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 unsigned long iopm_base
;
48 unsigned long msrpm_base
;
50 struct kvm_ldttss_desc
{
53 unsigned base1
: 8, type
: 5, dpl
: 2, p
: 1;
54 unsigned limit1
: 4, zero0
: 3, g
: 1, base2
: 8;
57 } __attribute__((packed
));
62 uint64_t asid_generation
;
65 struct kvm_ldttss_desc
*tss_desc
;
67 struct page
*save_area
;
70 static DEFINE_PER_CPU(struct svm_cpu_data
*, svm_data
);
72 struct svm_init_data
{
77 static u32 msrpm_ranges
[] = {0, 0xc0000000, 0xc0010000};
79 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
80 #define MSRS_RANGE_SIZE 2048
81 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
83 #define MAX_INST_SIZE 15
85 static unsigned get_addr_size(struct kvm_vcpu
*vcpu
)
87 struct vmcb_save_area
*sa
= &vcpu
->svm
->vmcb
->save
;
90 if (!(sa
->cr0
& CR0_PE_MASK
) || (sa
->rflags
& X86_EFLAGS_VM
))
93 cs_attrib
= sa
->cs
.attrib
;
95 return (cs_attrib
& SVM_SELECTOR_L_MASK
) ? 8 :
96 (cs_attrib
& SVM_SELECTOR_DB_MASK
) ? 4 : 2;
99 static inline u8
pop_irq(struct kvm_vcpu
*vcpu
)
101 int word_index
= __ffs(vcpu
->irq_summary
);
102 int bit_index
= __ffs(vcpu
->irq_pending
[word_index
]);
103 int irq
= word_index
* BITS_PER_LONG
+ bit_index
;
105 clear_bit(bit_index
, &vcpu
->irq_pending
[word_index
]);
106 if (!vcpu
->irq_pending
[word_index
])
107 clear_bit(word_index
, &vcpu
->irq_summary
);
111 static inline void push_irq(struct kvm_vcpu
*vcpu
, u8 irq
)
113 set_bit(irq
, vcpu
->irq_pending
);
114 set_bit(irq
/ BITS_PER_LONG
, &vcpu
->irq_summary
);
117 static inline void clgi(void)
119 asm volatile (SVM_CLGI
);
122 static inline void stgi(void)
124 asm volatile (SVM_STGI
);
127 static inline void invlpga(unsigned long addr
, u32 asid
)
129 asm volatile (SVM_INVLPGA :: "a"(addr
), "c"(asid
));
132 static inline unsigned long kvm_read_cr2(void)
136 asm volatile ("mov %%cr2, %0" : "=r" (cr2
));
140 static inline void kvm_write_cr2(unsigned long val
)
142 asm volatile ("mov %0, %%cr2" :: "r" (val
));
145 static inline unsigned long read_dr6(void)
149 asm volatile ("mov %%dr6, %0" : "=r" (dr6
));
153 static inline void write_dr6(unsigned long val
)
155 asm volatile ("mov %0, %%dr6" :: "r" (val
));
158 static inline unsigned long read_dr7(void)
162 asm volatile ("mov %%dr7, %0" : "=r" (dr7
));
166 static inline void write_dr7(unsigned long val
)
168 asm volatile ("mov %0, %%dr7" :: "r" (val
));
171 static inline void force_new_asid(struct kvm_vcpu
*vcpu
)
173 vcpu
->svm
->asid_generation
--;
176 static inline void flush_guest_tlb(struct kvm_vcpu
*vcpu
)
178 force_new_asid(vcpu
);
181 static void svm_set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
183 if (!(efer
& KVM_EFER_LMA
))
184 efer
&= ~KVM_EFER_LME
;
186 vcpu
->svm
->vmcb
->save
.efer
= efer
| MSR_EFER_SVME_MASK
;
187 vcpu
->shadow_efer
= efer
;
190 static void svm_inject_gp(struct kvm_vcpu
*vcpu
, unsigned error_code
)
192 vcpu
->svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
|
193 SVM_EVTINJ_VALID_ERR
|
194 SVM_EVTINJ_TYPE_EXEPT
|
196 vcpu
->svm
->vmcb
->control
.event_inj_err
= error_code
;
199 static void inject_ud(struct kvm_vcpu
*vcpu
)
201 vcpu
->svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
|
202 SVM_EVTINJ_TYPE_EXEPT
|
206 static void inject_db(struct kvm_vcpu
*vcpu
)
208 vcpu
->svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
|
209 SVM_EVTINJ_TYPE_EXEPT
|
213 static int is_page_fault(uint32_t info
)
215 info
&= SVM_EVTINJ_VEC_MASK
| SVM_EVTINJ_TYPE_MASK
| SVM_EVTINJ_VALID
;
216 return info
== (PF_VECTOR
| SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_EXEPT
);
219 static int is_external_interrupt(u32 info
)
221 info
&= SVM_EVTINJ_TYPE_MASK
| SVM_EVTINJ_VALID
;
222 return info
== (SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
);
225 static void skip_emulated_instruction(struct kvm_vcpu
*vcpu
)
227 if (!vcpu
->svm
->next_rip
) {
228 printk(KERN_DEBUG
"%s: NOP\n", __FUNCTION__
);
231 if (vcpu
->svm
->next_rip
- vcpu
->svm
->vmcb
->save
.rip
> 15) {
232 printk(KERN_ERR
"%s: ip 0x%llx next 0x%llx\n",
234 vcpu
->svm
->vmcb
->save
.rip
,
235 vcpu
->svm
->next_rip
);
238 vcpu
->rip
= vcpu
->svm
->vmcb
->save
.rip
= vcpu
->svm
->next_rip
;
239 vcpu
->svm
->vmcb
->control
.int_state
&= ~SVM_INTERRUPT_SHADOW_MASK
;
241 vcpu
->interrupt_window_open
= 1;
244 static int has_svm(void)
246 uint32_t eax
, ebx
, ecx
, edx
;
248 if (boot_cpu_data
.x86_vendor
!= X86_VENDOR_AMD
) {
249 printk(KERN_INFO
"has_svm: not amd\n");
253 cpuid(0x80000000, &eax
, &ebx
, &ecx
, &edx
);
254 if (eax
< SVM_CPUID_FUNC
) {
255 printk(KERN_INFO
"has_svm: can't execute cpuid_8000000a\n");
259 cpuid(0x80000001, &eax
, &ebx
, &ecx
, &edx
);
260 if (!(ecx
& (1 << SVM_CPUID_FEATURE_SHIFT
))) {
261 printk(KERN_DEBUG
"has_svm: svm not available\n");
267 static void svm_hardware_disable(void *garbage
)
269 struct svm_cpu_data
*svm_data
270 = per_cpu(svm_data
, raw_smp_processor_id());
275 wrmsrl(MSR_VM_HSAVE_PA
, 0);
276 rdmsrl(MSR_EFER
, efer
);
277 wrmsrl(MSR_EFER
, efer
& ~MSR_EFER_SVME_MASK
);
278 per_cpu(svm_data
, raw_smp_processor_id()) = NULL
;
279 __free_page(svm_data
->save_area
);
284 static void svm_hardware_enable(void *garbage
)
287 struct svm_cpu_data
*svm_data
;
290 struct desc_ptr gdt_descr
;
292 struct Xgt_desc_struct gdt_descr
;
294 struct desc_struct
*gdt
;
295 int me
= raw_smp_processor_id();
298 printk(KERN_ERR
"svm_cpu_init: err EOPNOTSUPP on %d\n", me
);
301 svm_data
= per_cpu(svm_data
, me
);
304 printk(KERN_ERR
"svm_cpu_init: svm_data is NULL on %d\n",
309 svm_data
->asid_generation
= 1;
310 svm_data
->max_asid
= cpuid_ebx(SVM_CPUID_FUNC
) - 1;
311 svm_data
->next_asid
= svm_data
->max_asid
+ 1;
313 asm volatile ( "sgdt %0" : "=m"(gdt_descr
) );
314 gdt
= (struct desc_struct
*)gdt_descr
.address
;
315 svm_data
->tss_desc
= (struct kvm_ldttss_desc
*)(gdt
+ GDT_ENTRY_TSS
);
317 rdmsrl(MSR_EFER
, efer
);
318 wrmsrl(MSR_EFER
, efer
| MSR_EFER_SVME_MASK
);
320 wrmsrl(MSR_VM_HSAVE_PA
,
321 page_to_pfn(svm_data
->save_area
) << PAGE_SHIFT
);
324 static int svm_cpu_init(int cpu
)
326 struct svm_cpu_data
*svm_data
;
329 svm_data
= kzalloc(sizeof(struct svm_cpu_data
), GFP_KERNEL
);
333 svm_data
->save_area
= alloc_page(GFP_KERNEL
);
335 if (!svm_data
->save_area
)
338 per_cpu(svm_data
, cpu
) = svm_data
;
348 static int set_msr_interception(u32
*msrpm
, unsigned msr
,
353 for (i
= 0; i
< NUM_MSR_MAPS
; i
++) {
354 if (msr
>= msrpm_ranges
[i
] &&
355 msr
< msrpm_ranges
[i
] + MSRS_IN_RANGE
) {
356 u32 msr_offset
= (i
* MSRS_IN_RANGE
+ msr
-
357 msrpm_ranges
[i
]) * 2;
359 u32
*base
= msrpm
+ (msr_offset
/ 32);
360 u32 msr_shift
= msr_offset
% 32;
361 u32 mask
= ((write
) ? 0 : 2) | ((read
) ? 0 : 1);
362 *base
= (*base
& ~(0x3 << msr_shift
)) |
367 printk(KERN_DEBUG
"%s: not found 0x%x\n", __FUNCTION__
, msr
);
371 static __init
int svm_hardware_setup(void)
374 struct page
*iopm_pages
;
375 struct page
*msrpm_pages
;
379 kvm_emulator_want_group7_invlpg();
381 iopm_pages
= alloc_pages(GFP_KERNEL
, IOPM_ALLOC_ORDER
);
385 memset(page_address(iopm_pages
), 0xff,
386 PAGE_SIZE
* (1 << IOPM_ALLOC_ORDER
));
387 iopm_base
= page_to_pfn(iopm_pages
) << PAGE_SHIFT
;
390 msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
396 msrpm_va
= page_address(msrpm_pages
);
397 memset(msrpm_va
, 0xff, PAGE_SIZE
* (1 << MSRPM_ALLOC_ORDER
));
398 msrpm_base
= page_to_pfn(msrpm_pages
) << PAGE_SHIFT
;
401 set_msr_interception(msrpm_va
, MSR_GS_BASE
, 1, 1);
402 set_msr_interception(msrpm_va
, MSR_FS_BASE
, 1, 1);
403 set_msr_interception(msrpm_va
, MSR_KERNEL_GS_BASE
, 1, 1);
404 set_msr_interception(msrpm_va
, MSR_LSTAR
, 1, 1);
405 set_msr_interception(msrpm_va
, MSR_CSTAR
, 1, 1);
406 set_msr_interception(msrpm_va
, MSR_SYSCALL_MASK
, 1, 1);
408 set_msr_interception(msrpm_va
, MSR_K6_STAR
, 1, 1);
409 set_msr_interception(msrpm_va
, MSR_IA32_SYSENTER_CS
, 1, 1);
410 set_msr_interception(msrpm_va
, MSR_IA32_SYSENTER_ESP
, 1, 1);
411 set_msr_interception(msrpm_va
, MSR_IA32_SYSENTER_EIP
, 1, 1);
413 for_each_online_cpu(cpu
) {
414 r
= svm_cpu_init(cpu
);
421 __free_pages(msrpm_pages
, MSRPM_ALLOC_ORDER
);
424 __free_pages(iopm_pages
, IOPM_ALLOC_ORDER
);
429 static __exit
void svm_hardware_unsetup(void)
431 __free_pages(pfn_to_page(msrpm_base
>> PAGE_SHIFT
), MSRPM_ALLOC_ORDER
);
432 __free_pages(pfn_to_page(iopm_base
>> PAGE_SHIFT
), IOPM_ALLOC_ORDER
);
433 iopm_base
= msrpm_base
= 0;
436 static void init_seg(struct vmcb_seg
*seg
)
439 seg
->attrib
= SVM_SELECTOR_P_MASK
| SVM_SELECTOR_S_MASK
|
440 SVM_SELECTOR_WRITE_MASK
; /* Read/Write Data Segment */
445 static void init_sys_seg(struct vmcb_seg
*seg
, uint32_t type
)
448 seg
->attrib
= SVM_SELECTOR_P_MASK
| type
;
453 static int svm_vcpu_setup(struct kvm_vcpu
*vcpu
)
458 static void init_vmcb(struct vmcb
*vmcb
)
460 struct vmcb_control_area
*control
= &vmcb
->control
;
461 struct vmcb_save_area
*save
= &vmcb
->save
;
464 control
->intercept_cr_read
= INTERCEPT_CR0_MASK
|
468 control
->intercept_cr_write
= INTERCEPT_CR0_MASK
|
472 control
->intercept_dr_read
= INTERCEPT_DR0_MASK
|
477 control
->intercept_dr_write
= INTERCEPT_DR0_MASK
|
484 control
->intercept_exceptions
= 1 << PF_VECTOR
;
487 control
->intercept
= (1ULL << INTERCEPT_INTR
) |
488 (1ULL << INTERCEPT_NMI
) |
489 (1ULL << INTERCEPT_SMI
) |
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_INVLPGA
) |
504 (1ULL << INTERCEPT_IOIO_PROT
) |
505 (1ULL << INTERCEPT_MSR_PROT
) |
506 (1ULL << INTERCEPT_TASK_SWITCH
) |
507 (1ULL << INTERCEPT_SHUTDOWN
) |
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
) |
515 (1ULL << INTERCEPT_MONITOR
) |
516 (1ULL << INTERCEPT_MWAIT
);
518 control
->iopm_base_pa
= iopm_base
;
519 control
->msrpm_base_pa
= msrpm_base
;
521 control
->tsc_offset
= -tsc
;
522 control
->int_ctl
= V_INTR_MASKING_MASK
;
530 save
->cs
.selector
= 0xf000;
531 /* Executable/Readable Code Segment */
532 save
->cs
.attrib
= SVM_SELECTOR_READ_MASK
| SVM_SELECTOR_P_MASK
|
533 SVM_SELECTOR_S_MASK
| SVM_SELECTOR_CODE_MASK
;
534 save
->cs
.limit
= 0xffff;
536 * cs.base should really be 0xffff0000, but vmx can't handle that, so
537 * be consistent with it.
539 * Replace when we have real mode working for vmx.
541 save
->cs
.base
= 0xf0000;
543 save
->gdtr
.limit
= 0xffff;
544 save
->idtr
.limit
= 0xffff;
546 init_sys_seg(&save
->ldtr
, SEG_TYPE_LDT
);
547 init_sys_seg(&save
->tr
, SEG_TYPE_BUSY_TSS16
);
549 save
->efer
= MSR_EFER_SVME_MASK
;
551 save
->dr6
= 0xffff0ff0;
554 save
->rip
= 0x0000fff0;
557 * cr0 val on cpu init should be 0x60000010, we enable cpu
558 * cache by default. the orderly way is to enable cache in bios.
560 save
->cr0
= 0x00000010 | CR0_PG_MASK
| CR0_WP_MASK
;
561 save
->cr4
= CR4_PAE_MASK
;
565 static int svm_create_vcpu(struct kvm_vcpu
*vcpu
)
571 vcpu
->svm
= kzalloc(sizeof *vcpu
->svm
, GFP_KERNEL
);
574 page
= alloc_page(GFP_KERNEL
);
578 vcpu
->svm
->vmcb
= page_address(page
);
579 memset(vcpu
->svm
->vmcb
, 0, PAGE_SIZE
);
580 vcpu
->svm
->vmcb_pa
= page_to_pfn(page
) << PAGE_SHIFT
;
581 vcpu
->svm
->asid_generation
= 0;
582 memset(vcpu
->svm
->db_regs
, 0, sizeof(vcpu
->svm
->db_regs
));
583 init_vmcb(vcpu
->svm
->vmcb
);
586 vcpu
->apic_base
= 0xfee00000 |
587 /*for vcpu 0*/ MSR_IA32_APICBASE_BSP
|
588 MSR_IA32_APICBASE_ENABLE
;
598 static void svm_free_vcpu(struct kvm_vcpu
*vcpu
)
603 __free_page(pfn_to_page(vcpu
->svm
->vmcb_pa
>> PAGE_SHIFT
));
607 static void svm_vcpu_load(struct kvm_vcpu
*vcpu
)
612 static void svm_vcpu_put(struct kvm_vcpu
*vcpu
)
617 static void svm_vcpu_decache(struct kvm_vcpu
*vcpu
)
621 static void svm_cache_regs(struct kvm_vcpu
*vcpu
)
623 vcpu
->regs
[VCPU_REGS_RAX
] = vcpu
->svm
->vmcb
->save
.rax
;
624 vcpu
->regs
[VCPU_REGS_RSP
] = vcpu
->svm
->vmcb
->save
.rsp
;
625 vcpu
->rip
= vcpu
->svm
->vmcb
->save
.rip
;
628 static void svm_decache_regs(struct kvm_vcpu
*vcpu
)
630 vcpu
->svm
->vmcb
->save
.rax
= vcpu
->regs
[VCPU_REGS_RAX
];
631 vcpu
->svm
->vmcb
->save
.rsp
= vcpu
->regs
[VCPU_REGS_RSP
];
632 vcpu
->svm
->vmcb
->save
.rip
= vcpu
->rip
;
635 static unsigned long svm_get_rflags(struct kvm_vcpu
*vcpu
)
637 return vcpu
->svm
->vmcb
->save
.rflags
;
640 static void svm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
642 vcpu
->svm
->vmcb
->save
.rflags
= rflags
;
645 static struct vmcb_seg
*svm_seg(struct kvm_vcpu
*vcpu
, int seg
)
647 struct vmcb_save_area
*save
= &vcpu
->svm
->vmcb
->save
;
650 case VCPU_SREG_CS
: return &save
->cs
;
651 case VCPU_SREG_DS
: return &save
->ds
;
652 case VCPU_SREG_ES
: return &save
->es
;
653 case VCPU_SREG_FS
: return &save
->fs
;
654 case VCPU_SREG_GS
: return &save
->gs
;
655 case VCPU_SREG_SS
: return &save
->ss
;
656 case VCPU_SREG_TR
: return &save
->tr
;
657 case VCPU_SREG_LDTR
: return &save
->ldtr
;
663 static u64
svm_get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
665 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
670 static void svm_get_segment(struct kvm_vcpu
*vcpu
,
671 struct kvm_segment
*var
, int seg
)
673 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
676 var
->limit
= s
->limit
;
677 var
->selector
= s
->selector
;
678 var
->type
= s
->attrib
& SVM_SELECTOR_TYPE_MASK
;
679 var
->s
= (s
->attrib
>> SVM_SELECTOR_S_SHIFT
) & 1;
680 var
->dpl
= (s
->attrib
>> SVM_SELECTOR_DPL_SHIFT
) & 3;
681 var
->present
= (s
->attrib
>> SVM_SELECTOR_P_SHIFT
) & 1;
682 var
->avl
= (s
->attrib
>> SVM_SELECTOR_AVL_SHIFT
) & 1;
683 var
->l
= (s
->attrib
>> SVM_SELECTOR_L_SHIFT
) & 1;
684 var
->db
= (s
->attrib
>> SVM_SELECTOR_DB_SHIFT
) & 1;
685 var
->g
= (s
->attrib
>> SVM_SELECTOR_G_SHIFT
) & 1;
686 var
->unusable
= !var
->present
;
689 static void svm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
691 struct vmcb_seg
*s
= svm_seg(vcpu
, VCPU_SREG_CS
);
693 *db
= (s
->attrib
>> SVM_SELECTOR_DB_SHIFT
) & 1;
694 *l
= (s
->attrib
>> SVM_SELECTOR_L_SHIFT
) & 1;
697 static void svm_get_idt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
699 dt
->limit
= vcpu
->svm
->vmcb
->save
.idtr
.limit
;
700 dt
->base
= vcpu
->svm
->vmcb
->save
.idtr
.base
;
703 static void svm_set_idt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
705 vcpu
->svm
->vmcb
->save
.idtr
.limit
= dt
->limit
;
706 vcpu
->svm
->vmcb
->save
.idtr
.base
= dt
->base
;
709 static void svm_get_gdt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
711 dt
->limit
= vcpu
->svm
->vmcb
->save
.gdtr
.limit
;
712 dt
->base
= vcpu
->svm
->vmcb
->save
.gdtr
.base
;
715 static void svm_set_gdt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
717 vcpu
->svm
->vmcb
->save
.gdtr
.limit
= dt
->limit
;
718 vcpu
->svm
->vmcb
->save
.gdtr
.base
= dt
->base
;
721 static void svm_decache_cr0_cr4_guest_bits(struct kvm_vcpu
*vcpu
)
725 static void svm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
728 if (vcpu
->shadow_efer
& KVM_EFER_LME
) {
729 if (!is_paging(vcpu
) && (cr0
& CR0_PG_MASK
)) {
730 vcpu
->shadow_efer
|= KVM_EFER_LMA
;
731 vcpu
->svm
->vmcb
->save
.efer
|= KVM_EFER_LMA
| KVM_EFER_LME
;
734 if (is_paging(vcpu
) && !(cr0
& CR0_PG_MASK
) ) {
735 vcpu
->shadow_efer
&= ~KVM_EFER_LMA
;
736 vcpu
->svm
->vmcb
->save
.efer
&= ~(KVM_EFER_LMA
| KVM_EFER_LME
);
741 cr0
|= CR0_PG_MASK
| CR0_WP_MASK
;
742 cr0
&= ~(CR0_CD_MASK
| CR0_NW_MASK
);
743 vcpu
->svm
->vmcb
->save
.cr0
= cr0
;
746 static void svm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
749 vcpu
->svm
->vmcb
->save
.cr4
= cr4
| CR4_PAE_MASK
;
752 static void svm_set_segment(struct kvm_vcpu
*vcpu
,
753 struct kvm_segment
*var
, int seg
)
755 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
758 s
->limit
= var
->limit
;
759 s
->selector
= var
->selector
;
763 s
->attrib
= (var
->type
& SVM_SELECTOR_TYPE_MASK
);
764 s
->attrib
|= (var
->s
& 1) << SVM_SELECTOR_S_SHIFT
;
765 s
->attrib
|= (var
->dpl
& 3) << SVM_SELECTOR_DPL_SHIFT
;
766 s
->attrib
|= (var
->present
& 1) << SVM_SELECTOR_P_SHIFT
;
767 s
->attrib
|= (var
->avl
& 1) << SVM_SELECTOR_AVL_SHIFT
;
768 s
->attrib
|= (var
->l
& 1) << SVM_SELECTOR_L_SHIFT
;
769 s
->attrib
|= (var
->db
& 1) << SVM_SELECTOR_DB_SHIFT
;
770 s
->attrib
|= (var
->g
& 1) << SVM_SELECTOR_G_SHIFT
;
772 if (seg
== VCPU_SREG_CS
)
773 vcpu
->svm
->vmcb
->save
.cpl
774 = (vcpu
->svm
->vmcb
->save
.cs
.attrib
775 >> SVM_SELECTOR_DPL_SHIFT
) & 3;
781 vcpu->svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
782 vcpu->svm->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
786 static int svm_guest_debug(struct kvm_vcpu
*vcpu
, struct kvm_debug_guest
*dbg
)
791 static void load_host_msrs(struct kvm_vcpu
*vcpu
)
795 for ( i
= 0; i
< NR_HOST_SAVE_MSRS
; i
++)
796 wrmsrl(host_save_msrs
[i
], vcpu
->svm
->host_msrs
[i
]);
799 static void save_host_msrs(struct kvm_vcpu
*vcpu
)
803 for ( i
= 0; i
< NR_HOST_SAVE_MSRS
; i
++)
804 rdmsrl(host_save_msrs
[i
], vcpu
->svm
->host_msrs
[i
]);
807 static void new_asid(struct kvm_vcpu
*vcpu
, struct svm_cpu_data
*svm_data
)
809 if (svm_data
->next_asid
> svm_data
->max_asid
) {
810 ++svm_data
->asid_generation
;
811 svm_data
->next_asid
= 1;
812 vcpu
->svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ALL_ASID
;
815 vcpu
->cpu
= svm_data
->cpu
;
816 vcpu
->svm
->asid_generation
= svm_data
->asid_generation
;
817 vcpu
->svm
->vmcb
->control
.asid
= svm_data
->next_asid
++;
820 static void svm_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
822 invlpga(address
, vcpu
->svm
->vmcb
->control
.asid
); // is needed?
825 static unsigned long svm_get_dr(struct kvm_vcpu
*vcpu
, int dr
)
827 return vcpu
->svm
->db_regs
[dr
];
830 static void svm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long value
,
835 if (vcpu
->svm
->vmcb
->save
.dr7
& DR7_GD_MASK
) {
836 vcpu
->svm
->vmcb
->save
.dr7
&= ~DR7_GD_MASK
;
837 vcpu
->svm
->vmcb
->save
.dr6
|= DR6_BD_MASK
;
838 *exception
= DB_VECTOR
;
844 vcpu
->svm
->db_regs
[dr
] = value
;
847 if (vcpu
->cr4
& CR4_DE_MASK
) {
848 *exception
= UD_VECTOR
;
852 if (value
& ~((1ULL << 32) - 1)) {
853 *exception
= GP_VECTOR
;
856 vcpu
->svm
->vmcb
->save
.dr7
= value
;
860 printk(KERN_DEBUG
"%s: unexpected dr %u\n",
862 *exception
= UD_VECTOR
;
867 static int pf_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
869 u32 exit_int_info
= vcpu
->svm
->vmcb
->control
.exit_int_info
;
872 enum emulation_result er
;
875 if (is_external_interrupt(exit_int_info
))
876 push_irq(vcpu
, exit_int_info
& SVM_EVTINJ_VEC_MASK
);
878 spin_lock(&vcpu
->kvm
->lock
);
880 fault_address
= vcpu
->svm
->vmcb
->control
.exit_info_2
;
881 error_code
= vcpu
->svm
->vmcb
->control
.exit_info_1
;
882 r
= kvm_mmu_page_fault(vcpu
, fault_address
, error_code
);
884 spin_unlock(&vcpu
->kvm
->lock
);
888 spin_unlock(&vcpu
->kvm
->lock
);
891 er
= emulate_instruction(vcpu
, kvm_run
, fault_address
, error_code
);
892 spin_unlock(&vcpu
->kvm
->lock
);
897 case EMULATE_DO_MMIO
:
898 ++kvm_stat
.mmio_exits
;
899 kvm_run
->exit_reason
= KVM_EXIT_MMIO
;
902 vcpu_printf(vcpu
, "%s: emulate fail\n", __FUNCTION__
);
908 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
912 static int shutdown_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
915 * VMCB is undefined after a SHUTDOWN intercept
916 * so reinitialize it.
918 memset(vcpu
->svm
->vmcb
, 0, PAGE_SIZE
);
919 init_vmcb(vcpu
->svm
->vmcb
);
921 kvm_run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
925 static int io_get_override(struct kvm_vcpu
*vcpu
,
926 struct vmcb_seg
**seg
,
929 u8 inst
[MAX_INST_SIZE
];
934 rip
= vcpu
->svm
->vmcb
->save
.rip
;
935 ins_length
= vcpu
->svm
->next_rip
- rip
;
936 rip
+= vcpu
->svm
->vmcb
->save
.cs
.base
;
938 if (ins_length
> MAX_INST_SIZE
)
940 "%s: inst length err, cs base 0x%llx rip 0x%llx "
941 "next rip 0x%llx ins_length %u\n",
943 vcpu
->svm
->vmcb
->save
.cs
.base
,
944 vcpu
->svm
->vmcb
->save
.rip
,
945 vcpu
->svm
->vmcb
->control
.exit_info_2
,
948 if (kvm_read_guest(vcpu
, rip
, ins_length
, inst
) != ins_length
)
954 for (i
= 0; i
< ins_length
; i
++)
965 *seg
= &vcpu
->svm
->vmcb
->save
.cs
;
968 *seg
= &vcpu
->svm
->vmcb
->save
.ss
;
971 *seg
= &vcpu
->svm
->vmcb
->save
.ds
;
974 *seg
= &vcpu
->svm
->vmcb
->save
.es
;
977 *seg
= &vcpu
->svm
->vmcb
->save
.fs
;
980 *seg
= &vcpu
->svm
->vmcb
->save
.gs
;
985 printk(KERN_DEBUG
"%s: unexpected\n", __FUNCTION__
);
989 static unsigned long io_adress(struct kvm_vcpu
*vcpu
, int ins
, gva_t
*address
)
991 unsigned long addr_mask
;
993 struct vmcb_seg
*seg
;
995 struct vmcb_save_area
*save_area
= &vcpu
->svm
->vmcb
->save
;
996 u16 cs_attrib
= save_area
->cs
.attrib
;
997 unsigned addr_size
= get_addr_size(vcpu
);
999 if (!io_get_override(vcpu
, &seg
, &addr_override
))
1003 addr_size
= (addr_size
== 2) ? 4: (addr_size
>> 1);
1006 reg
= &vcpu
->regs
[VCPU_REGS_RDI
];
1007 seg
= &vcpu
->svm
->vmcb
->save
.es
;
1009 reg
= &vcpu
->regs
[VCPU_REGS_RSI
];
1010 seg
= (seg
) ? seg
: &vcpu
->svm
->vmcb
->save
.ds
;
1013 addr_mask
= ~0ULL >> (64 - (addr_size
* 8));
1015 if ((cs_attrib
& SVM_SELECTOR_L_MASK
) &&
1016 !(vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_VM
)) {
1017 *address
= (*reg
& addr_mask
);
1021 if (!(seg
->attrib
& SVM_SELECTOR_P_SHIFT
)) {
1022 svm_inject_gp(vcpu
, 0);
1026 *address
= (*reg
& addr_mask
) + seg
->base
;
1030 static int io_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1032 u32 io_info
= vcpu
->svm
->vmcb
->control
.exit_info_1
; //address size bug?
1033 int size
, down
, in
, string
, rep
;
1035 unsigned long count
;
1038 ++kvm_stat
.io_exits
;
1040 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->control
.exit_info_2
;
1042 in
= (io_info
& SVM_IOIO_TYPE_MASK
) != 0;
1043 port
= io_info
>> 16;
1044 size
= (io_info
& SVM_IOIO_SIZE_MASK
) >> SVM_IOIO_SIZE_SHIFT
;
1045 string
= (io_info
& SVM_IOIO_STR_MASK
) != 0;
1046 rep
= (io_info
& SVM_IOIO_REP_MASK
) != 0;
1048 down
= (vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_DF
) != 0;
1053 addr_mask
= io_adress(vcpu
, in
, &address
);
1055 printk(KERN_DEBUG
"%s: get io address failed\n",
1061 count
= vcpu
->regs
[VCPU_REGS_RCX
] & addr_mask
;
1063 return kvm_setup_pio(vcpu
, kvm_run
, in
, size
, count
, string
, down
,
1064 address
, rep
, port
);
1067 static int nop_on_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1072 static int halt_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1074 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 1;
1075 skip_emulated_instruction(vcpu
);
1076 if (vcpu
->irq_summary
)
1079 kvm_run
->exit_reason
= KVM_EXIT_HLT
;
1080 ++kvm_stat
.halt_exits
;
1084 static int vmmcall_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1086 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 3;
1087 skip_emulated_instruction(vcpu
);
1088 return kvm_hypercall(vcpu
, kvm_run
);
1091 static int invalid_op_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1097 static int task_switch_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1099 printk(KERN_DEBUG
"%s: task swiche is unsupported\n", __FUNCTION__
);
1100 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
1104 static int cpuid_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1106 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 2;
1107 kvm_emulate_cpuid(vcpu
);
1111 static int emulate_on_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1113 if (emulate_instruction(vcpu
, NULL
, 0, 0) != EMULATE_DONE
)
1114 printk(KERN_ERR
"%s: failed\n", __FUNCTION__
);
1118 static int svm_get_msr(struct kvm_vcpu
*vcpu
, unsigned ecx
, u64
*data
)
1121 case MSR_IA32_TIME_STAMP_COUNTER
: {
1125 *data
= vcpu
->svm
->vmcb
->control
.tsc_offset
+ tsc
;
1129 *data
= vcpu
->svm
->vmcb
->save
.star
;
1131 #ifdef CONFIG_X86_64
1133 *data
= vcpu
->svm
->vmcb
->save
.lstar
;
1136 *data
= vcpu
->svm
->vmcb
->save
.cstar
;
1138 case MSR_KERNEL_GS_BASE
:
1139 *data
= vcpu
->svm
->vmcb
->save
.kernel_gs_base
;
1141 case MSR_SYSCALL_MASK
:
1142 *data
= vcpu
->svm
->vmcb
->save
.sfmask
;
1145 case MSR_IA32_SYSENTER_CS
:
1146 *data
= vcpu
->svm
->vmcb
->save
.sysenter_cs
;
1148 case MSR_IA32_SYSENTER_EIP
:
1149 *data
= vcpu
->svm
->vmcb
->save
.sysenter_eip
;
1151 case MSR_IA32_SYSENTER_ESP
:
1152 *data
= vcpu
->svm
->vmcb
->save
.sysenter_esp
;
1155 return kvm_get_msr_common(vcpu
, ecx
, data
);
1160 static int rdmsr_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1162 u32 ecx
= vcpu
->regs
[VCPU_REGS_RCX
];
1165 if (svm_get_msr(vcpu
, ecx
, &data
))
1166 svm_inject_gp(vcpu
, 0);
1168 vcpu
->svm
->vmcb
->save
.rax
= data
& 0xffffffff;
1169 vcpu
->regs
[VCPU_REGS_RDX
] = data
>> 32;
1170 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 2;
1171 skip_emulated_instruction(vcpu
);
1176 static int svm_set_msr(struct kvm_vcpu
*vcpu
, unsigned ecx
, u64 data
)
1179 case MSR_IA32_TIME_STAMP_COUNTER
: {
1183 vcpu
->svm
->vmcb
->control
.tsc_offset
= data
- tsc
;
1187 vcpu
->svm
->vmcb
->save
.star
= data
;
1189 #ifdef CONFIG_X86_64
1191 vcpu
->svm
->vmcb
->save
.lstar
= data
;
1194 vcpu
->svm
->vmcb
->save
.cstar
= data
;
1196 case MSR_KERNEL_GS_BASE
:
1197 vcpu
->svm
->vmcb
->save
.kernel_gs_base
= data
;
1199 case MSR_SYSCALL_MASK
:
1200 vcpu
->svm
->vmcb
->save
.sfmask
= data
;
1203 case MSR_IA32_SYSENTER_CS
:
1204 vcpu
->svm
->vmcb
->save
.sysenter_cs
= data
;
1206 case MSR_IA32_SYSENTER_EIP
:
1207 vcpu
->svm
->vmcb
->save
.sysenter_eip
= data
;
1209 case MSR_IA32_SYSENTER_ESP
:
1210 vcpu
->svm
->vmcb
->save
.sysenter_esp
= data
;
1213 return kvm_set_msr_common(vcpu
, ecx
, data
);
1218 static int wrmsr_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1220 u32 ecx
= vcpu
->regs
[VCPU_REGS_RCX
];
1221 u64 data
= (vcpu
->svm
->vmcb
->save
.rax
& -1u)
1222 | ((u64
)(vcpu
->regs
[VCPU_REGS_RDX
] & -1u) << 32);
1223 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 2;
1224 if (svm_set_msr(vcpu
, ecx
, data
))
1225 svm_inject_gp(vcpu
, 0);
1227 skip_emulated_instruction(vcpu
);
1231 static int msr_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1233 if (vcpu
->svm
->vmcb
->control
.exit_info_1
)
1234 return wrmsr_interception(vcpu
, kvm_run
);
1236 return rdmsr_interception(vcpu
, kvm_run
);
1239 static int interrupt_window_interception(struct kvm_vcpu
*vcpu
,
1240 struct kvm_run
*kvm_run
)
1243 * If the user space waits to inject interrupts, exit as soon as
1246 if (kvm_run
->request_interrupt_window
&&
1247 !vcpu
->irq_summary
) {
1248 ++kvm_stat
.irq_window_exits
;
1249 kvm_run
->exit_reason
= KVM_EXIT_IRQ_WINDOW_OPEN
;
1256 static int (*svm_exit_handlers
[])(struct kvm_vcpu
*vcpu
,
1257 struct kvm_run
*kvm_run
) = {
1258 [SVM_EXIT_READ_CR0
] = emulate_on_interception
,
1259 [SVM_EXIT_READ_CR3
] = emulate_on_interception
,
1260 [SVM_EXIT_READ_CR4
] = emulate_on_interception
,
1262 [SVM_EXIT_WRITE_CR0
] = emulate_on_interception
,
1263 [SVM_EXIT_WRITE_CR3
] = emulate_on_interception
,
1264 [SVM_EXIT_WRITE_CR4
] = emulate_on_interception
,
1265 [SVM_EXIT_READ_DR0
] = emulate_on_interception
,
1266 [SVM_EXIT_READ_DR1
] = emulate_on_interception
,
1267 [SVM_EXIT_READ_DR2
] = emulate_on_interception
,
1268 [SVM_EXIT_READ_DR3
] = emulate_on_interception
,
1269 [SVM_EXIT_WRITE_DR0
] = emulate_on_interception
,
1270 [SVM_EXIT_WRITE_DR1
] = emulate_on_interception
,
1271 [SVM_EXIT_WRITE_DR2
] = emulate_on_interception
,
1272 [SVM_EXIT_WRITE_DR3
] = emulate_on_interception
,
1273 [SVM_EXIT_WRITE_DR5
] = emulate_on_interception
,
1274 [SVM_EXIT_WRITE_DR7
] = emulate_on_interception
,
1275 [SVM_EXIT_EXCP_BASE
+ PF_VECTOR
] = pf_interception
,
1276 [SVM_EXIT_INTR
] = nop_on_interception
,
1277 [SVM_EXIT_NMI
] = nop_on_interception
,
1278 [SVM_EXIT_SMI
] = nop_on_interception
,
1279 [SVM_EXIT_INIT
] = nop_on_interception
,
1280 [SVM_EXIT_VINTR
] = interrupt_window_interception
,
1281 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
1282 [SVM_EXIT_CPUID
] = cpuid_interception
,
1283 [SVM_EXIT_HLT
] = halt_interception
,
1284 [SVM_EXIT_INVLPG
] = emulate_on_interception
,
1285 [SVM_EXIT_INVLPGA
] = invalid_op_interception
,
1286 [SVM_EXIT_IOIO
] = io_interception
,
1287 [SVM_EXIT_MSR
] = msr_interception
,
1288 [SVM_EXIT_TASK_SWITCH
] = task_switch_interception
,
1289 [SVM_EXIT_SHUTDOWN
] = shutdown_interception
,
1290 [SVM_EXIT_VMRUN
] = invalid_op_interception
,
1291 [SVM_EXIT_VMMCALL
] = vmmcall_interception
,
1292 [SVM_EXIT_VMLOAD
] = invalid_op_interception
,
1293 [SVM_EXIT_VMSAVE
] = invalid_op_interception
,
1294 [SVM_EXIT_STGI
] = invalid_op_interception
,
1295 [SVM_EXIT_CLGI
] = invalid_op_interception
,
1296 [SVM_EXIT_SKINIT
] = invalid_op_interception
,
1297 [SVM_EXIT_MONITOR
] = invalid_op_interception
,
1298 [SVM_EXIT_MWAIT
] = invalid_op_interception
,
1302 static int handle_exit(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1304 u32 exit_code
= vcpu
->svm
->vmcb
->control
.exit_code
;
1306 if (is_external_interrupt(vcpu
->svm
->vmcb
->control
.exit_int_info
) &&
1307 exit_code
!= SVM_EXIT_EXCP_BASE
+ PF_VECTOR
)
1308 printk(KERN_ERR
"%s: unexpected exit_ini_info 0x%x "
1310 __FUNCTION__
, vcpu
->svm
->vmcb
->control
.exit_int_info
,
1313 if (exit_code
>= ARRAY_SIZE(svm_exit_handlers
)
1314 || svm_exit_handlers
[exit_code
] == 0) {
1315 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
1316 printk(KERN_ERR
"%s: 0x%x @ 0x%llx cr0 0x%lx rflags 0x%llx\n",
1319 vcpu
->svm
->vmcb
->save
.rip
,
1321 vcpu
->svm
->vmcb
->save
.rflags
);
1325 return svm_exit_handlers
[exit_code
](vcpu
, kvm_run
);
1328 static void reload_tss(struct kvm_vcpu
*vcpu
)
1330 int cpu
= raw_smp_processor_id();
1332 struct svm_cpu_data
*svm_data
= per_cpu(svm_data
, cpu
);
1333 svm_data
->tss_desc
->type
= 9; //available 32/64-bit TSS
1337 static void pre_svm_run(struct kvm_vcpu
*vcpu
)
1339 int cpu
= raw_smp_processor_id();
1341 struct svm_cpu_data
*svm_data
= per_cpu(svm_data
, cpu
);
1343 vcpu
->svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_DO_NOTHING
;
1344 if (vcpu
->cpu
!= cpu
||
1345 vcpu
->svm
->asid_generation
!= svm_data
->asid_generation
)
1346 new_asid(vcpu
, svm_data
);
1350 static inline void kvm_do_inject_irq(struct kvm_vcpu
*vcpu
)
1352 struct vmcb_control_area
*control
;
1354 control
= &vcpu
->svm
->vmcb
->control
;
1355 control
->int_vector
= pop_irq(vcpu
);
1356 control
->int_ctl
&= ~V_INTR_PRIO_MASK
;
1357 control
->int_ctl
|= V_IRQ_MASK
|
1358 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT
);
1361 static void kvm_reput_irq(struct kvm_vcpu
*vcpu
)
1363 struct vmcb_control_area
*control
= &vcpu
->svm
->vmcb
->control
;
1365 if (control
->int_ctl
& V_IRQ_MASK
) {
1366 control
->int_ctl
&= ~V_IRQ_MASK
;
1367 push_irq(vcpu
, control
->int_vector
);
1370 vcpu
->interrupt_window_open
=
1371 !(control
->int_state
& SVM_INTERRUPT_SHADOW_MASK
);
1374 static void do_interrupt_requests(struct kvm_vcpu
*vcpu
,
1375 struct kvm_run
*kvm_run
)
1377 struct vmcb_control_area
*control
= &vcpu
->svm
->vmcb
->control
;
1379 vcpu
->interrupt_window_open
=
1380 (!(control
->int_state
& SVM_INTERRUPT_SHADOW_MASK
) &&
1381 (vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_IF
));
1383 if (vcpu
->interrupt_window_open
&& vcpu
->irq_summary
)
1385 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1387 kvm_do_inject_irq(vcpu
);
1390 * Interrupts blocked. Wait for unblock.
1392 if (!vcpu
->interrupt_window_open
&&
1393 (vcpu
->irq_summary
|| kvm_run
->request_interrupt_window
)) {
1394 control
->intercept
|= 1ULL << INTERCEPT_VINTR
;
1396 control
->intercept
&= ~(1ULL << INTERCEPT_VINTR
);
1399 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
,
1400 struct kvm_run
*kvm_run
)
1402 kvm_run
->ready_for_interrupt_injection
= (vcpu
->interrupt_window_open
&&
1403 vcpu
->irq_summary
== 0);
1404 kvm_run
->if_flag
= (vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_IF
) != 0;
1405 kvm_run
->cr8
= vcpu
->cr8
;
1406 kvm_run
->apic_base
= vcpu
->apic_base
;
1410 * Check if userspace requested an interrupt window, and that the
1411 * interrupt window is open.
1413 * No need to exit to userspace if we already have an interrupt queued.
1415 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
,
1416 struct kvm_run
*kvm_run
)
1418 return (!vcpu
->irq_summary
&&
1419 kvm_run
->request_interrupt_window
&&
1420 vcpu
->interrupt_window_open
&&
1421 (vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_IF
));
1424 static void save_db_regs(unsigned long *db_regs
)
1426 asm volatile ("mov %%dr0, %0" : "=r"(db_regs
[0]));
1427 asm volatile ("mov %%dr1, %0" : "=r"(db_regs
[1]));
1428 asm volatile ("mov %%dr2, %0" : "=r"(db_regs
[2]));
1429 asm volatile ("mov %%dr3, %0" : "=r"(db_regs
[3]));
1432 static void load_db_regs(unsigned long *db_regs
)
1434 asm volatile ("mov %0, %%dr0" : : "r"(db_regs
[0]));
1435 asm volatile ("mov %0, %%dr1" : : "r"(db_regs
[1]));
1436 asm volatile ("mov %0, %%dr2" : : "r"(db_regs
[2]));
1437 asm volatile ("mov %0, %%dr3" : : "r"(db_regs
[3]));
1440 static int svm_vcpu_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1448 if (!vcpu
->mmio_read_completed
)
1449 do_interrupt_requests(vcpu
, kvm_run
);
1455 save_host_msrs(vcpu
);
1456 fs_selector
= read_fs();
1457 gs_selector
= read_gs();
1458 ldt_selector
= read_ldt();
1459 vcpu
->svm
->host_cr2
= kvm_read_cr2();
1460 vcpu
->svm
->host_dr6
= read_dr6();
1461 vcpu
->svm
->host_dr7
= read_dr7();
1462 vcpu
->svm
->vmcb
->save
.cr2
= vcpu
->cr2
;
1464 if (vcpu
->svm
->vmcb
->save
.dr7
& 0xff) {
1466 save_db_regs(vcpu
->svm
->host_db_regs
);
1467 load_db_regs(vcpu
->svm
->db_regs
);
1470 fx_save(vcpu
->host_fx_image
);
1471 fx_restore(vcpu
->guest_fx_image
);
1474 #ifdef CONFIG_X86_64
1475 "push %%rbx; push %%rcx; push %%rdx;"
1476 "push %%rsi; push %%rdi; push %%rbp;"
1477 "push %%r8; push %%r9; push %%r10; push %%r11;"
1478 "push %%r12; push %%r13; push %%r14; push %%r15;"
1480 "push %%ebx; push %%ecx; push %%edx;"
1481 "push %%esi; push %%edi; push %%ebp;"
1484 #ifdef CONFIG_X86_64
1485 "mov %c[rbx](%[vcpu]), %%rbx \n\t"
1486 "mov %c[rcx](%[vcpu]), %%rcx \n\t"
1487 "mov %c[rdx](%[vcpu]), %%rdx \n\t"
1488 "mov %c[rsi](%[vcpu]), %%rsi \n\t"
1489 "mov %c[rdi](%[vcpu]), %%rdi \n\t"
1490 "mov %c[rbp](%[vcpu]), %%rbp \n\t"
1491 "mov %c[r8](%[vcpu]), %%r8 \n\t"
1492 "mov %c[r9](%[vcpu]), %%r9 \n\t"
1493 "mov %c[r10](%[vcpu]), %%r10 \n\t"
1494 "mov %c[r11](%[vcpu]), %%r11 \n\t"
1495 "mov %c[r12](%[vcpu]), %%r12 \n\t"
1496 "mov %c[r13](%[vcpu]), %%r13 \n\t"
1497 "mov %c[r14](%[vcpu]), %%r14 \n\t"
1498 "mov %c[r15](%[vcpu]), %%r15 \n\t"
1500 "mov %c[rbx](%[vcpu]), %%ebx \n\t"
1501 "mov %c[rcx](%[vcpu]), %%ecx \n\t"
1502 "mov %c[rdx](%[vcpu]), %%edx \n\t"
1503 "mov %c[rsi](%[vcpu]), %%esi \n\t"
1504 "mov %c[rdi](%[vcpu]), %%edi \n\t"
1505 "mov %c[rbp](%[vcpu]), %%ebp \n\t"
1508 #ifdef CONFIG_X86_64
1509 /* Enter guest mode */
1511 "mov %c[svm](%[vcpu]), %%rax \n\t"
1512 "mov %c[vmcb](%%rax), %%rax \n\t"
1518 /* Enter guest mode */
1520 "mov %c[svm](%[vcpu]), %%eax \n\t"
1521 "mov %c[vmcb](%%eax), %%eax \n\t"
1528 /* Save guest registers, load host registers */
1529 #ifdef CONFIG_X86_64
1530 "mov %%rbx, %c[rbx](%[vcpu]) \n\t"
1531 "mov %%rcx, %c[rcx](%[vcpu]) \n\t"
1532 "mov %%rdx, %c[rdx](%[vcpu]) \n\t"
1533 "mov %%rsi, %c[rsi](%[vcpu]) \n\t"
1534 "mov %%rdi, %c[rdi](%[vcpu]) \n\t"
1535 "mov %%rbp, %c[rbp](%[vcpu]) \n\t"
1536 "mov %%r8, %c[r8](%[vcpu]) \n\t"
1537 "mov %%r9, %c[r9](%[vcpu]) \n\t"
1538 "mov %%r10, %c[r10](%[vcpu]) \n\t"
1539 "mov %%r11, %c[r11](%[vcpu]) \n\t"
1540 "mov %%r12, %c[r12](%[vcpu]) \n\t"
1541 "mov %%r13, %c[r13](%[vcpu]) \n\t"
1542 "mov %%r14, %c[r14](%[vcpu]) \n\t"
1543 "mov %%r15, %c[r15](%[vcpu]) \n\t"
1545 "pop %%r15; pop %%r14; pop %%r13; pop %%r12;"
1546 "pop %%r11; pop %%r10; pop %%r9; pop %%r8;"
1547 "pop %%rbp; pop %%rdi; pop %%rsi;"
1548 "pop %%rdx; pop %%rcx; pop %%rbx; \n\t"
1550 "mov %%ebx, %c[rbx](%[vcpu]) \n\t"
1551 "mov %%ecx, %c[rcx](%[vcpu]) \n\t"
1552 "mov %%edx, %c[rdx](%[vcpu]) \n\t"
1553 "mov %%esi, %c[rsi](%[vcpu]) \n\t"
1554 "mov %%edi, %c[rdi](%[vcpu]) \n\t"
1555 "mov %%ebp, %c[rbp](%[vcpu]) \n\t"
1557 "pop %%ebp; pop %%edi; pop %%esi;"
1558 "pop %%edx; pop %%ecx; pop %%ebx; \n\t"
1562 [svm
]"i"(offsetof(struct kvm_vcpu
, svm
)),
1563 [vmcb
]"i"(offsetof(struct vcpu_svm
, vmcb_pa
)),
1564 [rbx
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RBX
])),
1565 [rcx
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RCX
])),
1566 [rdx
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RDX
])),
1567 [rsi
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RSI
])),
1568 [rdi
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RDI
])),
1569 [rbp
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RBP
]))
1570 #ifdef CONFIG_X86_64
1571 ,[r8
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R8
])),
1572 [r9
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R9
])),
1573 [r10
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R10
])),
1574 [r11
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R11
])),
1575 [r12
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R12
])),
1576 [r13
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R13
])),
1577 [r14
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R14
])),
1578 [r15
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R15
]))
1582 fx_save(vcpu
->guest_fx_image
);
1583 fx_restore(vcpu
->host_fx_image
);
1585 if ((vcpu
->svm
->vmcb
->save
.dr7
& 0xff))
1586 load_db_regs(vcpu
->svm
->host_db_regs
);
1588 vcpu
->cr2
= vcpu
->svm
->vmcb
->save
.cr2
;
1590 write_dr6(vcpu
->svm
->host_dr6
);
1591 write_dr7(vcpu
->svm
->host_dr7
);
1592 kvm_write_cr2(vcpu
->svm
->host_cr2
);
1594 load_fs(fs_selector
);
1595 load_gs(gs_selector
);
1596 load_ldt(ldt_selector
);
1597 load_host_msrs(vcpu
);
1602 * Profile KVM exit RIPs:
1604 if (unlikely(prof_on
== KVM_PROFILING
))
1605 profile_hit(KVM_PROFILING
,
1606 (void *)(unsigned long)vcpu
->svm
->vmcb
->save
.rip
);
1610 kvm_reput_irq(vcpu
);
1612 vcpu
->svm
->next_rip
= 0;
1614 if (vcpu
->svm
->vmcb
->control
.exit_code
== SVM_EXIT_ERR
) {
1615 kvm_run
->exit_reason
= KVM_EXIT_FAIL_ENTRY
;
1616 kvm_run
->fail_entry
.hardware_entry_failure_reason
1617 = vcpu
->svm
->vmcb
->control
.exit_code
;
1618 post_kvm_run_save(vcpu
, kvm_run
);
1622 r
= handle_exit(vcpu
, kvm_run
);
1624 if (signal_pending(current
)) {
1625 ++kvm_stat
.signal_exits
;
1626 post_kvm_run_save(vcpu
, kvm_run
);
1627 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
1631 if (dm_request_for_irq_injection(vcpu
, kvm_run
)) {
1632 ++kvm_stat
.request_irq_exits
;
1633 post_kvm_run_save(vcpu
, kvm_run
);
1634 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
1640 post_kvm_run_save(vcpu
, kvm_run
);
1644 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
)
1646 force_new_asid(vcpu
);
1649 static void svm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
1651 vcpu
->svm
->vmcb
->save
.cr3
= root
;
1652 force_new_asid(vcpu
);
1655 static void svm_inject_page_fault(struct kvm_vcpu
*vcpu
,
1659 uint32_t exit_int_info
= vcpu
->svm
->vmcb
->control
.exit_int_info
;
1661 ++kvm_stat
.pf_guest
;
1663 if (is_page_fault(exit_int_info
)) {
1665 vcpu
->svm
->vmcb
->control
.event_inj_err
= 0;
1666 vcpu
->svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
|
1667 SVM_EVTINJ_VALID_ERR
|
1668 SVM_EVTINJ_TYPE_EXEPT
|
1673 vcpu
->svm
->vmcb
->save
.cr2
= addr
;
1674 vcpu
->svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
|
1675 SVM_EVTINJ_VALID_ERR
|
1676 SVM_EVTINJ_TYPE_EXEPT
|
1678 vcpu
->svm
->vmcb
->control
.event_inj_err
= err_code
;
1682 static int is_disabled(void)
1688 svm_patch_hypercall(struct kvm_vcpu
*vcpu
, unsigned char *hypercall
)
1691 * Patch in the VMMCALL instruction:
1693 hypercall
[0] = 0x0f;
1694 hypercall
[1] = 0x01;
1695 hypercall
[2] = 0xd9;
1696 hypercall
[3] = 0xc3;
1699 static struct kvm_arch_ops svm_arch_ops
= {
1700 .cpu_has_kvm_support
= has_svm
,
1701 .disabled_by_bios
= is_disabled
,
1702 .hardware_setup
= svm_hardware_setup
,
1703 .hardware_unsetup
= svm_hardware_unsetup
,
1704 .hardware_enable
= svm_hardware_enable
,
1705 .hardware_disable
= svm_hardware_disable
,
1707 .vcpu_create
= svm_create_vcpu
,
1708 .vcpu_free
= svm_free_vcpu
,
1710 .vcpu_load
= svm_vcpu_load
,
1711 .vcpu_put
= svm_vcpu_put
,
1712 .vcpu_decache
= svm_vcpu_decache
,
1714 .set_guest_debug
= svm_guest_debug
,
1715 .get_msr
= svm_get_msr
,
1716 .set_msr
= svm_set_msr
,
1717 .get_segment_base
= svm_get_segment_base
,
1718 .get_segment
= svm_get_segment
,
1719 .set_segment
= svm_set_segment
,
1720 .get_cs_db_l_bits
= svm_get_cs_db_l_bits
,
1721 .decache_cr0_cr4_guest_bits
= svm_decache_cr0_cr4_guest_bits
,
1722 .set_cr0
= svm_set_cr0
,
1723 .set_cr3
= svm_set_cr3
,
1724 .set_cr4
= svm_set_cr4
,
1725 .set_efer
= svm_set_efer
,
1726 .get_idt
= svm_get_idt
,
1727 .set_idt
= svm_set_idt
,
1728 .get_gdt
= svm_get_gdt
,
1729 .set_gdt
= svm_set_gdt
,
1730 .get_dr
= svm_get_dr
,
1731 .set_dr
= svm_set_dr
,
1732 .cache_regs
= svm_cache_regs
,
1733 .decache_regs
= svm_decache_regs
,
1734 .get_rflags
= svm_get_rflags
,
1735 .set_rflags
= svm_set_rflags
,
1737 .invlpg
= svm_invlpg
,
1738 .tlb_flush
= svm_flush_tlb
,
1739 .inject_page_fault
= svm_inject_page_fault
,
1741 .inject_gp
= svm_inject_gp
,
1743 .run
= svm_vcpu_run
,
1744 .skip_emulated_instruction
= skip_emulated_instruction
,
1745 .vcpu_setup
= svm_vcpu_setup
,
1746 .patch_hypercall
= svm_patch_hypercall
,
1749 static int __init
svm_init(void)
1751 return kvm_init_arch(&svm_arch_ops
, THIS_MODULE
);
1754 static void __exit
svm_exit(void)
1759 module_init(svm_init
)
1760 module_exit(svm_exit
)