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)
49 #define SVM_FEATURE_PAUSE_FILTER (1 << 10)
51 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
52 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
53 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
55 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
57 static const u32 host_save_user_msrs
[] = {
59 MSR_STAR
, MSR_LSTAR
, MSR_CSTAR
, MSR_SYSCALL_MASK
, MSR_KERNEL_GS_BASE
,
62 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
65 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
74 /* These are the merged vectors */
77 /* gpa pointers to the real vectors */
80 /* A VMEXIT is required but not yet emulated */
83 /* cache for intercepts of the guest */
84 u16 intercept_cr_read
;
85 u16 intercept_cr_write
;
86 u16 intercept_dr_read
;
87 u16 intercept_dr_write
;
88 u32 intercept_exceptions
;
96 unsigned long vmcb_pa
;
97 struct svm_cpu_data
*svm_data
;
98 uint64_t asid_generation
;
99 uint64_t sysenter_esp
;
100 uint64_t sysenter_eip
;
104 u64 host_user_msrs
[NR_HOST_SAVE_USER_MSRS
];
109 struct nested_state nested
;
114 /* enable NPT for AMD64 and X86 with PAE */
115 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
116 static bool npt_enabled
= true;
118 static bool npt_enabled
= false;
122 module_param(npt
, int, S_IRUGO
);
124 static int nested
= 1;
125 module_param(nested
, int, S_IRUGO
);
127 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
);
128 static void svm_complete_interrupts(struct vcpu_svm
*svm
);
130 static int nested_svm_exit_handled(struct vcpu_svm
*svm
);
131 static int nested_svm_vmexit(struct vcpu_svm
*svm
);
132 static int nested_svm_check_exception(struct vcpu_svm
*svm
, unsigned nr
,
133 bool has_error_code
, u32 error_code
);
135 static inline struct vcpu_svm
*to_svm(struct kvm_vcpu
*vcpu
)
137 return container_of(vcpu
, struct vcpu_svm
, vcpu
);
140 static inline bool is_nested(struct vcpu_svm
*svm
)
142 return svm
->nested
.vmcb
;
145 static inline void enable_gif(struct vcpu_svm
*svm
)
147 svm
->vcpu
.arch
.hflags
|= HF_GIF_MASK
;
150 static inline void disable_gif(struct vcpu_svm
*svm
)
152 svm
->vcpu
.arch
.hflags
&= ~HF_GIF_MASK
;
155 static inline bool gif_set(struct vcpu_svm
*svm
)
157 return !!(svm
->vcpu
.arch
.hflags
& HF_GIF_MASK
);
160 static unsigned long iopm_base
;
162 struct kvm_ldttss_desc
{
165 unsigned base1
: 8, type
: 5, dpl
: 2, p
: 1;
166 unsigned limit1
: 4, zero0
: 3, g
: 1, base2
: 8;
169 } __attribute__((packed
));
171 struct svm_cpu_data
{
177 struct kvm_ldttss_desc
*tss_desc
;
179 struct page
*save_area
;
182 static DEFINE_PER_CPU(struct svm_cpu_data
*, svm_data
);
183 static uint32_t svm_features
;
185 struct svm_init_data
{
190 static u32 msrpm_ranges
[] = {0, 0xc0000000, 0xc0010000};
192 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
193 #define MSRS_RANGE_SIZE 2048
194 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
196 #define MAX_INST_SIZE 15
198 static inline u32
svm_has(u32 feat
)
200 return svm_features
& feat
;
203 static inline void clgi(void)
205 asm volatile (__ex(SVM_CLGI
));
208 static inline void stgi(void)
210 asm volatile (__ex(SVM_STGI
));
213 static inline void invlpga(unsigned long addr
, u32 asid
)
215 asm volatile (__ex(SVM_INVLPGA
) :: "a"(addr
), "c"(asid
));
218 static inline void force_new_asid(struct kvm_vcpu
*vcpu
)
220 to_svm(vcpu
)->asid_generation
--;
223 static inline void flush_guest_tlb(struct kvm_vcpu
*vcpu
)
225 force_new_asid(vcpu
);
228 static void svm_set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
230 if (!npt_enabled
&& !(efer
& EFER_LMA
))
233 to_svm(vcpu
)->vmcb
->save
.efer
= efer
| EFER_SVME
;
234 vcpu
->arch
.efer
= efer
;
237 static void svm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
,
238 bool has_error_code
, u32 error_code
)
240 struct vcpu_svm
*svm
= to_svm(vcpu
);
242 /* If we are within a nested VM we'd better #VMEXIT and let the
243 guest handle the exception */
244 if (nested_svm_check_exception(svm
, nr
, has_error_code
, error_code
))
247 svm
->vmcb
->control
.event_inj
= nr
249 | (has_error_code
? SVM_EVTINJ_VALID_ERR
: 0)
250 | SVM_EVTINJ_TYPE_EXEPT
;
251 svm
->vmcb
->control
.event_inj_err
= error_code
;
254 static int is_external_interrupt(u32 info
)
256 info
&= SVM_EVTINJ_TYPE_MASK
| SVM_EVTINJ_VALID
;
257 return info
== (SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
);
260 static u32
svm_get_interrupt_shadow(struct kvm_vcpu
*vcpu
, int mask
)
262 struct vcpu_svm
*svm
= to_svm(vcpu
);
265 if (svm
->vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
)
266 ret
|= X86_SHADOW_INT_STI
| X86_SHADOW_INT_MOV_SS
;
270 static void svm_set_interrupt_shadow(struct kvm_vcpu
*vcpu
, int mask
)
272 struct vcpu_svm
*svm
= to_svm(vcpu
);
275 svm
->vmcb
->control
.int_state
&= ~SVM_INTERRUPT_SHADOW_MASK
;
277 svm
->vmcb
->control
.int_state
|= SVM_INTERRUPT_SHADOW_MASK
;
281 static void skip_emulated_instruction(struct kvm_vcpu
*vcpu
)
283 struct vcpu_svm
*svm
= to_svm(vcpu
);
285 if (!svm
->next_rip
) {
286 if (emulate_instruction(vcpu
, 0, 0, EMULTYPE_SKIP
) !=
288 printk(KERN_DEBUG
"%s: NOP\n", __func__
);
291 if (svm
->next_rip
- kvm_rip_read(vcpu
) > MAX_INST_SIZE
)
292 printk(KERN_ERR
"%s: ip 0x%lx next 0x%llx\n",
293 __func__
, kvm_rip_read(vcpu
), svm
->next_rip
);
295 kvm_rip_write(vcpu
, svm
->next_rip
);
296 svm_set_interrupt_shadow(vcpu
, 0);
299 static int has_svm(void)
303 if (!cpu_has_svm(&msg
)) {
304 printk(KERN_INFO
"has_svm: %s\n", msg
);
311 static void svm_hardware_disable(void *garbage
)
316 static int svm_hardware_enable(void *garbage
)
319 struct svm_cpu_data
*sd
;
321 struct descriptor_table gdt_descr
;
322 struct desc_struct
*gdt
;
323 int me
= raw_smp_processor_id();
325 rdmsrl(MSR_EFER
, efer
);
326 if (efer
& EFER_SVME
)
330 printk(KERN_ERR
"svm_hardware_enable: err EOPNOTSUPP on %d\n",
334 sd
= per_cpu(svm_data
, me
);
337 printk(KERN_ERR
"svm_hardware_enable: svm_data is NULL on %d\n",
342 sd
->asid_generation
= 1;
343 sd
->max_asid
= cpuid_ebx(SVM_CPUID_FUNC
) - 1;
344 sd
->next_asid
= sd
->max_asid
+ 1;
346 kvm_get_gdt(&gdt_descr
);
347 gdt
= (struct desc_struct
*)gdt_descr
.base
;
348 sd
->tss_desc
= (struct kvm_ldttss_desc
*)(gdt
+ GDT_ENTRY_TSS
);
350 wrmsrl(MSR_EFER
, efer
| EFER_SVME
);
352 wrmsrl(MSR_VM_HSAVE_PA
, page_to_pfn(sd
->save_area
) << PAGE_SHIFT
);
357 static void svm_cpu_uninit(int cpu
)
359 struct svm_cpu_data
*sd
= per_cpu(svm_data
, raw_smp_processor_id());
364 per_cpu(svm_data
, raw_smp_processor_id()) = NULL
;
365 __free_page(sd
->save_area
);
369 static int svm_cpu_init(int cpu
)
371 struct svm_cpu_data
*sd
;
374 sd
= kzalloc(sizeof(struct svm_cpu_data
), GFP_KERNEL
);
378 sd
->save_area
= alloc_page(GFP_KERNEL
);
383 per_cpu(svm_data
, cpu
) = sd
;
393 static void set_msr_interception(u32
*msrpm
, unsigned msr
,
398 for (i
= 0; i
< NUM_MSR_MAPS
; i
++) {
399 if (msr
>= msrpm_ranges
[i
] &&
400 msr
< msrpm_ranges
[i
] + MSRS_IN_RANGE
) {
401 u32 msr_offset
= (i
* MSRS_IN_RANGE
+ msr
-
402 msrpm_ranges
[i
]) * 2;
404 u32
*base
= msrpm
+ (msr_offset
/ 32);
405 u32 msr_shift
= msr_offset
% 32;
406 u32 mask
= ((write
) ? 0 : 2) | ((read
) ? 0 : 1);
407 *base
= (*base
& ~(0x3 << msr_shift
)) |
415 static void svm_vcpu_init_msrpm(u32
*msrpm
)
417 memset(msrpm
, 0xff, PAGE_SIZE
* (1 << MSRPM_ALLOC_ORDER
));
420 set_msr_interception(msrpm
, MSR_GS_BASE
, 1, 1);
421 set_msr_interception(msrpm
, MSR_FS_BASE
, 1, 1);
422 set_msr_interception(msrpm
, MSR_KERNEL_GS_BASE
, 1, 1);
423 set_msr_interception(msrpm
, MSR_LSTAR
, 1, 1);
424 set_msr_interception(msrpm
, MSR_CSTAR
, 1, 1);
425 set_msr_interception(msrpm
, MSR_SYSCALL_MASK
, 1, 1);
427 set_msr_interception(msrpm
, MSR_K6_STAR
, 1, 1);
428 set_msr_interception(msrpm
, MSR_IA32_SYSENTER_CS
, 1, 1);
431 static void svm_enable_lbrv(struct vcpu_svm
*svm
)
433 u32
*msrpm
= svm
->msrpm
;
435 svm
->vmcb
->control
.lbr_ctl
= 1;
436 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHFROMIP
, 1, 1);
437 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHTOIP
, 1, 1);
438 set_msr_interception(msrpm
, MSR_IA32_LASTINTFROMIP
, 1, 1);
439 set_msr_interception(msrpm
, MSR_IA32_LASTINTTOIP
, 1, 1);
442 static void svm_disable_lbrv(struct vcpu_svm
*svm
)
444 u32
*msrpm
= svm
->msrpm
;
446 svm
->vmcb
->control
.lbr_ctl
= 0;
447 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHFROMIP
, 0, 0);
448 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHTOIP
, 0, 0);
449 set_msr_interception(msrpm
, MSR_IA32_LASTINTFROMIP
, 0, 0);
450 set_msr_interception(msrpm
, MSR_IA32_LASTINTTOIP
, 0, 0);
453 static __init
int svm_hardware_setup(void)
456 struct page
*iopm_pages
;
460 iopm_pages
= alloc_pages(GFP_KERNEL
, IOPM_ALLOC_ORDER
);
465 iopm_va
= page_address(iopm_pages
);
466 memset(iopm_va
, 0xff, PAGE_SIZE
* (1 << IOPM_ALLOC_ORDER
));
467 iopm_base
= page_to_pfn(iopm_pages
) << PAGE_SHIFT
;
469 if (boot_cpu_has(X86_FEATURE_NX
))
470 kvm_enable_efer_bits(EFER_NX
);
472 if (boot_cpu_has(X86_FEATURE_FXSR_OPT
))
473 kvm_enable_efer_bits(EFER_FFXSR
);
476 printk(KERN_INFO
"kvm: Nested Virtualization enabled\n");
477 kvm_enable_efer_bits(EFER_SVME
);
480 for_each_possible_cpu(cpu
) {
481 r
= svm_cpu_init(cpu
);
486 svm_features
= cpuid_edx(SVM_CPUID_FUNC
);
488 if (!svm_has(SVM_FEATURE_NPT
))
491 if (npt_enabled
&& !npt
) {
492 printk(KERN_INFO
"kvm: Nested Paging disabled\n");
497 printk(KERN_INFO
"kvm: Nested Paging enabled\n");
505 __free_pages(iopm_pages
, IOPM_ALLOC_ORDER
);
510 static __exit
void svm_hardware_unsetup(void)
514 for_each_possible_cpu(cpu
)
517 __free_pages(pfn_to_page(iopm_base
>> PAGE_SHIFT
), IOPM_ALLOC_ORDER
);
521 static void init_seg(struct vmcb_seg
*seg
)
524 seg
->attrib
= SVM_SELECTOR_P_MASK
| SVM_SELECTOR_S_MASK
|
525 SVM_SELECTOR_WRITE_MASK
; /* Read/Write Data Segment */
530 static void init_sys_seg(struct vmcb_seg
*seg
, uint32_t type
)
533 seg
->attrib
= SVM_SELECTOR_P_MASK
| type
;
538 static void init_vmcb(struct vcpu_svm
*svm
)
540 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
541 struct vmcb_save_area
*save
= &svm
->vmcb
->save
;
543 svm
->vcpu
.fpu_active
= 1;
545 control
->intercept_cr_read
= INTERCEPT_CR0_MASK
|
549 control
->intercept_cr_write
= INTERCEPT_CR0_MASK
|
554 control
->intercept_dr_read
= INTERCEPT_DR0_MASK
|
563 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_SELECTIVE_CR0
) |
581 (1ULL << INTERCEPT_CPUID
) |
582 (1ULL << INTERCEPT_INVD
) |
583 (1ULL << INTERCEPT_HLT
) |
584 (1ULL << INTERCEPT_INVLPG
) |
585 (1ULL << INTERCEPT_INVLPGA
) |
586 (1ULL << INTERCEPT_IOIO_PROT
) |
587 (1ULL << INTERCEPT_MSR_PROT
) |
588 (1ULL << INTERCEPT_TASK_SWITCH
) |
589 (1ULL << INTERCEPT_SHUTDOWN
) |
590 (1ULL << INTERCEPT_VMRUN
) |
591 (1ULL << INTERCEPT_VMMCALL
) |
592 (1ULL << INTERCEPT_VMLOAD
) |
593 (1ULL << INTERCEPT_VMSAVE
) |
594 (1ULL << INTERCEPT_STGI
) |
595 (1ULL << INTERCEPT_CLGI
) |
596 (1ULL << INTERCEPT_SKINIT
) |
597 (1ULL << INTERCEPT_WBINVD
) |
598 (1ULL << INTERCEPT_MONITOR
) |
599 (1ULL << INTERCEPT_MWAIT
);
601 control
->iopm_base_pa
= iopm_base
;
602 control
->msrpm_base_pa
= __pa(svm
->msrpm
);
603 control
->tsc_offset
= 0;
604 control
->int_ctl
= V_INTR_MASKING_MASK
;
612 save
->cs
.selector
= 0xf000;
613 /* Executable/Readable Code Segment */
614 save
->cs
.attrib
= SVM_SELECTOR_READ_MASK
| SVM_SELECTOR_P_MASK
|
615 SVM_SELECTOR_S_MASK
| SVM_SELECTOR_CODE_MASK
;
616 save
->cs
.limit
= 0xffff;
618 * cs.base should really be 0xffff0000, but vmx can't handle that, so
619 * be consistent with it.
621 * Replace when we have real mode working for vmx.
623 save
->cs
.base
= 0xf0000;
625 save
->gdtr
.limit
= 0xffff;
626 save
->idtr
.limit
= 0xffff;
628 init_sys_seg(&save
->ldtr
, SEG_TYPE_LDT
);
629 init_sys_seg(&save
->tr
, SEG_TYPE_BUSY_TSS16
);
631 save
->efer
= EFER_SVME
;
632 save
->dr6
= 0xffff0ff0;
635 save
->rip
= 0x0000fff0;
636 svm
->vcpu
.arch
.regs
[VCPU_REGS_RIP
] = save
->rip
;
638 /* This is the guest-visible cr0 value.
639 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
641 svm
->vcpu
.arch
.cr0
= X86_CR0_NW
| X86_CR0_CD
| X86_CR0_ET
;
642 kvm_set_cr0(&svm
->vcpu
, svm
->vcpu
.arch
.cr0
);
644 save
->cr4
= X86_CR4_PAE
;
648 /* Setup VMCB for Nested Paging */
649 control
->nested_ctl
= 1;
650 control
->intercept
&= ~((1ULL << INTERCEPT_TASK_SWITCH
) |
651 (1ULL << INTERCEPT_INVLPG
));
652 control
->intercept_exceptions
&= ~(1 << PF_VECTOR
);
653 control
->intercept_cr_read
&= ~INTERCEPT_CR3_MASK
;
654 control
->intercept_cr_write
&= ~INTERCEPT_CR3_MASK
;
655 save
->g_pat
= 0x0007040600070406ULL
;
659 force_new_asid(&svm
->vcpu
);
661 svm
->nested
.vmcb
= 0;
662 svm
->vcpu
.arch
.hflags
= 0;
664 if (svm_has(SVM_FEATURE_PAUSE_FILTER
)) {
665 control
->pause_filter_count
= 3000;
666 control
->intercept
|= (1ULL << INTERCEPT_PAUSE
);
672 static int svm_vcpu_reset(struct kvm_vcpu
*vcpu
)
674 struct vcpu_svm
*svm
= to_svm(vcpu
);
678 if (!kvm_vcpu_is_bsp(vcpu
)) {
679 kvm_rip_write(vcpu
, 0);
680 svm
->vmcb
->save
.cs
.base
= svm
->vcpu
.arch
.sipi_vector
<< 12;
681 svm
->vmcb
->save
.cs
.selector
= svm
->vcpu
.arch
.sipi_vector
<< 8;
683 vcpu
->arch
.regs_avail
= ~0;
684 vcpu
->arch
.regs_dirty
= ~0;
689 static struct kvm_vcpu
*svm_create_vcpu(struct kvm
*kvm
, unsigned int id
)
691 struct vcpu_svm
*svm
;
693 struct page
*msrpm_pages
;
694 struct page
*hsave_page
;
695 struct page
*nested_msrpm_pages
;
698 svm
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
704 err
= kvm_vcpu_init(&svm
->vcpu
, kvm
, id
);
708 page
= alloc_page(GFP_KERNEL
);
715 msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
719 nested_msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
720 if (!nested_msrpm_pages
)
723 svm
->msrpm
= page_address(msrpm_pages
);
724 svm_vcpu_init_msrpm(svm
->msrpm
);
726 hsave_page
= alloc_page(GFP_KERNEL
);
729 svm
->nested
.hsave
= page_address(hsave_page
);
731 svm
->nested
.msrpm
= page_address(nested_msrpm_pages
);
733 svm
->vmcb
= page_address(page
);
734 clear_page(svm
->vmcb
);
735 svm
->vmcb_pa
= page_to_pfn(page
) << PAGE_SHIFT
;
736 svm
->asid_generation
= 0;
740 svm
->vcpu
.arch
.apic_base
= 0xfee00000 | MSR_IA32_APICBASE_ENABLE
;
741 if (kvm_vcpu_is_bsp(&svm
->vcpu
))
742 svm
->vcpu
.arch
.apic_base
|= MSR_IA32_APICBASE_BSP
;
747 kvm_vcpu_uninit(&svm
->vcpu
);
749 kmem_cache_free(kvm_vcpu_cache
, svm
);
754 static void svm_free_vcpu(struct kvm_vcpu
*vcpu
)
756 struct vcpu_svm
*svm
= to_svm(vcpu
);
758 __free_page(pfn_to_page(svm
->vmcb_pa
>> PAGE_SHIFT
));
759 __free_pages(virt_to_page(svm
->msrpm
), MSRPM_ALLOC_ORDER
);
760 __free_page(virt_to_page(svm
->nested
.hsave
));
761 __free_pages(virt_to_page(svm
->nested
.msrpm
), MSRPM_ALLOC_ORDER
);
762 kvm_vcpu_uninit(vcpu
);
763 kmem_cache_free(kvm_vcpu_cache
, svm
);
766 static void svm_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
768 struct vcpu_svm
*svm
= to_svm(vcpu
);
771 if (unlikely(cpu
!= vcpu
->cpu
)) {
774 if (check_tsc_unstable()) {
776 * Make sure that the guest sees a monotonically
779 delta
= vcpu
->arch
.host_tsc
- native_read_tsc();
780 svm
->vmcb
->control
.tsc_offset
+= delta
;
782 svm
->nested
.hsave
->control
.tsc_offset
+= delta
;
785 kvm_migrate_timers(vcpu
);
786 svm
->asid_generation
= 0;
789 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
790 rdmsrl(host_save_user_msrs
[i
], svm
->host_user_msrs
[i
]);
793 static void svm_vcpu_put(struct kvm_vcpu
*vcpu
)
795 struct vcpu_svm
*svm
= to_svm(vcpu
);
798 ++vcpu
->stat
.host_state_reload
;
799 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
800 wrmsrl(host_save_user_msrs
[i
], svm
->host_user_msrs
[i
]);
802 vcpu
->arch
.host_tsc
= native_read_tsc();
805 static unsigned long svm_get_rflags(struct kvm_vcpu
*vcpu
)
807 return to_svm(vcpu
)->vmcb
->save
.rflags
;
810 static void svm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
812 to_svm(vcpu
)->vmcb
->save
.rflags
= rflags
;
815 static void svm_cache_reg(struct kvm_vcpu
*vcpu
, enum kvm_reg reg
)
818 case VCPU_EXREG_PDPTR
:
819 BUG_ON(!npt_enabled
);
820 load_pdptrs(vcpu
, vcpu
->arch
.cr3
);
827 static void svm_set_vintr(struct vcpu_svm
*svm
)
829 svm
->vmcb
->control
.intercept
|= 1ULL << INTERCEPT_VINTR
;
832 static void svm_clear_vintr(struct vcpu_svm
*svm
)
834 svm
->vmcb
->control
.intercept
&= ~(1ULL << INTERCEPT_VINTR
);
837 static struct vmcb_seg
*svm_seg(struct kvm_vcpu
*vcpu
, int seg
)
839 struct vmcb_save_area
*save
= &to_svm(vcpu
)->vmcb
->save
;
842 case VCPU_SREG_CS
: return &save
->cs
;
843 case VCPU_SREG_DS
: return &save
->ds
;
844 case VCPU_SREG_ES
: return &save
->es
;
845 case VCPU_SREG_FS
: return &save
->fs
;
846 case VCPU_SREG_GS
: return &save
->gs
;
847 case VCPU_SREG_SS
: return &save
->ss
;
848 case VCPU_SREG_TR
: return &save
->tr
;
849 case VCPU_SREG_LDTR
: return &save
->ldtr
;
855 static u64
svm_get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
857 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
862 static void svm_get_segment(struct kvm_vcpu
*vcpu
,
863 struct kvm_segment
*var
, int seg
)
865 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
868 var
->limit
= s
->limit
;
869 var
->selector
= s
->selector
;
870 var
->type
= s
->attrib
& SVM_SELECTOR_TYPE_MASK
;
871 var
->s
= (s
->attrib
>> SVM_SELECTOR_S_SHIFT
) & 1;
872 var
->dpl
= (s
->attrib
>> SVM_SELECTOR_DPL_SHIFT
) & 3;
873 var
->present
= (s
->attrib
>> SVM_SELECTOR_P_SHIFT
) & 1;
874 var
->avl
= (s
->attrib
>> SVM_SELECTOR_AVL_SHIFT
) & 1;
875 var
->l
= (s
->attrib
>> SVM_SELECTOR_L_SHIFT
) & 1;
876 var
->db
= (s
->attrib
>> SVM_SELECTOR_DB_SHIFT
) & 1;
877 var
->g
= (s
->attrib
>> SVM_SELECTOR_G_SHIFT
) & 1;
879 /* AMD's VMCB does not have an explicit unusable field, so emulate it
880 * for cross vendor migration purposes by "not present"
882 var
->unusable
= !var
->present
|| (var
->type
== 0);
887 * SVM always stores 0 for the 'G' bit in the CS selector in
888 * the VMCB on a VMEXIT. This hurts cross-vendor migration:
889 * Intel's VMENTRY has a check on the 'G' bit.
891 var
->g
= s
->limit
> 0xfffff;
895 * Work around a bug where the busy flag in the tr selector
905 * The accessed bit must always be set in the segment
906 * descriptor cache, although it can be cleared in the
907 * descriptor, the cached bit always remains at 1. Since
908 * Intel has a check on this, set it here to support
909 * cross-vendor migration.
915 /* On AMD CPUs sometimes the DB bit in the segment
916 * descriptor is left as 1, although the whole segment has
917 * been made unusable. Clear it here to pass an Intel VMX
918 * entry check when cross vendor migrating.
926 static int svm_get_cpl(struct kvm_vcpu
*vcpu
)
928 struct vmcb_save_area
*save
= &to_svm(vcpu
)->vmcb
->save
;
933 static void svm_get_idt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
935 struct vcpu_svm
*svm
= to_svm(vcpu
);
937 dt
->limit
= svm
->vmcb
->save
.idtr
.limit
;
938 dt
->base
= svm
->vmcb
->save
.idtr
.base
;
941 static void svm_set_idt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
943 struct vcpu_svm
*svm
= to_svm(vcpu
);
945 svm
->vmcb
->save
.idtr
.limit
= dt
->limit
;
946 svm
->vmcb
->save
.idtr
.base
= dt
->base
;
949 static void svm_get_gdt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
951 struct vcpu_svm
*svm
= to_svm(vcpu
);
953 dt
->limit
= svm
->vmcb
->save
.gdtr
.limit
;
954 dt
->base
= svm
->vmcb
->save
.gdtr
.base
;
957 static void svm_set_gdt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
959 struct vcpu_svm
*svm
= to_svm(vcpu
);
961 svm
->vmcb
->save
.gdtr
.limit
= dt
->limit
;
962 svm
->vmcb
->save
.gdtr
.base
= dt
->base
;
965 static void svm_decache_cr0_guest_bits(struct kvm_vcpu
*vcpu
)
969 static void svm_decache_cr4_guest_bits(struct kvm_vcpu
*vcpu
)
973 static void update_cr0_intercept(struct vcpu_svm
*svm
)
975 ulong gcr0
= svm
->vcpu
.arch
.cr0
;
976 u64
*hcr0
= &svm
->vmcb
->save
.cr0
;
978 if (!svm
->vcpu
.fpu_active
)
979 *hcr0
|= SVM_CR0_SELECTIVE_MASK
;
981 *hcr0
= (*hcr0
& ~SVM_CR0_SELECTIVE_MASK
)
982 | (gcr0
& SVM_CR0_SELECTIVE_MASK
);
985 if (gcr0
== *hcr0
&& svm
->vcpu
.fpu_active
) {
986 svm
->vmcb
->control
.intercept_cr_read
&= ~INTERCEPT_CR0_MASK
;
987 svm
->vmcb
->control
.intercept_cr_write
&= ~INTERCEPT_CR0_MASK
;
989 svm
->vmcb
->control
.intercept_cr_read
|= INTERCEPT_CR0_MASK
;
990 svm
->vmcb
->control
.intercept_cr_write
|= INTERCEPT_CR0_MASK
;
994 static void svm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
996 struct vcpu_svm
*svm
= to_svm(vcpu
);
999 if (vcpu
->arch
.efer
& EFER_LME
) {
1000 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
1001 vcpu
->arch
.efer
|= EFER_LMA
;
1002 svm
->vmcb
->save
.efer
|= EFER_LMA
| EFER_LME
;
1005 if (is_paging(vcpu
) && !(cr0
& X86_CR0_PG
)) {
1006 vcpu
->arch
.efer
&= ~EFER_LMA
;
1007 svm
->vmcb
->save
.efer
&= ~(EFER_LMA
| EFER_LME
);
1011 vcpu
->arch
.cr0
= cr0
;
1014 cr0
|= X86_CR0_PG
| X86_CR0_WP
;
1016 if (!vcpu
->fpu_active
)
1019 * re-enable caching here because the QEMU bios
1020 * does not do it - this results in some delay at
1023 cr0
&= ~(X86_CR0_CD
| X86_CR0_NW
);
1024 svm
->vmcb
->save
.cr0
= cr0
;
1025 update_cr0_intercept(svm
);
1028 static void svm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
1030 unsigned long host_cr4_mce
= read_cr4() & X86_CR4_MCE
;
1031 unsigned long old_cr4
= to_svm(vcpu
)->vmcb
->save
.cr4
;
1033 if (npt_enabled
&& ((old_cr4
^ cr4
) & X86_CR4_PGE
))
1034 force_new_asid(vcpu
);
1036 vcpu
->arch
.cr4
= cr4
;
1039 cr4
|= host_cr4_mce
;
1040 to_svm(vcpu
)->vmcb
->save
.cr4
= cr4
;
1043 static void svm_set_segment(struct kvm_vcpu
*vcpu
,
1044 struct kvm_segment
*var
, int seg
)
1046 struct vcpu_svm
*svm
= to_svm(vcpu
);
1047 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
1049 s
->base
= var
->base
;
1050 s
->limit
= var
->limit
;
1051 s
->selector
= var
->selector
;
1055 s
->attrib
= (var
->type
& SVM_SELECTOR_TYPE_MASK
);
1056 s
->attrib
|= (var
->s
& 1) << SVM_SELECTOR_S_SHIFT
;
1057 s
->attrib
|= (var
->dpl
& 3) << SVM_SELECTOR_DPL_SHIFT
;
1058 s
->attrib
|= (var
->present
& 1) << SVM_SELECTOR_P_SHIFT
;
1059 s
->attrib
|= (var
->avl
& 1) << SVM_SELECTOR_AVL_SHIFT
;
1060 s
->attrib
|= (var
->l
& 1) << SVM_SELECTOR_L_SHIFT
;
1061 s
->attrib
|= (var
->db
& 1) << SVM_SELECTOR_DB_SHIFT
;
1062 s
->attrib
|= (var
->g
& 1) << SVM_SELECTOR_G_SHIFT
;
1064 if (seg
== VCPU_SREG_CS
)
1066 = (svm
->vmcb
->save
.cs
.attrib
1067 >> SVM_SELECTOR_DPL_SHIFT
) & 3;
1071 static void update_db_intercept(struct kvm_vcpu
*vcpu
)
1073 struct vcpu_svm
*svm
= to_svm(vcpu
);
1075 svm
->vmcb
->control
.intercept_exceptions
&=
1076 ~((1 << DB_VECTOR
) | (1 << BP_VECTOR
));
1078 if (svm
->nmi_singlestep
)
1079 svm
->vmcb
->control
.intercept_exceptions
|= (1 << DB_VECTOR
);
1081 if (vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
) {
1082 if (vcpu
->guest_debug
&
1083 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
))
1084 svm
->vmcb
->control
.intercept_exceptions
|=
1086 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_SW_BP
)
1087 svm
->vmcb
->control
.intercept_exceptions
|=
1090 vcpu
->guest_debug
= 0;
1093 static void svm_guest_debug(struct kvm_vcpu
*vcpu
, struct kvm_guest_debug
*dbg
)
1095 struct vcpu_svm
*svm
= to_svm(vcpu
);
1097 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
1098 svm
->vmcb
->save
.dr7
= dbg
->arch
.debugreg
[7];
1100 svm
->vmcb
->save
.dr7
= vcpu
->arch
.dr7
;
1102 update_db_intercept(vcpu
);
1105 static void load_host_msrs(struct kvm_vcpu
*vcpu
)
1107 #ifdef CONFIG_X86_64
1108 wrmsrl(MSR_GS_BASE
, to_svm(vcpu
)->host_gs_base
);
1112 static void save_host_msrs(struct kvm_vcpu
*vcpu
)
1114 #ifdef CONFIG_X86_64
1115 rdmsrl(MSR_GS_BASE
, to_svm(vcpu
)->host_gs_base
);
1119 static void new_asid(struct vcpu_svm
*svm
, struct svm_cpu_data
*sd
)
1121 if (sd
->next_asid
> sd
->max_asid
) {
1122 ++sd
->asid_generation
;
1124 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ALL_ASID
;
1127 svm
->asid_generation
= sd
->asid_generation
;
1128 svm
->vmcb
->control
.asid
= sd
->next_asid
++;
1131 static int svm_get_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long *dest
)
1133 struct vcpu_svm
*svm
= to_svm(vcpu
);
1137 *dest
= vcpu
->arch
.db
[dr
];
1140 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
1141 return EMULATE_FAIL
; /* will re-inject UD */
1144 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
1145 *dest
= vcpu
->arch
.dr6
;
1147 *dest
= svm
->vmcb
->save
.dr6
;
1150 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
1151 return EMULATE_FAIL
; /* will re-inject UD */
1154 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
1155 *dest
= vcpu
->arch
.dr7
;
1157 *dest
= svm
->vmcb
->save
.dr7
;
1161 return EMULATE_DONE
;
1164 static int svm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long value
)
1166 struct vcpu_svm
*svm
= to_svm(vcpu
);
1170 vcpu
->arch
.db
[dr
] = value
;
1171 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
))
1172 vcpu
->arch
.eff_db
[dr
] = value
;
1175 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
1176 return EMULATE_FAIL
; /* will re-inject UD */
1179 vcpu
->arch
.dr6
= (value
& DR6_VOLATILE
) | DR6_FIXED_1
;
1182 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
1183 return EMULATE_FAIL
; /* will re-inject UD */
1186 vcpu
->arch
.dr7
= (value
& DR7_VOLATILE
) | DR7_FIXED_1
;
1187 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)) {
1188 svm
->vmcb
->save
.dr7
= vcpu
->arch
.dr7
;
1189 vcpu
->arch
.switch_db_regs
= (value
& DR7_BP_EN_MASK
);
1194 return EMULATE_DONE
;
1197 static int pf_interception(struct vcpu_svm
*svm
)
1202 fault_address
= svm
->vmcb
->control
.exit_info_2
;
1203 error_code
= svm
->vmcb
->control
.exit_info_1
;
1205 trace_kvm_page_fault(fault_address
, error_code
);
1206 if (!npt_enabled
&& kvm_event_needs_reinjection(&svm
->vcpu
))
1207 kvm_mmu_unprotect_page_virt(&svm
->vcpu
, fault_address
);
1208 return kvm_mmu_page_fault(&svm
->vcpu
, fault_address
, error_code
);
1211 static int db_interception(struct vcpu_svm
*svm
)
1213 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
1215 if (!(svm
->vcpu
.guest_debug
&
1216 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
)) &&
1217 !svm
->nmi_singlestep
) {
1218 kvm_queue_exception(&svm
->vcpu
, DB_VECTOR
);
1222 if (svm
->nmi_singlestep
) {
1223 svm
->nmi_singlestep
= false;
1224 if (!(svm
->vcpu
.guest_debug
& KVM_GUESTDBG_SINGLESTEP
))
1225 svm
->vmcb
->save
.rflags
&=
1226 ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
1227 update_db_intercept(&svm
->vcpu
);
1230 if (svm
->vcpu
.guest_debug
&
1231 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
)){
1232 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
1233 kvm_run
->debug
.arch
.pc
=
1234 svm
->vmcb
->save
.cs
.base
+ svm
->vmcb
->save
.rip
;
1235 kvm_run
->debug
.arch
.exception
= DB_VECTOR
;
1242 static int bp_interception(struct vcpu_svm
*svm
)
1244 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
1246 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
1247 kvm_run
->debug
.arch
.pc
= svm
->vmcb
->save
.cs
.base
+ svm
->vmcb
->save
.rip
;
1248 kvm_run
->debug
.arch
.exception
= BP_VECTOR
;
1252 static int ud_interception(struct vcpu_svm
*svm
)
1256 er
= emulate_instruction(&svm
->vcpu
, 0, 0, EMULTYPE_TRAP_UD
);
1257 if (er
!= EMULATE_DONE
)
1258 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
1262 static void svm_fpu_activate(struct kvm_vcpu
*vcpu
)
1264 struct vcpu_svm
*svm
= to_svm(vcpu
);
1265 svm
->vmcb
->control
.intercept_exceptions
&= ~(1 << NM_VECTOR
);
1266 svm
->vcpu
.fpu_active
= 1;
1267 update_cr0_intercept(svm
);
1270 static int nm_interception(struct vcpu_svm
*svm
)
1272 svm_fpu_activate(&svm
->vcpu
);
1276 static int mc_interception(struct vcpu_svm
*svm
)
1279 * On an #MC intercept the MCE handler is not called automatically in
1280 * the host. So do it by hand here.
1284 /* not sure if we ever come back to this point */
1289 static int shutdown_interception(struct vcpu_svm
*svm
)
1291 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
1294 * VMCB is undefined after a SHUTDOWN intercept
1295 * so reinitialize it.
1297 clear_page(svm
->vmcb
);
1300 kvm_run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
1304 static int io_interception(struct vcpu_svm
*svm
)
1306 u32 io_info
= svm
->vmcb
->control
.exit_info_1
; /* address size bug? */
1307 int size
, in
, string
;
1310 ++svm
->vcpu
.stat
.io_exits
;
1312 svm
->next_rip
= svm
->vmcb
->control
.exit_info_2
;
1314 string
= (io_info
& SVM_IOIO_STR_MASK
) != 0;
1317 if (emulate_instruction(&svm
->vcpu
,
1318 0, 0, 0) == EMULATE_DO_MMIO
)
1323 in
= (io_info
& SVM_IOIO_TYPE_MASK
) != 0;
1324 port
= io_info
>> 16;
1325 size
= (io_info
& SVM_IOIO_SIZE_MASK
) >> SVM_IOIO_SIZE_SHIFT
;
1327 skip_emulated_instruction(&svm
->vcpu
);
1328 return kvm_emulate_pio(&svm
->vcpu
, in
, size
, port
);
1331 static int nmi_interception(struct vcpu_svm
*svm
)
1336 static int intr_interception(struct vcpu_svm
*svm
)
1338 ++svm
->vcpu
.stat
.irq_exits
;
1342 static int nop_on_interception(struct vcpu_svm
*svm
)
1347 static int halt_interception(struct vcpu_svm
*svm
)
1349 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 1;
1350 skip_emulated_instruction(&svm
->vcpu
);
1351 return kvm_emulate_halt(&svm
->vcpu
);
1354 static int vmmcall_interception(struct vcpu_svm
*svm
)
1356 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1357 skip_emulated_instruction(&svm
->vcpu
);
1358 kvm_emulate_hypercall(&svm
->vcpu
);
1362 static int nested_svm_check_permissions(struct vcpu_svm
*svm
)
1364 if (!(svm
->vcpu
.arch
.efer
& EFER_SVME
)
1365 || !is_paging(&svm
->vcpu
)) {
1366 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
1370 if (svm
->vmcb
->save
.cpl
) {
1371 kvm_inject_gp(&svm
->vcpu
, 0);
1378 static int nested_svm_check_exception(struct vcpu_svm
*svm
, unsigned nr
,
1379 bool has_error_code
, u32 error_code
)
1381 if (!is_nested(svm
))
1384 svm
->vmcb
->control
.exit_code
= SVM_EXIT_EXCP_BASE
+ nr
;
1385 svm
->vmcb
->control
.exit_code_hi
= 0;
1386 svm
->vmcb
->control
.exit_info_1
= error_code
;
1387 svm
->vmcb
->control
.exit_info_2
= svm
->vcpu
.arch
.cr2
;
1389 return nested_svm_exit_handled(svm
);
1392 static inline int nested_svm_intr(struct vcpu_svm
*svm
)
1394 if (!is_nested(svm
))
1397 if (!(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
))
1400 if (!(svm
->vcpu
.arch
.hflags
& HF_HIF_MASK
))
1403 svm
->vmcb
->control
.exit_code
= SVM_EXIT_INTR
;
1405 if (svm
->nested
.intercept
& 1ULL) {
1407 * The #vmexit can't be emulated here directly because this
1408 * code path runs with irqs and preemtion disabled. A
1409 * #vmexit emulation might sleep. Only signal request for
1412 svm
->nested
.exit_required
= true;
1413 trace_kvm_nested_intr_vmexit(svm
->vmcb
->save
.rip
);
1420 static void *nested_svm_map(struct vcpu_svm
*svm
, u64 gpa
, enum km_type idx
)
1424 page
= gfn_to_page(svm
->vcpu
.kvm
, gpa
>> PAGE_SHIFT
);
1425 if (is_error_page(page
))
1428 return kmap_atomic(page
, idx
);
1431 kvm_release_page_clean(page
);
1432 kvm_inject_gp(&svm
->vcpu
, 0);
1437 static void nested_svm_unmap(void *addr
, enum km_type idx
)
1444 page
= kmap_atomic_to_page(addr
);
1446 kunmap_atomic(addr
, idx
);
1447 kvm_release_page_dirty(page
);
1450 static bool nested_svm_exit_handled_msr(struct vcpu_svm
*svm
)
1452 u32 param
= svm
->vmcb
->control
.exit_info_1
& 1;
1453 u32 msr
= svm
->vcpu
.arch
.regs
[VCPU_REGS_RCX
];
1458 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_MSR_PROT
)))
1461 msrpm
= nested_svm_map(svm
, svm
->nested
.vmcb_msrpm
, KM_USER0
);
1471 case 0xc0000000 ... 0xc0001fff:
1472 t0
= (8192 + msr
- 0xc0000000) * 2;
1476 case 0xc0010000 ... 0xc0011fff:
1477 t0
= (16384 + msr
- 0xc0010000) * 2;
1486 ret
= msrpm
[t1
] & ((1 << param
) << t0
);
1489 nested_svm_unmap(msrpm
, KM_USER0
);
1494 static int nested_svm_exit_special(struct vcpu_svm
*svm
)
1496 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
1498 switch (exit_code
) {
1501 return NESTED_EXIT_HOST
;
1502 /* For now we are always handling NPFs when using them */
1505 return NESTED_EXIT_HOST
;
1507 /* When we're shadowing, trap PFs */
1508 case SVM_EXIT_EXCP_BASE
+ PF_VECTOR
:
1510 return NESTED_EXIT_HOST
;
1516 return NESTED_EXIT_CONTINUE
;
1520 * If this function returns true, this #vmexit was already handled
1522 static int nested_svm_exit_handled(struct vcpu_svm
*svm
)
1524 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
1525 int vmexit
= NESTED_EXIT_HOST
;
1527 switch (exit_code
) {
1529 vmexit
= nested_svm_exit_handled_msr(svm
);
1531 case SVM_EXIT_READ_CR0
... SVM_EXIT_READ_CR8
: {
1532 u32 cr_bits
= 1 << (exit_code
- SVM_EXIT_READ_CR0
);
1533 if (svm
->nested
.intercept_cr_read
& cr_bits
)
1534 vmexit
= NESTED_EXIT_DONE
;
1537 case SVM_EXIT_WRITE_CR0
... SVM_EXIT_WRITE_CR8
: {
1538 u32 cr_bits
= 1 << (exit_code
- SVM_EXIT_WRITE_CR0
);
1539 if (svm
->nested
.intercept_cr_write
& cr_bits
)
1540 vmexit
= NESTED_EXIT_DONE
;
1543 case SVM_EXIT_READ_DR0
... SVM_EXIT_READ_DR7
: {
1544 u32 dr_bits
= 1 << (exit_code
- SVM_EXIT_READ_DR0
);
1545 if (svm
->nested
.intercept_dr_read
& dr_bits
)
1546 vmexit
= NESTED_EXIT_DONE
;
1549 case SVM_EXIT_WRITE_DR0
... SVM_EXIT_WRITE_DR7
: {
1550 u32 dr_bits
= 1 << (exit_code
- SVM_EXIT_WRITE_DR0
);
1551 if (svm
->nested
.intercept_dr_write
& dr_bits
)
1552 vmexit
= NESTED_EXIT_DONE
;
1555 case SVM_EXIT_EXCP_BASE
... SVM_EXIT_EXCP_BASE
+ 0x1f: {
1556 u32 excp_bits
= 1 << (exit_code
- SVM_EXIT_EXCP_BASE
);
1557 if (svm
->nested
.intercept_exceptions
& excp_bits
)
1558 vmexit
= NESTED_EXIT_DONE
;
1562 u64 exit_bits
= 1ULL << (exit_code
- SVM_EXIT_INTR
);
1563 if (svm
->nested
.intercept
& exit_bits
)
1564 vmexit
= NESTED_EXIT_DONE
;
1568 if (vmexit
== NESTED_EXIT_DONE
) {
1569 nested_svm_vmexit(svm
);
1575 static inline void copy_vmcb_control_area(struct vmcb
*dst_vmcb
, struct vmcb
*from_vmcb
)
1577 struct vmcb_control_area
*dst
= &dst_vmcb
->control
;
1578 struct vmcb_control_area
*from
= &from_vmcb
->control
;
1580 dst
->intercept_cr_read
= from
->intercept_cr_read
;
1581 dst
->intercept_cr_write
= from
->intercept_cr_write
;
1582 dst
->intercept_dr_read
= from
->intercept_dr_read
;
1583 dst
->intercept_dr_write
= from
->intercept_dr_write
;
1584 dst
->intercept_exceptions
= from
->intercept_exceptions
;
1585 dst
->intercept
= from
->intercept
;
1586 dst
->iopm_base_pa
= from
->iopm_base_pa
;
1587 dst
->msrpm_base_pa
= from
->msrpm_base_pa
;
1588 dst
->tsc_offset
= from
->tsc_offset
;
1589 dst
->asid
= from
->asid
;
1590 dst
->tlb_ctl
= from
->tlb_ctl
;
1591 dst
->int_ctl
= from
->int_ctl
;
1592 dst
->int_vector
= from
->int_vector
;
1593 dst
->int_state
= from
->int_state
;
1594 dst
->exit_code
= from
->exit_code
;
1595 dst
->exit_code_hi
= from
->exit_code_hi
;
1596 dst
->exit_info_1
= from
->exit_info_1
;
1597 dst
->exit_info_2
= from
->exit_info_2
;
1598 dst
->exit_int_info
= from
->exit_int_info
;
1599 dst
->exit_int_info_err
= from
->exit_int_info_err
;
1600 dst
->nested_ctl
= from
->nested_ctl
;
1601 dst
->event_inj
= from
->event_inj
;
1602 dst
->event_inj_err
= from
->event_inj_err
;
1603 dst
->nested_cr3
= from
->nested_cr3
;
1604 dst
->lbr_ctl
= from
->lbr_ctl
;
1607 static int nested_svm_vmexit(struct vcpu_svm
*svm
)
1609 struct vmcb
*nested_vmcb
;
1610 struct vmcb
*hsave
= svm
->nested
.hsave
;
1611 struct vmcb
*vmcb
= svm
->vmcb
;
1613 trace_kvm_nested_vmexit_inject(vmcb
->control
.exit_code
,
1614 vmcb
->control
.exit_info_1
,
1615 vmcb
->control
.exit_info_2
,
1616 vmcb
->control
.exit_int_info
,
1617 vmcb
->control
.exit_int_info_err
);
1619 nested_vmcb
= nested_svm_map(svm
, svm
->nested
.vmcb
, KM_USER0
);
1623 /* Give the current vmcb to the guest */
1626 nested_vmcb
->save
.es
= vmcb
->save
.es
;
1627 nested_vmcb
->save
.cs
= vmcb
->save
.cs
;
1628 nested_vmcb
->save
.ss
= vmcb
->save
.ss
;
1629 nested_vmcb
->save
.ds
= vmcb
->save
.ds
;
1630 nested_vmcb
->save
.gdtr
= vmcb
->save
.gdtr
;
1631 nested_vmcb
->save
.idtr
= vmcb
->save
.idtr
;
1633 nested_vmcb
->save
.cr3
= vmcb
->save
.cr3
;
1634 nested_vmcb
->save
.cr2
= vmcb
->save
.cr2
;
1635 nested_vmcb
->save
.rflags
= vmcb
->save
.rflags
;
1636 nested_vmcb
->save
.rip
= vmcb
->save
.rip
;
1637 nested_vmcb
->save
.rsp
= vmcb
->save
.rsp
;
1638 nested_vmcb
->save
.rax
= vmcb
->save
.rax
;
1639 nested_vmcb
->save
.dr7
= vmcb
->save
.dr7
;
1640 nested_vmcb
->save
.dr6
= vmcb
->save
.dr6
;
1641 nested_vmcb
->save
.cpl
= vmcb
->save
.cpl
;
1643 nested_vmcb
->control
.int_ctl
= vmcb
->control
.int_ctl
;
1644 nested_vmcb
->control
.int_vector
= vmcb
->control
.int_vector
;
1645 nested_vmcb
->control
.int_state
= vmcb
->control
.int_state
;
1646 nested_vmcb
->control
.exit_code
= vmcb
->control
.exit_code
;
1647 nested_vmcb
->control
.exit_code_hi
= vmcb
->control
.exit_code_hi
;
1648 nested_vmcb
->control
.exit_info_1
= vmcb
->control
.exit_info_1
;
1649 nested_vmcb
->control
.exit_info_2
= vmcb
->control
.exit_info_2
;
1650 nested_vmcb
->control
.exit_int_info
= vmcb
->control
.exit_int_info
;
1651 nested_vmcb
->control
.exit_int_info_err
= vmcb
->control
.exit_int_info_err
;
1654 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
1655 * to make sure that we do not lose injected events. So check event_inj
1656 * here and copy it to exit_int_info if it is valid.
1657 * Exit_int_info and event_inj can't be both valid because the case
1658 * below only happens on a VMRUN instruction intercept which has
1659 * no valid exit_int_info set.
1661 if (vmcb
->control
.event_inj
& SVM_EVTINJ_VALID
) {
1662 struct vmcb_control_area
*nc
= &nested_vmcb
->control
;
1664 nc
->exit_int_info
= vmcb
->control
.event_inj
;
1665 nc
->exit_int_info_err
= vmcb
->control
.event_inj_err
;
1668 nested_vmcb
->control
.tlb_ctl
= 0;
1669 nested_vmcb
->control
.event_inj
= 0;
1670 nested_vmcb
->control
.event_inj_err
= 0;
1672 /* We always set V_INTR_MASKING and remember the old value in hflags */
1673 if (!(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
))
1674 nested_vmcb
->control
.int_ctl
&= ~V_INTR_MASKING_MASK
;
1676 /* Restore the original control entries */
1677 copy_vmcb_control_area(vmcb
, hsave
);
1679 kvm_clear_exception_queue(&svm
->vcpu
);
1680 kvm_clear_interrupt_queue(&svm
->vcpu
);
1682 /* Restore selected save entries */
1683 svm
->vmcb
->save
.es
= hsave
->save
.es
;
1684 svm
->vmcb
->save
.cs
= hsave
->save
.cs
;
1685 svm
->vmcb
->save
.ss
= hsave
->save
.ss
;
1686 svm
->vmcb
->save
.ds
= hsave
->save
.ds
;
1687 svm
->vmcb
->save
.gdtr
= hsave
->save
.gdtr
;
1688 svm
->vmcb
->save
.idtr
= hsave
->save
.idtr
;
1689 svm
->vmcb
->save
.rflags
= hsave
->save
.rflags
;
1690 svm_set_efer(&svm
->vcpu
, hsave
->save
.efer
);
1691 svm_set_cr0(&svm
->vcpu
, hsave
->save
.cr0
| X86_CR0_PE
);
1692 svm_set_cr4(&svm
->vcpu
, hsave
->save
.cr4
);
1694 svm
->vmcb
->save
.cr3
= hsave
->save
.cr3
;
1695 svm
->vcpu
.arch
.cr3
= hsave
->save
.cr3
;
1697 kvm_set_cr3(&svm
->vcpu
, hsave
->save
.cr3
);
1699 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
, hsave
->save
.rax
);
1700 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RSP
, hsave
->save
.rsp
);
1701 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RIP
, hsave
->save
.rip
);
1702 svm
->vmcb
->save
.dr7
= 0;
1703 svm
->vmcb
->save
.cpl
= 0;
1704 svm
->vmcb
->control
.exit_int_info
= 0;
1706 /* Exit nested SVM mode */
1707 svm
->nested
.vmcb
= 0;
1709 nested_svm_unmap(nested_vmcb
, KM_USER0
);
1711 kvm_mmu_reset_context(&svm
->vcpu
);
1712 kvm_mmu_load(&svm
->vcpu
);
1717 static bool nested_svm_vmrun_msrpm(struct vcpu_svm
*svm
)
1722 nested_msrpm
= nested_svm_map(svm
, svm
->nested
.vmcb_msrpm
, KM_USER0
);
1726 for (i
=0; i
< PAGE_SIZE
* (1 << MSRPM_ALLOC_ORDER
) / 4; i
++)
1727 svm
->nested
.msrpm
[i
] = svm
->msrpm
[i
] | nested_msrpm
[i
];
1729 svm
->vmcb
->control
.msrpm_base_pa
= __pa(svm
->nested
.msrpm
);
1731 nested_svm_unmap(nested_msrpm
, KM_USER0
);
1736 static bool nested_svm_vmrun(struct vcpu_svm
*svm
)
1738 struct vmcb
*nested_vmcb
;
1739 struct vmcb
*hsave
= svm
->nested
.hsave
;
1740 struct vmcb
*vmcb
= svm
->vmcb
;
1742 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, KM_USER0
);
1746 /* nested_vmcb is our indicator if nested SVM is activated */
1747 svm
->nested
.vmcb
= svm
->vmcb
->save
.rax
;
1749 trace_kvm_nested_vmrun(svm
->vmcb
->save
.rip
- 3, svm
->nested
.vmcb
,
1750 nested_vmcb
->save
.rip
,
1751 nested_vmcb
->control
.int_ctl
,
1752 nested_vmcb
->control
.event_inj
,
1753 nested_vmcb
->control
.nested_ctl
);
1755 /* Clear internal status */
1756 kvm_clear_exception_queue(&svm
->vcpu
);
1757 kvm_clear_interrupt_queue(&svm
->vcpu
);
1759 /* Save the old vmcb, so we don't need to pick what we save, but
1760 can restore everything when a VMEXIT occurs */
1761 hsave
->save
.es
= vmcb
->save
.es
;
1762 hsave
->save
.cs
= vmcb
->save
.cs
;
1763 hsave
->save
.ss
= vmcb
->save
.ss
;
1764 hsave
->save
.ds
= vmcb
->save
.ds
;
1765 hsave
->save
.gdtr
= vmcb
->save
.gdtr
;
1766 hsave
->save
.idtr
= vmcb
->save
.idtr
;
1767 hsave
->save
.efer
= svm
->vcpu
.arch
.efer
;
1768 hsave
->save
.cr0
= kvm_read_cr0(&svm
->vcpu
);
1769 hsave
->save
.cr4
= svm
->vcpu
.arch
.cr4
;
1770 hsave
->save
.rflags
= vmcb
->save
.rflags
;
1771 hsave
->save
.rip
= svm
->next_rip
;
1772 hsave
->save
.rsp
= vmcb
->save
.rsp
;
1773 hsave
->save
.rax
= vmcb
->save
.rax
;
1775 hsave
->save
.cr3
= vmcb
->save
.cr3
;
1777 hsave
->save
.cr3
= svm
->vcpu
.arch
.cr3
;
1779 copy_vmcb_control_area(hsave
, vmcb
);
1781 if (svm
->vmcb
->save
.rflags
& X86_EFLAGS_IF
)
1782 svm
->vcpu
.arch
.hflags
|= HF_HIF_MASK
;
1784 svm
->vcpu
.arch
.hflags
&= ~HF_HIF_MASK
;
1786 /* Load the nested guest state */
1787 svm
->vmcb
->save
.es
= nested_vmcb
->save
.es
;
1788 svm
->vmcb
->save
.cs
= nested_vmcb
->save
.cs
;
1789 svm
->vmcb
->save
.ss
= nested_vmcb
->save
.ss
;
1790 svm
->vmcb
->save
.ds
= nested_vmcb
->save
.ds
;
1791 svm
->vmcb
->save
.gdtr
= nested_vmcb
->save
.gdtr
;
1792 svm
->vmcb
->save
.idtr
= nested_vmcb
->save
.idtr
;
1793 svm
->vmcb
->save
.rflags
= nested_vmcb
->save
.rflags
;
1794 svm_set_efer(&svm
->vcpu
, nested_vmcb
->save
.efer
);
1795 svm_set_cr0(&svm
->vcpu
, nested_vmcb
->save
.cr0
);
1796 svm_set_cr4(&svm
->vcpu
, nested_vmcb
->save
.cr4
);
1798 svm
->vmcb
->save
.cr3
= nested_vmcb
->save
.cr3
;
1799 svm
->vcpu
.arch
.cr3
= nested_vmcb
->save
.cr3
;
1801 kvm_set_cr3(&svm
->vcpu
, nested_vmcb
->save
.cr3
);
1802 kvm_mmu_reset_context(&svm
->vcpu
);
1804 svm
->vmcb
->save
.cr2
= svm
->vcpu
.arch
.cr2
= nested_vmcb
->save
.cr2
;
1805 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
, nested_vmcb
->save
.rax
);
1806 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RSP
, nested_vmcb
->save
.rsp
);
1807 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RIP
, nested_vmcb
->save
.rip
);
1808 /* In case we don't even reach vcpu_run, the fields are not updated */
1809 svm
->vmcb
->save
.rax
= nested_vmcb
->save
.rax
;
1810 svm
->vmcb
->save
.rsp
= nested_vmcb
->save
.rsp
;
1811 svm
->vmcb
->save
.rip
= nested_vmcb
->save
.rip
;
1812 svm
->vmcb
->save
.dr7
= nested_vmcb
->save
.dr7
;
1813 svm
->vmcb
->save
.dr6
= nested_vmcb
->save
.dr6
;
1814 svm
->vmcb
->save
.cpl
= nested_vmcb
->save
.cpl
;
1816 /* We don't want a nested guest to be more powerful than the guest,
1817 so all intercepts are ORed */
1818 svm
->vmcb
->control
.intercept_cr_read
|=
1819 nested_vmcb
->control
.intercept_cr_read
;
1820 svm
->vmcb
->control
.intercept_cr_write
|=
1821 nested_vmcb
->control
.intercept_cr_write
;
1822 svm
->vmcb
->control
.intercept_dr_read
|=
1823 nested_vmcb
->control
.intercept_dr_read
;
1824 svm
->vmcb
->control
.intercept_dr_write
|=
1825 nested_vmcb
->control
.intercept_dr_write
;
1826 svm
->vmcb
->control
.intercept_exceptions
|=
1827 nested_vmcb
->control
.intercept_exceptions
;
1829 svm
->vmcb
->control
.intercept
|= nested_vmcb
->control
.intercept
;
1831 svm
->nested
.vmcb_msrpm
= nested_vmcb
->control
.msrpm_base_pa
;
1833 /* cache intercepts */
1834 svm
->nested
.intercept_cr_read
= nested_vmcb
->control
.intercept_cr_read
;
1835 svm
->nested
.intercept_cr_write
= nested_vmcb
->control
.intercept_cr_write
;
1836 svm
->nested
.intercept_dr_read
= nested_vmcb
->control
.intercept_dr_read
;
1837 svm
->nested
.intercept_dr_write
= nested_vmcb
->control
.intercept_dr_write
;
1838 svm
->nested
.intercept_exceptions
= nested_vmcb
->control
.intercept_exceptions
;
1839 svm
->nested
.intercept
= nested_vmcb
->control
.intercept
;
1841 force_new_asid(&svm
->vcpu
);
1842 svm
->vmcb
->control
.int_ctl
= nested_vmcb
->control
.int_ctl
| V_INTR_MASKING_MASK
;
1843 if (nested_vmcb
->control
.int_ctl
& V_INTR_MASKING_MASK
)
1844 svm
->vcpu
.arch
.hflags
|= HF_VINTR_MASK
;
1846 svm
->vcpu
.arch
.hflags
&= ~HF_VINTR_MASK
;
1848 svm
->vmcb
->control
.int_vector
= nested_vmcb
->control
.int_vector
;
1849 svm
->vmcb
->control
.int_state
= nested_vmcb
->control
.int_state
;
1850 svm
->vmcb
->control
.tsc_offset
+= nested_vmcb
->control
.tsc_offset
;
1851 svm
->vmcb
->control
.event_inj
= nested_vmcb
->control
.event_inj
;
1852 svm
->vmcb
->control
.event_inj_err
= nested_vmcb
->control
.event_inj_err
;
1854 nested_svm_unmap(nested_vmcb
, KM_USER0
);
1861 static void nested_svm_vmloadsave(struct vmcb
*from_vmcb
, struct vmcb
*to_vmcb
)
1863 to_vmcb
->save
.fs
= from_vmcb
->save
.fs
;
1864 to_vmcb
->save
.gs
= from_vmcb
->save
.gs
;
1865 to_vmcb
->save
.tr
= from_vmcb
->save
.tr
;
1866 to_vmcb
->save
.ldtr
= from_vmcb
->save
.ldtr
;
1867 to_vmcb
->save
.kernel_gs_base
= from_vmcb
->save
.kernel_gs_base
;
1868 to_vmcb
->save
.star
= from_vmcb
->save
.star
;
1869 to_vmcb
->save
.lstar
= from_vmcb
->save
.lstar
;
1870 to_vmcb
->save
.cstar
= from_vmcb
->save
.cstar
;
1871 to_vmcb
->save
.sfmask
= from_vmcb
->save
.sfmask
;
1872 to_vmcb
->save
.sysenter_cs
= from_vmcb
->save
.sysenter_cs
;
1873 to_vmcb
->save
.sysenter_esp
= from_vmcb
->save
.sysenter_esp
;
1874 to_vmcb
->save
.sysenter_eip
= from_vmcb
->save
.sysenter_eip
;
1877 static int vmload_interception(struct vcpu_svm
*svm
)
1879 struct vmcb
*nested_vmcb
;
1881 if (nested_svm_check_permissions(svm
))
1884 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1885 skip_emulated_instruction(&svm
->vcpu
);
1887 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, KM_USER0
);
1891 nested_svm_vmloadsave(nested_vmcb
, svm
->vmcb
);
1892 nested_svm_unmap(nested_vmcb
, KM_USER0
);
1897 static int vmsave_interception(struct vcpu_svm
*svm
)
1899 struct vmcb
*nested_vmcb
;
1901 if (nested_svm_check_permissions(svm
))
1904 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1905 skip_emulated_instruction(&svm
->vcpu
);
1907 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, KM_USER0
);
1911 nested_svm_vmloadsave(svm
->vmcb
, nested_vmcb
);
1912 nested_svm_unmap(nested_vmcb
, KM_USER0
);
1917 static int vmrun_interception(struct vcpu_svm
*svm
)
1919 if (nested_svm_check_permissions(svm
))
1922 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1923 skip_emulated_instruction(&svm
->vcpu
);
1925 if (!nested_svm_vmrun(svm
))
1928 if (!nested_svm_vmrun_msrpm(svm
))
1935 svm
->vmcb
->control
.exit_code
= SVM_EXIT_ERR
;
1936 svm
->vmcb
->control
.exit_code_hi
= 0;
1937 svm
->vmcb
->control
.exit_info_1
= 0;
1938 svm
->vmcb
->control
.exit_info_2
= 0;
1940 nested_svm_vmexit(svm
);
1945 static int stgi_interception(struct vcpu_svm
*svm
)
1947 if (nested_svm_check_permissions(svm
))
1950 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1951 skip_emulated_instruction(&svm
->vcpu
);
1958 static int clgi_interception(struct vcpu_svm
*svm
)
1960 if (nested_svm_check_permissions(svm
))
1963 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1964 skip_emulated_instruction(&svm
->vcpu
);
1968 /* After a CLGI no interrupts should come */
1969 svm_clear_vintr(svm
);
1970 svm
->vmcb
->control
.int_ctl
&= ~V_IRQ_MASK
;
1975 static int invlpga_interception(struct vcpu_svm
*svm
)
1977 struct kvm_vcpu
*vcpu
= &svm
->vcpu
;
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 trace_kvm_msr_read_ex(ecx
);
2182 kvm_inject_gp(&svm
->vcpu
, 0);
2184 trace_kvm_msr_read(ecx
, data
);
2186 svm
->vcpu
.arch
.regs
[VCPU_REGS_RAX
] = data
& 0xffffffff;
2187 svm
->vcpu
.arch
.regs
[VCPU_REGS_RDX
] = data
>> 32;
2188 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
2189 skip_emulated_instruction(&svm
->vcpu
);
2194 static int svm_set_msr(struct kvm_vcpu
*vcpu
, unsigned ecx
, u64 data
)
2196 struct vcpu_svm
*svm
= to_svm(vcpu
);
2199 case MSR_IA32_TSC
: {
2200 u64 tsc_offset
= data
- native_read_tsc();
2201 u64 g_tsc_offset
= 0;
2203 if (is_nested(svm
)) {
2204 g_tsc_offset
= svm
->vmcb
->control
.tsc_offset
-
2205 svm
->nested
.hsave
->control
.tsc_offset
;
2206 svm
->nested
.hsave
->control
.tsc_offset
= tsc_offset
;
2209 svm
->vmcb
->control
.tsc_offset
= tsc_offset
+ g_tsc_offset
;
2214 svm
->vmcb
->save
.star
= data
;
2216 #ifdef CONFIG_X86_64
2218 svm
->vmcb
->save
.lstar
= data
;
2221 svm
->vmcb
->save
.cstar
= data
;
2223 case MSR_KERNEL_GS_BASE
:
2224 svm
->vmcb
->save
.kernel_gs_base
= data
;
2226 case MSR_SYSCALL_MASK
:
2227 svm
->vmcb
->save
.sfmask
= data
;
2230 case MSR_IA32_SYSENTER_CS
:
2231 svm
->vmcb
->save
.sysenter_cs
= data
;
2233 case MSR_IA32_SYSENTER_EIP
:
2234 svm
->sysenter_eip
= data
;
2235 svm
->vmcb
->save
.sysenter_eip
= data
;
2237 case MSR_IA32_SYSENTER_ESP
:
2238 svm
->sysenter_esp
= data
;
2239 svm
->vmcb
->save
.sysenter_esp
= data
;
2241 case MSR_IA32_DEBUGCTLMSR
:
2242 if (!svm_has(SVM_FEATURE_LBRV
)) {
2243 pr_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2247 if (data
& DEBUGCTL_RESERVED_BITS
)
2250 svm
->vmcb
->save
.dbgctl
= data
;
2251 if (data
& (1ULL<<0))
2252 svm_enable_lbrv(svm
);
2254 svm_disable_lbrv(svm
);
2256 case MSR_VM_HSAVE_PA
:
2257 svm
->nested
.hsave_msr
= data
;
2261 pr_unimpl(vcpu
, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx
, data
);
2264 return kvm_set_msr_common(vcpu
, ecx
, data
);
2269 static int wrmsr_interception(struct vcpu_svm
*svm
)
2271 u32 ecx
= svm
->vcpu
.arch
.regs
[VCPU_REGS_RCX
];
2272 u64 data
= (svm
->vcpu
.arch
.regs
[VCPU_REGS_RAX
] & -1u)
2273 | ((u64
)(svm
->vcpu
.arch
.regs
[VCPU_REGS_RDX
] & -1u) << 32);
2276 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
2277 if (svm_set_msr(&svm
->vcpu
, ecx
, data
)) {
2278 trace_kvm_msr_write_ex(ecx
, data
);
2279 kvm_inject_gp(&svm
->vcpu
, 0);
2281 trace_kvm_msr_write(ecx
, data
);
2282 skip_emulated_instruction(&svm
->vcpu
);
2287 static int msr_interception(struct vcpu_svm
*svm
)
2289 if (svm
->vmcb
->control
.exit_info_1
)
2290 return wrmsr_interception(svm
);
2292 return rdmsr_interception(svm
);
2295 static int interrupt_window_interception(struct vcpu_svm
*svm
)
2297 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
2299 svm_clear_vintr(svm
);
2300 svm
->vmcb
->control
.int_ctl
&= ~V_IRQ_MASK
;
2302 * If the user space waits to inject interrupts, exit as soon as
2305 if (!irqchip_in_kernel(svm
->vcpu
.kvm
) &&
2306 kvm_run
->request_interrupt_window
&&
2307 !kvm_cpu_has_interrupt(&svm
->vcpu
)) {
2308 ++svm
->vcpu
.stat
.irq_window_exits
;
2309 kvm_run
->exit_reason
= KVM_EXIT_IRQ_WINDOW_OPEN
;
2316 static int pause_interception(struct vcpu_svm
*svm
)
2318 kvm_vcpu_on_spin(&(svm
->vcpu
));
2322 static int (*svm_exit_handlers
[])(struct vcpu_svm
*svm
) = {
2323 [SVM_EXIT_READ_CR0
] = emulate_on_interception
,
2324 [SVM_EXIT_READ_CR3
] = emulate_on_interception
,
2325 [SVM_EXIT_READ_CR4
] = emulate_on_interception
,
2326 [SVM_EXIT_READ_CR8
] = emulate_on_interception
,
2327 [SVM_EXIT_CR0_SEL_WRITE
] = emulate_on_interception
,
2328 [SVM_EXIT_WRITE_CR0
] = emulate_on_interception
,
2329 [SVM_EXIT_WRITE_CR3
] = emulate_on_interception
,
2330 [SVM_EXIT_WRITE_CR4
] = emulate_on_interception
,
2331 [SVM_EXIT_WRITE_CR8
] = cr8_write_interception
,
2332 [SVM_EXIT_READ_DR0
] = emulate_on_interception
,
2333 [SVM_EXIT_READ_DR1
] = emulate_on_interception
,
2334 [SVM_EXIT_READ_DR2
] = emulate_on_interception
,
2335 [SVM_EXIT_READ_DR3
] = emulate_on_interception
,
2336 [SVM_EXIT_READ_DR4
] = emulate_on_interception
,
2337 [SVM_EXIT_READ_DR5
] = emulate_on_interception
,
2338 [SVM_EXIT_READ_DR6
] = emulate_on_interception
,
2339 [SVM_EXIT_READ_DR7
] = emulate_on_interception
,
2340 [SVM_EXIT_WRITE_DR0
] = emulate_on_interception
,
2341 [SVM_EXIT_WRITE_DR1
] = emulate_on_interception
,
2342 [SVM_EXIT_WRITE_DR2
] = emulate_on_interception
,
2343 [SVM_EXIT_WRITE_DR3
] = emulate_on_interception
,
2344 [SVM_EXIT_WRITE_DR4
] = emulate_on_interception
,
2345 [SVM_EXIT_WRITE_DR5
] = emulate_on_interception
,
2346 [SVM_EXIT_WRITE_DR6
] = emulate_on_interception
,
2347 [SVM_EXIT_WRITE_DR7
] = emulate_on_interception
,
2348 [SVM_EXIT_EXCP_BASE
+ DB_VECTOR
] = db_interception
,
2349 [SVM_EXIT_EXCP_BASE
+ BP_VECTOR
] = bp_interception
,
2350 [SVM_EXIT_EXCP_BASE
+ UD_VECTOR
] = ud_interception
,
2351 [SVM_EXIT_EXCP_BASE
+ PF_VECTOR
] = pf_interception
,
2352 [SVM_EXIT_EXCP_BASE
+ NM_VECTOR
] = nm_interception
,
2353 [SVM_EXIT_EXCP_BASE
+ MC_VECTOR
] = mc_interception
,
2354 [SVM_EXIT_INTR
] = intr_interception
,
2355 [SVM_EXIT_NMI
] = nmi_interception
,
2356 [SVM_EXIT_SMI
] = nop_on_interception
,
2357 [SVM_EXIT_INIT
] = nop_on_interception
,
2358 [SVM_EXIT_VINTR
] = interrupt_window_interception
,
2359 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
2360 [SVM_EXIT_CPUID
] = cpuid_interception
,
2361 [SVM_EXIT_IRET
] = iret_interception
,
2362 [SVM_EXIT_INVD
] = emulate_on_interception
,
2363 [SVM_EXIT_PAUSE
] = pause_interception
,
2364 [SVM_EXIT_HLT
] = halt_interception
,
2365 [SVM_EXIT_INVLPG
] = invlpg_interception
,
2366 [SVM_EXIT_INVLPGA
] = invlpga_interception
,
2367 [SVM_EXIT_IOIO
] = io_interception
,
2368 [SVM_EXIT_MSR
] = msr_interception
,
2369 [SVM_EXIT_TASK_SWITCH
] = task_switch_interception
,
2370 [SVM_EXIT_SHUTDOWN
] = shutdown_interception
,
2371 [SVM_EXIT_VMRUN
] = vmrun_interception
,
2372 [SVM_EXIT_VMMCALL
] = vmmcall_interception
,
2373 [SVM_EXIT_VMLOAD
] = vmload_interception
,
2374 [SVM_EXIT_VMSAVE
] = vmsave_interception
,
2375 [SVM_EXIT_STGI
] = stgi_interception
,
2376 [SVM_EXIT_CLGI
] = clgi_interception
,
2377 [SVM_EXIT_SKINIT
] = skinit_interception
,
2378 [SVM_EXIT_WBINVD
] = emulate_on_interception
,
2379 [SVM_EXIT_MONITOR
] = invalid_op_interception
,
2380 [SVM_EXIT_MWAIT
] = invalid_op_interception
,
2381 [SVM_EXIT_NPF
] = pf_interception
,
2384 static int handle_exit(struct kvm_vcpu
*vcpu
)
2386 struct vcpu_svm
*svm
= to_svm(vcpu
);
2387 struct kvm_run
*kvm_run
= vcpu
->run
;
2388 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
2390 trace_kvm_exit(exit_code
, svm
->vmcb
->save
.rip
);
2392 if (unlikely(svm
->nested
.exit_required
)) {
2393 nested_svm_vmexit(svm
);
2394 svm
->nested
.exit_required
= false;
2399 if (is_nested(svm
)) {
2402 trace_kvm_nested_vmexit(svm
->vmcb
->save
.rip
, exit_code
,
2403 svm
->vmcb
->control
.exit_info_1
,
2404 svm
->vmcb
->control
.exit_info_2
,
2405 svm
->vmcb
->control
.exit_int_info
,
2406 svm
->vmcb
->control
.exit_int_info_err
);
2408 vmexit
= nested_svm_exit_special(svm
);
2410 if (vmexit
== NESTED_EXIT_CONTINUE
)
2411 vmexit
= nested_svm_exit_handled(svm
);
2413 if (vmexit
== NESTED_EXIT_DONE
)
2417 svm_complete_interrupts(svm
);
2419 if (!(svm
->vmcb
->control
.intercept_cr_write
& INTERCEPT_CR0_MASK
))
2420 vcpu
->arch
.cr0
= svm
->vmcb
->save
.cr0
;
2422 vcpu
->arch
.cr3
= svm
->vmcb
->save
.cr3
;
2424 if (svm
->vmcb
->control
.exit_code
== SVM_EXIT_ERR
) {
2425 kvm_run
->exit_reason
= KVM_EXIT_FAIL_ENTRY
;
2426 kvm_run
->fail_entry
.hardware_entry_failure_reason
2427 = svm
->vmcb
->control
.exit_code
;
2431 if (is_external_interrupt(svm
->vmcb
->control
.exit_int_info
) &&
2432 exit_code
!= SVM_EXIT_EXCP_BASE
+ PF_VECTOR
&&
2433 exit_code
!= SVM_EXIT_NPF
&& exit_code
!= SVM_EXIT_TASK_SWITCH
)
2434 printk(KERN_ERR
"%s: unexpected exit_ini_info 0x%x "
2436 __func__
, svm
->vmcb
->control
.exit_int_info
,
2439 if (exit_code
>= ARRAY_SIZE(svm_exit_handlers
)
2440 || !svm_exit_handlers
[exit_code
]) {
2441 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
2442 kvm_run
->hw
.hardware_exit_reason
= exit_code
;
2446 return svm_exit_handlers
[exit_code
](svm
);
2449 static void reload_tss(struct kvm_vcpu
*vcpu
)
2451 int cpu
= raw_smp_processor_id();
2453 struct svm_cpu_data
*sd
= per_cpu(svm_data
, cpu
);
2454 sd
->tss_desc
->type
= 9; /* available 32/64-bit TSS */
2458 static void pre_svm_run(struct vcpu_svm
*svm
)
2460 int cpu
= raw_smp_processor_id();
2462 struct svm_cpu_data
*sd
= per_cpu(svm_data
, cpu
);
2464 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_DO_NOTHING
;
2465 /* FIXME: handle wraparound of asid_generation */
2466 if (svm
->asid_generation
!= sd
->asid_generation
)
2470 static void svm_inject_nmi(struct kvm_vcpu
*vcpu
)
2472 struct vcpu_svm
*svm
= to_svm(vcpu
);
2474 svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_NMI
;
2475 vcpu
->arch
.hflags
|= HF_NMI_MASK
;
2476 svm
->vmcb
->control
.intercept
|= (1UL << INTERCEPT_IRET
);
2477 ++vcpu
->stat
.nmi_injections
;
2480 static inline void svm_inject_irq(struct vcpu_svm
*svm
, int irq
)
2482 struct vmcb_control_area
*control
;
2484 trace_kvm_inj_virq(irq
);
2486 ++svm
->vcpu
.stat
.irq_injections
;
2487 control
= &svm
->vmcb
->control
;
2488 control
->int_vector
= irq
;
2489 control
->int_ctl
&= ~V_INTR_PRIO_MASK
;
2490 control
->int_ctl
|= V_IRQ_MASK
|
2491 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT
);
2494 static void svm_set_irq(struct kvm_vcpu
*vcpu
)
2496 struct vcpu_svm
*svm
= to_svm(vcpu
);
2498 BUG_ON(!(gif_set(svm
)));
2500 svm
->vmcb
->control
.event_inj
= vcpu
->arch
.interrupt
.nr
|
2501 SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
;
2504 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
, int tpr
, int irr
)
2506 struct vcpu_svm
*svm
= to_svm(vcpu
);
2512 svm
->vmcb
->control
.intercept_cr_write
|= INTERCEPT_CR8_MASK
;
2515 static int svm_nmi_allowed(struct kvm_vcpu
*vcpu
)
2517 struct vcpu_svm
*svm
= to_svm(vcpu
);
2518 struct vmcb
*vmcb
= svm
->vmcb
;
2519 return !(vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
) &&
2520 !(svm
->vcpu
.arch
.hflags
& HF_NMI_MASK
);
2523 static bool svm_get_nmi_mask(struct kvm_vcpu
*vcpu
)
2525 struct vcpu_svm
*svm
= to_svm(vcpu
);
2527 return !!(svm
->vcpu
.arch
.hflags
& HF_NMI_MASK
);
2530 static void svm_set_nmi_mask(struct kvm_vcpu
*vcpu
, bool masked
)
2532 struct vcpu_svm
*svm
= to_svm(vcpu
);
2535 svm
->vcpu
.arch
.hflags
|= HF_NMI_MASK
;
2536 svm
->vmcb
->control
.intercept
|= (1UL << INTERCEPT_IRET
);
2538 svm
->vcpu
.arch
.hflags
&= ~HF_NMI_MASK
;
2539 svm
->vmcb
->control
.intercept
&= ~(1UL << INTERCEPT_IRET
);
2543 static int svm_interrupt_allowed(struct kvm_vcpu
*vcpu
)
2545 struct vcpu_svm
*svm
= to_svm(vcpu
);
2546 struct vmcb
*vmcb
= svm
->vmcb
;
2549 if (!gif_set(svm
) ||
2550 (vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
))
2553 ret
= !!(vmcb
->save
.rflags
& X86_EFLAGS_IF
);
2556 return ret
&& !(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
);
2561 static void enable_irq_window(struct kvm_vcpu
*vcpu
)
2563 struct vcpu_svm
*svm
= to_svm(vcpu
);
2565 nested_svm_intr(svm
);
2567 /* In case GIF=0 we can't rely on the CPU to tell us when
2568 * GIF becomes 1, because that's a separate STGI/VMRUN intercept.
2569 * The next time we get that intercept, this function will be
2570 * called again though and we'll get the vintr intercept. */
2573 svm_inject_irq(svm
, 0x0);
2577 static void enable_nmi_window(struct kvm_vcpu
*vcpu
)
2579 struct vcpu_svm
*svm
= to_svm(vcpu
);
2581 if ((svm
->vcpu
.arch
.hflags
& (HF_NMI_MASK
| HF_IRET_MASK
))
2583 return; /* IRET will cause a vm exit */
2585 /* Something prevents NMI from been injected. Single step over
2586 possible problem (IRET or exception injection or interrupt
2588 svm
->nmi_singlestep
= true;
2589 svm
->vmcb
->save
.rflags
|= (X86_EFLAGS_TF
| X86_EFLAGS_RF
);
2590 update_db_intercept(vcpu
);
2593 static int svm_set_tss_addr(struct kvm
*kvm
, unsigned int addr
)
2598 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
)
2600 force_new_asid(vcpu
);
2603 static void svm_prepare_guest_switch(struct kvm_vcpu
*vcpu
)
2607 static inline void sync_cr8_to_lapic(struct kvm_vcpu
*vcpu
)
2609 struct vcpu_svm
*svm
= to_svm(vcpu
);
2611 if (!(svm
->vmcb
->control
.intercept_cr_write
& INTERCEPT_CR8_MASK
)) {
2612 int cr8
= svm
->vmcb
->control
.int_ctl
& V_TPR_MASK
;
2613 kvm_set_cr8(vcpu
, cr8
);
2617 static inline void sync_lapic_to_cr8(struct kvm_vcpu
*vcpu
)
2619 struct vcpu_svm
*svm
= to_svm(vcpu
);
2622 cr8
= kvm_get_cr8(vcpu
);
2623 svm
->vmcb
->control
.int_ctl
&= ~V_TPR_MASK
;
2624 svm
->vmcb
->control
.int_ctl
|= cr8
& V_TPR_MASK
;
2627 static void svm_complete_interrupts(struct vcpu_svm
*svm
)
2631 u32 exitintinfo
= svm
->vmcb
->control
.exit_int_info
;
2633 if (svm
->vcpu
.arch
.hflags
& HF_IRET_MASK
)
2634 svm
->vcpu
.arch
.hflags
&= ~(HF_NMI_MASK
| HF_IRET_MASK
);
2636 svm
->vcpu
.arch
.nmi_injected
= false;
2637 kvm_clear_exception_queue(&svm
->vcpu
);
2638 kvm_clear_interrupt_queue(&svm
->vcpu
);
2640 if (!(exitintinfo
& SVM_EXITINTINFO_VALID
))
2643 vector
= exitintinfo
& SVM_EXITINTINFO_VEC_MASK
;
2644 type
= exitintinfo
& SVM_EXITINTINFO_TYPE_MASK
;
2647 case SVM_EXITINTINFO_TYPE_NMI
:
2648 svm
->vcpu
.arch
.nmi_injected
= true;
2650 case SVM_EXITINTINFO_TYPE_EXEPT
:
2651 /* In case of software exception do not reinject an exception
2652 vector, but re-execute and instruction instead */
2655 if (kvm_exception_is_soft(vector
))
2657 if (exitintinfo
& SVM_EXITINTINFO_VALID_ERR
) {
2658 u32 err
= svm
->vmcb
->control
.exit_int_info_err
;
2659 kvm_queue_exception_e(&svm
->vcpu
, vector
, err
);
2662 kvm_queue_exception(&svm
->vcpu
, vector
);
2664 case SVM_EXITINTINFO_TYPE_INTR
:
2665 kvm_queue_interrupt(&svm
->vcpu
, vector
, false);
2672 #ifdef CONFIG_X86_64
2678 static void svm_vcpu_run(struct kvm_vcpu
*vcpu
)
2680 struct vcpu_svm
*svm
= to_svm(vcpu
);
2686 * A vmexit emulation is required before the vcpu can be executed
2689 if (unlikely(svm
->nested
.exit_required
))
2692 svm
->vmcb
->save
.rax
= vcpu
->arch
.regs
[VCPU_REGS_RAX
];
2693 svm
->vmcb
->save
.rsp
= vcpu
->arch
.regs
[VCPU_REGS_RSP
];
2694 svm
->vmcb
->save
.rip
= vcpu
->arch
.regs
[VCPU_REGS_RIP
];
2698 sync_lapic_to_cr8(vcpu
);
2700 save_host_msrs(vcpu
);
2701 fs_selector
= kvm_read_fs();
2702 gs_selector
= kvm_read_gs();
2703 ldt_selector
= kvm_read_ldt();
2704 svm
->vmcb
->save
.cr2
= vcpu
->arch
.cr2
;
2705 /* required for live migration with NPT */
2707 svm
->vmcb
->save
.cr3
= vcpu
->arch
.cr3
;
2714 "push %%"R
"bp; \n\t"
2715 "mov %c[rbx](%[svm]), %%"R
"bx \n\t"
2716 "mov %c[rcx](%[svm]), %%"R
"cx \n\t"
2717 "mov %c[rdx](%[svm]), %%"R
"dx \n\t"
2718 "mov %c[rsi](%[svm]), %%"R
"si \n\t"
2719 "mov %c[rdi](%[svm]), %%"R
"di \n\t"
2720 "mov %c[rbp](%[svm]), %%"R
"bp \n\t"
2721 #ifdef CONFIG_X86_64
2722 "mov %c[r8](%[svm]), %%r8 \n\t"
2723 "mov %c[r9](%[svm]), %%r9 \n\t"
2724 "mov %c[r10](%[svm]), %%r10 \n\t"
2725 "mov %c[r11](%[svm]), %%r11 \n\t"
2726 "mov %c[r12](%[svm]), %%r12 \n\t"
2727 "mov %c[r13](%[svm]), %%r13 \n\t"
2728 "mov %c[r14](%[svm]), %%r14 \n\t"
2729 "mov %c[r15](%[svm]), %%r15 \n\t"
2732 /* Enter guest mode */
2734 "mov %c[vmcb](%[svm]), %%"R
"ax \n\t"
2735 __ex(SVM_VMLOAD
) "\n\t"
2736 __ex(SVM_VMRUN
) "\n\t"
2737 __ex(SVM_VMSAVE
) "\n\t"
2740 /* Save guest registers, load host registers */
2741 "mov %%"R
"bx, %c[rbx](%[svm]) \n\t"
2742 "mov %%"R
"cx, %c[rcx](%[svm]) \n\t"
2743 "mov %%"R
"dx, %c[rdx](%[svm]) \n\t"
2744 "mov %%"R
"si, %c[rsi](%[svm]) \n\t"
2745 "mov %%"R
"di, %c[rdi](%[svm]) \n\t"
2746 "mov %%"R
"bp, %c[rbp](%[svm]) \n\t"
2747 #ifdef CONFIG_X86_64
2748 "mov %%r8, %c[r8](%[svm]) \n\t"
2749 "mov %%r9, %c[r9](%[svm]) \n\t"
2750 "mov %%r10, %c[r10](%[svm]) \n\t"
2751 "mov %%r11, %c[r11](%[svm]) \n\t"
2752 "mov %%r12, %c[r12](%[svm]) \n\t"
2753 "mov %%r13, %c[r13](%[svm]) \n\t"
2754 "mov %%r14, %c[r14](%[svm]) \n\t"
2755 "mov %%r15, %c[r15](%[svm]) \n\t"
2760 [vmcb
]"i"(offsetof(struct vcpu_svm
, vmcb_pa
)),
2761 [rbx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RBX
])),
2762 [rcx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RCX
])),
2763 [rdx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RDX
])),
2764 [rsi
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RSI
])),
2765 [rdi
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RDI
])),
2766 [rbp
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RBP
]))
2767 #ifdef CONFIG_X86_64
2768 , [r8
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R8
])),
2769 [r9
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R9
])),
2770 [r10
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R10
])),
2771 [r11
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R11
])),
2772 [r12
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R12
])),
2773 [r13
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R13
])),
2774 [r14
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R14
])),
2775 [r15
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R15
]))
2778 , R
"bx", R
"cx", R
"dx", R
"si", R
"di"
2779 #ifdef CONFIG_X86_64
2780 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
2784 vcpu
->arch
.cr2
= svm
->vmcb
->save
.cr2
;
2785 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = svm
->vmcb
->save
.rax
;
2786 vcpu
->arch
.regs
[VCPU_REGS_RSP
] = svm
->vmcb
->save
.rsp
;
2787 vcpu
->arch
.regs
[VCPU_REGS_RIP
] = svm
->vmcb
->save
.rip
;
2789 kvm_load_fs(fs_selector
);
2790 kvm_load_gs(gs_selector
);
2791 kvm_load_ldt(ldt_selector
);
2792 load_host_msrs(vcpu
);
2796 local_irq_disable();
2800 sync_cr8_to_lapic(vcpu
);
2805 vcpu
->arch
.regs_avail
&= ~(1 << VCPU_EXREG_PDPTR
);
2806 vcpu
->arch
.regs_dirty
&= ~(1 << VCPU_EXREG_PDPTR
);
2812 static void svm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
2814 struct vcpu_svm
*svm
= to_svm(vcpu
);
2817 svm
->vmcb
->control
.nested_cr3
= root
;
2818 force_new_asid(vcpu
);
2822 svm
->vmcb
->save
.cr3
= root
;
2823 force_new_asid(vcpu
);
2826 static int is_disabled(void)
2830 rdmsrl(MSR_VM_CR
, vm_cr
);
2831 if (vm_cr
& (1 << SVM_VM_CR_SVM_DISABLE
))
2838 svm_patch_hypercall(struct kvm_vcpu
*vcpu
, unsigned char *hypercall
)
2841 * Patch in the VMMCALL instruction:
2843 hypercall
[0] = 0x0f;
2844 hypercall
[1] = 0x01;
2845 hypercall
[2] = 0xd9;
2848 static void svm_check_processor_compat(void *rtn
)
2853 static bool svm_cpu_has_accelerated_tpr(void)
2858 static int get_npt_level(void)
2860 #ifdef CONFIG_X86_64
2861 return PT64_ROOT_LEVEL
;
2863 return PT32E_ROOT_LEVEL
;
2867 static u64
svm_get_mt_mask(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool is_mmio
)
2872 static void svm_cpuid_update(struct kvm_vcpu
*vcpu
)
2876 static const struct trace_print_flags svm_exit_reasons_str
[] = {
2877 { SVM_EXIT_READ_CR0
, "read_cr0" },
2878 { SVM_EXIT_READ_CR3
, "read_cr3" },
2879 { SVM_EXIT_READ_CR4
, "read_cr4" },
2880 { SVM_EXIT_READ_CR8
, "read_cr8" },
2881 { SVM_EXIT_WRITE_CR0
, "write_cr0" },
2882 { SVM_EXIT_WRITE_CR3
, "write_cr3" },
2883 { SVM_EXIT_WRITE_CR4
, "write_cr4" },
2884 { SVM_EXIT_WRITE_CR8
, "write_cr8" },
2885 { SVM_EXIT_READ_DR0
, "read_dr0" },
2886 { SVM_EXIT_READ_DR1
, "read_dr1" },
2887 { SVM_EXIT_READ_DR2
, "read_dr2" },
2888 { SVM_EXIT_READ_DR3
, "read_dr3" },
2889 { SVM_EXIT_WRITE_DR0
, "write_dr0" },
2890 { SVM_EXIT_WRITE_DR1
, "write_dr1" },
2891 { SVM_EXIT_WRITE_DR2
, "write_dr2" },
2892 { SVM_EXIT_WRITE_DR3
, "write_dr3" },
2893 { SVM_EXIT_WRITE_DR5
, "write_dr5" },
2894 { SVM_EXIT_WRITE_DR7
, "write_dr7" },
2895 { SVM_EXIT_EXCP_BASE
+ DB_VECTOR
, "DB excp" },
2896 { SVM_EXIT_EXCP_BASE
+ BP_VECTOR
, "BP excp" },
2897 { SVM_EXIT_EXCP_BASE
+ UD_VECTOR
, "UD excp" },
2898 { SVM_EXIT_EXCP_BASE
+ PF_VECTOR
, "PF excp" },
2899 { SVM_EXIT_EXCP_BASE
+ NM_VECTOR
, "NM excp" },
2900 { SVM_EXIT_EXCP_BASE
+ MC_VECTOR
, "MC excp" },
2901 { SVM_EXIT_INTR
, "interrupt" },
2902 { SVM_EXIT_NMI
, "nmi" },
2903 { SVM_EXIT_SMI
, "smi" },
2904 { SVM_EXIT_INIT
, "init" },
2905 { SVM_EXIT_VINTR
, "vintr" },
2906 { SVM_EXIT_CPUID
, "cpuid" },
2907 { SVM_EXIT_INVD
, "invd" },
2908 { SVM_EXIT_HLT
, "hlt" },
2909 { SVM_EXIT_INVLPG
, "invlpg" },
2910 { SVM_EXIT_INVLPGA
, "invlpga" },
2911 { SVM_EXIT_IOIO
, "io" },
2912 { SVM_EXIT_MSR
, "msr" },
2913 { SVM_EXIT_TASK_SWITCH
, "task_switch" },
2914 { SVM_EXIT_SHUTDOWN
, "shutdown" },
2915 { SVM_EXIT_VMRUN
, "vmrun" },
2916 { SVM_EXIT_VMMCALL
, "hypercall" },
2917 { SVM_EXIT_VMLOAD
, "vmload" },
2918 { SVM_EXIT_VMSAVE
, "vmsave" },
2919 { SVM_EXIT_STGI
, "stgi" },
2920 { SVM_EXIT_CLGI
, "clgi" },
2921 { SVM_EXIT_SKINIT
, "skinit" },
2922 { SVM_EXIT_WBINVD
, "wbinvd" },
2923 { SVM_EXIT_MONITOR
, "monitor" },
2924 { SVM_EXIT_MWAIT
, "mwait" },
2925 { SVM_EXIT_NPF
, "npf" },
2929 static int svm_get_lpage_level(void)
2931 return PT_PDPE_LEVEL
;
2934 static bool svm_rdtscp_supported(void)
2939 static void svm_fpu_deactivate(struct kvm_vcpu
*vcpu
)
2941 struct vcpu_svm
*svm
= to_svm(vcpu
);
2943 update_cr0_intercept(svm
);
2944 svm
->vmcb
->control
.intercept_exceptions
|= 1 << NM_VECTOR
;
2947 static struct kvm_x86_ops svm_x86_ops
= {
2948 .cpu_has_kvm_support
= has_svm
,
2949 .disabled_by_bios
= is_disabled
,
2950 .hardware_setup
= svm_hardware_setup
,
2951 .hardware_unsetup
= svm_hardware_unsetup
,
2952 .check_processor_compatibility
= svm_check_processor_compat
,
2953 .hardware_enable
= svm_hardware_enable
,
2954 .hardware_disable
= svm_hardware_disable
,
2955 .cpu_has_accelerated_tpr
= svm_cpu_has_accelerated_tpr
,
2957 .vcpu_create
= svm_create_vcpu
,
2958 .vcpu_free
= svm_free_vcpu
,
2959 .vcpu_reset
= svm_vcpu_reset
,
2961 .prepare_guest_switch
= svm_prepare_guest_switch
,
2962 .vcpu_load
= svm_vcpu_load
,
2963 .vcpu_put
= svm_vcpu_put
,
2965 .set_guest_debug
= svm_guest_debug
,
2966 .get_msr
= svm_get_msr
,
2967 .set_msr
= svm_set_msr
,
2968 .get_segment_base
= svm_get_segment_base
,
2969 .get_segment
= svm_get_segment
,
2970 .set_segment
= svm_set_segment
,
2971 .get_cpl
= svm_get_cpl
,
2972 .get_cs_db_l_bits
= kvm_get_cs_db_l_bits
,
2973 .decache_cr0_guest_bits
= svm_decache_cr0_guest_bits
,
2974 .decache_cr4_guest_bits
= svm_decache_cr4_guest_bits
,
2975 .set_cr0
= svm_set_cr0
,
2976 .set_cr3
= svm_set_cr3
,
2977 .set_cr4
= svm_set_cr4
,
2978 .set_efer
= svm_set_efer
,
2979 .get_idt
= svm_get_idt
,
2980 .set_idt
= svm_set_idt
,
2981 .get_gdt
= svm_get_gdt
,
2982 .set_gdt
= svm_set_gdt
,
2983 .get_dr
= svm_get_dr
,
2984 .set_dr
= svm_set_dr
,
2985 .cache_reg
= svm_cache_reg
,
2986 .get_rflags
= svm_get_rflags
,
2987 .set_rflags
= svm_set_rflags
,
2988 .fpu_activate
= svm_fpu_activate
,
2989 .fpu_deactivate
= svm_fpu_deactivate
,
2991 .tlb_flush
= svm_flush_tlb
,
2993 .run
= svm_vcpu_run
,
2994 .handle_exit
= handle_exit
,
2995 .skip_emulated_instruction
= skip_emulated_instruction
,
2996 .set_interrupt_shadow
= svm_set_interrupt_shadow
,
2997 .get_interrupt_shadow
= svm_get_interrupt_shadow
,
2998 .patch_hypercall
= svm_patch_hypercall
,
2999 .set_irq
= svm_set_irq
,
3000 .set_nmi
= svm_inject_nmi
,
3001 .queue_exception
= svm_queue_exception
,
3002 .interrupt_allowed
= svm_interrupt_allowed
,
3003 .nmi_allowed
= svm_nmi_allowed
,
3004 .get_nmi_mask
= svm_get_nmi_mask
,
3005 .set_nmi_mask
= svm_set_nmi_mask
,
3006 .enable_nmi_window
= enable_nmi_window
,
3007 .enable_irq_window
= enable_irq_window
,
3008 .update_cr8_intercept
= update_cr8_intercept
,
3010 .set_tss_addr
= svm_set_tss_addr
,
3011 .get_tdp_level
= get_npt_level
,
3012 .get_mt_mask
= svm_get_mt_mask
,
3014 .exit_reasons_str
= svm_exit_reasons_str
,
3015 .get_lpage_level
= svm_get_lpage_level
,
3017 .cpuid_update
= svm_cpuid_update
,
3019 .rdtscp_supported
= svm_rdtscp_supported
,
3022 static int __init
svm_init(void)
3024 return kvm_init(&svm_x86_ops
, sizeof(struct vcpu_svm
),
3028 static void __exit
svm_exit(void)
3033 module_init(svm_init
)
3034 module_exit(svm_exit
)