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 int is_page_fault(uint32_t info
)
208 info
&= SVM_EVTINJ_VEC_MASK
| SVM_EVTINJ_TYPE_MASK
| SVM_EVTINJ_VALID
;
209 return info
== (PF_VECTOR
| SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_EXEPT
);
212 static int is_external_interrupt(u32 info
)
214 info
&= SVM_EVTINJ_TYPE_MASK
| SVM_EVTINJ_VALID
;
215 return info
== (SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
);
218 static void skip_emulated_instruction(struct kvm_vcpu
*vcpu
)
220 if (!vcpu
->svm
->next_rip
) {
221 printk(KERN_DEBUG
"%s: NOP\n", __FUNCTION__
);
224 if (vcpu
->svm
->next_rip
- vcpu
->svm
->vmcb
->save
.rip
> 15) {
225 printk(KERN_ERR
"%s: ip 0x%llx next 0x%llx\n",
227 vcpu
->svm
->vmcb
->save
.rip
,
228 vcpu
->svm
->next_rip
);
231 vcpu
->rip
= vcpu
->svm
->vmcb
->save
.rip
= vcpu
->svm
->next_rip
;
232 vcpu
->svm
->vmcb
->control
.int_state
&= ~SVM_INTERRUPT_SHADOW_MASK
;
234 vcpu
->interrupt_window_open
= 1;
237 static int has_svm(void)
239 uint32_t eax
, ebx
, ecx
, edx
;
241 if (boot_cpu_data
.x86_vendor
!= X86_VENDOR_AMD
) {
242 printk(KERN_INFO
"has_svm: not amd\n");
246 cpuid(0x80000000, &eax
, &ebx
, &ecx
, &edx
);
247 if (eax
< SVM_CPUID_FUNC
) {
248 printk(KERN_INFO
"has_svm: can't execute cpuid_8000000a\n");
252 cpuid(0x80000001, &eax
, &ebx
, &ecx
, &edx
);
253 if (!(ecx
& (1 << SVM_CPUID_FEATURE_SHIFT
))) {
254 printk(KERN_DEBUG
"has_svm: svm not available\n");
260 static void svm_hardware_disable(void *garbage
)
262 struct svm_cpu_data
*svm_data
263 = per_cpu(svm_data
, raw_smp_processor_id());
268 wrmsrl(MSR_VM_HSAVE_PA
, 0);
269 rdmsrl(MSR_EFER
, efer
);
270 wrmsrl(MSR_EFER
, efer
& ~MSR_EFER_SVME_MASK
);
271 per_cpu(svm_data
, raw_smp_processor_id()) = NULL
;
272 __free_page(svm_data
->save_area
);
277 static void svm_hardware_enable(void *garbage
)
280 struct svm_cpu_data
*svm_data
;
283 struct desc_ptr gdt_descr
;
285 struct Xgt_desc_struct gdt_descr
;
287 struct desc_struct
*gdt
;
288 int me
= raw_smp_processor_id();
291 printk(KERN_ERR
"svm_cpu_init: err EOPNOTSUPP on %d\n", me
);
294 svm_data
= per_cpu(svm_data
, me
);
297 printk(KERN_ERR
"svm_cpu_init: svm_data is NULL on %d\n",
302 svm_data
->asid_generation
= 1;
303 svm_data
->max_asid
= cpuid_ebx(SVM_CPUID_FUNC
) - 1;
304 svm_data
->next_asid
= svm_data
->max_asid
+ 1;
306 asm volatile ( "sgdt %0" : "=m"(gdt_descr
) );
307 gdt
= (struct desc_struct
*)gdt_descr
.address
;
308 svm_data
->tss_desc
= (struct kvm_ldttss_desc
*)(gdt
+ GDT_ENTRY_TSS
);
310 rdmsrl(MSR_EFER
, efer
);
311 wrmsrl(MSR_EFER
, efer
| MSR_EFER_SVME_MASK
);
313 wrmsrl(MSR_VM_HSAVE_PA
,
314 page_to_pfn(svm_data
->save_area
) << PAGE_SHIFT
);
317 static int svm_cpu_init(int cpu
)
319 struct svm_cpu_data
*svm_data
;
322 svm_data
= kzalloc(sizeof(struct svm_cpu_data
), GFP_KERNEL
);
326 svm_data
->save_area
= alloc_page(GFP_KERNEL
);
328 if (!svm_data
->save_area
)
331 per_cpu(svm_data
, cpu
) = svm_data
;
341 static int set_msr_interception(u32
*msrpm
, unsigned msr
,
346 for (i
= 0; i
< NUM_MSR_MAPS
; i
++) {
347 if (msr
>= msrpm_ranges
[i
] &&
348 msr
< msrpm_ranges
[i
] + MSRS_IN_RANGE
) {
349 u32 msr_offset
= (i
* MSRS_IN_RANGE
+ msr
-
350 msrpm_ranges
[i
]) * 2;
352 u32
*base
= msrpm
+ (msr_offset
/ 32);
353 u32 msr_shift
= msr_offset
% 32;
354 u32 mask
= ((write
) ? 0 : 2) | ((read
) ? 0 : 1);
355 *base
= (*base
& ~(0x3 << msr_shift
)) |
360 printk(KERN_DEBUG
"%s: not found 0x%x\n", __FUNCTION__
, msr
);
364 static __init
int svm_hardware_setup(void)
367 struct page
*iopm_pages
;
368 struct page
*msrpm_pages
;
372 kvm_emulator_want_group7_invlpg();
374 iopm_pages
= alloc_pages(GFP_KERNEL
, IOPM_ALLOC_ORDER
);
378 memset(page_address(iopm_pages
), 0xff,
379 PAGE_SIZE
* (1 << IOPM_ALLOC_ORDER
));
380 iopm_base
= page_to_pfn(iopm_pages
) << PAGE_SHIFT
;
383 msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
389 msrpm_va
= page_address(msrpm_pages
);
390 memset(msrpm_va
, 0xff, PAGE_SIZE
* (1 << MSRPM_ALLOC_ORDER
));
391 msrpm_base
= page_to_pfn(msrpm_pages
) << PAGE_SHIFT
;
394 set_msr_interception(msrpm_va
, MSR_GS_BASE
, 1, 1);
395 set_msr_interception(msrpm_va
, MSR_FS_BASE
, 1, 1);
396 set_msr_interception(msrpm_va
, MSR_KERNEL_GS_BASE
, 1, 1);
397 set_msr_interception(msrpm_va
, MSR_LSTAR
, 1, 1);
398 set_msr_interception(msrpm_va
, MSR_CSTAR
, 1, 1);
399 set_msr_interception(msrpm_va
, MSR_SYSCALL_MASK
, 1, 1);
401 set_msr_interception(msrpm_va
, MSR_K6_STAR
, 1, 1);
402 set_msr_interception(msrpm_va
, MSR_IA32_SYSENTER_CS
, 1, 1);
403 set_msr_interception(msrpm_va
, MSR_IA32_SYSENTER_ESP
, 1, 1);
404 set_msr_interception(msrpm_va
, MSR_IA32_SYSENTER_EIP
, 1, 1);
406 for_each_online_cpu(cpu
) {
407 r
= svm_cpu_init(cpu
);
414 __free_pages(msrpm_pages
, MSRPM_ALLOC_ORDER
);
417 __free_pages(iopm_pages
, IOPM_ALLOC_ORDER
);
422 static __exit
void svm_hardware_unsetup(void)
424 __free_pages(pfn_to_page(msrpm_base
>> PAGE_SHIFT
), MSRPM_ALLOC_ORDER
);
425 __free_pages(pfn_to_page(iopm_base
>> PAGE_SHIFT
), IOPM_ALLOC_ORDER
);
426 iopm_base
= msrpm_base
= 0;
429 static void init_seg(struct vmcb_seg
*seg
)
432 seg
->attrib
= SVM_SELECTOR_P_MASK
| SVM_SELECTOR_S_MASK
|
433 SVM_SELECTOR_WRITE_MASK
; /* Read/Write Data Segment */
438 static void init_sys_seg(struct vmcb_seg
*seg
, uint32_t type
)
441 seg
->attrib
= SVM_SELECTOR_P_MASK
| type
;
446 static int svm_vcpu_setup(struct kvm_vcpu
*vcpu
)
451 static void init_vmcb(struct vmcb
*vmcb
)
453 struct vmcb_control_area
*control
= &vmcb
->control
;
454 struct vmcb_save_area
*save
= &vmcb
->save
;
456 control
->intercept_cr_read
= INTERCEPT_CR0_MASK
|
460 control
->intercept_cr_write
= INTERCEPT_CR0_MASK
|
464 control
->intercept_dr_read
= INTERCEPT_DR0_MASK
|
469 control
->intercept_dr_write
= INTERCEPT_DR0_MASK
|
476 control
->intercept_exceptions
= 1 << PF_VECTOR
;
479 control
->intercept
= (1ULL << INTERCEPT_INTR
) |
480 (1ULL << INTERCEPT_NMI
) |
481 (1ULL << INTERCEPT_SMI
) |
483 * selective cr0 intercept bug?
484 * 0: 0f 22 d8 mov %eax,%cr3
485 * 3: 0f 20 c0 mov %cr0,%eax
486 * 6: 0d 00 00 00 80 or $0x80000000,%eax
487 * b: 0f 22 c0 mov %eax,%cr0
488 * set cr3 ->interception
489 * get cr0 ->interception
490 * set cr0 -> no interception
492 /* (1ULL << INTERCEPT_SELECTIVE_CR0) | */
493 (1ULL << INTERCEPT_CPUID
) |
494 (1ULL << INTERCEPT_HLT
) |
495 (1ULL << INTERCEPT_INVLPGA
) |
496 (1ULL << INTERCEPT_IOIO_PROT
) |
497 (1ULL << INTERCEPT_MSR_PROT
) |
498 (1ULL << INTERCEPT_TASK_SWITCH
) |
499 (1ULL << INTERCEPT_SHUTDOWN
) |
500 (1ULL << INTERCEPT_VMRUN
) |
501 (1ULL << INTERCEPT_VMMCALL
) |
502 (1ULL << INTERCEPT_VMLOAD
) |
503 (1ULL << INTERCEPT_VMSAVE
) |
504 (1ULL << INTERCEPT_STGI
) |
505 (1ULL << INTERCEPT_CLGI
) |
506 (1ULL << INTERCEPT_SKINIT
) |
507 (1ULL << INTERCEPT_MONITOR
) |
508 (1ULL << INTERCEPT_MWAIT
);
510 control
->iopm_base_pa
= iopm_base
;
511 control
->msrpm_base_pa
= msrpm_base
;
512 control
->tsc_offset
= 0;
513 control
->int_ctl
= V_INTR_MASKING_MASK
;
521 save
->cs
.selector
= 0xf000;
522 /* Executable/Readable Code Segment */
523 save
->cs
.attrib
= SVM_SELECTOR_READ_MASK
| SVM_SELECTOR_P_MASK
|
524 SVM_SELECTOR_S_MASK
| SVM_SELECTOR_CODE_MASK
;
525 save
->cs
.limit
= 0xffff;
527 * cs.base should really be 0xffff0000, but vmx can't handle that, so
528 * be consistent with it.
530 * Replace when we have real mode working for vmx.
532 save
->cs
.base
= 0xf0000;
534 save
->gdtr
.limit
= 0xffff;
535 save
->idtr
.limit
= 0xffff;
537 init_sys_seg(&save
->ldtr
, SEG_TYPE_LDT
);
538 init_sys_seg(&save
->tr
, SEG_TYPE_BUSY_TSS16
);
540 save
->efer
= MSR_EFER_SVME_MASK
;
542 save
->dr6
= 0xffff0ff0;
545 save
->rip
= 0x0000fff0;
548 * cr0 val on cpu init should be 0x60000010, we enable cpu
549 * cache by default. the orderly way is to enable cache in bios.
551 save
->cr0
= 0x00000010 | CR0_PG_MASK
| CR0_WP_MASK
;
552 save
->cr4
= CR4_PAE_MASK
;
556 static int svm_create_vcpu(struct kvm_vcpu
*vcpu
)
562 vcpu
->svm
= kzalloc(sizeof *vcpu
->svm
, GFP_KERNEL
);
565 page
= alloc_page(GFP_KERNEL
);
569 vcpu
->svm
->vmcb
= page_address(page
);
570 memset(vcpu
->svm
->vmcb
, 0, PAGE_SIZE
);
571 vcpu
->svm
->vmcb_pa
= page_to_pfn(page
) << PAGE_SHIFT
;
572 vcpu
->svm
->asid_generation
= 0;
573 memset(vcpu
->svm
->db_regs
, 0, sizeof(vcpu
->svm
->db_regs
));
574 init_vmcb(vcpu
->svm
->vmcb
);
577 vcpu
->apic_base
= 0xfee00000 |
578 /*for vcpu 0*/ MSR_IA32_APICBASE_BSP
|
579 MSR_IA32_APICBASE_ENABLE
;
589 static void svm_free_vcpu(struct kvm_vcpu
*vcpu
)
594 __free_page(pfn_to_page(vcpu
->svm
->vmcb_pa
>> PAGE_SHIFT
));
598 static void svm_vcpu_load(struct kvm_vcpu
*vcpu
)
603 if (unlikely(cpu
!= vcpu
->cpu
)) {
607 * Make sure that the guest sees a monotonically
611 delta
= vcpu
->host_tsc
- tsc_this
;
612 vcpu
->svm
->vmcb
->control
.tsc_offset
+= delta
;
617 static void svm_vcpu_put(struct kvm_vcpu
*vcpu
)
619 rdtscll(vcpu
->host_tsc
);
623 static void svm_vcpu_decache(struct kvm_vcpu
*vcpu
)
627 static void svm_cache_regs(struct kvm_vcpu
*vcpu
)
629 vcpu
->regs
[VCPU_REGS_RAX
] = vcpu
->svm
->vmcb
->save
.rax
;
630 vcpu
->regs
[VCPU_REGS_RSP
] = vcpu
->svm
->vmcb
->save
.rsp
;
631 vcpu
->rip
= vcpu
->svm
->vmcb
->save
.rip
;
634 static void svm_decache_regs(struct kvm_vcpu
*vcpu
)
636 vcpu
->svm
->vmcb
->save
.rax
= vcpu
->regs
[VCPU_REGS_RAX
];
637 vcpu
->svm
->vmcb
->save
.rsp
= vcpu
->regs
[VCPU_REGS_RSP
];
638 vcpu
->svm
->vmcb
->save
.rip
= vcpu
->rip
;
641 static unsigned long svm_get_rflags(struct kvm_vcpu
*vcpu
)
643 return vcpu
->svm
->vmcb
->save
.rflags
;
646 static void svm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
648 vcpu
->svm
->vmcb
->save
.rflags
= rflags
;
651 static struct vmcb_seg
*svm_seg(struct kvm_vcpu
*vcpu
, int seg
)
653 struct vmcb_save_area
*save
= &vcpu
->svm
->vmcb
->save
;
656 case VCPU_SREG_CS
: return &save
->cs
;
657 case VCPU_SREG_DS
: return &save
->ds
;
658 case VCPU_SREG_ES
: return &save
->es
;
659 case VCPU_SREG_FS
: return &save
->fs
;
660 case VCPU_SREG_GS
: return &save
->gs
;
661 case VCPU_SREG_SS
: return &save
->ss
;
662 case VCPU_SREG_TR
: return &save
->tr
;
663 case VCPU_SREG_LDTR
: return &save
->ldtr
;
669 static u64
svm_get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
671 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
676 static void svm_get_segment(struct kvm_vcpu
*vcpu
,
677 struct kvm_segment
*var
, int seg
)
679 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
682 var
->limit
= s
->limit
;
683 var
->selector
= s
->selector
;
684 var
->type
= s
->attrib
& SVM_SELECTOR_TYPE_MASK
;
685 var
->s
= (s
->attrib
>> SVM_SELECTOR_S_SHIFT
) & 1;
686 var
->dpl
= (s
->attrib
>> SVM_SELECTOR_DPL_SHIFT
) & 3;
687 var
->present
= (s
->attrib
>> SVM_SELECTOR_P_SHIFT
) & 1;
688 var
->avl
= (s
->attrib
>> SVM_SELECTOR_AVL_SHIFT
) & 1;
689 var
->l
= (s
->attrib
>> SVM_SELECTOR_L_SHIFT
) & 1;
690 var
->db
= (s
->attrib
>> SVM_SELECTOR_DB_SHIFT
) & 1;
691 var
->g
= (s
->attrib
>> SVM_SELECTOR_G_SHIFT
) & 1;
692 var
->unusable
= !var
->present
;
695 static void svm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
697 struct vmcb_seg
*s
= svm_seg(vcpu
, VCPU_SREG_CS
);
699 *db
= (s
->attrib
>> SVM_SELECTOR_DB_SHIFT
) & 1;
700 *l
= (s
->attrib
>> SVM_SELECTOR_L_SHIFT
) & 1;
703 static void svm_get_idt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
705 dt
->limit
= vcpu
->svm
->vmcb
->save
.idtr
.limit
;
706 dt
->base
= vcpu
->svm
->vmcb
->save
.idtr
.base
;
709 static void svm_set_idt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
711 vcpu
->svm
->vmcb
->save
.idtr
.limit
= dt
->limit
;
712 vcpu
->svm
->vmcb
->save
.idtr
.base
= dt
->base
;
715 static void svm_get_gdt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
717 dt
->limit
= vcpu
->svm
->vmcb
->save
.gdtr
.limit
;
718 dt
->base
= vcpu
->svm
->vmcb
->save
.gdtr
.base
;
721 static void svm_set_gdt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
723 vcpu
->svm
->vmcb
->save
.gdtr
.limit
= dt
->limit
;
724 vcpu
->svm
->vmcb
->save
.gdtr
.base
= dt
->base
;
727 static void svm_decache_cr0_cr4_guest_bits(struct kvm_vcpu
*vcpu
)
731 static void svm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
734 if (vcpu
->shadow_efer
& KVM_EFER_LME
) {
735 if (!is_paging(vcpu
) && (cr0
& CR0_PG_MASK
)) {
736 vcpu
->shadow_efer
|= KVM_EFER_LMA
;
737 vcpu
->svm
->vmcb
->save
.efer
|= KVM_EFER_LMA
| KVM_EFER_LME
;
740 if (is_paging(vcpu
) && !(cr0
& CR0_PG_MASK
) ) {
741 vcpu
->shadow_efer
&= ~KVM_EFER_LMA
;
742 vcpu
->svm
->vmcb
->save
.efer
&= ~(KVM_EFER_LMA
| KVM_EFER_LME
);
747 cr0
|= CR0_PG_MASK
| CR0_WP_MASK
;
748 cr0
&= ~(CR0_CD_MASK
| CR0_NW_MASK
);
749 vcpu
->svm
->vmcb
->save
.cr0
= cr0
;
752 static void svm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
755 vcpu
->svm
->vmcb
->save
.cr4
= cr4
| CR4_PAE_MASK
;
758 static void svm_set_segment(struct kvm_vcpu
*vcpu
,
759 struct kvm_segment
*var
, int seg
)
761 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
764 s
->limit
= var
->limit
;
765 s
->selector
= var
->selector
;
769 s
->attrib
= (var
->type
& SVM_SELECTOR_TYPE_MASK
);
770 s
->attrib
|= (var
->s
& 1) << SVM_SELECTOR_S_SHIFT
;
771 s
->attrib
|= (var
->dpl
& 3) << SVM_SELECTOR_DPL_SHIFT
;
772 s
->attrib
|= (var
->present
& 1) << SVM_SELECTOR_P_SHIFT
;
773 s
->attrib
|= (var
->avl
& 1) << SVM_SELECTOR_AVL_SHIFT
;
774 s
->attrib
|= (var
->l
& 1) << SVM_SELECTOR_L_SHIFT
;
775 s
->attrib
|= (var
->db
& 1) << SVM_SELECTOR_DB_SHIFT
;
776 s
->attrib
|= (var
->g
& 1) << SVM_SELECTOR_G_SHIFT
;
778 if (seg
== VCPU_SREG_CS
)
779 vcpu
->svm
->vmcb
->save
.cpl
780 = (vcpu
->svm
->vmcb
->save
.cs
.attrib
781 >> SVM_SELECTOR_DPL_SHIFT
) & 3;
787 vcpu->svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
788 vcpu->svm->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
792 static int svm_guest_debug(struct kvm_vcpu
*vcpu
, struct kvm_debug_guest
*dbg
)
797 static void load_host_msrs(struct kvm_vcpu
*vcpu
)
801 for ( i
= 0; i
< NR_HOST_SAVE_MSRS
; i
++)
802 wrmsrl(host_save_msrs
[i
], vcpu
->svm
->host_msrs
[i
]);
805 static void save_host_msrs(struct kvm_vcpu
*vcpu
)
809 for ( i
= 0; i
< NR_HOST_SAVE_MSRS
; i
++)
810 rdmsrl(host_save_msrs
[i
], vcpu
->svm
->host_msrs
[i
]);
813 static void new_asid(struct kvm_vcpu
*vcpu
, struct svm_cpu_data
*svm_data
)
815 if (svm_data
->next_asid
> svm_data
->max_asid
) {
816 ++svm_data
->asid_generation
;
817 svm_data
->next_asid
= 1;
818 vcpu
->svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ALL_ASID
;
821 vcpu
->cpu
= svm_data
->cpu
;
822 vcpu
->svm
->asid_generation
= svm_data
->asid_generation
;
823 vcpu
->svm
->vmcb
->control
.asid
= svm_data
->next_asid
++;
826 static void svm_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
828 invlpga(address
, vcpu
->svm
->vmcb
->control
.asid
); // is needed?
831 static unsigned long svm_get_dr(struct kvm_vcpu
*vcpu
, int dr
)
833 return vcpu
->svm
->db_regs
[dr
];
836 static void svm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long value
,
841 if (vcpu
->svm
->vmcb
->save
.dr7
& DR7_GD_MASK
) {
842 vcpu
->svm
->vmcb
->save
.dr7
&= ~DR7_GD_MASK
;
843 vcpu
->svm
->vmcb
->save
.dr6
|= DR6_BD_MASK
;
844 *exception
= DB_VECTOR
;
850 vcpu
->svm
->db_regs
[dr
] = value
;
853 if (vcpu
->cr4
& CR4_DE_MASK
) {
854 *exception
= UD_VECTOR
;
858 if (value
& ~((1ULL << 32) - 1)) {
859 *exception
= GP_VECTOR
;
862 vcpu
->svm
->vmcb
->save
.dr7
= value
;
866 printk(KERN_DEBUG
"%s: unexpected dr %u\n",
868 *exception
= UD_VECTOR
;
873 static int pf_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
875 u32 exit_int_info
= vcpu
->svm
->vmcb
->control
.exit_int_info
;
878 enum emulation_result er
;
881 if (is_external_interrupt(exit_int_info
))
882 push_irq(vcpu
, exit_int_info
& SVM_EVTINJ_VEC_MASK
);
884 spin_lock(&vcpu
->kvm
->lock
);
886 fault_address
= vcpu
->svm
->vmcb
->control
.exit_info_2
;
887 error_code
= vcpu
->svm
->vmcb
->control
.exit_info_1
;
888 r
= kvm_mmu_page_fault(vcpu
, fault_address
, error_code
);
890 spin_unlock(&vcpu
->kvm
->lock
);
894 spin_unlock(&vcpu
->kvm
->lock
);
897 er
= emulate_instruction(vcpu
, kvm_run
, fault_address
, error_code
);
898 spin_unlock(&vcpu
->kvm
->lock
);
903 case EMULATE_DO_MMIO
:
904 ++kvm_stat
.mmio_exits
;
905 kvm_run
->exit_reason
= KVM_EXIT_MMIO
;
908 vcpu_printf(vcpu
, "%s: emulate fail\n", __FUNCTION__
);
914 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
918 static int shutdown_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
921 * VMCB is undefined after a SHUTDOWN intercept
922 * so reinitialize it.
924 memset(vcpu
->svm
->vmcb
, 0, PAGE_SIZE
);
925 init_vmcb(vcpu
->svm
->vmcb
);
927 kvm_run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
931 static int io_get_override(struct kvm_vcpu
*vcpu
,
932 struct vmcb_seg
**seg
,
935 u8 inst
[MAX_INST_SIZE
];
940 rip
= vcpu
->svm
->vmcb
->save
.rip
;
941 ins_length
= vcpu
->svm
->next_rip
- rip
;
942 rip
+= vcpu
->svm
->vmcb
->save
.cs
.base
;
944 if (ins_length
> MAX_INST_SIZE
)
946 "%s: inst length err, cs base 0x%llx rip 0x%llx "
947 "next rip 0x%llx ins_length %u\n",
949 vcpu
->svm
->vmcb
->save
.cs
.base
,
950 vcpu
->svm
->vmcb
->save
.rip
,
951 vcpu
->svm
->vmcb
->control
.exit_info_2
,
954 if (kvm_read_guest(vcpu
, rip
, ins_length
, inst
) != ins_length
)
960 for (i
= 0; i
< ins_length
; i
++)
971 *seg
= &vcpu
->svm
->vmcb
->save
.cs
;
974 *seg
= &vcpu
->svm
->vmcb
->save
.ss
;
977 *seg
= &vcpu
->svm
->vmcb
->save
.ds
;
980 *seg
= &vcpu
->svm
->vmcb
->save
.es
;
983 *seg
= &vcpu
->svm
->vmcb
->save
.fs
;
986 *seg
= &vcpu
->svm
->vmcb
->save
.gs
;
991 printk(KERN_DEBUG
"%s: unexpected\n", __FUNCTION__
);
995 static unsigned long io_adress(struct kvm_vcpu
*vcpu
, int ins
, gva_t
*address
)
997 unsigned long addr_mask
;
999 struct vmcb_seg
*seg
;
1001 struct vmcb_save_area
*save_area
= &vcpu
->svm
->vmcb
->save
;
1002 u16 cs_attrib
= save_area
->cs
.attrib
;
1003 unsigned addr_size
= get_addr_size(vcpu
);
1005 if (!io_get_override(vcpu
, &seg
, &addr_override
))
1009 addr_size
= (addr_size
== 2) ? 4: (addr_size
>> 1);
1012 reg
= &vcpu
->regs
[VCPU_REGS_RDI
];
1013 seg
= &vcpu
->svm
->vmcb
->save
.es
;
1015 reg
= &vcpu
->regs
[VCPU_REGS_RSI
];
1016 seg
= (seg
) ? seg
: &vcpu
->svm
->vmcb
->save
.ds
;
1019 addr_mask
= ~0ULL >> (64 - (addr_size
* 8));
1021 if ((cs_attrib
& SVM_SELECTOR_L_MASK
) &&
1022 !(vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_VM
)) {
1023 *address
= (*reg
& addr_mask
);
1027 if (!(seg
->attrib
& SVM_SELECTOR_P_SHIFT
)) {
1028 svm_inject_gp(vcpu
, 0);
1032 *address
= (*reg
& addr_mask
) + seg
->base
;
1036 static int io_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1038 u32 io_info
= vcpu
->svm
->vmcb
->control
.exit_info_1
; //address size bug?
1039 int size
, down
, in
, string
, rep
;
1041 unsigned long count
;
1044 ++kvm_stat
.io_exits
;
1046 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->control
.exit_info_2
;
1048 in
= (io_info
& SVM_IOIO_TYPE_MASK
) != 0;
1049 port
= io_info
>> 16;
1050 size
= (io_info
& SVM_IOIO_SIZE_MASK
) >> SVM_IOIO_SIZE_SHIFT
;
1051 string
= (io_info
& SVM_IOIO_STR_MASK
) != 0;
1052 rep
= (io_info
& SVM_IOIO_REP_MASK
) != 0;
1054 down
= (vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_DF
) != 0;
1059 addr_mask
= io_adress(vcpu
, in
, &address
);
1061 printk(KERN_DEBUG
"%s: get io address failed\n",
1067 count
= vcpu
->regs
[VCPU_REGS_RCX
] & addr_mask
;
1069 return kvm_setup_pio(vcpu
, kvm_run
, in
, size
, count
, string
, down
,
1070 address
, rep
, port
);
1073 static int nop_on_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1078 static int halt_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1080 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 1;
1081 skip_emulated_instruction(vcpu
);
1082 if (vcpu
->irq_summary
)
1085 kvm_run
->exit_reason
= KVM_EXIT_HLT
;
1086 ++kvm_stat
.halt_exits
;
1090 static int vmmcall_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1092 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 3;
1093 skip_emulated_instruction(vcpu
);
1094 return kvm_hypercall(vcpu
, kvm_run
);
1097 static int invalid_op_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1103 static int task_switch_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1105 printk(KERN_DEBUG
"%s: task swiche is unsupported\n", __FUNCTION__
);
1106 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
1110 static int cpuid_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1112 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 2;
1113 kvm_emulate_cpuid(vcpu
);
1117 static int emulate_on_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1119 if (emulate_instruction(vcpu
, NULL
, 0, 0) != EMULATE_DONE
)
1120 printk(KERN_ERR
"%s: failed\n", __FUNCTION__
);
1124 static int svm_get_msr(struct kvm_vcpu
*vcpu
, unsigned ecx
, u64
*data
)
1127 case MSR_IA32_TIME_STAMP_COUNTER
: {
1131 *data
= vcpu
->svm
->vmcb
->control
.tsc_offset
+ tsc
;
1135 *data
= vcpu
->svm
->vmcb
->save
.star
;
1137 #ifdef CONFIG_X86_64
1139 *data
= vcpu
->svm
->vmcb
->save
.lstar
;
1142 *data
= vcpu
->svm
->vmcb
->save
.cstar
;
1144 case MSR_KERNEL_GS_BASE
:
1145 *data
= vcpu
->svm
->vmcb
->save
.kernel_gs_base
;
1147 case MSR_SYSCALL_MASK
:
1148 *data
= vcpu
->svm
->vmcb
->save
.sfmask
;
1151 case MSR_IA32_SYSENTER_CS
:
1152 *data
= vcpu
->svm
->vmcb
->save
.sysenter_cs
;
1154 case MSR_IA32_SYSENTER_EIP
:
1155 *data
= vcpu
->svm
->vmcb
->save
.sysenter_eip
;
1157 case MSR_IA32_SYSENTER_ESP
:
1158 *data
= vcpu
->svm
->vmcb
->save
.sysenter_esp
;
1161 return kvm_get_msr_common(vcpu
, ecx
, data
);
1166 static int rdmsr_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1168 u32 ecx
= vcpu
->regs
[VCPU_REGS_RCX
];
1171 if (svm_get_msr(vcpu
, ecx
, &data
))
1172 svm_inject_gp(vcpu
, 0);
1174 vcpu
->svm
->vmcb
->save
.rax
= data
& 0xffffffff;
1175 vcpu
->regs
[VCPU_REGS_RDX
] = data
>> 32;
1176 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 2;
1177 skip_emulated_instruction(vcpu
);
1182 static int svm_set_msr(struct kvm_vcpu
*vcpu
, unsigned ecx
, u64 data
)
1185 case MSR_IA32_TIME_STAMP_COUNTER
: {
1189 vcpu
->svm
->vmcb
->control
.tsc_offset
= data
- tsc
;
1193 vcpu
->svm
->vmcb
->save
.star
= data
;
1195 #ifdef CONFIG_X86_64
1197 vcpu
->svm
->vmcb
->save
.lstar
= data
;
1200 vcpu
->svm
->vmcb
->save
.cstar
= data
;
1202 case MSR_KERNEL_GS_BASE
:
1203 vcpu
->svm
->vmcb
->save
.kernel_gs_base
= data
;
1205 case MSR_SYSCALL_MASK
:
1206 vcpu
->svm
->vmcb
->save
.sfmask
= data
;
1209 case MSR_IA32_SYSENTER_CS
:
1210 vcpu
->svm
->vmcb
->save
.sysenter_cs
= data
;
1212 case MSR_IA32_SYSENTER_EIP
:
1213 vcpu
->svm
->vmcb
->save
.sysenter_eip
= data
;
1215 case MSR_IA32_SYSENTER_ESP
:
1216 vcpu
->svm
->vmcb
->save
.sysenter_esp
= data
;
1219 return kvm_set_msr_common(vcpu
, ecx
, data
);
1224 static int wrmsr_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1226 u32 ecx
= vcpu
->regs
[VCPU_REGS_RCX
];
1227 u64 data
= (vcpu
->svm
->vmcb
->save
.rax
& -1u)
1228 | ((u64
)(vcpu
->regs
[VCPU_REGS_RDX
] & -1u) << 32);
1229 vcpu
->svm
->next_rip
= vcpu
->svm
->vmcb
->save
.rip
+ 2;
1230 if (svm_set_msr(vcpu
, ecx
, data
))
1231 svm_inject_gp(vcpu
, 0);
1233 skip_emulated_instruction(vcpu
);
1237 static int msr_interception(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1239 if (vcpu
->svm
->vmcb
->control
.exit_info_1
)
1240 return wrmsr_interception(vcpu
, kvm_run
);
1242 return rdmsr_interception(vcpu
, kvm_run
);
1245 static int interrupt_window_interception(struct kvm_vcpu
*vcpu
,
1246 struct kvm_run
*kvm_run
)
1249 * If the user space waits to inject interrupts, exit as soon as
1252 if (kvm_run
->request_interrupt_window
&&
1253 !vcpu
->irq_summary
) {
1254 ++kvm_stat
.irq_window_exits
;
1255 kvm_run
->exit_reason
= KVM_EXIT_IRQ_WINDOW_OPEN
;
1262 static int (*svm_exit_handlers
[])(struct kvm_vcpu
*vcpu
,
1263 struct kvm_run
*kvm_run
) = {
1264 [SVM_EXIT_READ_CR0
] = emulate_on_interception
,
1265 [SVM_EXIT_READ_CR3
] = emulate_on_interception
,
1266 [SVM_EXIT_READ_CR4
] = emulate_on_interception
,
1268 [SVM_EXIT_WRITE_CR0
] = emulate_on_interception
,
1269 [SVM_EXIT_WRITE_CR3
] = emulate_on_interception
,
1270 [SVM_EXIT_WRITE_CR4
] = emulate_on_interception
,
1271 [SVM_EXIT_READ_DR0
] = emulate_on_interception
,
1272 [SVM_EXIT_READ_DR1
] = emulate_on_interception
,
1273 [SVM_EXIT_READ_DR2
] = emulate_on_interception
,
1274 [SVM_EXIT_READ_DR3
] = emulate_on_interception
,
1275 [SVM_EXIT_WRITE_DR0
] = emulate_on_interception
,
1276 [SVM_EXIT_WRITE_DR1
] = emulate_on_interception
,
1277 [SVM_EXIT_WRITE_DR2
] = emulate_on_interception
,
1278 [SVM_EXIT_WRITE_DR3
] = emulate_on_interception
,
1279 [SVM_EXIT_WRITE_DR5
] = emulate_on_interception
,
1280 [SVM_EXIT_WRITE_DR7
] = emulate_on_interception
,
1281 [SVM_EXIT_EXCP_BASE
+ PF_VECTOR
] = pf_interception
,
1282 [SVM_EXIT_INTR
] = nop_on_interception
,
1283 [SVM_EXIT_NMI
] = nop_on_interception
,
1284 [SVM_EXIT_SMI
] = nop_on_interception
,
1285 [SVM_EXIT_INIT
] = nop_on_interception
,
1286 [SVM_EXIT_VINTR
] = interrupt_window_interception
,
1287 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
1288 [SVM_EXIT_CPUID
] = cpuid_interception
,
1289 [SVM_EXIT_HLT
] = halt_interception
,
1290 [SVM_EXIT_INVLPG
] = emulate_on_interception
,
1291 [SVM_EXIT_INVLPGA
] = invalid_op_interception
,
1292 [SVM_EXIT_IOIO
] = io_interception
,
1293 [SVM_EXIT_MSR
] = msr_interception
,
1294 [SVM_EXIT_TASK_SWITCH
] = task_switch_interception
,
1295 [SVM_EXIT_SHUTDOWN
] = shutdown_interception
,
1296 [SVM_EXIT_VMRUN
] = invalid_op_interception
,
1297 [SVM_EXIT_VMMCALL
] = vmmcall_interception
,
1298 [SVM_EXIT_VMLOAD
] = invalid_op_interception
,
1299 [SVM_EXIT_VMSAVE
] = invalid_op_interception
,
1300 [SVM_EXIT_STGI
] = invalid_op_interception
,
1301 [SVM_EXIT_CLGI
] = invalid_op_interception
,
1302 [SVM_EXIT_SKINIT
] = invalid_op_interception
,
1303 [SVM_EXIT_MONITOR
] = invalid_op_interception
,
1304 [SVM_EXIT_MWAIT
] = invalid_op_interception
,
1308 static int handle_exit(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1310 u32 exit_code
= vcpu
->svm
->vmcb
->control
.exit_code
;
1312 if (is_external_interrupt(vcpu
->svm
->vmcb
->control
.exit_int_info
) &&
1313 exit_code
!= SVM_EXIT_EXCP_BASE
+ PF_VECTOR
)
1314 printk(KERN_ERR
"%s: unexpected exit_ini_info 0x%x "
1316 __FUNCTION__
, vcpu
->svm
->vmcb
->control
.exit_int_info
,
1319 if (exit_code
>= ARRAY_SIZE(svm_exit_handlers
)
1320 || svm_exit_handlers
[exit_code
] == 0) {
1321 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
1322 printk(KERN_ERR
"%s: 0x%x @ 0x%llx cr0 0x%lx rflags 0x%llx\n",
1325 vcpu
->svm
->vmcb
->save
.rip
,
1327 vcpu
->svm
->vmcb
->save
.rflags
);
1331 return svm_exit_handlers
[exit_code
](vcpu
, kvm_run
);
1334 static void reload_tss(struct kvm_vcpu
*vcpu
)
1336 int cpu
= raw_smp_processor_id();
1338 struct svm_cpu_data
*svm_data
= per_cpu(svm_data
, cpu
);
1339 svm_data
->tss_desc
->type
= 9; //available 32/64-bit TSS
1343 static void pre_svm_run(struct kvm_vcpu
*vcpu
)
1345 int cpu
= raw_smp_processor_id();
1347 struct svm_cpu_data
*svm_data
= per_cpu(svm_data
, cpu
);
1349 vcpu
->svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_DO_NOTHING
;
1350 if (vcpu
->cpu
!= cpu
||
1351 vcpu
->svm
->asid_generation
!= svm_data
->asid_generation
)
1352 new_asid(vcpu
, svm_data
);
1356 static inline void kvm_do_inject_irq(struct kvm_vcpu
*vcpu
)
1358 struct vmcb_control_area
*control
;
1360 control
= &vcpu
->svm
->vmcb
->control
;
1361 control
->int_vector
= pop_irq(vcpu
);
1362 control
->int_ctl
&= ~V_INTR_PRIO_MASK
;
1363 control
->int_ctl
|= V_IRQ_MASK
|
1364 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT
);
1367 static void kvm_reput_irq(struct kvm_vcpu
*vcpu
)
1369 struct vmcb_control_area
*control
= &vcpu
->svm
->vmcb
->control
;
1371 if (control
->int_ctl
& V_IRQ_MASK
) {
1372 control
->int_ctl
&= ~V_IRQ_MASK
;
1373 push_irq(vcpu
, control
->int_vector
);
1376 vcpu
->interrupt_window_open
=
1377 !(control
->int_state
& SVM_INTERRUPT_SHADOW_MASK
);
1380 static void do_interrupt_requests(struct kvm_vcpu
*vcpu
,
1381 struct kvm_run
*kvm_run
)
1383 struct vmcb_control_area
*control
= &vcpu
->svm
->vmcb
->control
;
1385 vcpu
->interrupt_window_open
=
1386 (!(control
->int_state
& SVM_INTERRUPT_SHADOW_MASK
) &&
1387 (vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_IF
));
1389 if (vcpu
->interrupt_window_open
&& vcpu
->irq_summary
)
1391 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1393 kvm_do_inject_irq(vcpu
);
1396 * Interrupts blocked. Wait for unblock.
1398 if (!vcpu
->interrupt_window_open
&&
1399 (vcpu
->irq_summary
|| kvm_run
->request_interrupt_window
)) {
1400 control
->intercept
|= 1ULL << INTERCEPT_VINTR
;
1402 control
->intercept
&= ~(1ULL << INTERCEPT_VINTR
);
1405 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
,
1406 struct kvm_run
*kvm_run
)
1408 kvm_run
->ready_for_interrupt_injection
= (vcpu
->interrupt_window_open
&&
1409 vcpu
->irq_summary
== 0);
1410 kvm_run
->if_flag
= (vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_IF
) != 0;
1411 kvm_run
->cr8
= vcpu
->cr8
;
1412 kvm_run
->apic_base
= vcpu
->apic_base
;
1416 * Check if userspace requested an interrupt window, and that the
1417 * interrupt window is open.
1419 * No need to exit to userspace if we already have an interrupt queued.
1421 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
,
1422 struct kvm_run
*kvm_run
)
1424 return (!vcpu
->irq_summary
&&
1425 kvm_run
->request_interrupt_window
&&
1426 vcpu
->interrupt_window_open
&&
1427 (vcpu
->svm
->vmcb
->save
.rflags
& X86_EFLAGS_IF
));
1430 static void save_db_regs(unsigned long *db_regs
)
1432 asm volatile ("mov %%dr0, %0" : "=r"(db_regs
[0]));
1433 asm volatile ("mov %%dr1, %0" : "=r"(db_regs
[1]));
1434 asm volatile ("mov %%dr2, %0" : "=r"(db_regs
[2]));
1435 asm volatile ("mov %%dr3, %0" : "=r"(db_regs
[3]));
1438 static void load_db_regs(unsigned long *db_regs
)
1440 asm volatile ("mov %0, %%dr0" : : "r"(db_regs
[0]));
1441 asm volatile ("mov %0, %%dr1" : : "r"(db_regs
[1]));
1442 asm volatile ("mov %0, %%dr2" : : "r"(db_regs
[2]));
1443 asm volatile ("mov %0, %%dr3" : : "r"(db_regs
[3]));
1446 static int svm_vcpu_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1454 if (!vcpu
->mmio_read_completed
)
1455 do_interrupt_requests(vcpu
, kvm_run
);
1461 save_host_msrs(vcpu
);
1462 fs_selector
= read_fs();
1463 gs_selector
= read_gs();
1464 ldt_selector
= read_ldt();
1465 vcpu
->svm
->host_cr2
= kvm_read_cr2();
1466 vcpu
->svm
->host_dr6
= read_dr6();
1467 vcpu
->svm
->host_dr7
= read_dr7();
1468 vcpu
->svm
->vmcb
->save
.cr2
= vcpu
->cr2
;
1470 if (vcpu
->svm
->vmcb
->save
.dr7
& 0xff) {
1472 save_db_regs(vcpu
->svm
->host_db_regs
);
1473 load_db_regs(vcpu
->svm
->db_regs
);
1476 fx_save(vcpu
->host_fx_image
);
1477 fx_restore(vcpu
->guest_fx_image
);
1480 #ifdef CONFIG_X86_64
1481 "push %%rbx; push %%rcx; push %%rdx;"
1482 "push %%rsi; push %%rdi; push %%rbp;"
1483 "push %%r8; push %%r9; push %%r10; push %%r11;"
1484 "push %%r12; push %%r13; push %%r14; push %%r15;"
1486 "push %%ebx; push %%ecx; push %%edx;"
1487 "push %%esi; push %%edi; push %%ebp;"
1490 #ifdef CONFIG_X86_64
1491 "mov %c[rbx](%[vcpu]), %%rbx \n\t"
1492 "mov %c[rcx](%[vcpu]), %%rcx \n\t"
1493 "mov %c[rdx](%[vcpu]), %%rdx \n\t"
1494 "mov %c[rsi](%[vcpu]), %%rsi \n\t"
1495 "mov %c[rdi](%[vcpu]), %%rdi \n\t"
1496 "mov %c[rbp](%[vcpu]), %%rbp \n\t"
1497 "mov %c[r8](%[vcpu]), %%r8 \n\t"
1498 "mov %c[r9](%[vcpu]), %%r9 \n\t"
1499 "mov %c[r10](%[vcpu]), %%r10 \n\t"
1500 "mov %c[r11](%[vcpu]), %%r11 \n\t"
1501 "mov %c[r12](%[vcpu]), %%r12 \n\t"
1502 "mov %c[r13](%[vcpu]), %%r13 \n\t"
1503 "mov %c[r14](%[vcpu]), %%r14 \n\t"
1504 "mov %c[r15](%[vcpu]), %%r15 \n\t"
1506 "mov %c[rbx](%[vcpu]), %%ebx \n\t"
1507 "mov %c[rcx](%[vcpu]), %%ecx \n\t"
1508 "mov %c[rdx](%[vcpu]), %%edx \n\t"
1509 "mov %c[rsi](%[vcpu]), %%esi \n\t"
1510 "mov %c[rdi](%[vcpu]), %%edi \n\t"
1511 "mov %c[rbp](%[vcpu]), %%ebp \n\t"
1514 #ifdef CONFIG_X86_64
1515 /* Enter guest mode */
1517 "mov %c[svm](%[vcpu]), %%rax \n\t"
1518 "mov %c[vmcb](%%rax), %%rax \n\t"
1524 /* Enter guest mode */
1526 "mov %c[svm](%[vcpu]), %%eax \n\t"
1527 "mov %c[vmcb](%%eax), %%eax \n\t"
1534 /* Save guest registers, load host registers */
1535 #ifdef CONFIG_X86_64
1536 "mov %%rbx, %c[rbx](%[vcpu]) \n\t"
1537 "mov %%rcx, %c[rcx](%[vcpu]) \n\t"
1538 "mov %%rdx, %c[rdx](%[vcpu]) \n\t"
1539 "mov %%rsi, %c[rsi](%[vcpu]) \n\t"
1540 "mov %%rdi, %c[rdi](%[vcpu]) \n\t"
1541 "mov %%rbp, %c[rbp](%[vcpu]) \n\t"
1542 "mov %%r8, %c[r8](%[vcpu]) \n\t"
1543 "mov %%r9, %c[r9](%[vcpu]) \n\t"
1544 "mov %%r10, %c[r10](%[vcpu]) \n\t"
1545 "mov %%r11, %c[r11](%[vcpu]) \n\t"
1546 "mov %%r12, %c[r12](%[vcpu]) \n\t"
1547 "mov %%r13, %c[r13](%[vcpu]) \n\t"
1548 "mov %%r14, %c[r14](%[vcpu]) \n\t"
1549 "mov %%r15, %c[r15](%[vcpu]) \n\t"
1551 "pop %%r15; pop %%r14; pop %%r13; pop %%r12;"
1552 "pop %%r11; pop %%r10; pop %%r9; pop %%r8;"
1553 "pop %%rbp; pop %%rdi; pop %%rsi;"
1554 "pop %%rdx; pop %%rcx; pop %%rbx; \n\t"
1556 "mov %%ebx, %c[rbx](%[vcpu]) \n\t"
1557 "mov %%ecx, %c[rcx](%[vcpu]) \n\t"
1558 "mov %%edx, %c[rdx](%[vcpu]) \n\t"
1559 "mov %%esi, %c[rsi](%[vcpu]) \n\t"
1560 "mov %%edi, %c[rdi](%[vcpu]) \n\t"
1561 "mov %%ebp, %c[rbp](%[vcpu]) \n\t"
1563 "pop %%ebp; pop %%edi; pop %%esi;"
1564 "pop %%edx; pop %%ecx; pop %%ebx; \n\t"
1568 [svm
]"i"(offsetof(struct kvm_vcpu
, svm
)),
1569 [vmcb
]"i"(offsetof(struct vcpu_svm
, vmcb_pa
)),
1570 [rbx
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RBX
])),
1571 [rcx
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RCX
])),
1572 [rdx
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RDX
])),
1573 [rsi
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RSI
])),
1574 [rdi
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RDI
])),
1575 [rbp
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_RBP
]))
1576 #ifdef CONFIG_X86_64
1577 ,[r8
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R8
])),
1578 [r9
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R9
])),
1579 [r10
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R10
])),
1580 [r11
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R11
])),
1581 [r12
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R12
])),
1582 [r13
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R13
])),
1583 [r14
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R14
])),
1584 [r15
]"i"(offsetof(struct kvm_vcpu
, regs
[VCPU_REGS_R15
]))
1588 fx_save(vcpu
->guest_fx_image
);
1589 fx_restore(vcpu
->host_fx_image
);
1591 if ((vcpu
->svm
->vmcb
->save
.dr7
& 0xff))
1592 load_db_regs(vcpu
->svm
->host_db_regs
);
1594 vcpu
->cr2
= vcpu
->svm
->vmcb
->save
.cr2
;
1596 write_dr6(vcpu
->svm
->host_dr6
);
1597 write_dr7(vcpu
->svm
->host_dr7
);
1598 kvm_write_cr2(vcpu
->svm
->host_cr2
);
1600 load_fs(fs_selector
);
1601 load_gs(gs_selector
);
1602 load_ldt(ldt_selector
);
1603 load_host_msrs(vcpu
);
1608 * Profile KVM exit RIPs:
1610 if (unlikely(prof_on
== KVM_PROFILING
))
1611 profile_hit(KVM_PROFILING
,
1612 (void *)(unsigned long)vcpu
->svm
->vmcb
->save
.rip
);
1616 kvm_reput_irq(vcpu
);
1618 vcpu
->svm
->next_rip
= 0;
1620 if (vcpu
->svm
->vmcb
->control
.exit_code
== SVM_EXIT_ERR
) {
1621 kvm_run
->exit_reason
= KVM_EXIT_FAIL_ENTRY
;
1622 kvm_run
->fail_entry
.hardware_entry_failure_reason
1623 = vcpu
->svm
->vmcb
->control
.exit_code
;
1624 post_kvm_run_save(vcpu
, kvm_run
);
1628 r
= handle_exit(vcpu
, kvm_run
);
1630 if (signal_pending(current
)) {
1631 ++kvm_stat
.signal_exits
;
1632 post_kvm_run_save(vcpu
, kvm_run
);
1633 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
1637 if (dm_request_for_irq_injection(vcpu
, kvm_run
)) {
1638 ++kvm_stat
.request_irq_exits
;
1639 post_kvm_run_save(vcpu
, kvm_run
);
1640 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
1646 post_kvm_run_save(vcpu
, kvm_run
);
1650 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
)
1652 force_new_asid(vcpu
);
1655 static void svm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
1657 vcpu
->svm
->vmcb
->save
.cr3
= root
;
1658 force_new_asid(vcpu
);
1661 static void svm_inject_page_fault(struct kvm_vcpu
*vcpu
,
1665 uint32_t exit_int_info
= vcpu
->svm
->vmcb
->control
.exit_int_info
;
1667 ++kvm_stat
.pf_guest
;
1669 if (is_page_fault(exit_int_info
)) {
1671 vcpu
->svm
->vmcb
->control
.event_inj_err
= 0;
1672 vcpu
->svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
|
1673 SVM_EVTINJ_VALID_ERR
|
1674 SVM_EVTINJ_TYPE_EXEPT
|
1679 vcpu
->svm
->vmcb
->save
.cr2
= addr
;
1680 vcpu
->svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
|
1681 SVM_EVTINJ_VALID_ERR
|
1682 SVM_EVTINJ_TYPE_EXEPT
|
1684 vcpu
->svm
->vmcb
->control
.event_inj_err
= err_code
;
1688 static int is_disabled(void)
1694 svm_patch_hypercall(struct kvm_vcpu
*vcpu
, unsigned char *hypercall
)
1697 * Patch in the VMMCALL instruction:
1699 hypercall
[0] = 0x0f;
1700 hypercall
[1] = 0x01;
1701 hypercall
[2] = 0xd9;
1702 hypercall
[3] = 0xc3;
1705 static struct kvm_arch_ops svm_arch_ops
= {
1706 .cpu_has_kvm_support
= has_svm
,
1707 .disabled_by_bios
= is_disabled
,
1708 .hardware_setup
= svm_hardware_setup
,
1709 .hardware_unsetup
= svm_hardware_unsetup
,
1710 .hardware_enable
= svm_hardware_enable
,
1711 .hardware_disable
= svm_hardware_disable
,
1713 .vcpu_create
= svm_create_vcpu
,
1714 .vcpu_free
= svm_free_vcpu
,
1716 .vcpu_load
= svm_vcpu_load
,
1717 .vcpu_put
= svm_vcpu_put
,
1718 .vcpu_decache
= svm_vcpu_decache
,
1720 .set_guest_debug
= svm_guest_debug
,
1721 .get_msr
= svm_get_msr
,
1722 .set_msr
= svm_set_msr
,
1723 .get_segment_base
= svm_get_segment_base
,
1724 .get_segment
= svm_get_segment
,
1725 .set_segment
= svm_set_segment
,
1726 .get_cs_db_l_bits
= svm_get_cs_db_l_bits
,
1727 .decache_cr0_cr4_guest_bits
= svm_decache_cr0_cr4_guest_bits
,
1728 .set_cr0
= svm_set_cr0
,
1729 .set_cr3
= svm_set_cr3
,
1730 .set_cr4
= svm_set_cr4
,
1731 .set_efer
= svm_set_efer
,
1732 .get_idt
= svm_get_idt
,
1733 .set_idt
= svm_set_idt
,
1734 .get_gdt
= svm_get_gdt
,
1735 .set_gdt
= svm_set_gdt
,
1736 .get_dr
= svm_get_dr
,
1737 .set_dr
= svm_set_dr
,
1738 .cache_regs
= svm_cache_regs
,
1739 .decache_regs
= svm_decache_regs
,
1740 .get_rflags
= svm_get_rflags
,
1741 .set_rflags
= svm_set_rflags
,
1743 .invlpg
= svm_invlpg
,
1744 .tlb_flush
= svm_flush_tlb
,
1745 .inject_page_fault
= svm_inject_page_fault
,
1747 .inject_gp
= svm_inject_gp
,
1749 .run
= svm_vcpu_run
,
1750 .skip_emulated_instruction
= skip_emulated_instruction
,
1751 .vcpu_setup
= svm_vcpu_setup
,
1752 .patch_hypercall
= svm_patch_hypercall
,
1755 static int __init
svm_init(void)
1757 return kvm_init_arch(&svm_arch_ops
, THIS_MODULE
);
1760 static void __exit
svm_exit(void)
1765 module_init(svm_init
)
1766 module_exit(svm_exit
)