
|
205 SVM_EVTINJ_TYPE_EXEPT
|
209 static void inject_db(struct kvm_vcpu
*vcpu
)
211 vcpu
->svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
|
212 SVM_EVTINJ_TYPE_EXEPT
|
216 static int is_page_fault(uint32_t info
)
218 info
&= SVM_EVTINJ_VEC_MASK
| SVM_EVTINJ_TYPE_MASK
| SVM_EVTINJ_VALID
;
219 return info
== (PF_VECTOR
| SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_EXEPT
);
222 static int is_external_interrupt(u32 info
)
224 info
&= SVM_EVTINJ_TYPE_MASK
| SVM_EVTINJ_VALID
;
225 return info
== (SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
);
228 static void skip_emulated_instruction(struct kvm_vcpu
*vcpu
)
230 if (!vcpu
->svm
->next_rip
) {
231 printk(KERN_DEBUG
"%s: NOP\n", __FUNCTION__
);
234 if (vcpu
->svm
->next_rip
- vcpu
->svm
->vmcb
->save
.rip
> 15) {
235 printk(KERN_ERR
"%s: ip 0x%llx next 0x%llx\n",
237 vcpu
->svm
->vmcb
->save
.rip
,
238 vcpu
->svm
->next_rip
);
241 vcpu
->rip
= vcpu
->svm
->vmcb
->save
.rip
= vcpu
->svm
->next_rip
;
242 vcpu
->svm
->vmcb
->control
.int_state
&= ~SVM_INTERRUPT_SHADOW_MASK
;
245 static int has_svm(void)
247 uint32_t eax
, ebx
, ecx
, edx
;
249 if (current_cpu_data
.x86_vendor
!= X86_VENDOR_AMD
) {
250 printk(KERN_INFO
"has_svm: not amd\n");
254 cpuid(0x80000000, &eax
, &ebx
, &ecx
, &edx
);
255 if (eax
< SVM_CPUID_FUNC
) {
256 printk(KERN_INFO
"has_svm: can't execute cpuid_8000000a\n");
260 cpuid(0x80000001, &eax
, &ebx
, &ecx
, &edx
);
261 if (!(ecx
& (1 << SVM_CPUID_FEATURE_SHIFT
))) {
262 printk(KERN_DEBUG
"has_svm: svm not available\n");
268 static void svm_hardware_disable(void *garbage
)
270 struct svm_cpu_data
*svm_data
271 = per_cpu(svm_data
, raw_smp_processor_id());
276 wrmsrl(MSR_VM_HSAVE_PA
, 0);
277 rdmsrl(MSR_EFER
, efer
);
278 wrmsrl(MSR_EFER
, efer
& ~MSR_EFER_SVME_MASK
);
279 per_cpu(svm_data
, raw_smp_processor_id()) = 0;
280 __free_page(svm_data
->save_area
);
285 static void svm_hardware_enable(void *garbage
)
288 struct svm_cpu_data
*svm_data
;
291 struct desc_ptr gdt_descr
;
293 struct Xgt_desc_struct gdt_descr
;
295 struct desc_struct
*gdt
;
296 int me
= raw_smp_processor_id();
299 printk(KERN_ERR
"svm_cpu_init: err EOPNOTSUPP on %d\n", me
);
302 svm_data
= per_cpu(svm_data
, me
);
305 printk(KERN_ERR
"svm_cpu_init: svm_data is NULL on %d\n",
310 svm_data
->asid_generation
= 1;
311 svm_data
->max_asid
= cpuid_ebx(SVM_CPUID_FUNC
) - 1;
312 svm_data
->next_asid
= svm_data
->max_asid
+ 1;
314 asm volatile ( "sgdt %0" : "=m"(gdt_descr
) );
315 gdt
= (struct desc_struct
*)gdt_descr
.address
;
316 svm_data
->tss_desc
= (struct kvm_ldttss_desc
*)(gdt
+ GDT_ENTRY_TSS
);
318 rdmsrl(MSR_EFER
, efer
);
319 wrmsrl(MSR_EFER
, efer
| MSR_EFER_SVME_MASK
);
321 wrmsrl(MSR_VM_HSAVE_PA
,
322 page_to_pfn(svm_data
->save_area
) << PAGE_SHIFT
);
325 static int svm_cpu_init(int cpu
)
327 struct svm_cpu_data
*svm_data
;
330 svm_data
= kzalloc(sizeof(struct svm_cpu_data
), GFP_KERNEL
);
334 svm_data
->save_area
= alloc_page(GFP_KERNEL
);
336 if (!svm_data
->save_area
)
339 per_cpu(svm_data
, cpu
) = svm_data
;
349 static int set_msr_interception(u32
*msrpm
, unsigned msr
,
354 for (i
= 0; i
< NUM_MSR_MAPS
; i
++) {
355 if (msr
>= msrpm_ranges
[i
] &&
356 msr
< msrpm_ranges
[i
] + MSRS_IN_RANGE
) {
357 u32 msr_offset
= (i
* MSRS_IN_RANGE
+ msr
-
358 msrpm_ranges
[i
]) * 2;
360 u32
*base
= msrpm
+ (msr_offset
/ 32);
361 u32 msr_shift
= msr_offset
% 32;
362 u32 mask
= ((write
) ? 0 : 2) | ((read
) ? 0 : 1);
363 *base
= (*base
& ~(0x3 << msr_shift
)) |
368 printk(KERN_DEBUG
"%s: not found 0x%x\n", __FUNCTION__
, msr
);
372 static __init
int svm_hardware_setup(void)
375 struct page
*iopm_pages
;
376 struct page
*msrpm_pages
;
380 kvm_emulator_want_group7_invlpg();
382 iopm_pages
= alloc_pages(GFP_KERNEL
, IOPM_ALLOC_ORDER
);
386 memset(page_address(iopm_pages
), 0xff,
387 PAGE_SIZE
* (1 << IOPM_ALLOC_ORDER
));
388 iopm_base
= page_to_pfn(iopm_pages
) << PAGE_SHIFT
;
391 msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
397 msrpm_va
= page_address(msrpm_pages
);
398 memset(msrpm_va
, 0xff, PAGE_SIZE
* (1 << MSRPM_ALLOC_ORDER
));
399 msrpm_base
= page_to_pfn(msrpm_pages
) << PAGE_SHIFT
;
402 set_msr_interception(msrpm_va
, MSR_GS_BASE
, 1, 1);
403 set_msr_interception(msrpm_va
, MSR_FS_BASE
, 1, 1);
404 set_msr_interception(msrpm_va
, MSR_KERNEL_GS_BASE
, 1, 1);
405 set_msr_interception(msrpm_va
, MSR_STAR
, 1, 1);
406 set_msr_interception(msrpm_va
, MSR_LSTAR
, 1, 1);
407 set_msr_interception(msrpm_va
, MSR_CSTAR
, 1, 1);
408 set_msr_interception(msrpm_va
, MSR_SYSCALL_MASK
, 1, 1);
410 set_msr_interception(msrpm_va
, MSR_IA32_SYSENTER_CS
, 1, 1);
411 set_msr_interception(msrpm_va
, MSR_IA32_SYSENTER_ESP
, 1, 1);
412 set_msr_interception(msrpm_va
, MSR_IA32_SYSENTER_EIP
, 1, 1);
414 for_each_online_cpu(cpu
) {
415 r
= svm_cpu_init(cpu
);
422 __free_pages(msrpm_pages
, MSRPM_ALLOC_ORDER
);
425 __free_pages(iopm_pages
, IOPM_ALLOC_ORDER
);
430 static __exit
void svm_hardware_unsetup(void)
432 __free_pages(pfn_to_page(msrpm_base
>> PAGE_SHIFT
), MSRPM_ALLOC_ORDER
);
433 __free_pages(pfn_to_page(iopm_base
>> PAGE_SHIFT
), IOPM_ALLOC_ORDER
);
434 iopm_base
= msrpm_base
= 0;
437 static void init_seg(struct vmcb_seg
*seg
)
440 seg
->attrib
= SVM_SELECTOR_P_MASK
| SVM_SELECTOR_S_MASK
|
441 SVM_SELECTOR_WRITE_MASK
; /* Read/Write Data Segment */
446 static void init_sys_seg(struct vmcb_seg
*seg
, uint32_t type
)
449 seg
->attrib
= SVM_SELECTOR_P_MASK
| type
;
454 static int svm_vcpu_setup(struct kvm_vcpu
*vcpu
)
459 static void init_vmcb(struct vmcb
*vmcb
)
461 struct vmcb_control_area
*control
= &vmcb
->control
;
462 struct vmcb_save_area
*save
= &vmcb
->save
;
465 control
->intercept_cr_read
= INTERCEPT_CR0_MASK
|
469 control
->intercept_cr_write
= INTERCEPT_CR0_MASK
|
473 control
->intercept_dr_read
= INTERCEPT_DR0_MASK
|
478 control
->intercept_dr_write
= INTERCEPT_DR0_MASK
|
485 control
->intercept_exceptions
= 1 << PF_VECTOR
;
488 control
->intercept
= (1ULL << INTERCEPT_INTR
) |
489 (1ULL << INTERCEPT_NMI
) |
491 * selective cr0 intercept bug?
492 * 0: 0f 22 d8 mov %eax,%cr3
493 * 3: 0f 20 c0 mov %cr0,%eax
494 * 6: 0d 00 00 00 80 or $0x80000000,%eax
495 * b: 0f 22 c0 mov %eax,%cr0
496 * set cr3 ->interception
497 * get cr0 ->interception
498 * set cr0 -> no interception
500 /* (1ULL << INTERCEPT_SELECTIVE_CR0) | */
501 (1ULL << INTERCEPT_CPUID
) |
502 (1ULL << INTERCEPT_HLT
) |
503 (1ULL << INTERCEPT_INVLPG
) |
504 (1ULL << INTERCEPT_INVLPGA
) |
505 (1ULL << INTERCEPT_IOIO_PROT
) |
506 (1ULL << INTERCEPT_MSR_PROT
) |
507 (1ULL << INTERCEPT_TASK_SWITCH
) |
508 (1ULL << INTERCEPT_VMRUN
) |
509 (1ULL << INTERCEPT_VMMCALL
) |
510 (1ULL << INTERCEPT_VMLOAD
) |
511 (1ULL << INTERCEPT_VMSAVE
) |
512 (1ULL << INTERCEPT_STGI
) |
513 (1ULL << INTERCEPT_CLGI
) |
514 (1ULL << INTERCEPT_SKINIT
);
516 control
->iopm_base_pa
= iopm_base
;
517 control
->msrpm_base_pa
= msrpm_base
;
519 control
->tsc_offset
= -tsc
;
520 control
->int_ctl
= V_INTR_MASKING_MASK
;
528 save
->cs
.selector
= 0xf000;
529 /* Executable/Readable Code Segment */
530 save
->cs
.attrib
= SVM_SELECTOR_READ_MASK
| SVM_SELECTOR_P_MASK
|
531 SVM_SELECTOR_S_MASK
| SVM_SELECTOR_CODE_MASK
;
532 save
->cs
.limit
= 0xffff;
533 save
->cs
.base
= 0xffff0000;
535 save
->gdtr
.limit
= 0xffff;
536 save
->idtr
.limit
= 0xffff;
538 init_sys_seg(&save
->ldtr
, SEG_TYPE_LDT
);
539 init_sys_seg(&save
->tr
, SEG_TYPE_BUSY_TSS16
);
541 save
->efer
= MSR_EFER_SVME_MASK
;
543 save
->dr6
= 0xffff0ff0;
546 save
->rip
= 0x0000fff0;
549 * cr0 val on cpu init should be 0x60000010, we enable cpu
550 * cache by default. the orderly way is to enable cache in bios.
552 save
->cr0
= 0x00000010 | CR0_PG_MASK
;
553 save
->cr4
= CR4_PAE_MASK
;
557 static int svm_create_vcpu(struct kvm_vcpu
*vcpu
)
563 vcpu
->svm
= kzalloc(sizeof *vcpu
->svm
, GFP_KERNEL
);
566 page
= alloc_page(GFP_KERNEL
);
570 vcpu
->svm
->vmcb
= page_address(page
);
571 memset(vcpu
->svm
->vmcb
, 0, PAGE_SIZE
);
572 vcpu
->svm
->vmcb_pa
= page_to_pfn(page
) << PAGE_SHIFT
;
573 vcpu
->svm
->cr0
= 0x00000010;
574 vcpu
->svm
->asid_generation
= 0;
575 memset(vcpu
->svm
->db_regs
, 0, sizeof(vcpu
->svm
->db_regs
));
576 init_vmcb(vcpu
->svm
->vmcb
);
586 static void svm_free_vcpu(struct kvm_vcpu
*vcpu
)
591 __free_page(pfn_to_page(vcpu
->svm
->vmcb_pa
>> PAGE_SHIFT
));
595 static struct kvm_vcpu
*svm_vcpu_load(struct kvm_vcpu
*vcpu
)
601 static void svm_vcpu_put(struct kvm_vcpu
*vcpu
)
606 static void svm_cache_regs(struct kvm_vcpu
*vcpu
)
608 vcpu
->regs
[VCPU_REGS_RAX
] = vcpu
->svm
->vmcb
->save
.rax
;
609 vcpu
->regs
[VCPU_REGS_RSP
] = vcpu
->svm
->vmcb
->save
.rsp
;
610 vcpu
->rip
= vcpu
->svm
->vmcb
->save
.rip
;
613 static void svm_decache_regs(struct kvm_vcpu
*vcpu
)
615 vcpu
->svm
->vmcb
->save
.rax
= vcpu
->regs
[VCPU_REGS_RAX
];
616 vcpu
->svm
->vmcb
->save
.rsp
= vcpu
->regs
[VCPU_REGS_RSP
];
617 vcpu
->svm
->vmcb
->save
.rip
= vcpu
->rip
;
620 static unsigned long svm_get_rflags(struct kvm_vcpu
*vcpu
)
622 return vcpu
->svm
->vmcb
->save
.rflags
;
625 static void svm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
627 vcpu
->svm
->vmcb
->save
.rflags
= rflags
;
630 static struct vmcb_seg
*svm_seg(struct kvm_vcpu
*vcpu
, int seg
)
632 struct vmcb_save_area
*save
= &vcpu
->svm
->vmcb
->save
;
635 case VCPU_SREG_CS
: return &save
->cs
;
636 case VCPU_SREG_DS
: return &save
->ds
;
637 case VCPU_SREG_ES
: return &save
->es
;
638 case VCPU_SREG_FS
: return &save
->fs
;
639 case VCPU_SREG_GS
: return &save
->gs
;
640 case VCPU_SREG_SS
: return &save
->ss
;
641 case VCPU_SREG_TR
: return &save
->tr
;
642 case VCPU_SREG_LDTR
: return &save
->ldtr
;
648 static u64
svm_get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
650 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
655 static void svm_get_segment(struct kvm_vcpu
*vcpu
,
656 struct kvm_segment
*var
, int seg
)
658 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
661 var
->limit
= s
->limit
;
662 var
->selector
= s
->selector
;
663 var
->type
= s
->attrib
& SVM_SELECTOR_TYPE_MASK
;
664 var
->s
= (s
->attrib
>> SVM_SELECTOR_S_SHIFT
) & 1;
665 var
->dpl
= (s
->attrib
>> SVM_SELECTOR_DPL_SHIFT
) & 3;
666 var
->present
= (s
->attrib
>> SVM_SELECTOR_P_SHIFT
) & 1;
667 var
->avl
= (s
->attrib
>> SVM_SELECTOR_AVL_SHIFT
) & 1;
668 var
->l
= (s
->attrib
>> SVM_SELECTOR_L_SHIFT
) & 1;
669 var
->db
= (s
->attrib
>> SVM_SELECTOR_DB_SHIFT
) & 1;
670 var
->g
= (s
->attrib
>> SVM_SELECTOR_G_SHIFT
) & 1;
671 var
->unusable
= !var
->present
;
674 static void svm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
676 struct vmcb_seg
*s
= svm_seg(vcpu
, VCPU_SREG_CS
);
678 *db
= (s
->attrib
>> SVM_SELECTOR_DB_SHIFT
) & 1;
679 *l
= (s
->attrib
>> SVM_SELECTOR_L_SHIFT
) & 1;
682 static void svm_get_idt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
684 dt
->limit
= vcpu
->svm
->vmcb
->save
.ldtr
.limit
;
685 dt
->base
= vcpu
->svm
->vmcb
->save
.ldtr
.base
;
688 static void svm_set_idt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
690 vcpu
->svm
->vmcb
->save
.ldtr
.limit
= dt
->limit
;
691 vcpu
->svm
->vmcb
->save
.ldtr
.base
= dt
->base
;
694 static void svm_get_gdt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
696 dt
->limit
= vcpu
->svm
->vmcb
->save
.gdtr
.limit
;
697 dt
->base
= vcpu
->svm
->vmcb
->save
.gdtr
.base
;
700 static void svm_set_gdt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
702 vcpu
->svm
->vmcb
->save
.gdtr
.limit
= dt
->limit
;
703 vcpu
->svm
->vmcb
->save
.gdtr
.base
= dt
->base
;
706 static void svm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
709 if (vcpu
->shadow_efer
& KVM_EFER_LME
) {
710 if (!is_paging(vcpu
) && (cr0
& CR0_PG_MASK
)) {
711 vcpu
->shadow_efer
|= KVM_EFER_LMA
;
712 vcpu
->svm
->vmcb
->save
.efer
|= KVM_EFER_LMA
| KVM_EFER_LME
;
715 if (is_paging(vcpu
) && !(cr0
& CR0_PG_MASK
) ) {
716 vcpu
->shadow_efer
&= ~KVM_EFER_LMA
;
717 vcpu
->svm
->vmcb
->save
.efer
&= ~(KVM_EFER_LMA
| KVM_EFER_LME
);
721 vcpu
->svm
->cr0
= cr0
;
722 vcpu
->svm
->vmcb
->save
.cr0
= cr0
| CR0_PG_MASK
;
726 static void svm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
729 vcpu
->svm
->vmcb
->save
.cr4
= cr4
| CR4_PAE_MASK
;
732 static void svm_set_segment(struct kvm_vcpu
*vcpu
,
733 struct kvm_segment
*var
, int seg
)
735 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
738 s
->limit
= var
->limit
;
739 s
->selector
= var
->selector
;
743 s
->attrib
= (var
->type
& SVM_SELECTOR_TYPE_MASK
);
744 s
->attrib
|= (var
->s
& 1) << SVM_SELECTOR_S_SHIFT
;
745 s
->attrib
|= (var
->dpl
& 3) << SVM_SELECTOR_DPL_SHIFT
;
746 s
->attrib
|= (var
->present
& 1) << SVM_SELECTOR_P_SHIFT
;
747 s
->attrib
|= (var
->avl
& 1) << SVM_SELECTOR_AVL_SHIFT
;
748 s
->attrib
|= (var
->l
& 1) << SVM_SELECTOR_L_SHIFT
;
749 s
->attrib
|= (var
->db
& 1) << SVM_SELECTOR_DB_SHIFT
;
750 s
->attrib
|= (var
->g
& 1) << SVM_SELECTOR_G_SHIFT
;
752 if (seg
== VCPU_SREG_CS
)
753 vcpu
->svm
->vmcb
->save
.cpl
754 = (vcpu
->svm
->vmcb
->save
.cs
.attrib
755 >> SVM_SELECTOR_DPL_SHIFT
) & 3;
761 vcpu->svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
762 vcpu->svm->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
766 static int svm_guest_debug(struct kvm_vcpu
*vcpu
, struct kvm_debug_guest
*dbg
)
771 static void load_host_msrs(struct kvm_vcpu
*vcpu
)
775 for ( i
= 0; i
< NR_HOST_SAVE_MSRS
; i
++)
776 wrmsrl(host_save_msrs
[i
], vcpu
->svm
->host_msrs
[i
]);
779 static void save_host_msrs(struct kvm_vcpu
*vcpu
)
783 for ( i
= 0; i
< NR_HOST_SAVE_MSRS
; i
++)
784 rdmsrl(host_save_msrs
[i
], vcpu
->svm
->host_msrs
[i
]);
787 static void new_asid(struct kvm_vcpu
*vcpu
, struct svm_cpu_data
*svm_data
)
789 if (svm_data
->next_asid
> svm_data
->max_asid
) {
790 ++svm_data
->asid_generation
;
791 svm_data
->next_asid
= 1;
792 vcpu
->svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ALL_ASID
;
795 vcpu
->cpu
= svm_data
->cpu
;
796 vcpu
->svm
->asid_generation
= svm_data
->asid_generation
;
797 vcpu
->svm
->vmcb
->control
.asid
= svm_data
->next_asid
++;
800 static void svm_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
802 invlpga(address
, vcpu
->svm
->vmcb
->control
.asid
); // is needed?
805 static unsigned long svm_get_dr(struct kvm_vcpu
*vcpu
, int dr
)
807 return vcpu
->svm
->db_regs
[dr
];
810 static void svm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long value
,
815 if (vcpu
->svm
->vmcb
->save
.dr7
& DR7_GD_MASK
) {
816 vcpu
->svm
->vmcb
->save
.dr7
&= ~DR7_GD_MASK
;
817 vcpu
->svm
->vmcb
->save
.dr6
|= DR6_BD_MASK
;
818 *exception
= DB_VECTOR
;
824 vcpu
->svm
->db_regs
[dr
] = value
;
827 if (vcpu
->cr4
& CR4_DE_MASK
) {
828 *exception
= UD_VECTOR
;
832 if (value
& ~((1ULL << 32) - 1)) {
833 *exception
= GP_VECTOR
;
836 vcpu
->svm
->vmcb
->save
.dr7
= value
;
840 printk(KERN_DEBUG
"%s: unexpected dr %u\n",
842 *exception
= UD_VECTOR
;
847 static int pf_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
849 u32 exit_int_info
= vcpu
->svm
->vmcb
->control
.exit_int_info
;
852 enum emulation_result er
;
854 if (is_external_interrupt(exit_int_info
))
855 push_irq(vcpu
, exit_int_info
& SVM_EVTINJ_VEC_MASK
);
857 spin_lock(&vcpu
->kvm
->lock
);
859 fault_address
= vcpu
->svm
->vmcb
->control
.exit_info_2
;
860 error_code
= vcpu
->svm
->vmcb
->control
.exit_info_1
;
861 if (!vcpu
->mmu
.page_fault(vcpu
, fault_address
, error_code
)) {
862 spin_unlock(&vcpu
->kvm
->lock
);
865 er
= emulate_instruction(vcpu
, kvm_run
, fault_address
, error_code
);
866 spin_unlock(&vcpu
->kvm
->lock
);
871 case EMULATE_DO_MMIO
:
872 ++kvm_stat
.mmio_exits
;
873 kvm_run
->exit_reason
= KVM_EXIT_MMIO
;
876 vcpu_printf(vcpu
, "%s: emulate fail\n", __FUNCTION__
);
882 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
886 static int io_get_override(struct kvm_vcpu
*vcpu
,
887 struct vmcb_seg
**seg
,
890 u8 inst
[MAX_INST_SIZE
];
895 rip
= vcpu
->svm
->vmcb
->save
.rip
;
896 ins_length
= vcpu
->svm
->next_rip
- rip
;
897 rip
+= vcpu
->svm
->vmcb
->save
.cs
.base
;
899 if (ins_length
> MAX_INST_SIZE
)
901 "%s: inst length err, cs base 0x%llx rip 0x%llx "
902 "next rip 0x%llx ins_length %u\n",
904 vcpu
->svm
->vmcb
->save
.cs
.base
,
905 vcpu
->svm
->vmcb
->save
.rip
,
906 vcpu
->svm
->vmcb
->control
.exit_info_2
,
909 if (kvm_read_guest(vcpu
, rip
, ins_length
, inst
) != ins_length
)
915 for (i
= 0; i
< ins_length
; i
++)
926 *seg
= &vcpu
->svm
->vmcb
->save
.cs
;
929 *seg
= &vcpu
->svm
->vmcb
->save
.ss
;
932 *seg
= &vcpu
->svm
->vmcb
->save
.ds
;
935 *seg
= &vcpu
->svm
->vmcb
->save
.es
;
938 *seg
= &vcpu
->svm
->vmcb
->save
.fs
;
941 *seg
= &vcpu
->svm
->vmcb
->save
.gs
;
946 printk(KERN_DEBUG
"%s: unexpected\n", __FUNCTION__
);
950 static unsigned long io_adress(struct kvm_vcpu
*vcpu
, int ins
, u64
*address
)
952 unsigned long addr_mask
;
954 struct vmcb_seg
*seg
;
956 struct vmcb_save_area
*save_area
= &vcpu
->svm
->vmcb
->save
;
957 u16 cs_attrib
= save_area
->cs
.attrib
;
958 unsigned addr_size
= get_addr_size(vcpu
);
960 if (!io_get_override(vcpu
, &seg
, &addr_override
))
964 addr_size
= (addr_size
== 2) ? 4: (addr_size
>> 1);
967 reg
= &vcpu
->regs
[VCPU_REGS_RDI
];
968 seg
= &vcpu
->svm
->vmcb
->save
.es
;
970 reg
= &vcpu
->regs
[VCPU_REGS_RSI
];
971 seg
= (seg
) ? seg
: &vcpu
->svm
->vmcb
->save
.ds
;
974 addr_mask
= ~0ULL >> (64 - (addr_size
* 8));
976 if ((cs_attrib
& SVM_SELECTOR_L_MASK
) &&
977 !(vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_VM
)) {
978 *address
= (*reg
& addr_mask
);
982 if (!(seg
->attrib
& SVM_SELECTOR_P_SHIFT
)) {
983 svm_inject_gp(vcpu
, 0);
987 *address
= (*reg
& addr_mask
) + seg
->base
;
991 static int io_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
993 u32 io_info
= vcpu
->svm
->vmcb
->control
.exit_info_1
; //address size bug?
994 int _in
= io_info
& SVM_IOIO_TYPE_MASK
;
998 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->control
.exit_info_2
;
1000 kvm_run
->exit_reason
= KVM_EXIT_IO
;
1001 kvm_run
->io
.port
= io_info
>> 16;
1002 kvm_run
->io
.direction
= (_in
) ? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
1003 kvm_run
->io
.size
= ((io_info
& SVM_IOIO_SIZE_MASK
) >> SVM_IOIO_SIZE_SHIFT
);
1004 kvm_run
->io
.string
= (io_info
& SVM_IOIO_STR_MASK
) != 0;
1005 kvm_run
->io
.rep
= (io_info
& SVM_IOIO_REP_MASK
) != 0;
1007 if (kvm_run
->io
.string
) {
1010 addr_mask
= io_adress(vcpu
, _in
, &kvm_run
->io
.address
);
1012 printk(KERN_DEBUG
"%s: get io address failed\n", __FUNCTION__
);
1016 if (kvm_run
->io
.rep
) {
1017 kvm_run
->io
.count
= vcpu
->regs
[VCPU_REGS_RCX
] & addr_mask
;
1018 kvm_run
->io
.string_down
= (vcpu
->svm
->vmcb
->save
.rflags
1019 & X86_EFLAGS_DF
) != 0;
1022 kvm_run
->io
.value
= vcpu
->svm
->vmcb
->save
.rax
;
1028 static int nop_on_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1033 static int halt_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1035 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 1;
1036 skip_emulated_instruction(vcpu
);
1037 if (vcpu
->irq_summary
&& (vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_IF
))
1040 kvm_run
->exit_reason
= KVM_EXIT_HLT
;
1044 static int invalid_op_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1050 static int task_switch_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1052 printk(KERN_DEBUG
"%s: task swiche is unsupported\n", __FUNCTION__
);
1053 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
1057 static int cpuid_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1059 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 2;
1060 kvm_run
->exit_reason
= KVM_EXIT_CPUID
;
1064 static int emulate_on_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1066 if (emulate_instruction(vcpu
, 0, 0, 0) != EMULATE_DONE
)
1067 printk(KERN_ERR
"%s: failed\n", __FUNCTION__
);
1071 static int svm_get_msr(struct kvm_vcpu
*vcpu
, unsigned ecx
, u64
*data
)
1074 case MSR_IA32_MC0_CTL
:
1075 case MSR_IA32_MCG_STATUS
:
1076 case MSR_IA32_MCG_CAP
:
1077 case MSR_IA32_MC0_MISC
:
1078 case MSR_IA32_MC0_MISC
+4:
1079 case MSR_IA32_MC0_MISC
+8:
1080 case MSR_IA32_MC0_MISC
+12:
1081 case MSR_IA32_MC0_MISC
+16:
1082 case MSR_IA32_UCODE_REV
:
1083 /* MTRR registers */
1085 case 0x200 ... 0x2ff:
1088 case MSR_IA32_TIME_STAMP_COUNTER
: {
1092 *data
= vcpu
->svm
->vmcb
->control
.tsc_offset
+ tsc
;
1096 *data
= vcpu
->shadow_efer
;
1098 case MSR_IA32_APICBASE
:
1099 *data
= vcpu
->apic_base
;
1101 #ifdef CONFIG_X86_64
1103 *data
= vcpu
->svm
->vmcb
->save
.star
;
1106 *data
= vcpu
->svm
->vmcb
->save
.lstar
;
1109 *data
= vcpu
->svm
->vmcb
->save
.cstar
;
1111 case MSR_KERNEL_GS_BASE
:
1112 *data
= vcpu
->svm
->vmcb
->save
.kernel_gs_base
;
1114 case MSR_SYSCALL_MASK
:
1115 *data
= vcpu
->svm
->vmcb
->save
.sfmask
;
1118 case MSR_IA32_SYSENTER_CS
:
1119 *data
= vcpu
->svm
->vmcb
->save
.sysenter_cs
;
1121 case MSR_IA32_SYSENTER_EIP
:
1122 *data
= vcpu
->svm
->vmcb
->save
.sysenter_eip
;
1124 case MSR_IA32_SYSENTER_ESP
:
1125 *data
= vcpu
->svm
->vmcb
->save
.sysenter_esp
;
1128 printk(KERN_ERR
"kvm: unhandled rdmsr: 0x%x\n", ecx
);
1134 static int rdmsr_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1136 u32 ecx
= vcpu
->regs
[VCPU_REGS_RCX
];
1139 if (svm_get_msr(vcpu
, ecx
, &data
))
1140 svm_inject_gp(vcpu
, 0);
1142 vcpu
->svm
->vmcb
->save
.rax
= data
& 0xffffffff;
1143 vcpu
->regs
[VCPU_REGS_RDX
] = data
>> 32;
1144 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 2;
1145 skip_emulated_instruction(vcpu
);
1150 static int svm_set_msr(struct kvm_vcpu
*vcpu
, unsigned ecx
, u64 data
)
1153 #ifdef CONFIG_X86_64
1155 set_efer(vcpu
, data
);
1158 case MSR_IA32_MC0_STATUS
:
1159 printk(KERN_WARNING
"%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n"
1160 , __FUNCTION__
, data
);
1162 case MSR_IA32_TIME_STAMP_COUNTER
: {
1166 vcpu
->svm
->vmcb
->control
.tsc_offset
= data
- tsc
;
1169 case MSR_IA32_UCODE_REV
:
1170 case MSR_IA32_UCODE_WRITE
:
1171 case 0x200 ... 0x2ff: /* MTRRs */
1173 case MSR_IA32_APICBASE
:
1174 vcpu
->apic_base
= data
;
1176 #ifdef CONFIG_X86_64_
1178 vcpu
->svm
->vmcb
->save
.star
= data
;
1181 vcpu
->svm
->vmcb
->save
.lstar
= data
;
1184 vcpu
->svm
->vmcb
->save
.cstar
= data
;
1186 case MSR_KERNEL_GS_BASE
:
1187 vcpu
->svm
->vmcb
->save
.kernel_gs_base
= data
;
1189 case MSR_SYSCALL_MASK
:
1190 vcpu
->svm
->vmcb
->save
.sfmask
= data
;
1193 case MSR_IA32_SYSENTER_CS
:
1194 vcpu
->svm
->vmcb
->save
.sysenter_cs
= data
;
1196 case MSR_IA32_SYSENTER_EIP
:
1197 vcpu
->svm
->vmcb
->save
.sysenter_eip
= data
;
1199 case MSR_IA32_SYSENTER_ESP
:
1200 vcpu
->svm
->vmcb
->save
.sysenter_esp
= data
;
1203 printk(KERN_ERR
"kvm: unhandled wrmsr: %x\n", ecx
);
1209 static int wrmsr_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1211 u32 ecx
= vcpu
->regs
[VCPU_REGS_RCX
];
1212 u64 data
= (vcpu
->svm
->vmcb
->save
.rax
& -1u)
1213 | ((u64
)(vcpu
->regs
[VCPU_REGS_RDX
] & -1u) << 32);
1214 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 2;
1215 if (svm_set_msr(vcpu
, ecx
, data
))
1216 svm_inject_gp(vcpu
, 0);
1218 skip_emulated_instruction(vcpu
);
1222 static int msr_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1224 if (vcpu
->svm
->vmcb
->control
.exit_info_1
)
1225 return wrmsr_interception(vcpu
, kvm_run
);
1227 return rdmsr_interception(vcpu
, kvm_run
);
1230 static int (*svm_exit_handlers
[])(struct kvm_vcpu
*vcpu
,
1231 struct kvm_run
*kvm_run
) = {
1232 [SVM_EXIT_READ_CR0
] = emulate_on_interception
,
1233 [SVM_EXIT_READ_CR3
] = emulate_on_interception
,
1234 [SVM_EXIT_READ_CR4
] = emulate_on_interception
,
1236 [SVM_EXIT_WRITE_CR0
] = emulate_on_interception
,
1237 [SVM_EXIT_WRITE_CR3
] = emulate_on_interception
,
1238 [SVM_EXIT_WRITE_CR4
] = emulate_on_interception
,
1239 [SVM_EXIT_READ_DR0
] = emulate_on_interception
,
1240 [SVM_EXIT_READ_DR1
] = emulate_on_interception
,
1241 [SVM_EXIT_READ_DR2
] = emulate_on_interception
,
1242 [SVM_EXIT_READ_DR3
] = emulate_on_interception
,
1243 [SVM_EXIT_WRITE_DR0
] = emulate_on_interception
,
1244 [SVM_EXIT_WRITE_DR1
] = emulate_on_interception
,
1245 [SVM_EXIT_WRITE_DR2
] = emulate_on_interception
,
1246 [SVM_EXIT_WRITE_DR3
] = emulate_on_interception
,
1247 [SVM_EXIT_WRITE_DR5
] = emulate_on_interception
,
1248 [SVM_EXIT_WRITE_DR7
] = emulate_on_interception
,
1249 [SVM_EXIT_EXCP_BASE
+ PF_VECTOR
] = pf_interception
,
1250 [SVM_EXIT_INTR
] = nop_on_interception
,
1251 [SVM_EXIT_NMI
] = nop_on_interception
,
1252 [SVM_EXIT_SMI
] = nop_on_interception
,
1253 [SVM_EXIT_INIT
] = nop_on_interception
,
1254 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
1255 [SVM_EXIT_CPUID
] = cpuid_interception
,
1256 [SVM_EXIT_HLT
] = halt_interception
,
1257 [SVM_EXIT_INVLPG
] = emulate_on_interception
,
1258 [SVM_EXIT_INVLPGA
] = invalid_op_interception
,
1259 [SVM_EXIT_IOIO
] = io_interception
,
1260 [SVM_EXIT_MSR
] = msr_interception
,
1261 [SVM_EXIT_TASK_SWITCH
] = task_switch_interception
,
1262 [SVM_EXIT_VMRUN
] = invalid_op_interception
,
1263 [SVM_EXIT_VMMCALL
] = invalid_op_interception
,
1264 [SVM_EXIT_VMLOAD
] = invalid_op_interception
,
1265 [SVM_EXIT_VMSAVE
] = invalid_op_interception
,
1266 [SVM_EXIT_STGI
] = invalid_op_interception
,
1267 [SVM_EXIT_CLGI
] = invalid_op_interception
,
1268 [SVM_EXIT_SKINIT
] = invalid_op_interception
,
1272 static int handle_exit(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1274 u32 exit_code
= vcpu
->svm
->vmcb
->control
.exit_code
;
1276 kvm_run
->exit_type
= KVM_EXIT_TYPE_VM_EXIT
;
1278 if (is_external_interrupt(vcpu
->svm
->vmcb
->control
.exit_int_info
) &&
1279 exit_code
!= SVM_EXIT_EXCP_BASE
+ PF_VECTOR
)
1280 printk(KERN_ERR
"%s: unexpected exit_ini_info 0x%x "
1282 __FUNCTION__
, vcpu
->svm
->vmcb
->control
.exit_int_info
,
1285 if (exit_code
>= sizeof(svm_exit_handlers
) / sizeof(*svm_exit_handlers
)
1286 || svm_exit_handlers
[exit_code
] == 0) {
1287 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
1288 printk(KERN_ERR
"%s: 0x%x @ 0x%llx cr0 0x%lx rflags 0x%llx\n",
1291 vcpu
->svm
->vmcb
->save
.rip
,
1293 vcpu
->svm
->vmcb
->save
.rflags
);
1297 return svm_exit_handlers
[exit_code
](vcpu
, kvm_run
);
1300 static void reload_tss(struct kvm_vcpu
*vcpu
)
1302 int cpu
= raw_smp_processor_id();
1304 struct svm_cpu_data
*svm_data
= per_cpu(svm_data
, cpu
);
1305 svm_data
->tss_desc
->type
= 9; //available 32/64-bit TSS
1309 static void pre_svm_run(struct kvm_vcpu
*vcpu
)
1311 int cpu
= raw_smp_processor_id();
1313 struct svm_cpu_data
*svm_data
= per_cpu(svm_data
, cpu
);
1315 vcpu
->svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_DO_NOTHING
;
1316 if (vcpu
->cpu
!= cpu
||
1317 vcpu
->svm
->asid_generation
!= svm_data
->asid_generation
)
1318 new_asid(vcpu
, svm_data
);
1322 static inline void kvm_try_inject_irq(struct kvm_vcpu
*vcpu
)
1324 struct vmcb_control_area
*control
;
1326 if (!vcpu
->irq_summary
)
1329 control
= &vcpu
->svm
->vmcb
->control
;
1331 control
->int_vector
= pop_irq(vcpu
);
1332 control
->int_ctl
&= ~V_INTR_PRIO_MASK
;
1333 control
->int_ctl
|= V_IRQ_MASK
|
1334 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT
);
1337 static void kvm_reput_irq(struct kvm_vcpu
*vcpu
)
1339 struct vmcb_control_area
*control
= &vcpu
->svm
->vmcb
->control
;
1341 if (control
->int_ctl
& V_IRQ_MASK
) {
1342 control
->int_ctl
&= ~V_IRQ_MASK
;
1343 push_irq(vcpu
, control
->int_vector
);
1347 static void save_db_regs(unsigned long *db_regs
)
1349 asm volatile ("mov %%dr0, %0" : "=r"(db_regs
[0]));
1350 asm volatile ("mov %%dr1, %0" : "=r"(db_regs
[1]));
1351 asm volatile ("mov %%dr2, %0" : "=r"(db_regs
[2]));
1352 asm volatile ("mov %%dr3, %0" : "=r"(db_regs
[3]));
1355 static void load_db_regs(unsigned long *db_regs
)
1357 asm volatile ("mov %0, %%dr0" : : "r"(db_regs
[0]));
1358 asm volatile ("mov %0, %%dr1" : : "r"(db_regs
[1]));
1359 asm volatile ("mov %0, %%dr2" : : "r"(db_regs
[2]));
1360 asm volatile ("mov %0, %%dr3" : : "r"(db_regs
[3]));
1363 static int svm_vcpu_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1370 kvm_try_inject_irq(vcpu
);
1376 save_host_msrs(vcpu
);
1377 fs_selector
= read_fs();
1378 gs_selector
= read_gs();
1379 ldt_selector
= read_ldt();
1380 vcpu
->svm
->host_cr2
= kvm_read_cr2();
1381 vcpu
->svm
->host_dr6
= read_dr6();
1382 vcpu
->svm
->host_dr7
= read_dr7();
1383 vcpu
->svm
->vmcb
->save
.cr2
= vcpu
->cr2
;
1385 if (vcpu
->svm
->vmcb
->save
.dr7
& 0xff) {
1387 save_db_regs(vcpu
->svm
->host_db_regs
);
1388 load_db_regs(vcpu
->svm
->db_regs
);
1391 #ifdef CONFIG_X86_64
1392 "push %%rbx; push %%rcx; push %%rdx;"
1393 "push %%rsi; push %%rdi; push %%rbp;"
1394 "push %%r8; push %%r9; push %%r10; push %%r11;"
1395 "push %%r12; push %%r13; push %%r14; push %%r15;"
1397 "push %%ebx; push %%ecx; push %%edx;"
1398 "push %%esi; push %%edi; push %%ebp;"
1401 #ifdef CONFIG_X86_64
1402 "mov %c[rbx](%[vcpu]), %%rbx \n\t"
1403 "mov %c[rcx](%[vcpu]), %%rcx \n\t"
1404 "mov %c[rdx](%[vcpu]), %%rdx \n\t"
1405 "mov %c[rsi](%[vcpu]), %%rsi \n\t"
1406 "mov %c[rdi](%[vcpu]), %%rdi \n\t"
1407 "mov %c[rbp](%[vcpu]), %%rbp \n\t"
1408 "mov %c[r8](%[vcpu]), %%r8 \n\t"
1409 "mov %c[r9](%[vcpu]), %%r9 \n\t"
1410 "mov %c[r10](%[vcpu]), %%r10 \n\t"
1411 "mov %c[r11](%[vcpu]), %%r11 \n\t"
1412 "mov %c[r12](%[vcpu]), %%r12 \n\t"
1413 "mov %c[r13](%[vcpu]), %%r13 \n\t"
1414 "mov %c[r14](%[vcpu]), %%r14 \n\t"
1415 "mov %c[r15](%[vcpu]), %%r15 \n\t"
1417 "mov %c[rbx](%[vcpu]), %%ebx \n\t"
1418 "mov %c[rcx](%[vcpu]), %%ecx \n\t"
1419 "mov %c[rdx](%[vcpu]), %%edx \n\t"
1420 "mov %c[rsi](%[vcpu]), %%esi \n\t"
1421 "mov %c[rdi](%[vcpu]), %%edi \n\t"
1422 "mov %c[rbp](%[vcpu]), %%ebp \n\t"
1425 #ifdef CONFIG_X86_64
1426 /* Enter guest mode */
1428 "mov %c[svm](%[vcpu]), %%rax \n\t"
1429 "mov %c[vmcb](%%rax), %%rax \n\t"
1435 /* Enter guest mode */
1437 "mov %c[svm](%[vcpu]), %%eax \n\t"
1438 "mov %c[vmcb](%%eax), %%eax \n\t"
1445 /* Save guest registers, load host registers */
1446 #ifdef CONFIG_X86_64
1447 "mov %%rbx, %c[rbx](%[vcpu]) \n\t"
1448 "mov %%rcx, %c[rcx](%[vcpu]) \n\t"
1449 "mov %%rdx, %c[rdx](%[vcpu]) \n\t"
1450 "mov %%rsi, %c[rsi](%[vcpu]) \n\t"
1451 "mov %%rdi, %c[rdi](%[vcpu]) \n\t"
1452 "mov %%rbp, %c[rbp](%[vcpu]) \n\t"
1453 "mov %%r8, %c[r8](%[vcpu]) \n\t"
1454 "mov %%r9, %c[r9](%[vcpu]) \n\t"
1455 "mov %%r10, %c[r10](%[vcpu]) \n\t"
1456 "mov %%r11, %c[r11](%[vcpu]) \n\t"
1457 "mov %%r12, %c[r12](%[vcpu]) \n\t"
1458 "mov %%r13, %c[r13](%[vcpu]) \n\t"
1459 "mov %%r14, %c[r14](%[vcpu]) \n\t"
1460 "mov %%r15, %c[r15](%[vcpu]) \n\t"
1462 "pop %%r15; pop %%r14; pop %%r13; pop %%r12;"
1463 "pop %%r11; pop %%r10; pop %%r9; pop %%r8;"
1464 "pop %%rbp; pop %%rdi; pop %%rsi;"
1465 "pop %%rdx; pop %%rcx; pop %%rbx; \n\t"
1467 "mov %%ebx, %c[rbx](%[vcpu]) \n\t"
1468 "mov %%ecx, %c[rcx](%[vcpu]) \n\t"
1469 "mov %%edx, %c[rdx](%[vcpu]) \n\t"
1470 "mov %%esi, %c[rsi](%[vcpu]) \n\t"
1471 "mov %%edi, %c[rdi](%[vcpu]) \n\t"
1472 "mov %%ebp, %c[rbp](%[vcpu]) \n\t"
1474 "pop %%ebp; pop %%edi; pop %%esi;"
1475 "pop %%edx; pop %%ecx; pop %%ebx; \n\t"
1479 [svm
]"i"(offsetof(struct kvm_vcpu
, svm
)),
1480 [vmcb
]"i"(offsetof(struct vcpu_svm
, vmcb_pa
)),
1481 [rbx
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RBX
])),
1482 [rcx
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RCX
])),
1483 [rdx
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RDX
])),
1484 [rsi
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RSI
])),
1485 [rdi
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RDI
])),
1486 [rbp
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RBP
]))
1487 #ifdef CONFIG_X86_64
1488 ,[r8
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R8
])),
1489 [r9
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R9
])),
1490 [r10
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R10
])),
1491 [r11
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R11
])),
1492 [r12
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R12
])),
1493 [r13
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R13
])),
1494 [r14
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R14
])),
1495 [r15
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R15
]))
1499 if ((vcpu
->svm
->vmcb
->save
.dr7
& 0xff))
1500 load_db_regs(vcpu
->svm
->host_db_regs
);
1502 vcpu
->cr2
= vcpu
->svm
->vmcb
->save
.cr2
;
1504 write_dr6(vcpu
->svm
->host_dr6
);
1505 write_dr7(vcpu
->svm
->host_dr7
);
1506 kvm_write_cr2(vcpu
->svm
->host_cr2
);
1508 load_fs(fs_selector
);
1509 load_gs(gs_selector
);
1510 load_ldt(ldt_selector
);
1511 load_host_msrs(vcpu
);
1517 kvm_reput_irq(vcpu
);
1519 vcpu
->svm
->next_rip
= 0;
1521 if (vcpu
->svm
->vmcb
->control
.exit_code
== SVM_EXIT_ERR
) {
1522 kvm_run
->exit_type
= KVM_EXIT_TYPE_FAIL_ENTRY
;
1523 kvm_run
->exit_reason
= vcpu
->svm
->vmcb
->control
.exit_code
;
1527 if (handle_exit(vcpu
, kvm_run
)) {
1528 if (signal_pending(current
)) {
1529 ++kvm_stat
.signal_exits
;
1538 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
)
1540 force_new_asid(vcpu
);
1543 static void svm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
1545 vcpu
->svm
->vmcb
->save
.cr3
= root
;
1546 force_new_asid(vcpu
);
1549 static void svm_inject_page_fault(struct kvm_vcpu
*vcpu
,
1553 uint32_t exit_int_info
= vcpu
->svm
->vmcb
->control
.exit_int_info
;
1555 ++kvm_stat
.pf_guest
;
1557 if (is_page_fault(exit_int_info
)) {
1559 vcpu
->svm
->vmcb
->control
.event_inj_err
= 0;
1560 vcpu
->svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
|
1561 SVM_EVTINJ_VALID_ERR
|
1562 SVM_EVTINJ_TYPE_EXEPT
|
1567 vcpu
->svm
->vmcb
->save
.cr2
= addr
;
1568 vcpu
->svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
|
1569 SVM_EVTINJ_VALID_ERR
|
1570 SVM_EVTINJ_TYPE_EXEPT
|
1572 vcpu
->svm
->vmcb
->control
.event_inj_err
= err_code
;
1576 static int is_disabled(void)
1581 static struct kvm_arch_ops svm_arch_ops
= {
1582 .cpu_has_kvm_support
= has_svm
,
1583 .disabled_by_bios
= is_disabled
,
1584 .hardware_setup
= svm_hardware_setup
,
1585 .hardware_unsetup
= svm_hardware_unsetup
,
1586 .hardware_enable
= svm_hardware_enable
,
1587 .hardware_disable
= svm_hardware_disable
,
1589 .vcpu_create
= svm_create_vcpu
,
1590 .vcpu_free
= svm_free_vcpu
,
1592 .vcpu_load
= svm_vcpu_load
,
1593 .vcpu_put
= svm_vcpu_put
,
1595 .set_guest_debug
= svm_guest_debug
,
1596 .get_msr
= svm_get_msr
,
1597 .set_msr
= svm_set_msr
,
1598 .get_segment_base
= svm_get_segment_base
,
1599 .get_segment
= svm_get_segment
,
1600 .set_segment
= svm_set_segment
,
1601 .is_long_mode
= svm_is_long_mode
,
1602 .get_cs_db_l_bits
= svm_get_cs_db_l_bits
,
1603 .set_cr0
= svm_set_cr0
,
1604 .set_cr0_no_modeswitch
= svm_set_cr0
,
1605 .set_cr3
= svm_set_cr3
,
1606 .set_cr4
= svm_set_cr4
,
1607 .set_efer
= svm_set_efer
,
1608 .get_idt
= svm_get_idt
,
1609 .set_idt
= svm_set_idt
,
1610 .get_gdt
= svm_get_gdt
,
1611 .set_gdt
= svm_set_gdt
,
1612 .get_dr
= svm_get_dr
,
1613 .set_dr
= svm_set_dr
,
1614 .cache_regs
= svm_cache_regs
,
1615 .decache_regs
= svm_decache_regs
,
1616 .get_rflags
= svm_get_rflags
,
1617 .set_rflags
= svm_set_rflags
,
1619 .invlpg
= svm_invlpg
,
1620 .tlb_flush
= svm_flush_tlb
,
1621 .inject_page_fault
= svm_inject_page_fault
,
1623 .inject_gp
= svm_inject_gp
,
1625 .run
= svm_vcpu_run
,
1626 .skip_emulated_instruction
= skip_emulated_instruction
,
1627 .vcpu_setup
= svm_vcpu_setup
,
1630 static int __init
svm_init(void)
1632 return kvm_init_arch(&svm_arch_ops
, THIS_MODULE
);
1635 static void __exit
svm_exit(void)
1640 module_init(svm_init
)
1641 module_exit(svm_exit
)