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/kernel.h>
19 #include <linux/vmalloc.h>
20 #include <linux/highmem.h>
21 #include <linux/profile.h>
25 #include "x86_emulate.h"
27 MODULE_AUTHOR("Qumranet");
28 MODULE_LICENSE("GPL");
30 #define IOPM_ALLOC_ORDER 2
31 #define MSRPM_ALLOC_ORDER 1
37 #define DR7_GD_MASK (1 << 13)
38 #define DR6_BD_MASK (1 << 13)
39 #define CR4_DE_MASK (1UL << 3)
41 #define SEG_TYPE_LDT 2
42 #define SEG_TYPE_BUSY_TSS16 3
44 #define KVM_EFER_LMA (1 << 10)
45 #define KVM_EFER_LME (1 << 8)
47 #define SVM_FEATURE_NPT (1 << 0)
48 #define SVM_FEATURE_LBRV (1 << 1)
49 #define SVM_DEATURE_SVML (1 << 2)
51 unsigned long iopm_base
;
52 unsigned long msrpm_base
;
54 struct kvm_ldttss_desc
{
57 unsigned base1
: 8, type
: 5, dpl
: 2, p
: 1;
58 unsigned limit1
: 4, zero0
: 3, g
: 1, base2
: 8;
61 } __attribute__((packed
));
69 struct kvm_ldttss_desc
*tss_desc
;
71 struct page
*save_area
;
74 static DEFINE_PER_CPU(struct svm_cpu_data
*, svm_data
);
75 static uint32_t svm_features
;
77 struct svm_init_data
{
82 static u32 msrpm_ranges
[] = {0, 0xc0000000, 0xc0010000};
84 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
85 #define MSRS_RANGE_SIZE 2048
86 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
88 #define MAX_INST_SIZE 15
90 static inline u32
svm_has(u32 feat
)
92 return svm_features
& feat
;
95 static unsigned get_addr_size(struct kvm_vcpu
*vcpu
)
97 struct vmcb_save_area
*sa
= &vcpu
->svm
->vmcb
->save
;
100 if (!(sa
->cr0
& CR0_PE_MASK
) || (sa
->rflags
& X86_EFLAGS_VM
))
103 cs_attrib
= sa
->cs
.attrib
;
105 return (cs_attrib
& SVM_SELECTOR_L_MASK
) ? 8 :
106 (cs_attrib
& SVM_SELECTOR_DB_MASK
) ? 4 : 2;
109 static inline u8
pop_irq(struct kvm_vcpu
*vcpu
)
111 int word_index
= __ffs(vcpu
->irq_summary
);
112 int bit_index
= __ffs(vcpu
->irq_pending
[word_index
]);
113 int irq
= word_index
* BITS_PER_LONG
+ bit_index
;
115 clear_bit(bit_index
, &vcpu
->irq_pending
[word_index
]);
116 if (!vcpu
->irq_pending
[word_index
])
117 clear_bit(word_index
, &vcpu
->irq_summary
);
121 static inline void push_irq(struct kvm_vcpu
*vcpu
, u8 irq
)
123 set_bit(irq
, vcpu
->irq_pending
);
124 set_bit(irq
/ BITS_PER_LONG
, &vcpu
->irq_summary
);
127 static inline void clgi(void)
129 asm volatile (SVM_CLGI
);
132 static inline void stgi(void)
134 asm volatile (SVM_STGI
);
137 static inline void invlpga(unsigned long addr
, u32 asid
)
139 asm volatile (SVM_INVLPGA :: "a"(addr
), "c"(asid
));
142 static inline unsigned long kvm_read_cr2(void)
146 asm volatile ("mov %%cr2, %0" : "=r" (cr2
));
150 static inline void kvm_write_cr2(unsigned long val
)
152 asm volatile ("mov %0, %%cr2" :: "r" (val
));
155 static inline unsigned long read_dr6(void)
159 asm volatile ("mov %%dr6, %0" : "=r" (dr6
));
163 static inline void write_dr6(unsigned long val
)
165 asm volatile ("mov %0, %%dr6" :: "r" (val
));
168 static inline unsigned long read_dr7(void)
172 asm volatile ("mov %%dr7, %0" : "=r" (dr7
));
176 static inline void write_dr7(unsigned long val
)
178 asm volatile ("mov %0, %%dr7" :: "r" (val
));
181 static inline void force_new_asid(struct kvm_vcpu
*vcpu
)
183 vcpu
->svm
->asid_generation
--;
186 static inline void flush_guest_tlb(struct kvm_vcpu
*vcpu
)
188 force_new_asid(vcpu
);
191 static void svm_set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
193 if (!(efer
& KVM_EFER_LMA
))
194 efer
&= ~KVM_EFER_LME
;
196 vcpu
->svm
->vmcb
->save
.efer
= efer
| MSR_EFER_SVME_MASK
;
197 vcpu
->shadow_efer
= efer
;
200 static void svm_inject_gp(struct kvm_vcpu
*vcpu
, unsigned error_code
)
202 vcpu
->svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
|
203 SVM_EVTINJ_VALID_ERR
|
204 SVM_EVTINJ_TYPE_EXEPT
|
206 vcpu
->svm
->vmcb
->control
.event_inj_err
= error_code
;
209 static void inject_ud(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
;
244 vcpu
->interrupt_window_open
= 1;
247 static int has_svm(void)
249 uint32_t eax
, ebx
, ecx
, edx
;
251 if (boot_cpu_data
.x86_vendor
!= X86_VENDOR_AMD
) {
252 printk(KERN_INFO
"has_svm: not amd\n");
256 cpuid(0x80000000, &eax
, &ebx
, &ecx
, &edx
);
257 if (eax
< SVM_CPUID_FUNC
) {
258 printk(KERN_INFO
"has_svm: can't execute cpuid_8000000a\n");
262 cpuid(0x80000001, &eax
, &ebx
, &ecx
, &edx
);
263 if (!(ecx
& (1 << SVM_CPUID_FEATURE_SHIFT
))) {
264 printk(KERN_DEBUG
"has_svm: svm not available\n");
270 static void svm_hardware_disable(void *garbage
)
272 struct svm_cpu_data
*svm_data
273 = per_cpu(svm_data
, raw_smp_processor_id());
278 wrmsrl(MSR_VM_HSAVE_PA
, 0);
279 rdmsrl(MSR_EFER
, efer
);
280 wrmsrl(MSR_EFER
, efer
& ~MSR_EFER_SVME_MASK
);
281 per_cpu(svm_data
, raw_smp_processor_id()) = NULL
;
282 __free_page(svm_data
->save_area
);
287 static void svm_hardware_enable(void *garbage
)
290 struct svm_cpu_data
*svm_data
;
293 struct desc_ptr gdt_descr
;
295 struct Xgt_desc_struct gdt_descr
;
297 struct desc_struct
*gdt
;
298 int me
= raw_smp_processor_id();
301 printk(KERN_ERR
"svm_cpu_init: err EOPNOTSUPP on %d\n", me
);
304 svm_data
= per_cpu(svm_data
, me
);
307 printk(KERN_ERR
"svm_cpu_init: svm_data is NULL on %d\n",
312 svm_data
->asid_generation
= 1;
313 svm_data
->max_asid
= cpuid_ebx(SVM_CPUID_FUNC
) - 1;
314 svm_data
->next_asid
= svm_data
->max_asid
+ 1;
315 svm_features
= cpuid_edx(SVM_CPUID_FUNC
);
317 asm volatile ( "sgdt %0" : "=m"(gdt_descr
) );
318 gdt
= (struct desc_struct
*)gdt_descr
.address
;
319 svm_data
->tss_desc
= (struct kvm_ldttss_desc
*)(gdt
+ GDT_ENTRY_TSS
);
321 rdmsrl(MSR_EFER
, efer
);
322 wrmsrl(MSR_EFER
, efer
| MSR_EFER_SVME_MASK
);
324 wrmsrl(MSR_VM_HSAVE_PA
,
325 page_to_pfn(svm_data
->save_area
) << PAGE_SHIFT
);
328 static int svm_cpu_init(int cpu
)
330 struct svm_cpu_data
*svm_data
;
333 svm_data
= kzalloc(sizeof(struct svm_cpu_data
), GFP_KERNEL
);
337 svm_data
->save_area
= alloc_page(GFP_KERNEL
);
339 if (!svm_data
->save_area
)
342 per_cpu(svm_data
, cpu
) = svm_data
;
352 static int set_msr_interception(u32
*msrpm
, unsigned msr
,
357 for (i
= 0; i
< NUM_MSR_MAPS
; i
++) {
358 if (msr
>= msrpm_ranges
[i
] &&
359 msr
< msrpm_ranges
[i
] + MSRS_IN_RANGE
) {
360 u32 msr_offset
= (i
* MSRS_IN_RANGE
+ msr
-
361 msrpm_ranges
[i
]) * 2;
363 u32
*base
= msrpm
+ (msr_offset
/ 32);
364 u32 msr_shift
= msr_offset
% 32;
365 u32 mask
= ((write
) ? 0 : 2) | ((read
) ? 0 : 1);
366 *base
= (*base
& ~(0x3 << msr_shift
)) |
371 printk(KERN_DEBUG
"%s: not found 0x%x\n", __FUNCTION__
, msr
);
375 static __init
int svm_hardware_setup(void)
378 struct page
*iopm_pages
;
379 struct page
*msrpm_pages
;
383 kvm_emulator_want_group7_invlpg();
385 iopm_pages
= alloc_pages(GFP_KERNEL
, IOPM_ALLOC_ORDER
);
389 memset(page_address(iopm_pages
), 0xff,
390 PAGE_SIZE
* (1 << IOPM_ALLOC_ORDER
));
391 iopm_base
= page_to_pfn(iopm_pages
) << PAGE_SHIFT
;
394 msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
400 msrpm_va
= page_address(msrpm_pages
);
401 memset(msrpm_va
, 0xff, PAGE_SIZE
* (1 << MSRPM_ALLOC_ORDER
));
402 msrpm_base
= page_to_pfn(msrpm_pages
) << PAGE_SHIFT
;
405 set_msr_interception(msrpm_va
, MSR_GS_BASE
, 1, 1);
406 set_msr_interception(msrpm_va
, MSR_FS_BASE
, 1, 1);
407 set_msr_interception(msrpm_va
, MSR_KERNEL_GS_BASE
, 1, 1);
408 set_msr_interception(msrpm_va
, MSR_LSTAR
, 1, 1);
409 set_msr_interception(msrpm_va
, MSR_CSTAR
, 1, 1);
410 set_msr_interception(msrpm_va
, MSR_SYSCALL_MASK
, 1, 1);
412 set_msr_interception(msrpm_va
, MSR_K6_STAR
, 1, 1);
413 set_msr_interception(msrpm_va
, MSR_IA32_SYSENTER_CS
, 1, 1);
414 set_msr_interception(msrpm_va
, MSR_IA32_SYSENTER_ESP
, 1, 1);
415 set_msr_interception(msrpm_va
, MSR_IA32_SYSENTER_EIP
, 1, 1);
417 for_each_online_cpu(cpu
) {
418 r
= svm_cpu_init(cpu
);
425 __free_pages(msrpm_pages
, MSRPM_ALLOC_ORDER
);
428 __free_pages(iopm_pages
, IOPM_ALLOC_ORDER
);
433 static __exit
void svm_hardware_unsetup(void)
435 __free_pages(pfn_to_page(msrpm_base
>> PAGE_SHIFT
), MSRPM_ALLOC_ORDER
);
436 __free_pages(pfn_to_page(iopm_base
>> PAGE_SHIFT
), IOPM_ALLOC_ORDER
);
437 iopm_base
= msrpm_base
= 0;
440 static void init_seg(struct vmcb_seg
*seg
)
443 seg
->attrib
= SVM_SELECTOR_P_MASK
| SVM_SELECTOR_S_MASK
|
444 SVM_SELECTOR_WRITE_MASK
; /* Read/Write Data Segment */
449 static void init_sys_seg(struct vmcb_seg
*seg
, uint32_t type
)
452 seg
->attrib
= SVM_SELECTOR_P_MASK
| type
;
457 static int svm_vcpu_setup(struct kvm_vcpu
*vcpu
)
462 static void init_vmcb(struct vmcb
*vmcb
)
464 struct vmcb_control_area
*control
= &vmcb
->control
;
465 struct vmcb_save_area
*save
= &vmcb
->save
;
467 control
->intercept_cr_read
= INTERCEPT_CR0_MASK
|
471 control
->intercept_cr_write
= INTERCEPT_CR0_MASK
|
475 control
->intercept_dr_read
= INTERCEPT_DR0_MASK
|
480 control
->intercept_dr_write
= INTERCEPT_DR0_MASK
|
487 control
->intercept_exceptions
= 1 << PF_VECTOR
;
490 control
->intercept
= (1ULL << INTERCEPT_INTR
) |
491 (1ULL << INTERCEPT_NMI
) |
492 (1ULL << INTERCEPT_SMI
) |
494 * selective cr0 intercept bug?
495 * 0: 0f 22 d8 mov %eax,%cr3
496 * 3: 0f 20 c0 mov %cr0,%eax
497 * 6: 0d 00 00 00 80 or $0x80000000,%eax
498 * b: 0f 22 c0 mov %eax,%cr0
499 * set cr3 ->interception
500 * get cr0 ->interception
501 * set cr0 -> no interception
503 /* (1ULL << INTERCEPT_SELECTIVE_CR0) | */
504 (1ULL << INTERCEPT_CPUID
) |
505 (1ULL << INTERCEPT_HLT
) |
506 (1ULL << INTERCEPT_INVLPGA
) |
507 (1ULL << INTERCEPT_IOIO_PROT
) |
508 (1ULL << INTERCEPT_MSR_PROT
) |
509 (1ULL << INTERCEPT_TASK_SWITCH
) |
510 (1ULL << INTERCEPT_SHUTDOWN
) |
511 (1ULL << INTERCEPT_VMRUN
) |
512 (1ULL << INTERCEPT_VMMCALL
) |
513 (1ULL << INTERCEPT_VMLOAD
) |
514 (1ULL << INTERCEPT_VMSAVE
) |
515 (1ULL << INTERCEPT_STGI
) |
516 (1ULL << INTERCEPT_CLGI
) |
517 (1ULL << INTERCEPT_SKINIT
) |
518 (1ULL << INTERCEPT_MONITOR
) |
519 (1ULL << INTERCEPT_MWAIT
);
521 control
->iopm_base_pa
= iopm_base
;
522 control
->msrpm_base_pa
= msrpm_base
;
523 control
->tsc_offset
= 0;
524 control
->int_ctl
= V_INTR_MASKING_MASK
;
525 if (svm_has(SVM_FEATURE_LBRV
))
526 control
->lbr_ctl
= 1ULL;
534 save
->cs
.selector
= 0xf000;
535 /* Executable/Readable Code Segment */
536 save
->cs
.attrib
= SVM_SELECTOR_READ_MASK
| SVM_SELECTOR_P_MASK
|
537 SVM_SELECTOR_S_MASK
| SVM_SELECTOR_CODE_MASK
;
538 save
->cs
.limit
= 0xffff;
540 * cs.base should really be 0xffff0000, but vmx can't handle that, so
541 * be consistent with it.
543 * Replace when we have real mode working for vmx.
545 save
->cs
.base
= 0xf0000;
547 save
->gdtr
.limit
= 0xffff;
548 save
->idtr
.limit
= 0xffff;
550 init_sys_seg(&save
->ldtr
, SEG_TYPE_LDT
);
551 init_sys_seg(&save
->tr
, SEG_TYPE_BUSY_TSS16
);
553 save
->efer
= MSR_EFER_SVME_MASK
;
555 save
->dr6
= 0xffff0ff0;
558 save
->rip
= 0x0000fff0;
561 * cr0 val on cpu init should be 0x60000010, we enable cpu
562 * cache by default. the orderly way is to enable cache in bios.
564 save
->cr0
= 0x00000010 | CR0_PG_MASK
| CR0_WP_MASK
;
565 save
->cr4
= CR4_PAE_MASK
;
569 static int svm_create_vcpu(struct kvm_vcpu
*vcpu
)
575 vcpu
->svm
= kzalloc(sizeof *vcpu
->svm
, GFP_KERNEL
);
578 page
= alloc_page(GFP_KERNEL
);
582 vcpu
->svm
->vmcb
= page_address(page
);
583 memset(vcpu
->svm
->vmcb
, 0, PAGE_SIZE
);
584 vcpu
->svm
->vmcb_pa
= page_to_pfn(page
) << PAGE_SHIFT
;
585 vcpu
->svm
->asid_generation
= 0;
586 memset(vcpu
->svm
->db_regs
, 0, sizeof(vcpu
->svm
->db_regs
));
587 init_vmcb(vcpu
->svm
->vmcb
);
590 vcpu
->fpu_active
= 1;
591 vcpu
->apic_base
= 0xfee00000 |
592 /*for vcpu 0*/ MSR_IA32_APICBASE_BSP
|
593 MSR_IA32_APICBASE_ENABLE
;
603 static void svm_free_vcpu(struct kvm_vcpu
*vcpu
)
608 __free_page(pfn_to_page(vcpu
->svm
->vmcb_pa
>> PAGE_SHIFT
));
612 static void svm_vcpu_load(struct kvm_vcpu
*vcpu
)
617 if (unlikely(cpu
!= vcpu
->cpu
)) {
621 * Make sure that the guest sees a monotonically
625 delta
= vcpu
->host_tsc
- tsc_this
;
626 vcpu
->svm
->vmcb
->control
.tsc_offset
+= delta
;
631 static void svm_vcpu_put(struct kvm_vcpu
*vcpu
)
633 rdtscll(vcpu
->host_tsc
);
637 static void svm_vcpu_decache(struct kvm_vcpu
*vcpu
)
641 static void svm_cache_regs(struct kvm_vcpu
*vcpu
)
643 vcpu
->regs
[VCPU_REGS_RAX
] = vcpu
->svm
->vmcb
->save
.rax
;
644 vcpu
->regs
[VCPU_REGS_RSP
] = vcpu
->svm
->vmcb
->save
.rsp
;
645 vcpu
->rip
= vcpu
->svm
->vmcb
->save
.rip
;
648 static void svm_decache_regs(struct kvm_vcpu
*vcpu
)
650 vcpu
->svm
->vmcb
->save
.rax
= vcpu
->regs
[VCPU_REGS_RAX
];
651 vcpu
->svm
->vmcb
->save
.rsp
= vcpu
->regs
[VCPU_REGS_RSP
];
652 vcpu
->svm
->vmcb
->save
.rip
= vcpu
->rip
;
655 static unsigned long svm_get_rflags(struct kvm_vcpu
*vcpu
)
657 return vcpu
->svm
->vmcb
->save
.rflags
;
660 static void svm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
662 vcpu
->svm
->vmcb
->save
.rflags
= rflags
;
665 static struct vmcb_seg
*svm_seg(struct kvm_vcpu
*vcpu
, int seg
)
667 struct vmcb_save_area
*save
= &vcpu
->svm
->vmcb
->save
;
670 case VCPU_SREG_CS
: return &save
->cs
;
671 case VCPU_SREG_DS
: return &save
->ds
;
672 case VCPU_SREG_ES
: return &save
->es
;
673 case VCPU_SREG_FS
: return &save
->fs
;
674 case VCPU_SREG_GS
: return &save
->gs
;
675 case VCPU_SREG_SS
: return &save
->ss
;
676 case VCPU_SREG_TR
: return &save
->tr
;
677 case VCPU_SREG_LDTR
: return &save
->ldtr
;
683 static u64
svm_get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
685 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
690 static void svm_get_segment(struct kvm_vcpu
*vcpu
,
691 struct kvm_segment
*var
, int seg
)
693 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
696 var
->limit
= s
->limit
;
697 var
->selector
= s
->selector
;
698 var
->type
= s
->attrib
& SVM_SELECTOR_TYPE_MASK
;
699 var
->s
= (s
->attrib
>> SVM_SELECTOR_S_SHIFT
) & 1;
700 var
->dpl
= (s
->attrib
>> SVM_SELECTOR_DPL_SHIFT
) & 3;
701 var
->present
= (s
->attrib
>> SVM_SELECTOR_P_SHIFT
) & 1;
702 var
->avl
= (s
->attrib
>> SVM_SELECTOR_AVL_SHIFT
) & 1;
703 var
->l
= (s
->attrib
>> SVM_SELECTOR_L_SHIFT
) & 1;
704 var
->db
= (s
->attrib
>> SVM_SELECTOR_DB_SHIFT
) & 1;
705 var
->g
= (s
->attrib
>> SVM_SELECTOR_G_SHIFT
) & 1;
706 var
->unusable
= !var
->present
;
709 static void svm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
711 struct vmcb_seg
*s
= svm_seg(vcpu
, VCPU_SREG_CS
);
713 *db
= (s
->attrib
>> SVM_SELECTOR_DB_SHIFT
) & 1;
714 *l
= (s
->attrib
>> SVM_SELECTOR_L_SHIFT
) & 1;
717 static void svm_get_idt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
719 dt
->limit
= vcpu
->svm
->vmcb
->save
.idtr
.limit
;
720 dt
->base
= vcpu
->svm
->vmcb
->save
.idtr
.base
;
723 static void svm_set_idt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
725 vcpu
->svm
->vmcb
->save
.idtr
.limit
= dt
->limit
;
726 vcpu
->svm
->vmcb
->save
.idtr
.base
= dt
->base
;
729 static void svm_get_gdt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
731 dt
->limit
= vcpu
->svm
->vmcb
->save
.gdtr
.limit
;
732 dt
->base
= vcpu
->svm
->vmcb
->save
.gdtr
.base
;
735 static void svm_set_gdt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
737 vcpu
->svm
->vmcb
->save
.gdtr
.limit
= dt
->limit
;
738 vcpu
->svm
->vmcb
->save
.gdtr
.base
= dt
->base
;
741 static void svm_decache_cr4_guest_bits(struct kvm_vcpu
*vcpu
)
745 static void svm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
748 if (vcpu
->shadow_efer
& KVM_EFER_LME
) {
749 if (!is_paging(vcpu
) && (cr0
& CR0_PG_MASK
)) {
750 vcpu
->shadow_efer
|= KVM_EFER_LMA
;
751 vcpu
->svm
->vmcb
->save
.efer
|= KVM_EFER_LMA
| KVM_EFER_LME
;
754 if (is_paging(vcpu
) && !(cr0
& CR0_PG_MASK
) ) {
755 vcpu
->shadow_efer
&= ~KVM_EFER_LMA
;
756 vcpu
->svm
->vmcb
->save
.efer
&= ~(KVM_EFER_LMA
| KVM_EFER_LME
);
760 if ((vcpu
->cr0
& CR0_TS_MASK
) && !(cr0
& CR0_TS_MASK
)) {
761 vcpu
->svm
->vmcb
->control
.intercept_exceptions
&= ~(1 << NM_VECTOR
);
762 vcpu
->fpu_active
= 1;
766 cr0
|= CR0_PG_MASK
| CR0_WP_MASK
;
767 cr0
&= ~(CR0_CD_MASK
| CR0_NW_MASK
);
768 vcpu
->svm
->vmcb
->save
.cr0
= cr0
;
771 static void svm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
774 vcpu
->svm
->vmcb
->save
.cr4
= cr4
| CR4_PAE_MASK
;
777 static void svm_set_segment(struct kvm_vcpu
*vcpu
,
778 struct kvm_segment
*var
, int seg
)
780 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
783 s
->limit
= var
->limit
;
784 s
->selector
= var
->selector
;
788 s
->attrib
= (var
->type
& SVM_SELECTOR_TYPE_MASK
);
789 s
->attrib
|= (var
->s
& 1) << SVM_SELECTOR_S_SHIFT
;
790 s
->attrib
|= (var
->dpl
& 3) << SVM_SELECTOR_DPL_SHIFT
;
791 s
->attrib
|= (var
->present
& 1) << SVM_SELECTOR_P_SHIFT
;
792 s
->attrib
|= (var
->avl
& 1) << SVM_SELECTOR_AVL_SHIFT
;
793 s
->attrib
|= (var
->l
& 1) << SVM_SELECTOR_L_SHIFT
;
794 s
->attrib
|= (var
->db
& 1) << SVM_SELECTOR_DB_SHIFT
;
795 s
->attrib
|= (var
->g
& 1) << SVM_SELECTOR_G_SHIFT
;
797 if (seg
== VCPU_SREG_CS
)
798 vcpu
->svm
->vmcb
->save
.cpl
799 = (vcpu
->svm
->vmcb
->save
.cs
.attrib
800 >> SVM_SELECTOR_DPL_SHIFT
) & 3;
806 vcpu->svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
807 vcpu->svm->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
811 static int svm_guest_debug(struct kvm_vcpu
*vcpu
, struct kvm_debug_guest
*dbg
)
816 static void load_host_msrs(struct kvm_vcpu
*vcpu
)
820 for ( i
= 0; i
< NR_HOST_SAVE_MSRS
; i
++)
821 wrmsrl(host_save_msrs
[i
], vcpu
->svm
->host_msrs
[i
]);
824 static void save_host_msrs(struct kvm_vcpu
*vcpu
)
828 for ( i
= 0; i
< NR_HOST_SAVE_MSRS
; i
++)
829 rdmsrl(host_save_msrs
[i
], vcpu
->svm
->host_msrs
[i
]);
832 static void new_asid(struct kvm_vcpu
*vcpu
, struct svm_cpu_data
*svm_data
)
834 if (svm_data
->next_asid
> svm_data
->max_asid
) {
835 ++svm_data
->asid_generation
;
836 svm_data
->next_asid
= 1;
837 vcpu
->svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ALL_ASID
;
840 vcpu
->cpu
= svm_data
->cpu
;
841 vcpu
->svm
->asid_generation
= svm_data
->asid_generation
;
842 vcpu
->svm
->vmcb
->control
.asid
= svm_data
->next_asid
++;
845 static void svm_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
847 invlpga(address
, vcpu
->svm
->vmcb
->control
.asid
); // is needed?
850 static unsigned long svm_get_dr(struct kvm_vcpu
*vcpu
, int dr
)
852 return vcpu
->svm
->db_regs
[dr
];
855 static void svm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long value
,
860 if (vcpu
->svm
->vmcb
->save
.dr7
& DR7_GD_MASK
) {
861 vcpu
->svm
->vmcb
->save
.dr7
&= ~DR7_GD_MASK
;
862 vcpu
->svm
->vmcb
->save
.dr6
|= DR6_BD_MASK
;
863 *exception
= DB_VECTOR
;
869 vcpu
->svm
->db_regs
[dr
] = value
;
872 if (vcpu
->cr4
& CR4_DE_MASK
) {
873 *exception
= UD_VECTOR
;
877 if (value
& ~((1ULL << 32) - 1)) {
878 *exception
= GP_VECTOR
;
881 vcpu
->svm
->vmcb
->save
.dr7
= value
;
885 printk(KERN_DEBUG
"%s: unexpected dr %u\n",
887 *exception
= UD_VECTOR
;
892 static int pf_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
894 u32 exit_int_info
= vcpu
->svm
->vmcb
->control
.exit_int_info
;
897 enum emulation_result er
;
900 if (is_external_interrupt(exit_int_info
))
901 push_irq(vcpu
, exit_int_info
& SVM_EVTINJ_VEC_MASK
);
903 spin_lock(&vcpu
->kvm
->lock
);
905 fault_address
= vcpu
->svm
->vmcb
->control
.exit_info_2
;
906 error_code
= vcpu
->svm
->vmcb
->control
.exit_info_1
;
907 r
= kvm_mmu_page_fault(vcpu
, fault_address
, error_code
);
909 spin_unlock(&vcpu
->kvm
->lock
);
913 spin_unlock(&vcpu
->kvm
->lock
);
916 er
= emulate_instruction(vcpu
, kvm_run
, fault_address
, error_code
);
917 spin_unlock(&vcpu
->kvm
->lock
);
922 case EMULATE_DO_MMIO
:
923 ++vcpu
->stat
.mmio_exits
;
924 kvm_run
->exit_reason
= KVM_EXIT_MMIO
;
927 vcpu_printf(vcpu
, "%s: emulate fail\n", __FUNCTION__
);
933 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
937 static int nm_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
939 vcpu
->svm
->vmcb
->control
.intercept_exceptions
&= ~(1 << NM_VECTOR
);
940 if (!(vcpu
->cr0
& CR0_TS_MASK
))
941 vcpu
->svm
->vmcb
->save
.cr0
&= ~CR0_TS_MASK
;
942 vcpu
->fpu_active
= 1;
947 static int shutdown_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
950 * VMCB is undefined after a SHUTDOWN intercept
951 * so reinitialize it.
953 memset(vcpu
->svm
->vmcb
, 0, PAGE_SIZE
);
954 init_vmcb(vcpu
->svm
->vmcb
);
956 kvm_run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
960 static int io_get_override(struct kvm_vcpu
*vcpu
,
961 struct vmcb_seg
**seg
,
964 u8 inst
[MAX_INST_SIZE
];
969 rip
= vcpu
->svm
->vmcb
->save
.rip
;
970 ins_length
= vcpu
->svm
->next_rip
- rip
;
971 rip
+= vcpu
->svm
->vmcb
->save
.cs
.base
;
973 if (ins_length
> MAX_INST_SIZE
)
975 "%s: inst length err, cs base 0x%llx rip 0x%llx "
976 "next rip 0x%llx ins_length %u\n",
978 vcpu
->svm
->vmcb
->save
.cs
.base
,
979 vcpu
->svm
->vmcb
->save
.rip
,
980 vcpu
->svm
->vmcb
->control
.exit_info_2
,
983 if (kvm_read_guest(vcpu
, rip
, ins_length
, inst
) != ins_length
)
989 for (i
= 0; i
< ins_length
; i
++)
1000 *seg
= &vcpu
->svm
->vmcb
->save
.cs
;
1003 *seg
= &vcpu
->svm
->vmcb
->save
.ss
;
1006 *seg
= &vcpu
->svm
->vmcb
->save
.ds
;
1009 *seg
= &vcpu
->svm
->vmcb
->save
.es
;
1012 *seg
= &vcpu
->svm
->vmcb
->save
.fs
;
1015 *seg
= &vcpu
->svm
->vmcb
->save
.gs
;
1020 printk(KERN_DEBUG
"%s: unexpected\n", __FUNCTION__
);
1024 static unsigned long io_adress(struct kvm_vcpu
*vcpu
, int ins
, gva_t
*address
)
1026 unsigned long addr_mask
;
1028 struct vmcb_seg
*seg
;
1030 struct vmcb_save_area
*save_area
= &vcpu
->svm
->vmcb
->save
;
1031 u16 cs_attrib
= save_area
->cs
.attrib
;
1032 unsigned addr_size
= get_addr_size(vcpu
);
1034 if (!io_get_override(vcpu
, &seg
, &addr_override
))
1038 addr_size
= (addr_size
== 2) ? 4: (addr_size
>> 1);
1041 reg
= &vcpu
->regs
[VCPU_REGS_RDI
];
1042 seg
= &vcpu
->svm
->vmcb
->save
.es
;
1044 reg
= &vcpu
->regs
[VCPU_REGS_RSI
];
1045 seg
= (seg
) ? seg
: &vcpu
->svm
->vmcb
->save
.ds
;
1048 addr_mask
= ~0ULL >> (64 - (addr_size
* 8));
1050 if ((cs_attrib
& SVM_SELECTOR_L_MASK
) &&
1051 !(vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_VM
)) {
1052 *address
= (*reg
& addr_mask
);
1056 if (!(seg
->attrib
& SVM_SELECTOR_P_SHIFT
)) {
1057 svm_inject_gp(vcpu
, 0);
1061 *address
= (*reg
& addr_mask
) + seg
->base
;
1065 static int io_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1067 u32 io_info
= vcpu
->svm
->vmcb
->control
.exit_info_1
; //address size bug?
1068 int size
, down
, in
, string
, rep
;
1070 unsigned long count
;
1073 ++vcpu
->stat
.io_exits
;
1075 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->control
.exit_info_2
;
1077 in
= (io_info
& SVM_IOIO_TYPE_MASK
) != 0;
1078 port
= io_info
>> 16;
1079 size
= (io_info
& SVM_IOIO_SIZE_MASK
) >> SVM_IOIO_SIZE_SHIFT
;
1080 string
= (io_info
& SVM_IOIO_STR_MASK
) != 0;
1081 rep
= (io_info
& SVM_IOIO_REP_MASK
) != 0;
1083 down
= (vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_DF
) != 0;
1088 addr_mask
= io_adress(vcpu
, in
, &address
);
1090 printk(KERN_DEBUG
"%s: get io address failed\n",
1096 count
= vcpu
->regs
[VCPU_REGS_RCX
] & addr_mask
;
1098 return kvm_setup_pio(vcpu
, kvm_run
, in
, size
, count
, string
, down
,
1099 address
, rep
, port
);
1102 static int nop_on_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1107 static int halt_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1109 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 1;
1110 skip_emulated_instruction(vcpu
);
1111 if (vcpu
->irq_summary
)
1114 kvm_run
->exit_reason
= KVM_EXIT_HLT
;
1115 ++vcpu
->stat
.halt_exits
;
1119 static int vmmcall_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1121 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 3;
1122 skip_emulated_instruction(vcpu
);
1123 return kvm_hypercall(vcpu
, kvm_run
);
1126 static int invalid_op_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1132 static int task_switch_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1134 printk(KERN_DEBUG
"%s: task swiche is unsupported\n", __FUNCTION__
);
1135 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
1139 static int cpuid_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1141 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 2;
1142 kvm_emulate_cpuid(vcpu
);
1146 static int emulate_on_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1148 if (emulate_instruction(vcpu
, NULL
, 0, 0) != EMULATE_DONE
)
1149 printk(KERN_ERR
"%s: failed\n", __FUNCTION__
);
1153 static int svm_get_msr(struct kvm_vcpu
*vcpu
, unsigned ecx
, u64
*data
)
1156 case MSR_IA32_TIME_STAMP_COUNTER
: {
1160 *data
= vcpu
->svm
->vmcb
->control
.tsc_offset
+ tsc
;
1164 *data
= vcpu
->svm
->vmcb
->save
.star
;
1166 #ifdef CONFIG_X86_64
1168 *data
= vcpu
->svm
->vmcb
->save
.lstar
;
1171 *data
= vcpu
->svm
->vmcb
->save
.cstar
;
1173 case MSR_KERNEL_GS_BASE
:
1174 *data
= vcpu
->svm
->vmcb
->save
.kernel_gs_base
;
1176 case MSR_SYSCALL_MASK
:
1177 *data
= vcpu
->svm
->vmcb
->save
.sfmask
;
1180 case MSR_IA32_SYSENTER_CS
:
1181 *data
= vcpu
->svm
->vmcb
->save
.sysenter_cs
;
1183 case MSR_IA32_SYSENTER_EIP
:
1184 *data
= vcpu
->svm
->vmcb
->save
.sysenter_eip
;
1186 case MSR_IA32_SYSENTER_ESP
:
1187 *data
= vcpu
->svm
->vmcb
->save
.sysenter_esp
;
1190 return kvm_get_msr_common(vcpu
, ecx
, data
);
1195 static int rdmsr_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1197 u32 ecx
= vcpu
->regs
[VCPU_REGS_RCX
];
1200 if (svm_get_msr(vcpu
, ecx
, &data
))
1201 svm_inject_gp(vcpu
, 0);
1203 vcpu
->svm
->vmcb
->save
.rax
= data
& 0xffffffff;
1204 vcpu
->regs
[VCPU_REGS_RDX
] = data
>> 32;
1205 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 2;
1206 skip_emulated_instruction(vcpu
);
1211 static int svm_set_msr(struct kvm_vcpu
*vcpu
, unsigned ecx
, u64 data
)
1214 case MSR_IA32_TIME_STAMP_COUNTER
: {
1218 vcpu
->svm
->vmcb
->control
.tsc_offset
= data
- tsc
;
1222 vcpu
->svm
->vmcb
->save
.star
= data
;
1224 #ifdef CONFIG_X86_64
1226 vcpu
->svm
->vmcb
->save
.lstar
= data
;
1229 vcpu
->svm
->vmcb
->save
.cstar
= data
;
1231 case MSR_KERNEL_GS_BASE
:
1232 vcpu
->svm
->vmcb
->save
.kernel_gs_base
= data
;
1234 case MSR_SYSCALL_MASK
:
1235 vcpu
->svm
->vmcb
->save
.sfmask
= data
;
1238 case MSR_IA32_SYSENTER_CS
:
1239 vcpu
->svm
->vmcb
->save
.sysenter_cs
= data
;
1241 case MSR_IA32_SYSENTER_EIP
:
1242 vcpu
->svm
->vmcb
->save
.sysenter_eip
= data
;
1244 case MSR_IA32_SYSENTER_ESP
:
1245 vcpu
->svm
->vmcb
->save
.sysenter_esp
= data
;
1248 return kvm_set_msr_common(vcpu
, ecx
, data
);
1253 static int wrmsr_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1255 u32 ecx
= vcpu
->regs
[VCPU_REGS_RCX
];
1256 u64 data
= (vcpu
->svm
->vmcb
->save
.rax
& -1u)
1257 | ((u64
)(vcpu
->regs
[VCPU_REGS_RDX
] & -1u) << 32);
1258 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 2;
1259 if (svm_set_msr(vcpu
, ecx
, data
))
1260 svm_inject_gp(vcpu
, 0);
1262 skip_emulated_instruction(vcpu
);
1266 static int msr_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1268 if (vcpu
->svm
->vmcb
->control
.exit_info_1
)
1269 return wrmsr_interception(vcpu
, kvm_run
);
1271 return rdmsr_interception(vcpu
, kvm_run
);
1274 static int interrupt_window_interception(struct kvm_vcpu
*vcpu
,
1275 struct kvm_run
*kvm_run
)
1278 * If the user space waits to inject interrupts, exit as soon as
1281 if (kvm_run
->request_interrupt_window
&&
1282 !vcpu
->irq_summary
) {
1283 ++vcpu
->stat
.irq_window_exits
;
1284 kvm_run
->exit_reason
= KVM_EXIT_IRQ_WINDOW_OPEN
;
1291 static int (*svm_exit_handlers
[])(struct kvm_vcpu
*vcpu
,
1292 struct kvm_run
*kvm_run
) = {
1293 [SVM_EXIT_READ_CR0
] = emulate_on_interception
,
1294 [SVM_EXIT_READ_CR3
] = emulate_on_interception
,
1295 [SVM_EXIT_READ_CR4
] = emulate_on_interception
,
1297 [SVM_EXIT_WRITE_CR0
] = emulate_on_interception
,
1298 [SVM_EXIT_WRITE_CR3
] = emulate_on_interception
,
1299 [SVM_EXIT_WRITE_CR4
] = emulate_on_interception
,
1300 [SVM_EXIT_READ_DR0
] = emulate_on_interception
,
1301 [SVM_EXIT_READ_DR1
] = emulate_on_interception
,
1302 [SVM_EXIT_READ_DR2
] = emulate_on_interception
,
1303 [SVM_EXIT_READ_DR3
] = emulate_on_interception
,
1304 [SVM_EXIT_WRITE_DR0
] = emulate_on_interception
,
1305 [SVM_EXIT_WRITE_DR1
] = emulate_on_interception
,
1306 [SVM_EXIT_WRITE_DR2
] = emulate_on_interception
,
1307 [SVM_EXIT_WRITE_DR3
] = emulate_on_interception
,
1308 [SVM_EXIT_WRITE_DR5
] = emulate_on_interception
,
1309 [SVM_EXIT_WRITE_DR7
] = emulate_on_interception
,
1310 [SVM_EXIT_EXCP_BASE
+ PF_VECTOR
] = pf_interception
,
1311 [SVM_EXIT_EXCP_BASE
+ NM_VECTOR
] = nm_interception
,
1312 [SVM_EXIT_INTR
] = nop_on_interception
,
1313 [SVM_EXIT_NMI
] = nop_on_interception
,
1314 [SVM_EXIT_SMI
] = nop_on_interception
,
1315 [SVM_EXIT_INIT
] = nop_on_interception
,
1316 [SVM_EXIT_VINTR
] = interrupt_window_interception
,
1317 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
1318 [SVM_EXIT_CPUID
] = cpuid_interception
,
1319 [SVM_EXIT_HLT
] = halt_interception
,
1320 [SVM_EXIT_INVLPG
] = emulate_on_interception
,
1321 [SVM_EXIT_INVLPGA
] = invalid_op_interception
,
1322 [SVM_EXIT_IOIO
] = io_interception
,
1323 [SVM_EXIT_MSR
] = msr_interception
,
1324 [SVM_EXIT_TASK_SWITCH
] = task_switch_interception
,
1325 [SVM_EXIT_SHUTDOWN
] = shutdown_interception
,
1326 [SVM_EXIT_VMRUN
] = invalid_op_interception
,
1327 [SVM_EXIT_VMMCALL
] = vmmcall_interception
,
1328 [SVM_EXIT_VMLOAD
] = invalid_op_interception
,
1329 [SVM_EXIT_VMSAVE
] = invalid_op_interception
,
1330 [SVM_EXIT_STGI
] = invalid_op_interception
,
1331 [SVM_EXIT_CLGI
] = invalid_op_interception
,
1332 [SVM_EXIT_SKINIT
] = invalid_op_interception
,
1333 [SVM_EXIT_MONITOR
] = invalid_op_interception
,
1334 [SVM_EXIT_MWAIT
] = invalid_op_interception
,
1338 static int handle_exit(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1340 u32 exit_code
= vcpu
->svm
->vmcb
->control
.exit_code
;
1342 if (is_external_interrupt(vcpu
->svm
->vmcb
->control
.exit_int_info
) &&
1343 exit_code
!= SVM_EXIT_EXCP_BASE
+ PF_VECTOR
)
1344 printk(KERN_ERR
"%s: unexpected exit_ini_info 0x%x "
1346 __FUNCTION__
, vcpu
->svm
->vmcb
->control
.exit_int_info
,
1349 if (exit_code
>= ARRAY_SIZE(svm_exit_handlers
)
1350 || svm_exit_handlers
[exit_code
] == 0) {
1351 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
1352 kvm_run
->hw
.hardware_exit_reason
= exit_code
;
1356 return svm_exit_handlers
[exit_code
](vcpu
, kvm_run
);
1359 static void reload_tss(struct kvm_vcpu
*vcpu
)
1361 int cpu
= raw_smp_processor_id();
1363 struct svm_cpu_data
*svm_data
= per_cpu(svm_data
, cpu
);
1364 svm_data
->tss_desc
->type
= 9; //available 32/64-bit TSS
1368 static void pre_svm_run(struct kvm_vcpu
*vcpu
)
1370 int cpu
= raw_smp_processor_id();
1372 struct svm_cpu_data
*svm_data
= per_cpu(svm_data
, cpu
);
1374 vcpu
->svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_DO_NOTHING
;
1375 if (vcpu
->cpu
!= cpu
||
1376 vcpu
->svm
->asid_generation
!= svm_data
->asid_generation
)
1377 new_asid(vcpu
, svm_data
);
1381 static inline void kvm_do_inject_irq(struct kvm_vcpu
*vcpu
)
1383 struct vmcb_control_area
*control
;
1385 control
= &vcpu
->svm
->vmcb
->control
;
1386 control
->int_vector
= pop_irq(vcpu
);
1387 control
->int_ctl
&= ~V_INTR_PRIO_MASK
;
1388 control
->int_ctl
|= V_IRQ_MASK
|
1389 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT
);
1392 static void kvm_reput_irq(struct kvm_vcpu
*vcpu
)
1394 struct vmcb_control_area
*control
= &vcpu
->svm
->vmcb
->control
;
1396 if (control
->int_ctl
& V_IRQ_MASK
) {
1397 control
->int_ctl
&= ~V_IRQ_MASK
;
1398 push_irq(vcpu
, control
->int_vector
);
1401 vcpu
->interrupt_window_open
=
1402 !(control
->int_state
& SVM_INTERRUPT_SHADOW_MASK
);
1405 static void do_interrupt_requests(struct kvm_vcpu
*vcpu
,
1406 struct kvm_run
*kvm_run
)
1408 struct vmcb_control_area
*control
= &vcpu
->svm
->vmcb
->control
;
1410 vcpu
->interrupt_window_open
=
1411 (!(control
->int_state
& SVM_INTERRUPT_SHADOW_MASK
) &&
1412 (vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_IF
));
1414 if (vcpu
->interrupt_window_open
&& vcpu
->irq_summary
)
1416 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1418 kvm_do_inject_irq(vcpu
);
1421 * Interrupts blocked. Wait for unblock.
1423 if (!vcpu
->interrupt_window_open
&&
1424 (vcpu
->irq_summary
|| kvm_run
->request_interrupt_window
)) {
1425 control
->intercept
|= 1ULL << INTERCEPT_VINTR
;
1427 control
->intercept
&= ~(1ULL << INTERCEPT_VINTR
);
1430 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
,
1431 struct kvm_run
*kvm_run
)
1433 kvm_run
->ready_for_interrupt_injection
= (vcpu
->interrupt_window_open
&&
1434 vcpu
->irq_summary
== 0);
1435 kvm_run
->if_flag
= (vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_IF
) != 0;
1436 kvm_run
->cr8
= vcpu
->cr8
;
1437 kvm_run
->apic_base
= vcpu
->apic_base
;
1441 * Check if userspace requested an interrupt window, and that the
1442 * interrupt window is open.
1444 * No need to exit to userspace if we already have an interrupt queued.
1446 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
,
1447 struct kvm_run
*kvm_run
)
1449 return (!vcpu
->irq_summary
&&
1450 kvm_run
->request_interrupt_window
&&
1451 vcpu
->interrupt_window_open
&&
1452 (vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_IF
));
1455 static void save_db_regs(unsigned long *db_regs
)
1457 asm volatile ("mov %%dr0, %0" : "=r"(db_regs
[0]));
1458 asm volatile ("mov %%dr1, %0" : "=r"(db_regs
[1]));
1459 asm volatile ("mov %%dr2, %0" : "=r"(db_regs
[2]));
1460 asm volatile ("mov %%dr3, %0" : "=r"(db_regs
[3]));
1463 static void load_db_regs(unsigned long *db_regs
)
1465 asm volatile ("mov %0, %%dr0" : : "r"(db_regs
[0]));
1466 asm volatile ("mov %0, %%dr1" : : "r"(db_regs
[1]));
1467 asm volatile ("mov %0, %%dr2" : : "r"(db_regs
[2]));
1468 asm volatile ("mov %0, %%dr3" : : "r"(db_regs
[3]));
1471 static int svm_vcpu_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1479 if (!vcpu
->mmio_read_completed
)
1480 do_interrupt_requests(vcpu
, kvm_run
);
1486 save_host_msrs(vcpu
);
1487 fs_selector
= read_fs();
1488 gs_selector
= read_gs();
1489 ldt_selector
= read_ldt();
1490 vcpu
->svm
->host_cr2
= kvm_read_cr2();
1491 vcpu
->svm
->host_dr6
= read_dr6();
1492 vcpu
->svm
->host_dr7
= read_dr7();
1493 vcpu
->svm
->vmcb
->save
.cr2
= vcpu
->cr2
;
1495 if (vcpu
->svm
->vmcb
->save
.dr7
& 0xff) {
1497 save_db_regs(vcpu
->svm
->host_db_regs
);
1498 load_db_regs(vcpu
->svm
->db_regs
);
1501 if (vcpu
->fpu_active
) {
1502 fx_save(vcpu
->host_fx_image
);
1503 fx_restore(vcpu
->guest_fx_image
);
1507 #ifdef CONFIG_X86_64
1508 "push %%rbx; push %%rcx; push %%rdx;"
1509 "push %%rsi; push %%rdi; push %%rbp;"
1510 "push %%r8; push %%r9; push %%r10; push %%r11;"
1511 "push %%r12; push %%r13; push %%r14; push %%r15;"
1513 "push %%ebx; push %%ecx; push %%edx;"
1514 "push %%esi; push %%edi; push %%ebp;"
1517 #ifdef CONFIG_X86_64
1518 "mov %c[rbx](%[vcpu]), %%rbx \n\t"
1519 "mov %c[rcx](%[vcpu]), %%rcx \n\t"
1520 "mov %c[rdx](%[vcpu]), %%rdx \n\t"
1521 "mov %c[rsi](%[vcpu]), %%rsi \n\t"
1522 "mov %c[rdi](%[vcpu]), %%rdi \n\t"
1523 "mov %c[rbp](%[vcpu]), %%rbp \n\t"
1524 "mov %c[r8](%[vcpu]), %%r8 \n\t"
1525 "mov %c[r9](%[vcpu]), %%r9 \n\t"
1526 "mov %c[r10](%[vcpu]), %%r10 \n\t"
1527 "mov %c[r11](%[vcpu]), %%r11 \n\t"
1528 "mov %c[r12](%[vcpu]), %%r12 \n\t"
1529 "mov %c[r13](%[vcpu]), %%r13 \n\t"
1530 "mov %c[r14](%[vcpu]), %%r14 \n\t"
1531 "mov %c[r15](%[vcpu]), %%r15 \n\t"
1533 "mov %c[rbx](%[vcpu]), %%ebx \n\t"
1534 "mov %c[rcx](%[vcpu]), %%ecx \n\t"
1535 "mov %c[rdx](%[vcpu]), %%edx \n\t"
1536 "mov %c[rsi](%[vcpu]), %%esi \n\t"
1537 "mov %c[rdi](%[vcpu]), %%edi \n\t"
1538 "mov %c[rbp](%[vcpu]), %%ebp \n\t"
1541 #ifdef CONFIG_X86_64
1542 /* Enter guest mode */
1544 "mov %c[svm](%[vcpu]), %%rax \n\t"
1545 "mov %c[vmcb](%%rax), %%rax \n\t"
1551 /* Enter guest mode */
1553 "mov %c[svm](%[vcpu]), %%eax \n\t"
1554 "mov %c[vmcb](%%eax), %%eax \n\t"
1561 /* Save guest registers, load host registers */
1562 #ifdef CONFIG_X86_64
1563 "mov %%rbx, %c[rbx](%[vcpu]) \n\t"
1564 "mov %%rcx, %c[rcx](%[vcpu]) \n\t"
1565 "mov %%rdx, %c[rdx](%[vcpu]) \n\t"
1566 "mov %%rsi, %c[rsi](%[vcpu]) \n\t"
1567 "mov %%rdi, %c[rdi](%[vcpu]) \n\t"
1568 "mov %%rbp, %c[rbp](%[vcpu]) \n\t"
1569 "mov %%r8, %c[r8](%[vcpu]) \n\t"
1570 "mov %%r9, %c[r9](%[vcpu]) \n\t"
1571 "mov %%r10, %c[r10](%[vcpu]) \n\t"
1572 "mov %%r11, %c[r11](%[vcpu]) \n\t"
1573 "mov %%r12, %c[r12](%[vcpu]) \n\t"
1574 "mov %%r13, %c[r13](%[vcpu]) \n\t"
1575 "mov %%r14, %c[r14](%[vcpu]) \n\t"
1576 "mov %%r15, %c[r15](%[vcpu]) \n\t"
1578 "pop %%r15; pop %%r14; pop %%r13; pop %%r12;"
1579 "pop %%r11; pop %%r10; pop %%r9; pop %%r8;"
1580 "pop %%rbp; pop %%rdi; pop %%rsi;"
1581 "pop %%rdx; pop %%rcx; pop %%rbx; \n\t"
1583 "mov %%ebx, %c[rbx](%[vcpu]) \n\t"
1584 "mov %%ecx, %c[rcx](%[vcpu]) \n\t"
1585 "mov %%edx, %c[rdx](%[vcpu]) \n\t"
1586 "mov %%esi, %c[rsi](%[vcpu]) \n\t"
1587 "mov %%edi, %c[rdi](%[vcpu]) \n\t"
1588 "mov %%ebp, %c[rbp](%[vcpu]) \n\t"
1590 "pop %%ebp; pop %%edi; pop %%esi;"
1591 "pop %%edx; pop %%ecx; pop %%ebx; \n\t"
1595 [svm
]"i"(offsetof(struct kvm_vcpu
, svm
)),
1596 [vmcb
]"i"(offsetof(struct vcpu_svm
, vmcb_pa
)),
1597 [rbx
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RBX
])),
1598 [rcx
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RCX
])),
1599 [rdx
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RDX
])),
1600 [rsi
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RSI
])),
1601 [rdi
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RDI
])),
1602 [rbp
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RBP
]))
1603 #ifdef CONFIG_X86_64
1604 ,[r8
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R8
])),
1605 [r9
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R9
])),
1606 [r10
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R10
])),
1607 [r11
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R11
])),
1608 [r12
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R12
])),
1609 [r13
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R13
])),
1610 [r14
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R14
])),
1611 [r15
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R15
]))
1615 if (vcpu
->fpu_active
) {
1616 fx_save(vcpu
->guest_fx_image
);
1617 fx_restore(vcpu
->host_fx_image
);
1620 if ((vcpu
->svm
->vmcb
->save
.dr7
& 0xff))
1621 load_db_regs(vcpu
->svm
->host_db_regs
);
1623 vcpu
->cr2
= vcpu
->svm
->vmcb
->save
.cr2
;
1625 write_dr6(vcpu
->svm
->host_dr6
);
1626 write_dr7(vcpu
->svm
->host_dr7
);
1627 kvm_write_cr2(vcpu
->svm
->host_cr2
);
1629 load_fs(fs_selector
);
1630 load_gs(gs_selector
);
1631 load_ldt(ldt_selector
);
1632 load_host_msrs(vcpu
);
1637 * Profile KVM exit RIPs:
1639 if (unlikely(prof_on
== KVM_PROFILING
))
1640 profile_hit(KVM_PROFILING
,
1641 (void *)(unsigned long)vcpu
->svm
->vmcb
->save
.rip
);
1645 kvm_reput_irq(vcpu
);
1647 vcpu
->svm
->next_rip
= 0;
1649 if (vcpu
->svm
->vmcb
->control
.exit_code
== SVM_EXIT_ERR
) {
1650 kvm_run
->exit_reason
= KVM_EXIT_FAIL_ENTRY
;
1651 kvm_run
->fail_entry
.hardware_entry_failure_reason
1652 = vcpu
->svm
->vmcb
->control
.exit_code
;
1653 post_kvm_run_save(vcpu
, kvm_run
);
1657 r
= handle_exit(vcpu
, kvm_run
);
1659 if (signal_pending(current
)) {
1660 ++vcpu
->stat
.signal_exits
;
1661 post_kvm_run_save(vcpu
, kvm_run
);
1662 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
1666 if (dm_request_for_irq_injection(vcpu
, kvm_run
)) {
1667 ++vcpu
->stat
.request_irq_exits
;
1668 post_kvm_run_save(vcpu
, kvm_run
);
1669 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
1675 post_kvm_run_save(vcpu
, kvm_run
);
1679 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
)
1681 force_new_asid(vcpu
);
1684 static void svm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
1686 vcpu
->svm
->vmcb
->save
.cr3
= root
;
1687 force_new_asid(vcpu
);
1689 if (vcpu
->fpu_active
) {
1690 vcpu
->svm
->vmcb
->control
.intercept_exceptions
|= (1 << NM_VECTOR
);
1691 vcpu
->svm
->vmcb
->save
.cr0
|= CR0_TS_MASK
;
1692 vcpu
->fpu_active
= 0;
1696 static void svm_inject_page_fault(struct kvm_vcpu
*vcpu
,
1700 uint32_t exit_int_info
= vcpu
->svm
->vmcb
->control
.exit_int_info
;
1702 ++vcpu
->stat
.pf_guest
;
1704 if (is_page_fault(exit_int_info
)) {
1706 vcpu
->svm
->vmcb
->control
.event_inj_err
= 0;
1707 vcpu
->svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
|
1708 SVM_EVTINJ_VALID_ERR
|
1709 SVM_EVTINJ_TYPE_EXEPT
|
1714 vcpu
->svm
->vmcb
->save
.cr2
= addr
;
1715 vcpu
->svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
|
1716 SVM_EVTINJ_VALID_ERR
|
1717 SVM_EVTINJ_TYPE_EXEPT
|
1719 vcpu
->svm
->vmcb
->control
.event_inj_err
= err_code
;
1723 static int is_disabled(void)
1729 svm_patch_hypercall(struct kvm_vcpu
*vcpu
, unsigned char *hypercall
)
1732 * Patch in the VMMCALL instruction:
1734 hypercall
[0] = 0x0f;
1735 hypercall
[1] = 0x01;
1736 hypercall
[2] = 0xd9;
1737 hypercall
[3] = 0xc3;
1740 static struct kvm_arch_ops svm_arch_ops
= {
1741 .cpu_has_kvm_support
= has_svm
,
1742 .disabled_by_bios
= is_disabled
,
1743 .hardware_setup
= svm_hardware_setup
,
1744 .hardware_unsetup
= svm_hardware_unsetup
,
1745 .hardware_enable
= svm_hardware_enable
,
1746 .hardware_disable
= svm_hardware_disable
,
1748 .vcpu_create
= svm_create_vcpu
,
1749 .vcpu_free
= svm_free_vcpu
,
1751 .vcpu_load
= svm_vcpu_load
,
1752 .vcpu_put
= svm_vcpu_put
,
1753 .vcpu_decache
= svm_vcpu_decache
,
1755 .set_guest_debug
= svm_guest_debug
,
1756 .get_msr
= svm_get_msr
,
1757 .set_msr
= svm_set_msr
,
1758 .get_segment_base
= svm_get_segment_base
,
1759 .get_segment
= svm_get_segment
,
1760 .set_segment
= svm_set_segment
,
1761 .get_cs_db_l_bits
= svm_get_cs_db_l_bits
,
1762 .decache_cr4_guest_bits
= svm_decache_cr4_guest_bits
,
1763 .set_cr0
= svm_set_cr0
,
1764 .set_cr3
= svm_set_cr3
,
1765 .set_cr4
= svm_set_cr4
,
1766 .set_efer
= svm_set_efer
,
1767 .get_idt
= svm_get_idt
,
1768 .set_idt
= svm_set_idt
,
1769 .get_gdt
= svm_get_gdt
,
1770 .set_gdt
= svm_set_gdt
,
1771 .get_dr
= svm_get_dr
,
1772 .set_dr
= svm_set_dr
,
1773 .cache_regs
= svm_cache_regs
,
1774 .decache_regs
= svm_decache_regs
,
1775 .get_rflags
= svm_get_rflags
,
1776 .set_rflags
= svm_set_rflags
,
1778 .invlpg
= svm_invlpg
,
1779 .tlb_flush
= svm_flush_tlb
,
1780 .inject_page_fault
= svm_inject_page_fault
,
1782 .inject_gp
= svm_inject_gp
,
1784 .run
= svm_vcpu_run
,
1785 .skip_emulated_instruction
= skip_emulated_instruction
,
1786 .vcpu_setup
= svm_vcpu_setup
,
1787 .patch_hypercall
= svm_patch_hypercall
,
1790 static int __init
svm_init(void)
1792 return kvm_init_arch(&svm_arch_ops
, THIS_MODULE
);
1795 static void __exit
svm_exit(void)
1800 module_init(svm_init
)
1801 module_exit(svm_exit
)