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.
16 #include <linux/kvm_host.h>
20 #include "kvm_cache_regs.h"
23 #include <linux/module.h>
24 #include <linux/kernel.h>
25 #include <linux/vmalloc.h>
26 #include <linux/highmem.h>
27 #include <linux/sched.h>
28 #include <linux/ftrace_event.h>
32 #include <asm/virtext.h>
35 #define __ex(x) __kvm_handle_fault_on_reboot(x)
37 MODULE_AUTHOR("Qumranet");
38 MODULE_LICENSE("GPL");
40 #define IOPM_ALLOC_ORDER 2
41 #define MSRPM_ALLOC_ORDER 1
43 #define SEG_TYPE_LDT 2
44 #define SEG_TYPE_BUSY_TSS16 3
46 #define SVM_FEATURE_NPT (1 << 0)
47 #define SVM_FEATURE_LBRV (1 << 1)
48 #define SVM_FEATURE_SVML (1 << 2)
50 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
51 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
52 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
54 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
56 /* Turn on to get debugging output*/
57 /* #define NESTED_DEBUG */
60 #define nsvm_printk(fmt, args...) printk(KERN_INFO fmt, ## args)
62 #define nsvm_printk(fmt, args...) do {} while(0)
65 static const u32 host_save_user_msrs
[] = {
67 MSR_STAR
, MSR_LSTAR
, MSR_CSTAR
, MSR_SYSCALL_MASK
, MSR_KERNEL_GS_BASE
,
70 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
73 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
82 /* These are the merged vectors */
85 /* gpa pointers to the real vectors */
88 /* A VMEXIT is required but not yet emulated */
91 /* cache for intercepts of the guest */
92 u16 intercept_cr_read
;
93 u16 intercept_cr_write
;
94 u16 intercept_dr_read
;
95 u16 intercept_dr_write
;
96 u32 intercept_exceptions
;
102 struct kvm_vcpu vcpu
;
104 unsigned long vmcb_pa
;
105 struct svm_cpu_data
*svm_data
;
106 uint64_t asid_generation
;
107 uint64_t sysenter_esp
;
108 uint64_t sysenter_eip
;
112 u64 host_user_msrs
[NR_HOST_SAVE_USER_MSRS
];
117 struct nested_state nested
;
120 /* enable NPT for AMD64 and X86 with PAE */
121 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
122 static bool npt_enabled
= true;
124 static bool npt_enabled
= false;
128 module_param(npt
, int, S_IRUGO
);
130 static int nested
= 1;
131 module_param(nested
, int, S_IRUGO
);
133 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
);
134 static void svm_complete_interrupts(struct vcpu_svm
*svm
);
136 static int nested_svm_exit_handled(struct vcpu_svm
*svm
);
137 static int nested_svm_vmexit(struct vcpu_svm
*svm
);
138 static int nested_svm_check_exception(struct vcpu_svm
*svm
, unsigned nr
,
139 bool has_error_code
, u32 error_code
);
141 static inline struct vcpu_svm
*to_svm(struct kvm_vcpu
*vcpu
)
143 return container_of(vcpu
, struct vcpu_svm
, vcpu
);
146 static inline bool is_nested(struct vcpu_svm
*svm
)
148 return svm
->nested
.vmcb
;
151 static inline void enable_gif(struct vcpu_svm
*svm
)
153 svm
->vcpu
.arch
.hflags
|= HF_GIF_MASK
;
156 static inline void disable_gif(struct vcpu_svm
*svm
)
158 svm
->vcpu
.arch
.hflags
&= ~HF_GIF_MASK
;
161 static inline bool gif_set(struct vcpu_svm
*svm
)
163 return !!(svm
->vcpu
.arch
.hflags
& HF_GIF_MASK
);
166 static unsigned long iopm_base
;
168 struct kvm_ldttss_desc
{
171 unsigned base1
: 8, type
: 5, dpl
: 2, p
: 1;
172 unsigned limit1
: 4, zero0
: 3, g
: 1, base2
: 8;
175 } __attribute__((packed
));
177 struct svm_cpu_data
{
183 struct kvm_ldttss_desc
*tss_desc
;
185 struct page
*save_area
;
188 static DEFINE_PER_CPU(struct svm_cpu_data
*, svm_data
);
189 static uint32_t svm_features
;
191 struct svm_init_data
{
196 static u32 msrpm_ranges
[] = {0, 0xc0000000, 0xc0010000};
198 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
199 #define MSRS_RANGE_SIZE 2048
200 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
202 #define MAX_INST_SIZE 15
204 static inline u32
svm_has(u32 feat
)
206 return svm_features
& feat
;
209 static inline void clgi(void)
211 asm volatile (__ex(SVM_CLGI
));
214 static inline void stgi(void)
216 asm volatile (__ex(SVM_STGI
));
219 static inline void invlpga(unsigned long addr
, u32 asid
)
221 asm volatile (__ex(SVM_INVLPGA
) :: "a"(addr
), "c"(asid
));
224 static inline void force_new_asid(struct kvm_vcpu
*vcpu
)
226 to_svm(vcpu
)->asid_generation
--;
229 static inline void flush_guest_tlb(struct kvm_vcpu
*vcpu
)
231 force_new_asid(vcpu
);
234 static void svm_set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
236 if (!npt_enabled
&& !(efer
& EFER_LMA
))
239 to_svm(vcpu
)->vmcb
->save
.efer
= efer
| EFER_SVME
;
240 vcpu
->arch
.shadow_efer
= efer
;
243 static void svm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
,
244 bool has_error_code
, u32 error_code
)
246 struct vcpu_svm
*svm
= to_svm(vcpu
);
248 /* If we are within a nested VM we'd better #VMEXIT and let the
249 guest handle the exception */
250 if (nested_svm_check_exception(svm
, nr
, has_error_code
, error_code
))
253 svm
->vmcb
->control
.event_inj
= nr
255 | (has_error_code
? SVM_EVTINJ_VALID_ERR
: 0)
256 | SVM_EVTINJ_TYPE_EXEPT
;
257 svm
->vmcb
->control
.event_inj_err
= error_code
;
260 static int is_external_interrupt(u32 info
)
262 info
&= SVM_EVTINJ_TYPE_MASK
| SVM_EVTINJ_VALID
;
263 return info
== (SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
);
266 static u32
svm_get_interrupt_shadow(struct kvm_vcpu
*vcpu
, int mask
)
268 struct vcpu_svm
*svm
= to_svm(vcpu
);
271 if (svm
->vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
)
272 ret
|= X86_SHADOW_INT_STI
| X86_SHADOW_INT_MOV_SS
;
276 static void svm_set_interrupt_shadow(struct kvm_vcpu
*vcpu
, int mask
)
278 struct vcpu_svm
*svm
= to_svm(vcpu
);
281 svm
->vmcb
->control
.int_state
&= ~SVM_INTERRUPT_SHADOW_MASK
;
283 svm
->vmcb
->control
.int_state
|= SVM_INTERRUPT_SHADOW_MASK
;
287 static void skip_emulated_instruction(struct kvm_vcpu
*vcpu
)
289 struct vcpu_svm
*svm
= to_svm(vcpu
);
291 if (!svm
->next_rip
) {
292 if (emulate_instruction(vcpu
, 0, 0, EMULTYPE_SKIP
) !=
294 printk(KERN_DEBUG
"%s: NOP\n", __func__
);
297 if (svm
->next_rip
- kvm_rip_read(vcpu
) > MAX_INST_SIZE
)
298 printk(KERN_ERR
"%s: ip 0x%lx next 0x%llx\n",
299 __func__
, kvm_rip_read(vcpu
), svm
->next_rip
);
301 kvm_rip_write(vcpu
, svm
->next_rip
);
302 svm_set_interrupt_shadow(vcpu
, 0);
305 static int has_svm(void)
309 if (!cpu_has_svm(&msg
)) {
310 printk(KERN_INFO
"has_svm: %s\n", msg
);
317 static void svm_hardware_disable(void *garbage
)
322 static int svm_hardware_enable(void *garbage
)
325 struct svm_cpu_data
*svm_data
;
327 struct descriptor_table gdt_descr
;
328 struct desc_struct
*gdt
;
329 int me
= raw_smp_processor_id();
331 rdmsrl(MSR_EFER
, efer
);
332 if (efer
& EFER_SVME
)
336 printk(KERN_ERR
"svm_hardware_enable: err EOPNOTSUPP on %d\n",
340 svm_data
= per_cpu(svm_data
, me
);
343 printk(KERN_ERR
"svm_hardware_enable: svm_data is NULL on %d\n",
348 svm_data
->asid_generation
= 1;
349 svm_data
->max_asid
= cpuid_ebx(SVM_CPUID_FUNC
) - 1;
350 svm_data
->next_asid
= svm_data
->max_asid
+ 1;
352 kvm_get_gdt(&gdt_descr
);
353 gdt
= (struct desc_struct
*)gdt_descr
.base
;
354 svm_data
->tss_desc
= (struct kvm_ldttss_desc
*)(gdt
+ GDT_ENTRY_TSS
);
356 wrmsrl(MSR_EFER
, efer
| EFER_SVME
);
358 wrmsrl(MSR_VM_HSAVE_PA
,
359 page_to_pfn(svm_data
->save_area
) << PAGE_SHIFT
);
364 static void svm_cpu_uninit(int cpu
)
366 struct svm_cpu_data
*svm_data
367 = per_cpu(svm_data
, raw_smp_processor_id());
372 per_cpu(svm_data
, raw_smp_processor_id()) = NULL
;
373 __free_page(svm_data
->save_area
);
377 static int svm_cpu_init(int cpu
)
379 struct svm_cpu_data
*svm_data
;
382 svm_data
= kzalloc(sizeof(struct svm_cpu_data
), GFP_KERNEL
);
386 svm_data
->save_area
= alloc_page(GFP_KERNEL
);
388 if (!svm_data
->save_area
)
391 per_cpu(svm_data
, cpu
) = svm_data
;
401 static void set_msr_interception(u32
*msrpm
, unsigned msr
,
406 for (i
= 0; i
< NUM_MSR_MAPS
; i
++) {
407 if (msr
>= msrpm_ranges
[i
] &&
408 msr
< msrpm_ranges
[i
] + MSRS_IN_RANGE
) {
409 u32 msr_offset
= (i
* MSRS_IN_RANGE
+ msr
-
410 msrpm_ranges
[i
]) * 2;
412 u32
*base
= msrpm
+ (msr_offset
/ 32);
413 u32 msr_shift
= msr_offset
% 32;
414 u32 mask
= ((write
) ? 0 : 2) | ((read
) ? 0 : 1);
415 *base
= (*base
& ~(0x3 << msr_shift
)) |
423 static void svm_vcpu_init_msrpm(u32
*msrpm
)
425 memset(msrpm
, 0xff, PAGE_SIZE
* (1 << MSRPM_ALLOC_ORDER
));
428 set_msr_interception(msrpm
, MSR_GS_BASE
, 1, 1);
429 set_msr_interception(msrpm
, MSR_FS_BASE
, 1, 1);
430 set_msr_interception(msrpm
, MSR_KERNEL_GS_BASE
, 1, 1);
431 set_msr_interception(msrpm
, MSR_LSTAR
, 1, 1);
432 set_msr_interception(msrpm
, MSR_CSTAR
, 1, 1);
433 set_msr_interception(msrpm
, MSR_SYSCALL_MASK
, 1, 1);
435 set_msr_interception(msrpm
, MSR_K6_STAR
, 1, 1);
436 set_msr_interception(msrpm
, MSR_IA32_SYSENTER_CS
, 1, 1);
439 static void svm_enable_lbrv(struct vcpu_svm
*svm
)
441 u32
*msrpm
= svm
->msrpm
;
443 svm
->vmcb
->control
.lbr_ctl
= 1;
444 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHFROMIP
, 1, 1);
445 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHTOIP
, 1, 1);
446 set_msr_interception(msrpm
, MSR_IA32_LASTINTFROMIP
, 1, 1);
447 set_msr_interception(msrpm
, MSR_IA32_LASTINTTOIP
, 1, 1);
450 static void svm_disable_lbrv(struct vcpu_svm
*svm
)
452 u32
*msrpm
= svm
->msrpm
;
454 svm
->vmcb
->control
.lbr_ctl
= 0;
455 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHFROMIP
, 0, 0);
456 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHTOIP
, 0, 0);
457 set_msr_interception(msrpm
, MSR_IA32_LASTINTFROMIP
, 0, 0);
458 set_msr_interception(msrpm
, MSR_IA32_LASTINTTOIP
, 0, 0);
461 static __init
int svm_hardware_setup(void)
464 struct page
*iopm_pages
;
468 iopm_pages
= alloc_pages(GFP_KERNEL
, IOPM_ALLOC_ORDER
);
473 iopm_va
= page_address(iopm_pages
);
474 memset(iopm_va
, 0xff, PAGE_SIZE
* (1 << IOPM_ALLOC_ORDER
));
475 iopm_base
= page_to_pfn(iopm_pages
) << PAGE_SHIFT
;
477 if (boot_cpu_has(X86_FEATURE_NX
))
478 kvm_enable_efer_bits(EFER_NX
);
480 if (boot_cpu_has(X86_FEATURE_FXSR_OPT
))
481 kvm_enable_efer_bits(EFER_FFXSR
);
484 printk(KERN_INFO
"kvm: Nested Virtualization enabled\n");
485 kvm_enable_efer_bits(EFER_SVME
);
488 for_each_possible_cpu(cpu
) {
489 r
= svm_cpu_init(cpu
);
494 svm_features
= cpuid_edx(SVM_CPUID_FUNC
);
496 if (!svm_has(SVM_FEATURE_NPT
))
499 if (npt_enabled
&& !npt
) {
500 printk(KERN_INFO
"kvm: Nested Paging disabled\n");
505 printk(KERN_INFO
"kvm: Nested Paging enabled\n");
513 __free_pages(iopm_pages
, IOPM_ALLOC_ORDER
);
518 static __exit
void svm_hardware_unsetup(void)
522 for_each_possible_cpu(cpu
)
525 __free_pages(pfn_to_page(iopm_base
>> PAGE_SHIFT
), IOPM_ALLOC_ORDER
);
529 static void init_seg(struct vmcb_seg
*seg
)
532 seg
->attrib
= SVM_SELECTOR_P_MASK
| SVM_SELECTOR_S_MASK
|
533 SVM_SELECTOR_WRITE_MASK
; /* Read/Write Data Segment */
538 static void init_sys_seg(struct vmcb_seg
*seg
, uint32_t type
)
541 seg
->attrib
= SVM_SELECTOR_P_MASK
| type
;
546 static void init_vmcb(struct vcpu_svm
*svm
)
548 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
549 struct vmcb_save_area
*save
= &svm
->vmcb
->save
;
551 control
->intercept_cr_read
= INTERCEPT_CR0_MASK
|
555 control
->intercept_cr_write
= INTERCEPT_CR0_MASK
|
560 control
->intercept_dr_read
= INTERCEPT_DR0_MASK
|
565 control
->intercept_dr_write
= INTERCEPT_DR0_MASK
|
572 control
->intercept_exceptions
= (1 << PF_VECTOR
) |
577 control
->intercept
= (1ULL << INTERCEPT_INTR
) |
578 (1ULL << INTERCEPT_NMI
) |
579 (1ULL << INTERCEPT_SMI
) |
580 (1ULL << INTERCEPT_CPUID
) |
581 (1ULL << INTERCEPT_INVD
) |
582 (1ULL << INTERCEPT_HLT
) |
583 (1ULL << INTERCEPT_INVLPG
) |
584 (1ULL << INTERCEPT_INVLPGA
) |
585 (1ULL << INTERCEPT_IOIO_PROT
) |
586 (1ULL << INTERCEPT_MSR_PROT
) |
587 (1ULL << INTERCEPT_TASK_SWITCH
) |
588 (1ULL << INTERCEPT_SHUTDOWN
) |
589 (1ULL << INTERCEPT_VMRUN
) |
590 (1ULL << INTERCEPT_VMMCALL
) |
591 (1ULL << INTERCEPT_VMLOAD
) |
592 (1ULL << INTERCEPT_VMSAVE
) |
593 (1ULL << INTERCEPT_STGI
) |
594 (1ULL << INTERCEPT_CLGI
) |
595 (1ULL << INTERCEPT_SKINIT
) |
596 (1ULL << INTERCEPT_WBINVD
) |
597 (1ULL << INTERCEPT_MONITOR
) |
598 (1ULL << INTERCEPT_MWAIT
);
600 control
->iopm_base_pa
= iopm_base
;
601 control
->msrpm_base_pa
= __pa(svm
->msrpm
);
602 control
->tsc_offset
= 0;
603 control
->int_ctl
= V_INTR_MASKING_MASK
;
611 save
->cs
.selector
= 0xf000;
612 /* Executable/Readable Code Segment */
613 save
->cs
.attrib
= SVM_SELECTOR_READ_MASK
| SVM_SELECTOR_P_MASK
|
614 SVM_SELECTOR_S_MASK
| SVM_SELECTOR_CODE_MASK
;
615 save
->cs
.limit
= 0xffff;
617 * cs.base should really be 0xffff0000, but vmx can't handle that, so
618 * be consistent with it.
620 * Replace when we have real mode working for vmx.
622 save
->cs
.base
= 0xf0000;
624 save
->gdtr
.limit
= 0xffff;
625 save
->idtr
.limit
= 0xffff;
627 init_sys_seg(&save
->ldtr
, SEG_TYPE_LDT
);
628 init_sys_seg(&save
->tr
, SEG_TYPE_BUSY_TSS16
);
630 save
->efer
= EFER_SVME
;
631 save
->dr6
= 0xffff0ff0;
634 save
->rip
= 0x0000fff0;
635 svm
->vcpu
.arch
.regs
[VCPU_REGS_RIP
] = save
->rip
;
638 * cr0 val on cpu init should be 0x60000010, we enable cpu
639 * cache by default. the orderly way is to enable cache in bios.
641 save
->cr0
= 0x00000010 | X86_CR0_PG
| X86_CR0_WP
;
642 save
->cr4
= X86_CR4_PAE
;
646 /* Setup VMCB for Nested Paging */
647 control
->nested_ctl
= 1;
648 control
->intercept
&= ~((1ULL << INTERCEPT_TASK_SWITCH
) |
649 (1ULL << INTERCEPT_INVLPG
));
650 control
->intercept_exceptions
&= ~(1 << PF_VECTOR
);
651 control
->intercept_cr_read
&= ~(INTERCEPT_CR0_MASK
|
653 control
->intercept_cr_write
&= ~(INTERCEPT_CR0_MASK
|
655 save
->g_pat
= 0x0007040600070406ULL
;
656 /* enable caching because the QEMU Bios doesn't enable it */
657 save
->cr0
= X86_CR0_ET
;
661 force_new_asid(&svm
->vcpu
);
663 svm
->nested
.vmcb
= 0;
664 svm
->vcpu
.arch
.hflags
= 0;
669 static int svm_vcpu_reset(struct kvm_vcpu
*vcpu
)
671 struct vcpu_svm
*svm
= to_svm(vcpu
);
675 if (!kvm_vcpu_is_bsp(vcpu
)) {
676 kvm_rip_write(vcpu
, 0);
677 svm
->vmcb
->save
.cs
.base
= svm
->vcpu
.arch
.sipi_vector
<< 12;
678 svm
->vmcb
->save
.cs
.selector
= svm
->vcpu
.arch
.sipi_vector
<< 8;
680 vcpu
->arch
.regs_avail
= ~0;
681 vcpu
->arch
.regs_dirty
= ~0;
686 static struct kvm_vcpu
*svm_create_vcpu(struct kvm
*kvm
, unsigned int id
)
688 struct vcpu_svm
*svm
;
690 struct page
*msrpm_pages
;
691 struct page
*hsave_page
;
692 struct page
*nested_msrpm_pages
;
695 svm
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
701 err
= kvm_vcpu_init(&svm
->vcpu
, kvm
, id
);
705 page
= alloc_page(GFP_KERNEL
);
712 msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
716 nested_msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
717 if (!nested_msrpm_pages
)
720 svm
->msrpm
= page_address(msrpm_pages
);
721 svm_vcpu_init_msrpm(svm
->msrpm
);
723 hsave_page
= alloc_page(GFP_KERNEL
);
726 svm
->nested
.hsave
= page_address(hsave_page
);
728 svm
->nested
.msrpm
= page_address(nested_msrpm_pages
);
730 svm
->vmcb
= page_address(page
);
731 clear_page(svm
->vmcb
);
732 svm
->vmcb_pa
= page_to_pfn(page
) << PAGE_SHIFT
;
733 svm
->asid_generation
= 0;
737 svm
->vcpu
.fpu_active
= 1;
738 svm
->vcpu
.arch
.apic_base
= 0xfee00000 | MSR_IA32_APICBASE_ENABLE
;
739 if (kvm_vcpu_is_bsp(&svm
->vcpu
))
740 svm
->vcpu
.arch
.apic_base
|= MSR_IA32_APICBASE_BSP
;
745 kvm_vcpu_uninit(&svm
->vcpu
);
747 kmem_cache_free(kvm_vcpu_cache
, svm
);
752 static void svm_free_vcpu(struct kvm_vcpu
*vcpu
)
754 struct vcpu_svm
*svm
= to_svm(vcpu
);
756 __free_page(pfn_to_page(svm
->vmcb_pa
>> PAGE_SHIFT
));
757 __free_pages(virt_to_page(svm
->msrpm
), MSRPM_ALLOC_ORDER
);
758 __free_page(virt_to_page(svm
->nested
.hsave
));
759 __free_pages(virt_to_page(svm
->nested
.msrpm
), MSRPM_ALLOC_ORDER
);
760 kvm_vcpu_uninit(vcpu
);
761 kmem_cache_free(kvm_vcpu_cache
, svm
);
764 static void svm_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
766 struct vcpu_svm
*svm
= to_svm(vcpu
);
769 if (unlikely(cpu
!= vcpu
->cpu
)) {
773 * Make sure that the guest sees a monotonically
776 delta
= vcpu
->arch
.host_tsc
- native_read_tsc();
777 svm
->vmcb
->control
.tsc_offset
+= delta
;
779 svm
->nested
.hsave
->control
.tsc_offset
+= delta
;
781 kvm_migrate_timers(vcpu
);
782 svm
->asid_generation
= 0;
785 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
786 rdmsrl(host_save_user_msrs
[i
], svm
->host_user_msrs
[i
]);
789 static void svm_vcpu_put(struct kvm_vcpu
*vcpu
)
791 struct vcpu_svm
*svm
= to_svm(vcpu
);
794 ++vcpu
->stat
.host_state_reload
;
795 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
796 wrmsrl(host_save_user_msrs
[i
], svm
->host_user_msrs
[i
]);
798 vcpu
->arch
.host_tsc
= native_read_tsc();
801 static unsigned long svm_get_rflags(struct kvm_vcpu
*vcpu
)
803 return to_svm(vcpu
)->vmcb
->save
.rflags
;
806 static void svm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
808 to_svm(vcpu
)->vmcb
->save
.rflags
= rflags
;
811 static void svm_cache_reg(struct kvm_vcpu
*vcpu
, enum kvm_reg reg
)
814 case VCPU_EXREG_PDPTR
:
815 BUG_ON(!npt_enabled
);
816 load_pdptrs(vcpu
, vcpu
->arch
.cr3
);
823 static void svm_set_vintr(struct vcpu_svm
*svm
)
825 svm
->vmcb
->control
.intercept
|= 1ULL << INTERCEPT_VINTR
;
828 static void svm_clear_vintr(struct vcpu_svm
*svm
)
830 svm
->vmcb
->control
.intercept
&= ~(1ULL << INTERCEPT_VINTR
);
833 static struct vmcb_seg
*svm_seg(struct kvm_vcpu
*vcpu
, int seg
)
835 struct vmcb_save_area
*save
= &to_svm(vcpu
)->vmcb
->save
;
838 case VCPU_SREG_CS
: return &save
->cs
;
839 case VCPU_SREG_DS
: return &save
->ds
;
840 case VCPU_SREG_ES
: return &save
->es
;
841 case VCPU_SREG_FS
: return &save
->fs
;
842 case VCPU_SREG_GS
: return &save
->gs
;
843 case VCPU_SREG_SS
: return &save
->ss
;
844 case VCPU_SREG_TR
: return &save
->tr
;
845 case VCPU_SREG_LDTR
: return &save
->ldtr
;
851 static u64
svm_get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
853 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
858 static void svm_get_segment(struct kvm_vcpu
*vcpu
,
859 struct kvm_segment
*var
, int seg
)
861 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
864 var
->limit
= s
->limit
;
865 var
->selector
= s
->selector
;
866 var
->type
= s
->attrib
& SVM_SELECTOR_TYPE_MASK
;
867 var
->s
= (s
->attrib
>> SVM_SELECTOR_S_SHIFT
) & 1;
868 var
->dpl
= (s
->attrib
>> SVM_SELECTOR_DPL_SHIFT
) & 3;
869 var
->present
= (s
->attrib
>> SVM_SELECTOR_P_SHIFT
) & 1;
870 var
->avl
= (s
->attrib
>> SVM_SELECTOR_AVL_SHIFT
) & 1;
871 var
->l
= (s
->attrib
>> SVM_SELECTOR_L_SHIFT
) & 1;
872 var
->db
= (s
->attrib
>> SVM_SELECTOR_DB_SHIFT
) & 1;
873 var
->g
= (s
->attrib
>> SVM_SELECTOR_G_SHIFT
) & 1;
875 /* AMD's VMCB does not have an explicit unusable field, so emulate it
876 * for cross vendor migration purposes by "not present"
878 var
->unusable
= !var
->present
|| (var
->type
== 0);
883 * SVM always stores 0 for the 'G' bit in the CS selector in
884 * the VMCB on a VMEXIT. This hurts cross-vendor migration:
885 * Intel's VMENTRY has a check on the 'G' bit.
887 var
->g
= s
->limit
> 0xfffff;
891 * Work around a bug where the busy flag in the tr selector
901 * The accessed bit must always be set in the segment
902 * descriptor cache, although it can be cleared in the
903 * descriptor, the cached bit always remains at 1. Since
904 * Intel has a check on this, set it here to support
905 * cross-vendor migration.
911 /* On AMD CPUs sometimes the DB bit in the segment
912 * descriptor is left as 1, although the whole segment has
913 * been made unusable. Clear it here to pass an Intel VMX
914 * entry check when cross vendor migrating.
922 static int svm_get_cpl(struct kvm_vcpu
*vcpu
)
924 struct vmcb_save_area
*save
= &to_svm(vcpu
)->vmcb
->save
;
929 static void svm_get_idt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
931 struct vcpu_svm
*svm
= to_svm(vcpu
);
933 dt
->limit
= svm
->vmcb
->save
.idtr
.limit
;
934 dt
->base
= svm
->vmcb
->save
.idtr
.base
;
937 static void svm_set_idt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
939 struct vcpu_svm
*svm
= to_svm(vcpu
);
941 svm
->vmcb
->save
.idtr
.limit
= dt
->limit
;
942 svm
->vmcb
->save
.idtr
.base
= dt
->base
;
945 static void svm_get_gdt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
947 struct vcpu_svm
*svm
= to_svm(vcpu
);
949 dt
->limit
= svm
->vmcb
->save
.gdtr
.limit
;
950 dt
->base
= svm
->vmcb
->save
.gdtr
.base
;
953 static void svm_set_gdt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
955 struct vcpu_svm
*svm
= to_svm(vcpu
);
957 svm
->vmcb
->save
.gdtr
.limit
= dt
->limit
;
958 svm
->vmcb
->save
.gdtr
.base
= dt
->base
;
961 static void svm_decache_cr4_guest_bits(struct kvm_vcpu
*vcpu
)
965 static void svm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
967 struct vcpu_svm
*svm
= to_svm(vcpu
);
970 if (vcpu
->arch
.shadow_efer
& EFER_LME
) {
971 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
972 vcpu
->arch
.shadow_efer
|= EFER_LMA
;
973 svm
->vmcb
->save
.efer
|= EFER_LMA
| EFER_LME
;
976 if (is_paging(vcpu
) && !(cr0
& X86_CR0_PG
)) {
977 vcpu
->arch
.shadow_efer
&= ~EFER_LMA
;
978 svm
->vmcb
->save
.efer
&= ~(EFER_LMA
| EFER_LME
);
985 if ((vcpu
->arch
.cr0
& X86_CR0_TS
) && !(cr0
& X86_CR0_TS
)) {
986 svm
->vmcb
->control
.intercept_exceptions
&= ~(1 << NM_VECTOR
);
987 vcpu
->fpu_active
= 1;
990 vcpu
->arch
.cr0
= cr0
;
991 cr0
|= X86_CR0_PG
| X86_CR0_WP
;
992 if (!vcpu
->fpu_active
) {
993 svm
->vmcb
->control
.intercept_exceptions
|= (1 << NM_VECTOR
);
998 * re-enable caching here because the QEMU bios
999 * does not do it - this results in some delay at
1002 cr0
&= ~(X86_CR0_CD
| X86_CR0_NW
);
1003 svm
->vmcb
->save
.cr0
= cr0
;
1006 static void svm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
1008 unsigned long host_cr4_mce
= read_cr4() & X86_CR4_MCE
;
1009 unsigned long old_cr4
= to_svm(vcpu
)->vmcb
->save
.cr4
;
1011 if (npt_enabled
&& ((old_cr4
^ cr4
) & X86_CR4_PGE
))
1012 force_new_asid(vcpu
);
1014 vcpu
->arch
.cr4
= cr4
;
1017 cr4
|= host_cr4_mce
;
1018 to_svm(vcpu
)->vmcb
->save
.cr4
= cr4
;
1021 static void svm_set_segment(struct kvm_vcpu
*vcpu
,
1022 struct kvm_segment
*var
, int seg
)
1024 struct vcpu_svm
*svm
= to_svm(vcpu
);
1025 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
1027 s
->base
= var
->base
;
1028 s
->limit
= var
->limit
;
1029 s
->selector
= var
->selector
;
1033 s
->attrib
= (var
->type
& SVM_SELECTOR_TYPE_MASK
);
1034 s
->attrib
|= (var
->s
& 1) << SVM_SELECTOR_S_SHIFT
;
1035 s
->attrib
|= (var
->dpl
& 3) << SVM_SELECTOR_DPL_SHIFT
;
1036 s
->attrib
|= (var
->present
& 1) << SVM_SELECTOR_P_SHIFT
;
1037 s
->attrib
|= (var
->avl
& 1) << SVM_SELECTOR_AVL_SHIFT
;
1038 s
->attrib
|= (var
->l
& 1) << SVM_SELECTOR_L_SHIFT
;
1039 s
->attrib
|= (var
->db
& 1) << SVM_SELECTOR_DB_SHIFT
;
1040 s
->attrib
|= (var
->g
& 1) << SVM_SELECTOR_G_SHIFT
;
1042 if (seg
== VCPU_SREG_CS
)
1044 = (svm
->vmcb
->save
.cs
.attrib
1045 >> SVM_SELECTOR_DPL_SHIFT
) & 3;
1049 static void update_db_intercept(struct kvm_vcpu
*vcpu
)
1051 struct vcpu_svm
*svm
= to_svm(vcpu
);
1053 svm
->vmcb
->control
.intercept_exceptions
&=
1054 ~((1 << DB_VECTOR
) | (1 << BP_VECTOR
));
1056 if (vcpu
->arch
.singlestep
)
1057 svm
->vmcb
->control
.intercept_exceptions
|= (1 << DB_VECTOR
);
1059 if (vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
) {
1060 if (vcpu
->guest_debug
&
1061 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
))
1062 svm
->vmcb
->control
.intercept_exceptions
|=
1064 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_SW_BP
)
1065 svm
->vmcb
->control
.intercept_exceptions
|=
1068 vcpu
->guest_debug
= 0;
1071 static void svm_guest_debug(struct kvm_vcpu
*vcpu
, struct kvm_guest_debug
*dbg
)
1073 struct vcpu_svm
*svm
= to_svm(vcpu
);
1075 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
1076 svm
->vmcb
->save
.dr7
= dbg
->arch
.debugreg
[7];
1078 svm
->vmcb
->save
.dr7
= vcpu
->arch
.dr7
;
1080 update_db_intercept(vcpu
);
1083 static void load_host_msrs(struct kvm_vcpu
*vcpu
)
1085 #ifdef CONFIG_X86_64
1086 wrmsrl(MSR_GS_BASE
, to_svm(vcpu
)->host_gs_base
);
1090 static void save_host_msrs(struct kvm_vcpu
*vcpu
)
1092 #ifdef CONFIG_X86_64
1093 rdmsrl(MSR_GS_BASE
, to_svm(vcpu
)->host_gs_base
);
1097 static void new_asid(struct vcpu_svm
*svm
, struct svm_cpu_data
*svm_data
)
1099 if (svm_data
->next_asid
> svm_data
->max_asid
) {
1100 ++svm_data
->asid_generation
;
1101 svm_data
->next_asid
= 1;
1102 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ALL_ASID
;
1105 svm
->asid_generation
= svm_data
->asid_generation
;
1106 svm
->vmcb
->control
.asid
= svm_data
->next_asid
++;
1109 static unsigned long svm_get_dr(struct kvm_vcpu
*vcpu
, int dr
)
1111 struct vcpu_svm
*svm
= to_svm(vcpu
);
1116 val
= vcpu
->arch
.db
[dr
];
1119 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
1120 val
= vcpu
->arch
.dr6
;
1122 val
= svm
->vmcb
->save
.dr6
;
1125 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
1126 val
= vcpu
->arch
.dr7
;
1128 val
= svm
->vmcb
->save
.dr7
;
1137 static void svm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long value
,
1140 struct vcpu_svm
*svm
= to_svm(vcpu
);
1146 vcpu
->arch
.db
[dr
] = value
;
1147 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
))
1148 vcpu
->arch
.eff_db
[dr
] = value
;
1151 if (vcpu
->arch
.cr4
& X86_CR4_DE
)
1152 *exception
= UD_VECTOR
;
1155 if (value
& 0xffffffff00000000ULL
) {
1156 *exception
= GP_VECTOR
;
1159 vcpu
->arch
.dr6
= (value
& DR6_VOLATILE
) | DR6_FIXED_1
;
1162 if (value
& 0xffffffff00000000ULL
) {
1163 *exception
= GP_VECTOR
;
1166 vcpu
->arch
.dr7
= (value
& DR7_VOLATILE
) | DR7_FIXED_1
;
1167 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)) {
1168 svm
->vmcb
->save
.dr7
= vcpu
->arch
.dr7
;
1169 vcpu
->arch
.switch_db_regs
= (value
& DR7_BP_EN_MASK
);
1173 /* FIXME: Possible case? */
1174 printk(KERN_DEBUG
"%s: unexpected dr %u\n",
1176 *exception
= UD_VECTOR
;
1181 static int pf_interception(struct vcpu_svm
*svm
)
1186 fault_address
= svm
->vmcb
->control
.exit_info_2
;
1187 error_code
= svm
->vmcb
->control
.exit_info_1
;
1189 trace_kvm_page_fault(fault_address
, error_code
);
1190 if (!npt_enabled
&& kvm_event_needs_reinjection(&svm
->vcpu
))
1191 kvm_mmu_unprotect_page_virt(&svm
->vcpu
, fault_address
);
1192 return kvm_mmu_page_fault(&svm
->vcpu
, fault_address
, error_code
);
1195 static int db_interception(struct vcpu_svm
*svm
)
1197 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
1199 if (!(svm
->vcpu
.guest_debug
&
1200 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
)) &&
1201 !svm
->vcpu
.arch
.singlestep
) {
1202 kvm_queue_exception(&svm
->vcpu
, DB_VECTOR
);
1206 if (svm
->vcpu
.arch
.singlestep
) {
1207 svm
->vcpu
.arch
.singlestep
= false;
1208 if (!(svm
->vcpu
.guest_debug
& KVM_GUESTDBG_SINGLESTEP
))
1209 svm
->vmcb
->save
.rflags
&=
1210 ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
1211 update_db_intercept(&svm
->vcpu
);
1214 if (svm
->vcpu
.guest_debug
&
1215 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
)){
1216 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
1217 kvm_run
->debug
.arch
.pc
=
1218 svm
->vmcb
->save
.cs
.base
+ svm
->vmcb
->save
.rip
;
1219 kvm_run
->debug
.arch
.exception
= DB_VECTOR
;
1226 static int bp_interception(struct vcpu_svm
*svm
)
1228 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
1230 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
1231 kvm_run
->debug
.arch
.pc
= svm
->vmcb
->save
.cs
.base
+ svm
->vmcb
->save
.rip
;
1232 kvm_run
->debug
.arch
.exception
= BP_VECTOR
;
1236 static int ud_interception(struct vcpu_svm
*svm
)
1240 er
= emulate_instruction(&svm
->vcpu
, 0, 0, EMULTYPE_TRAP_UD
);
1241 if (er
!= EMULATE_DONE
)
1242 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
1246 static int nm_interception(struct vcpu_svm
*svm
)
1248 svm
->vmcb
->control
.intercept_exceptions
&= ~(1 << NM_VECTOR
);
1249 if (!(svm
->vcpu
.arch
.cr0
& X86_CR0_TS
))
1250 svm
->vmcb
->save
.cr0
&= ~X86_CR0_TS
;
1251 svm
->vcpu
.fpu_active
= 1;
1256 static int mc_interception(struct vcpu_svm
*svm
)
1259 * On an #MC intercept the MCE handler is not called automatically in
1260 * the host. So do it by hand here.
1264 /* not sure if we ever come back to this point */
1269 static int shutdown_interception(struct vcpu_svm
*svm
)
1271 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
1274 * VMCB is undefined after a SHUTDOWN intercept
1275 * so reinitialize it.
1277 clear_page(svm
->vmcb
);
1280 kvm_run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
1284 static int io_interception(struct vcpu_svm
*svm
)
1286 u32 io_info
= svm
->vmcb
->control
.exit_info_1
; /* address size bug? */
1287 int size
, in
, string
;
1290 ++svm
->vcpu
.stat
.io_exits
;
1292 svm
->next_rip
= svm
->vmcb
->control
.exit_info_2
;
1294 string
= (io_info
& SVM_IOIO_STR_MASK
) != 0;
1297 if (emulate_instruction(&svm
->vcpu
,
1298 0, 0, 0) == EMULATE_DO_MMIO
)
1303 in
= (io_info
& SVM_IOIO_TYPE_MASK
) != 0;
1304 port
= io_info
>> 16;
1305 size
= (io_info
& SVM_IOIO_SIZE_MASK
) >> SVM_IOIO_SIZE_SHIFT
;
1307 skip_emulated_instruction(&svm
->vcpu
);
1308 return kvm_emulate_pio(&svm
->vcpu
, in
, size
, port
);
1311 static int nmi_interception(struct vcpu_svm
*svm
)
1316 static int intr_interception(struct vcpu_svm
*svm
)
1318 ++svm
->vcpu
.stat
.irq_exits
;
1322 static int nop_on_interception(struct vcpu_svm
*svm
)
1327 static int halt_interception(struct vcpu_svm
*svm
)
1329 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 1;
1330 skip_emulated_instruction(&svm
->vcpu
);
1331 return kvm_emulate_halt(&svm
->vcpu
);
1334 static int vmmcall_interception(struct vcpu_svm
*svm
)
1336 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1337 skip_emulated_instruction(&svm
->vcpu
);
1338 kvm_emulate_hypercall(&svm
->vcpu
);
1342 static int nested_svm_check_permissions(struct vcpu_svm
*svm
)
1344 if (!(svm
->vcpu
.arch
.shadow_efer
& EFER_SVME
)
1345 || !is_paging(&svm
->vcpu
)) {
1346 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
1350 if (svm
->vmcb
->save
.cpl
) {
1351 kvm_inject_gp(&svm
->vcpu
, 0);
1358 static int nested_svm_check_exception(struct vcpu_svm
*svm
, unsigned nr
,
1359 bool has_error_code
, u32 error_code
)
1361 if (!is_nested(svm
))
1364 svm
->vmcb
->control
.exit_code
= SVM_EXIT_EXCP_BASE
+ nr
;
1365 svm
->vmcb
->control
.exit_code_hi
= 0;
1366 svm
->vmcb
->control
.exit_info_1
= error_code
;
1367 svm
->vmcb
->control
.exit_info_2
= svm
->vcpu
.arch
.cr2
;
1369 return nested_svm_exit_handled(svm
);
1372 static inline int nested_svm_intr(struct vcpu_svm
*svm
)
1374 if (!is_nested(svm
))
1377 if (!(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
))
1380 if (!(svm
->vcpu
.arch
.hflags
& HF_HIF_MASK
))
1383 svm
->vmcb
->control
.exit_code
= SVM_EXIT_INTR
;
1385 if (svm
->nested
.intercept
& 1ULL) {
1387 * The #vmexit can't be emulated here directly because this
1388 * code path runs with irqs and preemtion disabled. A
1389 * #vmexit emulation might sleep. Only signal request for
1392 svm
->nested
.exit_required
= true;
1393 trace_kvm_nested_intr_vmexit(svm
->vmcb
->save
.rip
);
1400 static void *nested_svm_map(struct vcpu_svm
*svm
, u64 gpa
, enum km_type idx
)
1404 page
= gfn_to_page(svm
->vcpu
.kvm
, gpa
>> PAGE_SHIFT
);
1405 if (is_error_page(page
))
1408 return kmap_atomic(page
, idx
);
1411 kvm_release_page_clean(page
);
1412 kvm_inject_gp(&svm
->vcpu
, 0);
1417 static void nested_svm_unmap(void *addr
, enum km_type idx
)
1424 page
= kmap_atomic_to_page(addr
);
1426 kunmap_atomic(addr
, idx
);
1427 kvm_release_page_dirty(page
);
1430 static bool nested_svm_exit_handled_msr(struct vcpu_svm
*svm
)
1432 u32 param
= svm
->vmcb
->control
.exit_info_1
& 1;
1433 u32 msr
= svm
->vcpu
.arch
.regs
[VCPU_REGS_RCX
];
1438 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_MSR_PROT
)))
1441 msrpm
= nested_svm_map(svm
, svm
->nested
.vmcb_msrpm
, KM_USER0
);
1451 case 0xc0000000 ... 0xc0001fff:
1452 t0
= (8192 + msr
- 0xc0000000) * 2;
1456 case 0xc0010000 ... 0xc0011fff:
1457 t0
= (16384 + msr
- 0xc0010000) * 2;
1466 ret
= msrpm
[t1
] & ((1 << param
) << t0
);
1469 nested_svm_unmap(msrpm
, KM_USER0
);
1474 static int nested_svm_exit_special(struct vcpu_svm
*svm
)
1476 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
1478 switch (exit_code
) {
1481 return NESTED_EXIT_HOST
;
1482 /* For now we are always handling NPFs when using them */
1485 return NESTED_EXIT_HOST
;
1487 /* When we're shadowing, trap PFs */
1488 case SVM_EXIT_EXCP_BASE
+ PF_VECTOR
:
1490 return NESTED_EXIT_HOST
;
1496 return NESTED_EXIT_CONTINUE
;
1500 * If this function returns true, this #vmexit was already handled
1502 static int nested_svm_exit_handled(struct vcpu_svm
*svm
)
1504 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
1505 int vmexit
= NESTED_EXIT_HOST
;
1507 switch (exit_code
) {
1509 vmexit
= nested_svm_exit_handled_msr(svm
);
1511 case SVM_EXIT_READ_CR0
... SVM_EXIT_READ_CR8
: {
1512 u32 cr_bits
= 1 << (exit_code
- SVM_EXIT_READ_CR0
);
1513 if (svm
->nested
.intercept_cr_read
& cr_bits
)
1514 vmexit
= NESTED_EXIT_DONE
;
1517 case SVM_EXIT_WRITE_CR0
... SVM_EXIT_WRITE_CR8
: {
1518 u32 cr_bits
= 1 << (exit_code
- SVM_EXIT_WRITE_CR0
);
1519 if (svm
->nested
.intercept_cr_write
& cr_bits
)
1520 vmexit
= NESTED_EXIT_DONE
;
1523 case SVM_EXIT_READ_DR0
... SVM_EXIT_READ_DR7
: {
1524 u32 dr_bits
= 1 << (exit_code
- SVM_EXIT_READ_DR0
);
1525 if (svm
->nested
.intercept_dr_read
& dr_bits
)
1526 vmexit
= NESTED_EXIT_DONE
;
1529 case SVM_EXIT_WRITE_DR0
... SVM_EXIT_WRITE_DR7
: {
1530 u32 dr_bits
= 1 << (exit_code
- SVM_EXIT_WRITE_DR0
);
1531 if (svm
->nested
.intercept_dr_write
& dr_bits
)
1532 vmexit
= NESTED_EXIT_DONE
;
1535 case SVM_EXIT_EXCP_BASE
... SVM_EXIT_EXCP_BASE
+ 0x1f: {
1536 u32 excp_bits
= 1 << (exit_code
- SVM_EXIT_EXCP_BASE
);
1537 if (svm
->nested
.intercept_exceptions
& excp_bits
)
1538 vmexit
= NESTED_EXIT_DONE
;
1542 u64 exit_bits
= 1ULL << (exit_code
- SVM_EXIT_INTR
);
1543 nsvm_printk("exit code: 0x%x\n", exit_code
);
1544 if (svm
->nested
.intercept
& exit_bits
)
1545 vmexit
= NESTED_EXIT_DONE
;
1549 if (vmexit
== NESTED_EXIT_DONE
) {
1550 nsvm_printk("#VMEXIT reason=%04x\n", exit_code
);
1551 nested_svm_vmexit(svm
);
1557 static inline void copy_vmcb_control_area(struct vmcb
*dst_vmcb
, struct vmcb
*from_vmcb
)
1559 struct vmcb_control_area
*dst
= &dst_vmcb
->control
;
1560 struct vmcb_control_area
*from
= &from_vmcb
->control
;
1562 dst
->intercept_cr_read
= from
->intercept_cr_read
;
1563 dst
->intercept_cr_write
= from
->intercept_cr_write
;
1564 dst
->intercept_dr_read
= from
->intercept_dr_read
;
1565 dst
->intercept_dr_write
= from
->intercept_dr_write
;
1566 dst
->intercept_exceptions
= from
->intercept_exceptions
;
1567 dst
->intercept
= from
->intercept
;
1568 dst
->iopm_base_pa
= from
->iopm_base_pa
;
1569 dst
->msrpm_base_pa
= from
->msrpm_base_pa
;
1570 dst
->tsc_offset
= from
->tsc_offset
;
1571 dst
->asid
= from
->asid
;
1572 dst
->tlb_ctl
= from
->tlb_ctl
;
1573 dst
->int_ctl
= from
->int_ctl
;
1574 dst
->int_vector
= from
->int_vector
;
1575 dst
->int_state
= from
->int_state
;
1576 dst
->exit_code
= from
->exit_code
;
1577 dst
->exit_code_hi
= from
->exit_code_hi
;
1578 dst
->exit_info_1
= from
->exit_info_1
;
1579 dst
->exit_info_2
= from
->exit_info_2
;
1580 dst
->exit_int_info
= from
->exit_int_info
;
1581 dst
->exit_int_info_err
= from
->exit_int_info_err
;
1582 dst
->nested_ctl
= from
->nested_ctl
;
1583 dst
->event_inj
= from
->event_inj
;
1584 dst
->event_inj_err
= from
->event_inj_err
;
1585 dst
->nested_cr3
= from
->nested_cr3
;
1586 dst
->lbr_ctl
= from
->lbr_ctl
;
1589 static int nested_svm_vmexit(struct vcpu_svm
*svm
)
1591 struct vmcb
*nested_vmcb
;
1592 struct vmcb
*hsave
= svm
->nested
.hsave
;
1593 struct vmcb
*vmcb
= svm
->vmcb
;
1595 trace_kvm_nested_vmexit_inject(vmcb
->control
.exit_code
,
1596 vmcb
->control
.exit_info_1
,
1597 vmcb
->control
.exit_info_2
,
1598 vmcb
->control
.exit_int_info
,
1599 vmcb
->control
.exit_int_info_err
);
1601 nested_vmcb
= nested_svm_map(svm
, svm
->nested
.vmcb
, KM_USER0
);
1605 /* Give the current vmcb to the guest */
1608 nested_vmcb
->save
.es
= vmcb
->save
.es
;
1609 nested_vmcb
->save
.cs
= vmcb
->save
.cs
;
1610 nested_vmcb
->save
.ss
= vmcb
->save
.ss
;
1611 nested_vmcb
->save
.ds
= vmcb
->save
.ds
;
1612 nested_vmcb
->save
.gdtr
= vmcb
->save
.gdtr
;
1613 nested_vmcb
->save
.idtr
= vmcb
->save
.idtr
;
1615 nested_vmcb
->save
.cr3
= vmcb
->save
.cr3
;
1616 nested_vmcb
->save
.cr2
= vmcb
->save
.cr2
;
1617 nested_vmcb
->save
.rflags
= vmcb
->save
.rflags
;
1618 nested_vmcb
->save
.rip
= vmcb
->save
.rip
;
1619 nested_vmcb
->save
.rsp
= vmcb
->save
.rsp
;
1620 nested_vmcb
->save
.rax
= vmcb
->save
.rax
;
1621 nested_vmcb
->save
.dr7
= vmcb
->save
.dr7
;
1622 nested_vmcb
->save
.dr6
= vmcb
->save
.dr6
;
1623 nested_vmcb
->save
.cpl
= vmcb
->save
.cpl
;
1625 nested_vmcb
->control
.int_ctl
= vmcb
->control
.int_ctl
;
1626 nested_vmcb
->control
.int_vector
= vmcb
->control
.int_vector
;
1627 nested_vmcb
->control
.int_state
= vmcb
->control
.int_state
;
1628 nested_vmcb
->control
.exit_code
= vmcb
->control
.exit_code
;
1629 nested_vmcb
->control
.exit_code_hi
= vmcb
->control
.exit_code_hi
;
1630 nested_vmcb
->control
.exit_info_1
= vmcb
->control
.exit_info_1
;
1631 nested_vmcb
->control
.exit_info_2
= vmcb
->control
.exit_info_2
;
1632 nested_vmcb
->control
.exit_int_info
= vmcb
->control
.exit_int_info
;
1633 nested_vmcb
->control
.exit_int_info_err
= vmcb
->control
.exit_int_info_err
;
1636 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
1637 * to make sure that we do not lose injected events. So check event_inj
1638 * here and copy it to exit_int_info if it is valid.
1639 * Exit_int_info and event_inj can't be both valid because the case
1640 * below only happens on a VMRUN instruction intercept which has
1641 * no valid exit_int_info set.
1643 if (vmcb
->control
.event_inj
& SVM_EVTINJ_VALID
) {
1644 struct vmcb_control_area
*nc
= &nested_vmcb
->control
;
1646 nc
->exit_int_info
= vmcb
->control
.event_inj
;
1647 nc
->exit_int_info_err
= vmcb
->control
.event_inj_err
;
1650 nested_vmcb
->control
.tlb_ctl
= 0;
1651 nested_vmcb
->control
.event_inj
= 0;
1652 nested_vmcb
->control
.event_inj_err
= 0;
1654 /* We always set V_INTR_MASKING and remember the old value in hflags */
1655 if (!(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
))
1656 nested_vmcb
->control
.int_ctl
&= ~V_INTR_MASKING_MASK
;
1658 /* Restore the original control entries */
1659 copy_vmcb_control_area(vmcb
, hsave
);
1661 /* Kill any pending exceptions */
1662 if (svm
->vcpu
.arch
.exception
.pending
== true)
1663 nsvm_printk("WARNING: Pending Exception\n");
1665 kvm_clear_exception_queue(&svm
->vcpu
);
1666 kvm_clear_interrupt_queue(&svm
->vcpu
);
1668 /* Restore selected save entries */
1669 svm
->vmcb
->save
.es
= hsave
->save
.es
;
1670 svm
->vmcb
->save
.cs
= hsave
->save
.cs
;
1671 svm
->vmcb
->save
.ss
= hsave
->save
.ss
;
1672 svm
->vmcb
->save
.ds
= hsave
->save
.ds
;
1673 svm
->vmcb
->save
.gdtr
= hsave
->save
.gdtr
;
1674 svm
->vmcb
->save
.idtr
= hsave
->save
.idtr
;
1675 svm
->vmcb
->save
.rflags
= hsave
->save
.rflags
;
1676 svm_set_efer(&svm
->vcpu
, hsave
->save
.efer
);
1677 svm_set_cr0(&svm
->vcpu
, hsave
->save
.cr0
| X86_CR0_PE
);
1678 svm_set_cr4(&svm
->vcpu
, hsave
->save
.cr4
);
1680 svm
->vmcb
->save
.cr3
= hsave
->save
.cr3
;
1681 svm
->vcpu
.arch
.cr3
= hsave
->save
.cr3
;
1683 kvm_set_cr3(&svm
->vcpu
, hsave
->save
.cr3
);
1685 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
, hsave
->save
.rax
);
1686 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RSP
, hsave
->save
.rsp
);
1687 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RIP
, hsave
->save
.rip
);
1688 svm
->vmcb
->save
.dr7
= 0;
1689 svm
->vmcb
->save
.cpl
= 0;
1690 svm
->vmcb
->control
.exit_int_info
= 0;
1692 /* Exit nested SVM mode */
1693 svm
->nested
.vmcb
= 0;
1695 nested_svm_unmap(nested_vmcb
, KM_USER0
);
1697 kvm_mmu_reset_context(&svm
->vcpu
);
1698 kvm_mmu_load(&svm
->vcpu
);
1703 static bool nested_svm_vmrun_msrpm(struct vcpu_svm
*svm
)
1708 nested_msrpm
= nested_svm_map(svm
, svm
->nested
.vmcb_msrpm
, KM_USER0
);
1712 for (i
=0; i
< PAGE_SIZE
* (1 << MSRPM_ALLOC_ORDER
) / 4; i
++)
1713 svm
->nested
.msrpm
[i
] = svm
->msrpm
[i
] | nested_msrpm
[i
];
1715 svm
->vmcb
->control
.msrpm_base_pa
= __pa(svm
->nested
.msrpm
);
1717 nested_svm_unmap(nested_msrpm
, KM_USER0
);
1722 static bool nested_svm_vmrun(struct vcpu_svm
*svm
)
1724 struct vmcb
*nested_vmcb
;
1725 struct vmcb
*hsave
= svm
->nested
.hsave
;
1726 struct vmcb
*vmcb
= svm
->vmcb
;
1728 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, KM_USER0
);
1732 /* nested_vmcb is our indicator if nested SVM is activated */
1733 svm
->nested
.vmcb
= svm
->vmcb
->save
.rax
;
1735 trace_kvm_nested_vmrun(svm
->vmcb
->save
.rip
- 3, svm
->nested
.vmcb
,
1736 nested_vmcb
->save
.rip
,
1737 nested_vmcb
->control
.int_ctl
,
1738 nested_vmcb
->control
.event_inj
,
1739 nested_vmcb
->control
.nested_ctl
);
1741 /* Clear internal status */
1742 kvm_clear_exception_queue(&svm
->vcpu
);
1743 kvm_clear_interrupt_queue(&svm
->vcpu
);
1745 /* Save the old vmcb, so we don't need to pick what we save, but
1746 can restore everything when a VMEXIT occurs */
1747 hsave
->save
.es
= vmcb
->save
.es
;
1748 hsave
->save
.cs
= vmcb
->save
.cs
;
1749 hsave
->save
.ss
= vmcb
->save
.ss
;
1750 hsave
->save
.ds
= vmcb
->save
.ds
;
1751 hsave
->save
.gdtr
= vmcb
->save
.gdtr
;
1752 hsave
->save
.idtr
= vmcb
->save
.idtr
;
1753 hsave
->save
.efer
= svm
->vcpu
.arch
.shadow_efer
;
1754 hsave
->save
.cr0
= svm
->vcpu
.arch
.cr0
;
1755 hsave
->save
.cr4
= svm
->vcpu
.arch
.cr4
;
1756 hsave
->save
.rflags
= vmcb
->save
.rflags
;
1757 hsave
->save
.rip
= svm
->next_rip
;
1758 hsave
->save
.rsp
= vmcb
->save
.rsp
;
1759 hsave
->save
.rax
= vmcb
->save
.rax
;
1761 hsave
->save
.cr3
= vmcb
->save
.cr3
;
1763 hsave
->save
.cr3
= svm
->vcpu
.arch
.cr3
;
1765 copy_vmcb_control_area(hsave
, vmcb
);
1767 if (svm
->vmcb
->save
.rflags
& X86_EFLAGS_IF
)
1768 svm
->vcpu
.arch
.hflags
|= HF_HIF_MASK
;
1770 svm
->vcpu
.arch
.hflags
&= ~HF_HIF_MASK
;
1772 /* Load the nested guest state */
1773 svm
->vmcb
->save
.es
= nested_vmcb
->save
.es
;
1774 svm
->vmcb
->save
.cs
= nested_vmcb
->save
.cs
;
1775 svm
->vmcb
->save
.ss
= nested_vmcb
->save
.ss
;
1776 svm
->vmcb
->save
.ds
= nested_vmcb
->save
.ds
;
1777 svm
->vmcb
->save
.gdtr
= nested_vmcb
->save
.gdtr
;
1778 svm
->vmcb
->save
.idtr
= nested_vmcb
->save
.idtr
;
1779 svm
->vmcb
->save
.rflags
= nested_vmcb
->save
.rflags
;
1780 svm_set_efer(&svm
->vcpu
, nested_vmcb
->save
.efer
);
1781 svm_set_cr0(&svm
->vcpu
, nested_vmcb
->save
.cr0
);
1782 svm_set_cr4(&svm
->vcpu
, nested_vmcb
->save
.cr4
);
1784 svm
->vmcb
->save
.cr3
= nested_vmcb
->save
.cr3
;
1785 svm
->vcpu
.arch
.cr3
= nested_vmcb
->save
.cr3
;
1787 kvm_set_cr3(&svm
->vcpu
, nested_vmcb
->save
.cr3
);
1788 kvm_mmu_reset_context(&svm
->vcpu
);
1790 svm
->vmcb
->save
.cr2
= svm
->vcpu
.arch
.cr2
= nested_vmcb
->save
.cr2
;
1791 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
, nested_vmcb
->save
.rax
);
1792 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RSP
, nested_vmcb
->save
.rsp
);
1793 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RIP
, nested_vmcb
->save
.rip
);
1794 /* In case we don't even reach vcpu_run, the fields are not updated */
1795 svm
->vmcb
->save
.rax
= nested_vmcb
->save
.rax
;
1796 svm
->vmcb
->save
.rsp
= nested_vmcb
->save
.rsp
;
1797 svm
->vmcb
->save
.rip
= nested_vmcb
->save
.rip
;
1798 svm
->vmcb
->save
.dr7
= nested_vmcb
->save
.dr7
;
1799 svm
->vmcb
->save
.dr6
= nested_vmcb
->save
.dr6
;
1800 svm
->vmcb
->save
.cpl
= nested_vmcb
->save
.cpl
;
1802 /* We don't want a nested guest to be more powerful than the guest,
1803 so all intercepts are ORed */
1804 svm
->vmcb
->control
.intercept_cr_read
|=
1805 nested_vmcb
->control
.intercept_cr_read
;
1806 svm
->vmcb
->control
.intercept_cr_write
|=
1807 nested_vmcb
->control
.intercept_cr_write
;
1808 svm
->vmcb
->control
.intercept_dr_read
|=
1809 nested_vmcb
->control
.intercept_dr_read
;
1810 svm
->vmcb
->control
.intercept_dr_write
|=
1811 nested_vmcb
->control
.intercept_dr_write
;
1812 svm
->vmcb
->control
.intercept_exceptions
|=
1813 nested_vmcb
->control
.intercept_exceptions
;
1815 svm
->vmcb
->control
.intercept
|= nested_vmcb
->control
.intercept
;
1817 svm
->nested
.vmcb_msrpm
= nested_vmcb
->control
.msrpm_base_pa
;
1819 /* cache intercepts */
1820 svm
->nested
.intercept_cr_read
= nested_vmcb
->control
.intercept_cr_read
;
1821 svm
->nested
.intercept_cr_write
= nested_vmcb
->control
.intercept_cr_write
;
1822 svm
->nested
.intercept_dr_read
= nested_vmcb
->control
.intercept_dr_read
;
1823 svm
->nested
.intercept_dr_write
= nested_vmcb
->control
.intercept_dr_write
;
1824 svm
->nested
.intercept_exceptions
= nested_vmcb
->control
.intercept_exceptions
;
1825 svm
->nested
.intercept
= nested_vmcb
->control
.intercept
;
1827 force_new_asid(&svm
->vcpu
);
1828 svm
->vmcb
->control
.int_ctl
= nested_vmcb
->control
.int_ctl
| V_INTR_MASKING_MASK
;
1829 if (nested_vmcb
->control
.int_ctl
& V_IRQ_MASK
) {
1830 nsvm_printk("nSVM Injecting Interrupt: 0x%x\n",
1831 nested_vmcb
->control
.int_ctl
);
1833 if (nested_vmcb
->control
.int_ctl
& V_INTR_MASKING_MASK
)
1834 svm
->vcpu
.arch
.hflags
|= HF_VINTR_MASK
;
1836 svm
->vcpu
.arch
.hflags
&= ~HF_VINTR_MASK
;
1838 nsvm_printk("nSVM exit_int_info: 0x%x | int_state: 0x%x\n",
1839 nested_vmcb
->control
.exit_int_info
,
1840 nested_vmcb
->control
.int_state
);
1842 svm
->vmcb
->control
.int_vector
= nested_vmcb
->control
.int_vector
;
1843 svm
->vmcb
->control
.int_state
= nested_vmcb
->control
.int_state
;
1844 svm
->vmcb
->control
.tsc_offset
+= nested_vmcb
->control
.tsc_offset
;
1845 if (nested_vmcb
->control
.event_inj
& SVM_EVTINJ_VALID
)
1846 nsvm_printk("Injecting Event: 0x%x\n",
1847 nested_vmcb
->control
.event_inj
);
1848 svm
->vmcb
->control
.event_inj
= nested_vmcb
->control
.event_inj
;
1849 svm
->vmcb
->control
.event_inj_err
= nested_vmcb
->control
.event_inj_err
;
1851 nested_svm_unmap(nested_vmcb
, KM_USER0
);
1858 static void nested_svm_vmloadsave(struct vmcb
*from_vmcb
, struct vmcb
*to_vmcb
)
1860 to_vmcb
->save
.fs
= from_vmcb
->save
.fs
;
1861 to_vmcb
->save
.gs
= from_vmcb
->save
.gs
;
1862 to_vmcb
->save
.tr
= from_vmcb
->save
.tr
;
1863 to_vmcb
->save
.ldtr
= from_vmcb
->save
.ldtr
;
1864 to_vmcb
->save
.kernel_gs_base
= from_vmcb
->save
.kernel_gs_base
;
1865 to_vmcb
->save
.star
= from_vmcb
->save
.star
;
1866 to_vmcb
->save
.lstar
= from_vmcb
->save
.lstar
;
1867 to_vmcb
->save
.cstar
= from_vmcb
->save
.cstar
;
1868 to_vmcb
->save
.sfmask
= from_vmcb
->save
.sfmask
;
1869 to_vmcb
->save
.sysenter_cs
= from_vmcb
->save
.sysenter_cs
;
1870 to_vmcb
->save
.sysenter_esp
= from_vmcb
->save
.sysenter_esp
;
1871 to_vmcb
->save
.sysenter_eip
= from_vmcb
->save
.sysenter_eip
;
1874 static int vmload_interception(struct vcpu_svm
*svm
)
1876 struct vmcb
*nested_vmcb
;
1878 if (nested_svm_check_permissions(svm
))
1881 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1882 skip_emulated_instruction(&svm
->vcpu
);
1884 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, KM_USER0
);
1888 nested_svm_vmloadsave(nested_vmcb
, svm
->vmcb
);
1889 nested_svm_unmap(nested_vmcb
, KM_USER0
);
1894 static int vmsave_interception(struct vcpu_svm
*svm
)
1896 struct vmcb
*nested_vmcb
;
1898 if (nested_svm_check_permissions(svm
))
1901 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1902 skip_emulated_instruction(&svm
->vcpu
);
1904 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, KM_USER0
);
1908 nested_svm_vmloadsave(svm
->vmcb
, nested_vmcb
);
1909 nested_svm_unmap(nested_vmcb
, KM_USER0
);
1914 static int vmrun_interception(struct vcpu_svm
*svm
)
1916 nsvm_printk("VMrun\n");
1918 if (nested_svm_check_permissions(svm
))
1921 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1922 skip_emulated_instruction(&svm
->vcpu
);
1924 if (!nested_svm_vmrun(svm
))
1927 if (!nested_svm_vmrun_msrpm(svm
))
1934 svm
->vmcb
->control
.exit_code
= SVM_EXIT_ERR
;
1935 svm
->vmcb
->control
.exit_code_hi
= 0;
1936 svm
->vmcb
->control
.exit_info_1
= 0;
1937 svm
->vmcb
->control
.exit_info_2
= 0;
1939 nested_svm_vmexit(svm
);
1944 static int stgi_interception(struct vcpu_svm
*svm
)
1946 if (nested_svm_check_permissions(svm
))
1949 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1950 skip_emulated_instruction(&svm
->vcpu
);
1957 static int clgi_interception(struct vcpu_svm
*svm
)
1959 if (nested_svm_check_permissions(svm
))
1962 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1963 skip_emulated_instruction(&svm
->vcpu
);
1967 /* After a CLGI no interrupts should come */
1968 svm_clear_vintr(svm
);
1969 svm
->vmcb
->control
.int_ctl
&= ~V_IRQ_MASK
;
1974 static int invlpga_interception(struct vcpu_svm
*svm
)
1976 struct kvm_vcpu
*vcpu
= &svm
->vcpu
;
1977 nsvm_printk("INVLPGA\n");
1979 trace_kvm_invlpga(svm
->vmcb
->save
.rip
, vcpu
->arch
.regs
[VCPU_REGS_RCX
],
1980 vcpu
->arch
.regs
[VCPU_REGS_RAX
]);
1982 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
1983 kvm_mmu_invlpg(vcpu
, vcpu
->arch
.regs
[VCPU_REGS_RAX
]);
1985 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1986 skip_emulated_instruction(&svm
->vcpu
);
1990 static int skinit_interception(struct vcpu_svm
*svm
)
1992 trace_kvm_skinit(svm
->vmcb
->save
.rip
, svm
->vcpu
.arch
.regs
[VCPU_REGS_RAX
]);
1994 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
1998 static int invalid_op_interception(struct vcpu_svm
*svm
)
2000 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
2004 static int task_switch_interception(struct vcpu_svm
*svm
)
2008 int int_type
= svm
->vmcb
->control
.exit_int_info
&
2009 SVM_EXITINTINFO_TYPE_MASK
;
2010 int int_vec
= svm
->vmcb
->control
.exit_int_info
& SVM_EVTINJ_VEC_MASK
;
2012 svm
->vmcb
->control
.exit_int_info
& SVM_EXITINTINFO_TYPE_MASK
;
2014 svm
->vmcb
->control
.exit_int_info
& SVM_EXITINTINFO_VALID
;
2016 tss_selector
= (u16
)svm
->vmcb
->control
.exit_info_1
;
2018 if (svm
->vmcb
->control
.exit_info_2
&
2019 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET
))
2020 reason
= TASK_SWITCH_IRET
;
2021 else if (svm
->vmcb
->control
.exit_info_2
&
2022 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP
))
2023 reason
= TASK_SWITCH_JMP
;
2025 reason
= TASK_SWITCH_GATE
;
2027 reason
= TASK_SWITCH_CALL
;
2029 if (reason
== TASK_SWITCH_GATE
) {
2031 case SVM_EXITINTINFO_TYPE_NMI
:
2032 svm
->vcpu
.arch
.nmi_injected
= false;
2034 case SVM_EXITINTINFO_TYPE_EXEPT
:
2035 kvm_clear_exception_queue(&svm
->vcpu
);
2037 case SVM_EXITINTINFO_TYPE_INTR
:
2038 kvm_clear_interrupt_queue(&svm
->vcpu
);
2045 if (reason
!= TASK_SWITCH_GATE
||
2046 int_type
== SVM_EXITINTINFO_TYPE_SOFT
||
2047 (int_type
== SVM_EXITINTINFO_TYPE_EXEPT
&&
2048 (int_vec
== OF_VECTOR
|| int_vec
== BP_VECTOR
)))
2049 skip_emulated_instruction(&svm
->vcpu
);
2051 return kvm_task_switch(&svm
->vcpu
, tss_selector
, reason
);
2054 static int cpuid_interception(struct vcpu_svm
*svm
)
2056 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
2057 kvm_emulate_cpuid(&svm
->vcpu
);
2061 static int iret_interception(struct vcpu_svm
*svm
)
2063 ++svm
->vcpu
.stat
.nmi_window_exits
;
2064 svm
->vmcb
->control
.intercept
&= ~(1UL << INTERCEPT_IRET
);
2065 svm
->vcpu
.arch
.hflags
|= HF_IRET_MASK
;
2069 static int invlpg_interception(struct vcpu_svm
*svm
)
2071 if (emulate_instruction(&svm
->vcpu
, 0, 0, 0) != EMULATE_DONE
)
2072 pr_unimpl(&svm
->vcpu
, "%s: failed\n", __func__
);
2076 static int emulate_on_interception(struct vcpu_svm
*svm
)
2078 if (emulate_instruction(&svm
->vcpu
, 0, 0, 0) != EMULATE_DONE
)
2079 pr_unimpl(&svm
->vcpu
, "%s: failed\n", __func__
);
2083 static int cr8_write_interception(struct vcpu_svm
*svm
)
2085 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
2087 u8 cr8_prev
= kvm_get_cr8(&svm
->vcpu
);
2088 /* instruction emulation calls kvm_set_cr8() */
2089 emulate_instruction(&svm
->vcpu
, 0, 0, 0);
2090 if (irqchip_in_kernel(svm
->vcpu
.kvm
)) {
2091 svm
->vmcb
->control
.intercept_cr_write
&= ~INTERCEPT_CR8_MASK
;
2094 if (cr8_prev
<= kvm_get_cr8(&svm
->vcpu
))
2096 kvm_run
->exit_reason
= KVM_EXIT_SET_TPR
;
2100 static int svm_get_msr(struct kvm_vcpu
*vcpu
, unsigned ecx
, u64
*data
)
2102 struct vcpu_svm
*svm
= to_svm(vcpu
);
2105 case MSR_IA32_TSC
: {
2109 tsc_offset
= svm
->nested
.hsave
->control
.tsc_offset
;
2111 tsc_offset
= svm
->vmcb
->control
.tsc_offset
;
2113 *data
= tsc_offset
+ native_read_tsc();
2117 *data
= svm
->vmcb
->save
.star
;
2119 #ifdef CONFIG_X86_64
2121 *data
= svm
->vmcb
->save
.lstar
;
2124 *data
= svm
->vmcb
->save
.cstar
;
2126 case MSR_KERNEL_GS_BASE
:
2127 *data
= svm
->vmcb
->save
.kernel_gs_base
;
2129 case MSR_SYSCALL_MASK
:
2130 *data
= svm
->vmcb
->save
.sfmask
;
2133 case MSR_IA32_SYSENTER_CS
:
2134 *data
= svm
->vmcb
->save
.sysenter_cs
;
2136 case MSR_IA32_SYSENTER_EIP
:
2137 *data
= svm
->sysenter_eip
;
2139 case MSR_IA32_SYSENTER_ESP
:
2140 *data
= svm
->sysenter_esp
;
2142 /* Nobody will change the following 5 values in the VMCB so
2143 we can safely return them on rdmsr. They will always be 0
2144 until LBRV is implemented. */
2145 case MSR_IA32_DEBUGCTLMSR
:
2146 *data
= svm
->vmcb
->save
.dbgctl
;
2148 case MSR_IA32_LASTBRANCHFROMIP
:
2149 *data
= svm
->vmcb
->save
.br_from
;
2151 case MSR_IA32_LASTBRANCHTOIP
:
2152 *data
= svm
->vmcb
->save
.br_to
;
2154 case MSR_IA32_LASTINTFROMIP
:
2155 *data
= svm
->vmcb
->save
.last_excp_from
;
2157 case MSR_IA32_LASTINTTOIP
:
2158 *data
= svm
->vmcb
->save
.last_excp_to
;
2160 case MSR_VM_HSAVE_PA
:
2161 *data
= svm
->nested
.hsave_msr
;
2166 case MSR_IA32_UCODE_REV
:
2170 return kvm_get_msr_common(vcpu
, ecx
, data
);
2175 static int rdmsr_interception(struct vcpu_svm
*svm
)
2177 u32 ecx
= svm
->vcpu
.arch
.regs
[VCPU_REGS_RCX
];
2180 if (svm_get_msr(&svm
->vcpu
, ecx
, &data
))
2181 kvm_inject_gp(&svm
->vcpu
, 0);
2183 trace_kvm_msr_read(ecx
, data
);
2185 svm
->vcpu
.arch
.regs
[VCPU_REGS_RAX
] = data
& 0xffffffff;
2186 svm
->vcpu
.arch
.regs
[VCPU_REGS_RDX
] = data
>> 32;
2187 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
2188 skip_emulated_instruction(&svm
->vcpu
);
2193 static int svm_set_msr(struct kvm_vcpu
*vcpu
, unsigned ecx
, u64 data
)
2195 struct vcpu_svm
*svm
= to_svm(vcpu
);
2198 case MSR_IA32_TSC
: {
2199 u64 tsc_offset
= data
- native_read_tsc();
2200 u64 g_tsc_offset
= 0;
2202 if (is_nested(svm
)) {
2203 g_tsc_offset
= svm
->vmcb
->control
.tsc_offset
-
2204 svm
->nested
.hsave
->control
.tsc_offset
;
2205 svm
->nested
.hsave
->control
.tsc_offset
= tsc_offset
;
2208 svm
->vmcb
->control
.tsc_offset
= tsc_offset
+ g_tsc_offset
;
2213 svm
->vmcb
->save
.star
= data
;
2215 #ifdef CONFIG_X86_64
2217 svm
->vmcb
->save
.lstar
= data
;
2220 svm
->vmcb
->save
.cstar
= data
;
2222 case MSR_KERNEL_GS_BASE
:
2223 svm
->vmcb
->save
.kernel_gs_base
= data
;
2225 case MSR_SYSCALL_MASK
:
2226 svm
->vmcb
->save
.sfmask
= data
;
2229 case MSR_IA32_SYSENTER_CS
:
2230 svm
->vmcb
->save
.sysenter_cs
= data
;
2232 case MSR_IA32_SYSENTER_EIP
:
2233 svm
->sysenter_eip
= data
;
2234 svm
->vmcb
->save
.sysenter_eip
= data
;
2236 case MSR_IA32_SYSENTER_ESP
:
2237 svm
->sysenter_esp
= data
;
2238 svm
->vmcb
->save
.sysenter_esp
= data
;
2240 case MSR_IA32_DEBUGCTLMSR
:
2241 if (!svm_has(SVM_FEATURE_LBRV
)) {
2242 pr_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2246 if (data
& DEBUGCTL_RESERVED_BITS
)
2249 svm
->vmcb
->save
.dbgctl
= data
;
2250 if (data
& (1ULL<<0))
2251 svm_enable_lbrv(svm
);
2253 svm_disable_lbrv(svm
);
2255 case MSR_VM_HSAVE_PA
:
2256 svm
->nested
.hsave_msr
= data
;
2260 pr_unimpl(vcpu
, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx
, data
);
2263 return kvm_set_msr_common(vcpu
, ecx
, data
);
2268 static int wrmsr_interception(struct vcpu_svm
*svm
)
2270 u32 ecx
= svm
->vcpu
.arch
.regs
[VCPU_REGS_RCX
];
2271 u64 data
= (svm
->vcpu
.arch
.regs
[VCPU_REGS_RAX
] & -1u)
2272 | ((u64
)(svm
->vcpu
.arch
.regs
[VCPU_REGS_RDX
] & -1u) << 32);
2274 trace_kvm_msr_write(ecx
, data
);
2276 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
2277 if (svm_set_msr(&svm
->vcpu
, ecx
, data
))
2278 kvm_inject_gp(&svm
->vcpu
, 0);
2280 skip_emulated_instruction(&svm
->vcpu
);
2284 static int msr_interception(struct vcpu_svm
*svm
)
2286 if (svm
->vmcb
->control
.exit_info_1
)
2287 return wrmsr_interception(svm
);
2289 return rdmsr_interception(svm
);
2292 static int interrupt_window_interception(struct vcpu_svm
*svm
)
2294 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
2296 svm_clear_vintr(svm
);
2297 svm
->vmcb
->control
.int_ctl
&= ~V_IRQ_MASK
;
2299 * If the user space waits to inject interrupts, exit as soon as
2302 if (!irqchip_in_kernel(svm
->vcpu
.kvm
) &&
2303 kvm_run
->request_interrupt_window
&&
2304 !kvm_cpu_has_interrupt(&svm
->vcpu
)) {
2305 ++svm
->vcpu
.stat
.irq_window_exits
;
2306 kvm_run
->exit_reason
= KVM_EXIT_IRQ_WINDOW_OPEN
;
2313 static int (*svm_exit_handlers
[])(struct vcpu_svm
*svm
) = {
2314 [SVM_EXIT_READ_CR0
] = emulate_on_interception
,
2315 [SVM_EXIT_READ_CR3
] = emulate_on_interception
,
2316 [SVM_EXIT_READ_CR4
] = emulate_on_interception
,
2317 [SVM_EXIT_READ_CR8
] = emulate_on_interception
,
2319 [SVM_EXIT_WRITE_CR0
] = emulate_on_interception
,
2320 [SVM_EXIT_WRITE_CR3
] = emulate_on_interception
,
2321 [SVM_EXIT_WRITE_CR4
] = emulate_on_interception
,
2322 [SVM_EXIT_WRITE_CR8
] = cr8_write_interception
,
2323 [SVM_EXIT_READ_DR0
] = emulate_on_interception
,
2324 [SVM_EXIT_READ_DR1
] = emulate_on_interception
,
2325 [SVM_EXIT_READ_DR2
] = emulate_on_interception
,
2326 [SVM_EXIT_READ_DR3
] = emulate_on_interception
,
2327 [SVM_EXIT_WRITE_DR0
] = emulate_on_interception
,
2328 [SVM_EXIT_WRITE_DR1
] = emulate_on_interception
,
2329 [SVM_EXIT_WRITE_DR2
] = emulate_on_interception
,
2330 [SVM_EXIT_WRITE_DR3
] = emulate_on_interception
,
2331 [SVM_EXIT_WRITE_DR5
] = emulate_on_interception
,
2332 [SVM_EXIT_WRITE_DR7
] = emulate_on_interception
,
2333 [SVM_EXIT_EXCP_BASE
+ DB_VECTOR
] = db_interception
,
2334 [SVM_EXIT_EXCP_BASE
+ BP_VECTOR
] = bp_interception
,
2335 [SVM_EXIT_EXCP_BASE
+ UD_VECTOR
] = ud_interception
,
2336 [SVM_EXIT_EXCP_BASE
+ PF_VECTOR
] = pf_interception
,
2337 [SVM_EXIT_EXCP_BASE
+ NM_VECTOR
] = nm_interception
,
2338 [SVM_EXIT_EXCP_BASE
+ MC_VECTOR
] = mc_interception
,
2339 [SVM_EXIT_INTR
] = intr_interception
,
2340 [SVM_EXIT_NMI
] = nmi_interception
,
2341 [SVM_EXIT_SMI
] = nop_on_interception
,
2342 [SVM_EXIT_INIT
] = nop_on_interception
,
2343 [SVM_EXIT_VINTR
] = interrupt_window_interception
,
2344 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
2345 [SVM_EXIT_CPUID
] = cpuid_interception
,
2346 [SVM_EXIT_IRET
] = iret_interception
,
2347 [SVM_EXIT_INVD
] = emulate_on_interception
,
2348 [SVM_EXIT_HLT
] = halt_interception
,
2349 [SVM_EXIT_INVLPG
] = invlpg_interception
,
2350 [SVM_EXIT_INVLPGA
] = invlpga_interception
,
2351 [SVM_EXIT_IOIO
] = io_interception
,
2352 [SVM_EXIT_MSR
] = msr_interception
,
2353 [SVM_EXIT_TASK_SWITCH
] = task_switch_interception
,
2354 [SVM_EXIT_SHUTDOWN
] = shutdown_interception
,
2355 [SVM_EXIT_VMRUN
] = vmrun_interception
,
2356 [SVM_EXIT_VMMCALL
] = vmmcall_interception
,
2357 [SVM_EXIT_VMLOAD
] = vmload_interception
,
2358 [SVM_EXIT_VMSAVE
] = vmsave_interception
,
2359 [SVM_EXIT_STGI
] = stgi_interception
,
2360 [SVM_EXIT_CLGI
] = clgi_interception
,
2361 [SVM_EXIT_SKINIT
] = skinit_interception
,
2362 [SVM_EXIT_WBINVD
] = emulate_on_interception
,
2363 [SVM_EXIT_MONITOR
] = invalid_op_interception
,
2364 [SVM_EXIT_MWAIT
] = invalid_op_interception
,
2365 [SVM_EXIT_NPF
] = pf_interception
,
2368 static int handle_exit(struct kvm_vcpu
*vcpu
)
2370 struct vcpu_svm
*svm
= to_svm(vcpu
);
2371 struct kvm_run
*kvm_run
= vcpu
->run
;
2372 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
2374 trace_kvm_exit(exit_code
, svm
->vmcb
->save
.rip
);
2376 if (unlikely(svm
->nested
.exit_required
)) {
2377 nested_svm_vmexit(svm
);
2378 svm
->nested
.exit_required
= false;
2383 if (is_nested(svm
)) {
2386 trace_kvm_nested_vmexit(svm
->vmcb
->save
.rip
, exit_code
,
2387 svm
->vmcb
->control
.exit_info_1
,
2388 svm
->vmcb
->control
.exit_info_2
,
2389 svm
->vmcb
->control
.exit_int_info
,
2390 svm
->vmcb
->control
.exit_int_info_err
);
2392 nsvm_printk("nested handle_exit: 0x%x | 0x%lx | 0x%lx | 0x%lx\n",
2393 exit_code
, svm
->vmcb
->control
.exit_info_1
,
2394 svm
->vmcb
->control
.exit_info_2
, svm
->vmcb
->save
.rip
);
2396 vmexit
= nested_svm_exit_special(svm
);
2398 if (vmexit
== NESTED_EXIT_CONTINUE
)
2399 vmexit
= nested_svm_exit_handled(svm
);
2401 if (vmexit
== NESTED_EXIT_DONE
)
2405 svm_complete_interrupts(svm
);
2409 if ((vcpu
->arch
.cr0
^ svm
->vmcb
->save
.cr0
) & X86_CR0_PG
) {
2410 svm_set_cr0(vcpu
, svm
->vmcb
->save
.cr0
);
2413 vcpu
->arch
.cr0
= svm
->vmcb
->save
.cr0
;
2414 vcpu
->arch
.cr3
= svm
->vmcb
->save
.cr3
;
2416 kvm_mmu_reset_context(vcpu
);
2422 if (svm
->vmcb
->control
.exit_code
== SVM_EXIT_ERR
) {
2423 kvm_run
->exit_reason
= KVM_EXIT_FAIL_ENTRY
;
2424 kvm_run
->fail_entry
.hardware_entry_failure_reason
2425 = svm
->vmcb
->control
.exit_code
;
2429 if (is_external_interrupt(svm
->vmcb
->control
.exit_int_info
) &&
2430 exit_code
!= SVM_EXIT_EXCP_BASE
+ PF_VECTOR
&&
2431 exit_code
!= SVM_EXIT_NPF
&& exit_code
!= SVM_EXIT_TASK_SWITCH
)
2432 printk(KERN_ERR
"%s: unexpected exit_ini_info 0x%x "
2434 __func__
, svm
->vmcb
->control
.exit_int_info
,
2437 if (exit_code
>= ARRAY_SIZE(svm_exit_handlers
)
2438 || !svm_exit_handlers
[exit_code
]) {
2439 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
2440 kvm_run
->hw
.hardware_exit_reason
= exit_code
;
2444 return svm_exit_handlers
[exit_code
](svm
);
2447 static void reload_tss(struct kvm_vcpu
*vcpu
)
2449 int cpu
= raw_smp_processor_id();
2451 struct svm_cpu_data
*svm_data
= per_cpu(svm_data
, cpu
);
2452 svm_data
->tss_desc
->type
= 9; /* available 32/64-bit TSS */
2456 static void pre_svm_run(struct vcpu_svm
*svm
)
2458 int cpu
= raw_smp_processor_id();
2460 struct svm_cpu_data
*svm_data
= per_cpu(svm_data
, cpu
);
2462 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_DO_NOTHING
;
2463 /* FIXME: handle wraparound of asid_generation */
2464 if (svm
->asid_generation
!= svm_data
->asid_generation
)
2465 new_asid(svm
, svm_data
);
2468 static void svm_inject_nmi(struct kvm_vcpu
*vcpu
)
2470 struct vcpu_svm
*svm
= to_svm(vcpu
);
2472 svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_NMI
;
2473 vcpu
->arch
.hflags
|= HF_NMI_MASK
;
2474 svm
->vmcb
->control
.intercept
|= (1UL << INTERCEPT_IRET
);
2475 ++vcpu
->stat
.nmi_injections
;
2478 static inline void svm_inject_irq(struct vcpu_svm
*svm
, int irq
)
2480 struct vmcb_control_area
*control
;
2482 trace_kvm_inj_virq(irq
);
2484 ++svm
->vcpu
.stat
.irq_injections
;
2485 control
= &svm
->vmcb
->control
;
2486 control
->int_vector
= irq
;
2487 control
->int_ctl
&= ~V_INTR_PRIO_MASK
;
2488 control
->int_ctl
|= V_IRQ_MASK
|
2489 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT
);
2492 static void svm_set_irq(struct kvm_vcpu
*vcpu
)
2494 struct vcpu_svm
*svm
= to_svm(vcpu
);
2496 BUG_ON(!(gif_set(svm
)));
2498 svm
->vmcb
->control
.event_inj
= vcpu
->arch
.interrupt
.nr
|
2499 SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
;
2502 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
, int tpr
, int irr
)
2504 struct vcpu_svm
*svm
= to_svm(vcpu
);
2510 svm
->vmcb
->control
.intercept_cr_write
|= INTERCEPT_CR8_MASK
;
2513 static int svm_nmi_allowed(struct kvm_vcpu
*vcpu
)
2515 struct vcpu_svm
*svm
= to_svm(vcpu
);
2516 struct vmcb
*vmcb
= svm
->vmcb
;
2517 return !(vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
) &&
2518 !(svm
->vcpu
.arch
.hflags
& HF_NMI_MASK
);
2521 static int svm_interrupt_allowed(struct kvm_vcpu
*vcpu
)
2523 struct vcpu_svm
*svm
= to_svm(vcpu
);
2524 struct vmcb
*vmcb
= svm
->vmcb
;
2527 if (!gif_set(svm
) ||
2528 (vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
))
2531 ret
= !!(vmcb
->save
.rflags
& X86_EFLAGS_IF
);
2534 return ret
&& !(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
);
2539 static void enable_irq_window(struct kvm_vcpu
*vcpu
)
2541 struct vcpu_svm
*svm
= to_svm(vcpu
);
2542 nsvm_printk("Trying to open IRQ window\n");
2544 nested_svm_intr(svm
);
2546 /* In case GIF=0 we can't rely on the CPU to tell us when
2547 * GIF becomes 1, because that's a separate STGI/VMRUN intercept.
2548 * The next time we get that intercept, this function will be
2549 * called again though and we'll get the vintr intercept. */
2552 svm_inject_irq(svm
, 0x0);
2556 static void enable_nmi_window(struct kvm_vcpu
*vcpu
)
2558 struct vcpu_svm
*svm
= to_svm(vcpu
);
2560 if ((svm
->vcpu
.arch
.hflags
& (HF_NMI_MASK
| HF_IRET_MASK
))
2562 return; /* IRET will cause a vm exit */
2564 /* Something prevents NMI from been injected. Single step over
2565 possible problem (IRET or exception injection or interrupt
2567 vcpu
->arch
.singlestep
= true;
2568 svm
->vmcb
->save
.rflags
|= (X86_EFLAGS_TF
| X86_EFLAGS_RF
);
2569 update_db_intercept(vcpu
);
2572 static int svm_set_tss_addr(struct kvm
*kvm
, unsigned int addr
)
2577 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
)
2579 force_new_asid(vcpu
);
2582 static void svm_prepare_guest_switch(struct kvm_vcpu
*vcpu
)
2586 static inline void sync_cr8_to_lapic(struct kvm_vcpu
*vcpu
)
2588 struct vcpu_svm
*svm
= to_svm(vcpu
);
2590 if (!(svm
->vmcb
->control
.intercept_cr_write
& INTERCEPT_CR8_MASK
)) {
2591 int cr8
= svm
->vmcb
->control
.int_ctl
& V_TPR_MASK
;
2592 kvm_set_cr8(vcpu
, cr8
);
2596 static inline void sync_lapic_to_cr8(struct kvm_vcpu
*vcpu
)
2598 struct vcpu_svm
*svm
= to_svm(vcpu
);
2601 cr8
= kvm_get_cr8(vcpu
);
2602 svm
->vmcb
->control
.int_ctl
&= ~V_TPR_MASK
;
2603 svm
->vmcb
->control
.int_ctl
|= cr8
& V_TPR_MASK
;
2606 static void svm_complete_interrupts(struct vcpu_svm
*svm
)
2610 u32 exitintinfo
= svm
->vmcb
->control
.exit_int_info
;
2612 if (svm
->vcpu
.arch
.hflags
& HF_IRET_MASK
)
2613 svm
->vcpu
.arch
.hflags
&= ~(HF_NMI_MASK
| HF_IRET_MASK
);
2615 svm
->vcpu
.arch
.nmi_injected
= false;
2616 kvm_clear_exception_queue(&svm
->vcpu
);
2617 kvm_clear_interrupt_queue(&svm
->vcpu
);
2619 if (!(exitintinfo
& SVM_EXITINTINFO_VALID
))
2622 vector
= exitintinfo
& SVM_EXITINTINFO_VEC_MASK
;
2623 type
= exitintinfo
& SVM_EXITINTINFO_TYPE_MASK
;
2626 case SVM_EXITINTINFO_TYPE_NMI
:
2627 svm
->vcpu
.arch
.nmi_injected
= true;
2629 case SVM_EXITINTINFO_TYPE_EXEPT
:
2630 /* In case of software exception do not reinject an exception
2631 vector, but re-execute and instruction instead */
2634 if (kvm_exception_is_soft(vector
))
2636 if (exitintinfo
& SVM_EXITINTINFO_VALID_ERR
) {
2637 u32 err
= svm
->vmcb
->control
.exit_int_info_err
;
2638 kvm_queue_exception_e(&svm
->vcpu
, vector
, err
);
2641 kvm_queue_exception(&svm
->vcpu
, vector
);
2643 case SVM_EXITINTINFO_TYPE_INTR
:
2644 kvm_queue_interrupt(&svm
->vcpu
, vector
, false);
2651 #ifdef CONFIG_X86_64
2657 static void svm_vcpu_run(struct kvm_vcpu
*vcpu
)
2659 struct vcpu_svm
*svm
= to_svm(vcpu
);
2665 * A vmexit emulation is required before the vcpu can be executed
2668 if (unlikely(svm
->nested
.exit_required
))
2671 svm
->vmcb
->save
.rax
= vcpu
->arch
.regs
[VCPU_REGS_RAX
];
2672 svm
->vmcb
->save
.rsp
= vcpu
->arch
.regs
[VCPU_REGS_RSP
];
2673 svm
->vmcb
->save
.rip
= vcpu
->arch
.regs
[VCPU_REGS_RIP
];
2677 sync_lapic_to_cr8(vcpu
);
2679 save_host_msrs(vcpu
);
2680 fs_selector
= kvm_read_fs();
2681 gs_selector
= kvm_read_gs();
2682 ldt_selector
= kvm_read_ldt();
2683 svm
->vmcb
->save
.cr2
= vcpu
->arch
.cr2
;
2684 /* required for live migration with NPT */
2686 svm
->vmcb
->save
.cr3
= vcpu
->arch
.cr3
;
2693 "push %%"R
"bp; \n\t"
2694 "mov %c[rbx](%[svm]), %%"R
"bx \n\t"
2695 "mov %c[rcx](%[svm]), %%"R
"cx \n\t"
2696 "mov %c[rdx](%[svm]), %%"R
"dx \n\t"
2697 "mov %c[rsi](%[svm]), %%"R
"si \n\t"
2698 "mov %c[rdi](%[svm]), %%"R
"di \n\t"
2699 "mov %c[rbp](%[svm]), %%"R
"bp \n\t"
2700 #ifdef CONFIG_X86_64
2701 "mov %c[r8](%[svm]), %%r8 \n\t"
2702 "mov %c[r9](%[svm]), %%r9 \n\t"
2703 "mov %c[r10](%[svm]), %%r10 \n\t"
2704 "mov %c[r11](%[svm]), %%r11 \n\t"
2705 "mov %c[r12](%[svm]), %%r12 \n\t"
2706 "mov %c[r13](%[svm]), %%r13 \n\t"
2707 "mov %c[r14](%[svm]), %%r14 \n\t"
2708 "mov %c[r15](%[svm]), %%r15 \n\t"
2711 /* Enter guest mode */
2713 "mov %c[vmcb](%[svm]), %%"R
"ax \n\t"
2714 __ex(SVM_VMLOAD
) "\n\t"
2715 __ex(SVM_VMRUN
) "\n\t"
2716 __ex(SVM_VMSAVE
) "\n\t"
2719 /* Save guest registers, load host registers */
2720 "mov %%"R
"bx, %c[rbx](%[svm]) \n\t"
2721 "mov %%"R
"cx, %c[rcx](%[svm]) \n\t"
2722 "mov %%"R
"dx, %c[rdx](%[svm]) \n\t"
2723 "mov %%"R
"si, %c[rsi](%[svm]) \n\t"
2724 "mov %%"R
"di, %c[rdi](%[svm]) \n\t"
2725 "mov %%"R
"bp, %c[rbp](%[svm]) \n\t"
2726 #ifdef CONFIG_X86_64
2727 "mov %%r8, %c[r8](%[svm]) \n\t"
2728 "mov %%r9, %c[r9](%[svm]) \n\t"
2729 "mov %%r10, %c[r10](%[svm]) \n\t"
2730 "mov %%r11, %c[r11](%[svm]) \n\t"
2731 "mov %%r12, %c[r12](%[svm]) \n\t"
2732 "mov %%r13, %c[r13](%[svm]) \n\t"
2733 "mov %%r14, %c[r14](%[svm]) \n\t"
2734 "mov %%r15, %c[r15](%[svm]) \n\t"
2739 [vmcb
]"i"(offsetof(struct vcpu_svm
, vmcb_pa
)),
2740 [rbx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RBX
])),
2741 [rcx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RCX
])),
2742 [rdx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RDX
])),
2743 [rsi
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RSI
])),
2744 [rdi
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RDI
])),
2745 [rbp
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RBP
]))
2746 #ifdef CONFIG_X86_64
2747 , [r8
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R8
])),
2748 [r9
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R9
])),
2749 [r10
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R10
])),
2750 [r11
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R11
])),
2751 [r12
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R12
])),
2752 [r13
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R13
])),
2753 [r14
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R14
])),
2754 [r15
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R15
]))
2757 , R
"bx", R
"cx", R
"dx", R
"si", R
"di"
2758 #ifdef CONFIG_X86_64
2759 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
2763 vcpu
->arch
.cr2
= svm
->vmcb
->save
.cr2
;
2764 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = svm
->vmcb
->save
.rax
;
2765 vcpu
->arch
.regs
[VCPU_REGS_RSP
] = svm
->vmcb
->save
.rsp
;
2766 vcpu
->arch
.regs
[VCPU_REGS_RIP
] = svm
->vmcb
->save
.rip
;
2768 kvm_load_fs(fs_selector
);
2769 kvm_load_gs(gs_selector
);
2770 kvm_load_ldt(ldt_selector
);
2771 load_host_msrs(vcpu
);
2775 local_irq_disable();
2779 sync_cr8_to_lapic(vcpu
);
2784 vcpu
->arch
.regs_avail
&= ~(1 << VCPU_EXREG_PDPTR
);
2785 vcpu
->arch
.regs_dirty
&= ~(1 << VCPU_EXREG_PDPTR
);
2791 static void svm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
2793 struct vcpu_svm
*svm
= to_svm(vcpu
);
2796 svm
->vmcb
->control
.nested_cr3
= root
;
2797 force_new_asid(vcpu
);
2801 svm
->vmcb
->save
.cr3
= root
;
2802 force_new_asid(vcpu
);
2804 if (vcpu
->fpu_active
) {
2805 svm
->vmcb
->control
.intercept_exceptions
|= (1 << NM_VECTOR
);
2806 svm
->vmcb
->save
.cr0
|= X86_CR0_TS
;
2807 vcpu
->fpu_active
= 0;
2811 static int is_disabled(void)
2815 rdmsrl(MSR_VM_CR
, vm_cr
);
2816 if (vm_cr
& (1 << SVM_VM_CR_SVM_DISABLE
))
2823 svm_patch_hypercall(struct kvm_vcpu
*vcpu
, unsigned char *hypercall
)
2826 * Patch in the VMMCALL instruction:
2828 hypercall
[0] = 0x0f;
2829 hypercall
[1] = 0x01;
2830 hypercall
[2] = 0xd9;
2833 static void svm_check_processor_compat(void *rtn
)
2838 static bool svm_cpu_has_accelerated_tpr(void)
2843 static int get_npt_level(void)
2845 #ifdef CONFIG_X86_64
2846 return PT64_ROOT_LEVEL
;
2848 return PT32E_ROOT_LEVEL
;
2852 static u64
svm_get_mt_mask(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool is_mmio
)
2857 static const struct trace_print_flags svm_exit_reasons_str
[] = {
2858 { SVM_EXIT_READ_CR0
, "read_cr0" },
2859 { SVM_EXIT_READ_CR3
, "read_cr3" },
2860 { SVM_EXIT_READ_CR4
, "read_cr4" },
2861 { SVM_EXIT_READ_CR8
, "read_cr8" },
2862 { SVM_EXIT_WRITE_CR0
, "write_cr0" },
2863 { SVM_EXIT_WRITE_CR3
, "write_cr3" },
2864 { SVM_EXIT_WRITE_CR4
, "write_cr4" },
2865 { SVM_EXIT_WRITE_CR8
, "write_cr8" },
2866 { SVM_EXIT_READ_DR0
, "read_dr0" },
2867 { SVM_EXIT_READ_DR1
, "read_dr1" },
2868 { SVM_EXIT_READ_DR2
, "read_dr2" },
2869 { SVM_EXIT_READ_DR3
, "read_dr3" },
2870 { SVM_EXIT_WRITE_DR0
, "write_dr0" },
2871 { SVM_EXIT_WRITE_DR1
, "write_dr1" },
2872 { SVM_EXIT_WRITE_DR2
, "write_dr2" },
2873 { SVM_EXIT_WRITE_DR3
, "write_dr3" },
2874 { SVM_EXIT_WRITE_DR5
, "write_dr5" },
2875 { SVM_EXIT_WRITE_DR7
, "write_dr7" },
2876 { SVM_EXIT_EXCP_BASE
+ DB_VECTOR
, "DB excp" },
2877 { SVM_EXIT_EXCP_BASE
+ BP_VECTOR
, "BP excp" },
2878 { SVM_EXIT_EXCP_BASE
+ UD_VECTOR
, "UD excp" },
2879 { SVM_EXIT_EXCP_BASE
+ PF_VECTOR
, "PF excp" },
2880 { SVM_EXIT_EXCP_BASE
+ NM_VECTOR
, "NM excp" },
2881 { SVM_EXIT_EXCP_BASE
+ MC_VECTOR
, "MC excp" },
2882 { SVM_EXIT_INTR
, "interrupt" },
2883 { SVM_EXIT_NMI
, "nmi" },
2884 { SVM_EXIT_SMI
, "smi" },
2885 { SVM_EXIT_INIT
, "init" },
2886 { SVM_EXIT_VINTR
, "vintr" },
2887 { SVM_EXIT_CPUID
, "cpuid" },
2888 { SVM_EXIT_INVD
, "invd" },
2889 { SVM_EXIT_HLT
, "hlt" },
2890 { SVM_EXIT_INVLPG
, "invlpg" },
2891 { SVM_EXIT_INVLPGA
, "invlpga" },
2892 { SVM_EXIT_IOIO
, "io" },
2893 { SVM_EXIT_MSR
, "msr" },
2894 { SVM_EXIT_TASK_SWITCH
, "task_switch" },
2895 { SVM_EXIT_SHUTDOWN
, "shutdown" },
2896 { SVM_EXIT_VMRUN
, "vmrun" },
2897 { SVM_EXIT_VMMCALL
, "hypercall" },
2898 { SVM_EXIT_VMLOAD
, "vmload" },
2899 { SVM_EXIT_VMSAVE
, "vmsave" },
2900 { SVM_EXIT_STGI
, "stgi" },
2901 { SVM_EXIT_CLGI
, "clgi" },
2902 { SVM_EXIT_SKINIT
, "skinit" },
2903 { SVM_EXIT_WBINVD
, "wbinvd" },
2904 { SVM_EXIT_MONITOR
, "monitor" },
2905 { SVM_EXIT_MWAIT
, "mwait" },
2906 { SVM_EXIT_NPF
, "npf" },
2910 static bool svm_gb_page_enable(void)
2915 static struct kvm_x86_ops svm_x86_ops
= {
2916 .cpu_has_kvm_support
= has_svm
,
2917 .disabled_by_bios
= is_disabled
,
2918 .hardware_setup
= svm_hardware_setup
,
2919 .hardware_unsetup
= svm_hardware_unsetup
,
2920 .check_processor_compatibility
= svm_check_processor_compat
,
2921 .hardware_enable
= svm_hardware_enable
,
2922 .hardware_disable
= svm_hardware_disable
,
2923 .cpu_has_accelerated_tpr
= svm_cpu_has_accelerated_tpr
,
2925 .vcpu_create
= svm_create_vcpu
,
2926 .vcpu_free
= svm_free_vcpu
,
2927 .vcpu_reset
= svm_vcpu_reset
,
2929 .prepare_guest_switch
= svm_prepare_guest_switch
,
2930 .vcpu_load
= svm_vcpu_load
,
2931 .vcpu_put
= svm_vcpu_put
,
2933 .set_guest_debug
= svm_guest_debug
,
2934 .get_msr
= svm_get_msr
,
2935 .set_msr
= svm_set_msr
,
2936 .get_segment_base
= svm_get_segment_base
,
2937 .get_segment
= svm_get_segment
,
2938 .set_segment
= svm_set_segment
,
2939 .get_cpl
= svm_get_cpl
,
2940 .get_cs_db_l_bits
= kvm_get_cs_db_l_bits
,
2941 .decache_cr4_guest_bits
= svm_decache_cr4_guest_bits
,
2942 .set_cr0
= svm_set_cr0
,
2943 .set_cr3
= svm_set_cr3
,
2944 .set_cr4
= svm_set_cr4
,
2945 .set_efer
= svm_set_efer
,
2946 .get_idt
= svm_get_idt
,
2947 .set_idt
= svm_set_idt
,
2948 .get_gdt
= svm_get_gdt
,
2949 .set_gdt
= svm_set_gdt
,
2950 .get_dr
= svm_get_dr
,
2951 .set_dr
= svm_set_dr
,
2952 .cache_reg
= svm_cache_reg
,
2953 .get_rflags
= svm_get_rflags
,
2954 .set_rflags
= svm_set_rflags
,
2956 .tlb_flush
= svm_flush_tlb
,
2958 .run
= svm_vcpu_run
,
2959 .handle_exit
= handle_exit
,
2960 .skip_emulated_instruction
= skip_emulated_instruction
,
2961 .set_interrupt_shadow
= svm_set_interrupt_shadow
,
2962 .get_interrupt_shadow
= svm_get_interrupt_shadow
,
2963 .patch_hypercall
= svm_patch_hypercall
,
2964 .set_irq
= svm_set_irq
,
2965 .set_nmi
= svm_inject_nmi
,
2966 .queue_exception
= svm_queue_exception
,
2967 .interrupt_allowed
= svm_interrupt_allowed
,
2968 .nmi_allowed
= svm_nmi_allowed
,
2969 .enable_nmi_window
= enable_nmi_window
,
2970 .enable_irq_window
= enable_irq_window
,
2971 .update_cr8_intercept
= update_cr8_intercept
,
2973 .set_tss_addr
= svm_set_tss_addr
,
2974 .get_tdp_level
= get_npt_level
,
2975 .get_mt_mask
= svm_get_mt_mask
,
2977 .exit_reasons_str
= svm_exit_reasons_str
,
2978 .gb_page_enable
= svm_gb_page_enable
,
2981 static int __init
svm_init(void)
2983 return kvm_init(&svm_x86_ops
, sizeof(struct vcpu_svm
),
2987 static void __exit
svm_exit(void)
2992 module_init(svm_init
)
2993 module_exit(svm_exit
)