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>
21 #include "kvm_cache_regs.h"
24 #include <linux/module.h>
25 #include <linux/kernel.h>
26 #include <linux/vmalloc.h>
27 #include <linux/highmem.h>
28 #include <linux/sched.h>
32 #include <asm/virtext.h>
34 #define __ex(x) __kvm_handle_fault_on_reboot(x)
36 MODULE_AUTHOR("Qumranet");
37 MODULE_LICENSE("GPL");
39 #define IOPM_ALLOC_ORDER 2
40 #define MSRPM_ALLOC_ORDER 1
42 #define SEG_TYPE_LDT 2
43 #define SEG_TYPE_BUSY_TSS16 3
45 #define SVM_FEATURE_NPT (1 << 0)
46 #define SVM_FEATURE_LBRV (1 << 1)
47 #define SVM_FEATURE_SVML (1 << 2)
49 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
51 /* Turn on to get debugging output*/
52 /* #define NESTED_DEBUG */
55 #define nsvm_printk(fmt, args...) printk(KERN_INFO fmt, ## args)
57 #define nsvm_printk(fmt, args...) do {} while(0)
60 /* enable NPT for AMD64 and X86 with PAE */
61 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
62 static bool npt_enabled
= true;
64 static bool npt_enabled
= false;
68 module_param(npt
, int, S_IRUGO
);
70 static int nested
= 0;
71 module_param(nested
, int, S_IRUGO
);
73 static void kvm_reput_irq(struct vcpu_svm
*svm
);
74 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
);
76 static int nested_svm_exit_handled(struct vcpu_svm
*svm
, bool kvm_override
);
77 static int nested_svm_vmexit(struct vcpu_svm
*svm
);
78 static int nested_svm_vmsave(struct vcpu_svm
*svm
, void *nested_vmcb
,
79 void *arg2
, void *opaque
);
80 static int nested_svm_check_exception(struct vcpu_svm
*svm
, unsigned nr
,
81 bool has_error_code
, u32 error_code
);
83 static inline struct vcpu_svm
*to_svm(struct kvm_vcpu
*vcpu
)
85 return container_of(vcpu
, struct vcpu_svm
, vcpu
);
88 static inline bool is_nested(struct vcpu_svm
*svm
)
90 return svm
->nested_vmcb
;
93 static unsigned long iopm_base
;
95 struct kvm_ldttss_desc
{
98 unsigned base1
: 8, type
: 5, dpl
: 2, p
: 1;
99 unsigned limit1
: 4, zero0
: 3, g
: 1, base2
: 8;
102 } __attribute__((packed
));
104 struct svm_cpu_data
{
110 struct kvm_ldttss_desc
*tss_desc
;
112 struct page
*save_area
;
115 static DEFINE_PER_CPU(struct svm_cpu_data
*, svm_data
);
116 static uint32_t svm_features
;
118 struct svm_init_data
{
123 static u32 msrpm_ranges
[] = {0, 0xc0000000, 0xc0010000};
125 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
126 #define MSRS_RANGE_SIZE 2048
127 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
129 #define MAX_INST_SIZE 15
131 static inline u32
svm_has(u32 feat
)
133 return svm_features
& feat
;
136 static inline void clgi(void)
138 asm volatile (__ex(SVM_CLGI
));
141 static inline void stgi(void)
143 asm volatile (__ex(SVM_STGI
));
146 static inline void invlpga(unsigned long addr
, u32 asid
)
148 asm volatile (__ex(SVM_INVLPGA
) :: "a"(addr
), "c"(asid
));
151 static inline unsigned long kvm_read_cr2(void)
155 asm volatile ("mov %%cr2, %0" : "=r" (cr2
));
159 static inline void kvm_write_cr2(unsigned long val
)
161 asm volatile ("mov %0, %%cr2" :: "r" (val
));
164 static inline void force_new_asid(struct kvm_vcpu
*vcpu
)
166 to_svm(vcpu
)->asid_generation
--;
169 static inline void flush_guest_tlb(struct kvm_vcpu
*vcpu
)
171 force_new_asid(vcpu
);
174 static void svm_set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
176 if (!npt_enabled
&& !(efer
& EFER_LMA
))
179 to_svm(vcpu
)->vmcb
->save
.efer
= efer
| EFER_SVME
;
180 vcpu
->arch
.shadow_efer
= efer
;
183 static void svm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
,
184 bool has_error_code
, u32 error_code
)
186 struct vcpu_svm
*svm
= to_svm(vcpu
);
188 /* If we are within a nested VM we'd better #VMEXIT and let the
189 guest handle the exception */
190 if (nested_svm_check_exception(svm
, nr
, has_error_code
, error_code
))
193 svm
->vmcb
->control
.event_inj
= nr
195 | (has_error_code
? SVM_EVTINJ_VALID_ERR
: 0)
196 | SVM_EVTINJ_TYPE_EXEPT
;
197 svm
->vmcb
->control
.event_inj_err
= error_code
;
200 static bool svm_exception_injected(struct kvm_vcpu
*vcpu
)
202 struct vcpu_svm
*svm
= to_svm(vcpu
);
204 return !(svm
->vmcb
->control
.exit_int_info
& SVM_EXITINTINFO_VALID
);
207 static int is_external_interrupt(u32 info
)
209 info
&= SVM_EVTINJ_TYPE_MASK
| SVM_EVTINJ_VALID
;
210 return info
== (SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
);
213 static void skip_emulated_instruction(struct kvm_vcpu
*vcpu
)
215 struct vcpu_svm
*svm
= to_svm(vcpu
);
217 if (!svm
->next_rip
) {
218 printk(KERN_DEBUG
"%s: NOP\n", __func__
);
221 if (svm
->next_rip
- kvm_rip_read(vcpu
) > MAX_INST_SIZE
)
222 printk(KERN_ERR
"%s: ip 0x%lx next 0x%llx\n",
223 __func__
, kvm_rip_read(vcpu
), svm
->next_rip
);
225 kvm_rip_write(vcpu
, svm
->next_rip
);
226 svm
->vmcb
->control
.int_state
&= ~SVM_INTERRUPT_SHADOW_MASK
;
228 vcpu
->arch
.interrupt_window_open
= (svm
->vcpu
.arch
.hflags
& HF_GIF_MASK
);
231 static int has_svm(void)
235 if (!cpu_has_svm(&msg
)) {
236 printk(KERN_INFO
"has_svm: %s\n", msg
);
243 static void svm_hardware_disable(void *garbage
)
248 static void svm_hardware_enable(void *garbage
)
251 struct svm_cpu_data
*svm_data
;
253 struct desc_ptr gdt_descr
;
254 struct desc_struct
*gdt
;
255 int me
= raw_smp_processor_id();
258 printk(KERN_ERR
"svm_cpu_init: err EOPNOTSUPP on %d\n", me
);
261 svm_data
= per_cpu(svm_data
, me
);
264 printk(KERN_ERR
"svm_cpu_init: svm_data is NULL on %d\n",
269 svm_data
->asid_generation
= 1;
270 svm_data
->max_asid
= cpuid_ebx(SVM_CPUID_FUNC
) - 1;
271 svm_data
->next_asid
= svm_data
->max_asid
+ 1;
273 asm volatile ("sgdt %0" : "=m"(gdt_descr
));
274 gdt
= (struct desc_struct
*)gdt_descr
.address
;
275 svm_data
->tss_desc
= (struct kvm_ldttss_desc
*)(gdt
+ GDT_ENTRY_TSS
);
277 rdmsrl(MSR_EFER
, efer
);
278 wrmsrl(MSR_EFER
, efer
| EFER_SVME
);
280 wrmsrl(MSR_VM_HSAVE_PA
,
281 page_to_pfn(svm_data
->save_area
) << PAGE_SHIFT
);
284 static void svm_cpu_uninit(int cpu
)
286 struct svm_cpu_data
*svm_data
287 = per_cpu(svm_data
, raw_smp_processor_id());
292 per_cpu(svm_data
, raw_smp_processor_id()) = NULL
;
293 __free_page(svm_data
->save_area
);
297 static int svm_cpu_init(int cpu
)
299 struct svm_cpu_data
*svm_data
;
302 svm_data
= kzalloc(sizeof(struct svm_cpu_data
), GFP_KERNEL
);
306 svm_data
->save_area
= alloc_page(GFP_KERNEL
);
308 if (!svm_data
->save_area
)
311 per_cpu(svm_data
, cpu
) = svm_data
;
321 static void set_msr_interception(u32
*msrpm
, unsigned msr
,
326 for (i
= 0; i
< NUM_MSR_MAPS
; i
++) {
327 if (msr
>= msrpm_ranges
[i
] &&
328 msr
< msrpm_ranges
[i
] + MSRS_IN_RANGE
) {
329 u32 msr_offset
= (i
* MSRS_IN_RANGE
+ msr
-
330 msrpm_ranges
[i
]) * 2;
332 u32
*base
= msrpm
+ (msr_offset
/ 32);
333 u32 msr_shift
= msr_offset
% 32;
334 u32 mask
= ((write
) ? 0 : 2) | ((read
) ? 0 : 1);
335 *base
= (*base
& ~(0x3 << msr_shift
)) |
343 static void svm_vcpu_init_msrpm(u32
*msrpm
)
345 memset(msrpm
, 0xff, PAGE_SIZE
* (1 << MSRPM_ALLOC_ORDER
));
348 set_msr_interception(msrpm
, MSR_GS_BASE
, 1, 1);
349 set_msr_interception(msrpm
, MSR_FS_BASE
, 1, 1);
350 set_msr_interception(msrpm
, MSR_KERNEL_GS_BASE
, 1, 1);
351 set_msr_interception(msrpm
, MSR_LSTAR
, 1, 1);
352 set_msr_interception(msrpm
, MSR_CSTAR
, 1, 1);
353 set_msr_interception(msrpm
, MSR_SYSCALL_MASK
, 1, 1);
355 set_msr_interception(msrpm
, MSR_K6_STAR
, 1, 1);
356 set_msr_interception(msrpm
, MSR_IA32_SYSENTER_CS
, 1, 1);
357 set_msr_interception(msrpm
, MSR_IA32_SYSENTER_ESP
, 1, 1);
358 set_msr_interception(msrpm
, MSR_IA32_SYSENTER_EIP
, 1, 1);
361 static void svm_enable_lbrv(struct vcpu_svm
*svm
)
363 u32
*msrpm
= svm
->msrpm
;
365 svm
->vmcb
->control
.lbr_ctl
= 1;
366 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHFROMIP
, 1, 1);
367 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHTOIP
, 1, 1);
368 set_msr_interception(msrpm
, MSR_IA32_LASTINTFROMIP
, 1, 1);
369 set_msr_interception(msrpm
, MSR_IA32_LASTINTTOIP
, 1, 1);
372 static void svm_disable_lbrv(struct vcpu_svm
*svm
)
374 u32
*msrpm
= svm
->msrpm
;
376 svm
->vmcb
->control
.lbr_ctl
= 0;
377 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHFROMIP
, 0, 0);
378 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHTOIP
, 0, 0);
379 set_msr_interception(msrpm
, MSR_IA32_LASTINTFROMIP
, 0, 0);
380 set_msr_interception(msrpm
, MSR_IA32_LASTINTTOIP
, 0, 0);
383 static __init
int svm_hardware_setup(void)
386 struct page
*iopm_pages
;
390 iopm_pages
= alloc_pages(GFP_KERNEL
, IOPM_ALLOC_ORDER
);
395 iopm_va
= page_address(iopm_pages
);
396 memset(iopm_va
, 0xff, PAGE_SIZE
* (1 << IOPM_ALLOC_ORDER
));
397 iopm_base
= page_to_pfn(iopm_pages
) << PAGE_SHIFT
;
399 if (boot_cpu_has(X86_FEATURE_NX
))
400 kvm_enable_efer_bits(EFER_NX
);
402 if (boot_cpu_has(X86_FEATURE_FXSR_OPT
))
403 kvm_enable_efer_bits(EFER_FFXSR
);
406 printk(KERN_INFO
"kvm: Nested Virtualization enabled\n");
407 kvm_enable_efer_bits(EFER_SVME
);
410 for_each_online_cpu(cpu
) {
411 r
= svm_cpu_init(cpu
);
416 svm_features
= cpuid_edx(SVM_CPUID_FUNC
);
418 if (!svm_has(SVM_FEATURE_NPT
))
421 if (npt_enabled
&& !npt
) {
422 printk(KERN_INFO
"kvm: Nested Paging disabled\n");
427 printk(KERN_INFO
"kvm: Nested Paging enabled\n");
435 __free_pages(iopm_pages
, IOPM_ALLOC_ORDER
);
440 static __exit
void svm_hardware_unsetup(void)
444 for_each_online_cpu(cpu
)
447 __free_pages(pfn_to_page(iopm_base
>> PAGE_SHIFT
), IOPM_ALLOC_ORDER
);
451 static void init_seg(struct vmcb_seg
*seg
)
454 seg
->attrib
= SVM_SELECTOR_P_MASK
| SVM_SELECTOR_S_MASK
|
455 SVM_SELECTOR_WRITE_MASK
; /* Read/Write Data Segment */
460 static void init_sys_seg(struct vmcb_seg
*seg
, uint32_t type
)
463 seg
->attrib
= SVM_SELECTOR_P_MASK
| type
;
468 static void init_vmcb(struct vcpu_svm
*svm
)
470 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
471 struct vmcb_save_area
*save
= &svm
->vmcb
->save
;
473 control
->intercept_cr_read
= INTERCEPT_CR0_MASK
|
477 control
->intercept_cr_write
= INTERCEPT_CR0_MASK
|
482 control
->intercept_dr_read
= INTERCEPT_DR0_MASK
|
487 control
->intercept_dr_write
= INTERCEPT_DR0_MASK
|
494 control
->intercept_exceptions
= (1 << PF_VECTOR
) |
499 control
->intercept
= (1ULL << INTERCEPT_INTR
) |
500 (1ULL << INTERCEPT_NMI
) |
501 (1ULL << INTERCEPT_SMI
) |
502 (1ULL << INTERCEPT_CPUID
) |
503 (1ULL << INTERCEPT_INVD
) |
504 (1ULL << INTERCEPT_HLT
) |
505 (1ULL << INTERCEPT_INVLPG
) |
506 (1ULL << INTERCEPT_INVLPGA
) |
507 (1ULL << INTERCEPT_IOIO_PROT
) |
508 (1ULL << INTERCEPT_MSR_PROT
) |
509 (1ULL << INTERCEPT_TASK_SWITCH
) |
510 (1ULL << INTERCEPT_SHUTDOWN
) |
511 (1ULL << INTERCEPT_VMRUN
) |
512 (1ULL << INTERCEPT_VMMCALL
) |
513 (1ULL << INTERCEPT_VMLOAD
) |
514 (1ULL << INTERCEPT_VMSAVE
) |
515 (1ULL << INTERCEPT_STGI
) |
516 (1ULL << INTERCEPT_CLGI
) |
517 (1ULL << INTERCEPT_SKINIT
) |
518 (1ULL << INTERCEPT_WBINVD
) |
519 (1ULL << INTERCEPT_MONITOR
) |
520 (1ULL << INTERCEPT_MWAIT
);
522 control
->iopm_base_pa
= iopm_base
;
523 control
->msrpm_base_pa
= __pa(svm
->msrpm
);
524 control
->tsc_offset
= 0;
525 control
->int_ctl
= V_INTR_MASKING_MASK
;
533 save
->cs
.selector
= 0xf000;
534 /* Executable/Readable Code Segment */
535 save
->cs
.attrib
= SVM_SELECTOR_READ_MASK
| SVM_SELECTOR_P_MASK
|
536 SVM_SELECTOR_S_MASK
| SVM_SELECTOR_CODE_MASK
;
537 save
->cs
.limit
= 0xffff;
539 * cs.base should really be 0xffff0000, but vmx can't handle that, so
540 * be consistent with it.
542 * Replace when we have real mode working for vmx.
544 save
->cs
.base
= 0xf0000;
546 save
->gdtr
.limit
= 0xffff;
547 save
->idtr
.limit
= 0xffff;
549 init_sys_seg(&save
->ldtr
, SEG_TYPE_LDT
);
550 init_sys_seg(&save
->tr
, SEG_TYPE_BUSY_TSS16
);
552 save
->efer
= EFER_SVME
;
553 save
->dr6
= 0xffff0ff0;
556 save
->rip
= 0x0000fff0;
557 svm
->vcpu
.arch
.regs
[VCPU_REGS_RIP
] = save
->rip
;
560 * cr0 val on cpu init should be 0x60000010, we enable cpu
561 * cache by default. the orderly way is to enable cache in bios.
563 save
->cr0
= 0x00000010 | X86_CR0_PG
| X86_CR0_WP
;
564 save
->cr4
= X86_CR4_PAE
;
568 /* Setup VMCB for Nested Paging */
569 control
->nested_ctl
= 1;
570 control
->intercept
&= ~((1ULL << INTERCEPT_TASK_SWITCH
) |
571 (1ULL << INTERCEPT_INVLPG
));
572 control
->intercept_exceptions
&= ~(1 << PF_VECTOR
);
573 control
->intercept_cr_read
&= ~(INTERCEPT_CR0_MASK
|
575 control
->intercept_cr_write
&= ~(INTERCEPT_CR0_MASK
|
577 save
->g_pat
= 0x0007040600070406ULL
;
578 /* enable caching because the QEMU Bios doesn't enable it */
579 save
->cr0
= X86_CR0_ET
;
583 force_new_asid(&svm
->vcpu
);
585 svm
->nested_vmcb
= 0;
586 svm
->vcpu
.arch
.hflags
= HF_GIF_MASK
;
589 static int svm_vcpu_reset(struct kvm_vcpu
*vcpu
)
591 struct vcpu_svm
*svm
= to_svm(vcpu
);
595 if (vcpu
->vcpu_id
!= 0) {
596 kvm_rip_write(vcpu
, 0);
597 svm
->vmcb
->save
.cs
.base
= svm
->vcpu
.arch
.sipi_vector
<< 12;
598 svm
->vmcb
->save
.cs
.selector
= svm
->vcpu
.arch
.sipi_vector
<< 8;
600 vcpu
->arch
.regs_avail
= ~0;
601 vcpu
->arch
.regs_dirty
= ~0;
606 static struct kvm_vcpu
*svm_create_vcpu(struct kvm
*kvm
, unsigned int id
)
608 struct vcpu_svm
*svm
;
610 struct page
*msrpm_pages
;
611 struct page
*hsave_page
;
612 struct page
*nested_msrpm_pages
;
615 svm
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
621 err
= kvm_vcpu_init(&svm
->vcpu
, kvm
, id
);
625 page
= alloc_page(GFP_KERNEL
);
632 msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
636 nested_msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
637 if (!nested_msrpm_pages
)
640 svm
->msrpm
= page_address(msrpm_pages
);
641 svm_vcpu_init_msrpm(svm
->msrpm
);
643 hsave_page
= alloc_page(GFP_KERNEL
);
646 svm
->hsave
= page_address(hsave_page
);
648 svm
->nested_msrpm
= page_address(nested_msrpm_pages
);
650 svm
->vmcb
= page_address(page
);
651 clear_page(svm
->vmcb
);
652 svm
->vmcb_pa
= page_to_pfn(page
) << PAGE_SHIFT
;
653 svm
->asid_generation
= 0;
657 svm
->vcpu
.fpu_active
= 1;
658 svm
->vcpu
.arch
.apic_base
= 0xfee00000 | MSR_IA32_APICBASE_ENABLE
;
659 if (svm
->vcpu
.vcpu_id
== 0)
660 svm
->vcpu
.arch
.apic_base
|= MSR_IA32_APICBASE_BSP
;
665 kvm_vcpu_uninit(&svm
->vcpu
);
667 kmem_cache_free(kvm_vcpu_cache
, svm
);
672 static void svm_free_vcpu(struct kvm_vcpu
*vcpu
)
674 struct vcpu_svm
*svm
= to_svm(vcpu
);
676 __free_page(pfn_to_page(svm
->vmcb_pa
>> PAGE_SHIFT
));
677 __free_pages(virt_to_page(svm
->msrpm
), MSRPM_ALLOC_ORDER
);
678 __free_page(virt_to_page(svm
->hsave
));
679 __free_pages(virt_to_page(svm
->nested_msrpm
), MSRPM_ALLOC_ORDER
);
680 kvm_vcpu_uninit(vcpu
);
681 kmem_cache_free(kvm_vcpu_cache
, svm
);
684 static void svm_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
686 struct vcpu_svm
*svm
= to_svm(vcpu
);
689 if (unlikely(cpu
!= vcpu
->cpu
)) {
693 * Make sure that the guest sees a monotonically
697 delta
= vcpu
->arch
.host_tsc
- tsc_this
;
698 svm
->vmcb
->control
.tsc_offset
+= delta
;
700 kvm_migrate_timers(vcpu
);
703 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
704 rdmsrl(host_save_user_msrs
[i
], svm
->host_user_msrs
[i
]);
707 static void svm_vcpu_put(struct kvm_vcpu
*vcpu
)
709 struct vcpu_svm
*svm
= to_svm(vcpu
);
712 ++vcpu
->stat
.host_state_reload
;
713 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
714 wrmsrl(host_save_user_msrs
[i
], svm
->host_user_msrs
[i
]);
716 rdtscll(vcpu
->arch
.host_tsc
);
719 static unsigned long svm_get_rflags(struct kvm_vcpu
*vcpu
)
721 return to_svm(vcpu
)->vmcb
->save
.rflags
;
724 static void svm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
726 to_svm(vcpu
)->vmcb
->save
.rflags
= rflags
;
729 static void svm_set_vintr(struct vcpu_svm
*svm
)
731 svm
->vmcb
->control
.intercept
|= 1ULL << INTERCEPT_VINTR
;
734 static void svm_clear_vintr(struct vcpu_svm
*svm
)
736 svm
->vmcb
->control
.intercept
&= ~(1ULL << INTERCEPT_VINTR
);
739 static struct vmcb_seg
*svm_seg(struct kvm_vcpu
*vcpu
, int seg
)
741 struct vmcb_save_area
*save
= &to_svm(vcpu
)->vmcb
->save
;
744 case VCPU_SREG_CS
: return &save
->cs
;
745 case VCPU_SREG_DS
: return &save
->ds
;
746 case VCPU_SREG_ES
: return &save
->es
;
747 case VCPU_SREG_FS
: return &save
->fs
;
748 case VCPU_SREG_GS
: return &save
->gs
;
749 case VCPU_SREG_SS
: return &save
->ss
;
750 case VCPU_SREG_TR
: return &save
->tr
;
751 case VCPU_SREG_LDTR
: return &save
->ldtr
;
757 static u64
svm_get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
759 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
764 static void svm_get_segment(struct kvm_vcpu
*vcpu
,
765 struct kvm_segment
*var
, int seg
)
767 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
770 var
->limit
= s
->limit
;
771 var
->selector
= s
->selector
;
772 var
->type
= s
->attrib
& SVM_SELECTOR_TYPE_MASK
;
773 var
->s
= (s
->attrib
>> SVM_SELECTOR_S_SHIFT
) & 1;
774 var
->dpl
= (s
->attrib
>> SVM_SELECTOR_DPL_SHIFT
) & 3;
775 var
->present
= (s
->attrib
>> SVM_SELECTOR_P_SHIFT
) & 1;
776 var
->avl
= (s
->attrib
>> SVM_SELECTOR_AVL_SHIFT
) & 1;
777 var
->l
= (s
->attrib
>> SVM_SELECTOR_L_SHIFT
) & 1;
778 var
->db
= (s
->attrib
>> SVM_SELECTOR_DB_SHIFT
) & 1;
779 var
->g
= (s
->attrib
>> SVM_SELECTOR_G_SHIFT
) & 1;
781 /* AMD's VMCB does not have an explicit unusable field, so emulate it
782 * for cross vendor migration purposes by "not present"
784 var
->unusable
= !var
->present
|| (var
->type
== 0);
789 * SVM always stores 0 for the 'G' bit in the CS selector in
790 * the VMCB on a VMEXIT. This hurts cross-vendor migration:
791 * Intel's VMENTRY has a check on the 'G' bit.
793 var
->g
= s
->limit
> 0xfffff;
797 * Work around a bug where the busy flag in the tr selector
807 * The accessed bit must always be set in the segment
808 * descriptor cache, although it can be cleared in the
809 * descriptor, the cached bit always remains at 1. Since
810 * Intel has a check on this, set it here to support
811 * cross-vendor migration.
819 static int svm_get_cpl(struct kvm_vcpu
*vcpu
)
821 struct vmcb_save_area
*save
= &to_svm(vcpu
)->vmcb
->save
;
826 static void svm_get_idt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
828 struct vcpu_svm
*svm
= to_svm(vcpu
);
830 dt
->limit
= svm
->vmcb
->save
.idtr
.limit
;
831 dt
->base
= svm
->vmcb
->save
.idtr
.base
;
834 static void svm_set_idt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
836 struct vcpu_svm
*svm
= to_svm(vcpu
);
838 svm
->vmcb
->save
.idtr
.limit
= dt
->limit
;
839 svm
->vmcb
->save
.idtr
.base
= dt
->base
;
842 static void svm_get_gdt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
844 struct vcpu_svm
*svm
= to_svm(vcpu
);
846 dt
->limit
= svm
->vmcb
->save
.gdtr
.limit
;
847 dt
->base
= svm
->vmcb
->save
.gdtr
.base
;
850 static void svm_set_gdt(struct kvm_vcpu
*vcpu
, struct descriptor_table
*dt
)
852 struct vcpu_svm
*svm
= to_svm(vcpu
);
854 svm
->vmcb
->save
.gdtr
.limit
= dt
->limit
;
855 svm
->vmcb
->save
.gdtr
.base
= dt
->base
;
858 static void svm_decache_cr4_guest_bits(struct kvm_vcpu
*vcpu
)
862 static void svm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
864 struct vcpu_svm
*svm
= to_svm(vcpu
);
867 if (vcpu
->arch
.shadow_efer
& EFER_LME
) {
868 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
869 vcpu
->arch
.shadow_efer
|= EFER_LMA
;
870 svm
->vmcb
->save
.efer
|= EFER_LMA
| EFER_LME
;
873 if (is_paging(vcpu
) && !(cr0
& X86_CR0_PG
)) {
874 vcpu
->arch
.shadow_efer
&= ~EFER_LMA
;
875 svm
->vmcb
->save
.efer
&= ~(EFER_LMA
| EFER_LME
);
882 if ((vcpu
->arch
.cr0
& X86_CR0_TS
) && !(cr0
& X86_CR0_TS
)) {
883 svm
->vmcb
->control
.intercept_exceptions
&= ~(1 << NM_VECTOR
);
884 vcpu
->fpu_active
= 1;
887 vcpu
->arch
.cr0
= cr0
;
888 cr0
|= X86_CR0_PG
| X86_CR0_WP
;
889 if (!vcpu
->fpu_active
) {
890 svm
->vmcb
->control
.intercept_exceptions
|= (1 << NM_VECTOR
);
895 * re-enable caching here because the QEMU bios
896 * does not do it - this results in some delay at
899 cr0
&= ~(X86_CR0_CD
| X86_CR0_NW
);
900 svm
->vmcb
->save
.cr0
= cr0
;
903 static void svm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
905 unsigned long host_cr4_mce
= read_cr4() & X86_CR4_MCE
;
906 unsigned long old_cr4
= to_svm(vcpu
)->vmcb
->save
.cr4
;
908 if (npt_enabled
&& ((old_cr4
^ cr4
) & X86_CR4_PGE
))
909 force_new_asid(vcpu
);
911 vcpu
->arch
.cr4
= cr4
;
915 to_svm(vcpu
)->vmcb
->save
.cr4
= cr4
;
918 static void svm_set_segment(struct kvm_vcpu
*vcpu
,
919 struct kvm_segment
*var
, int seg
)
921 struct vcpu_svm
*svm
= to_svm(vcpu
);
922 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
925 s
->limit
= var
->limit
;
926 s
->selector
= var
->selector
;
930 s
->attrib
= (var
->type
& SVM_SELECTOR_TYPE_MASK
);
931 s
->attrib
|= (var
->s
& 1) << SVM_SELECTOR_S_SHIFT
;
932 s
->attrib
|= (var
->dpl
& 3) << SVM_SELECTOR_DPL_SHIFT
;
933 s
->attrib
|= (var
->present
& 1) << SVM_SELECTOR_P_SHIFT
;
934 s
->attrib
|= (var
->avl
& 1) << SVM_SELECTOR_AVL_SHIFT
;
935 s
->attrib
|= (var
->l
& 1) << SVM_SELECTOR_L_SHIFT
;
936 s
->attrib
|= (var
->db
& 1) << SVM_SELECTOR_DB_SHIFT
;
937 s
->attrib
|= (var
->g
& 1) << SVM_SELECTOR_G_SHIFT
;
939 if (seg
== VCPU_SREG_CS
)
941 = (svm
->vmcb
->save
.cs
.attrib
942 >> SVM_SELECTOR_DPL_SHIFT
) & 3;
946 static int svm_guest_debug(struct kvm_vcpu
*vcpu
, struct kvm_guest_debug
*dbg
)
948 int old_debug
= vcpu
->guest_debug
;
949 struct vcpu_svm
*svm
= to_svm(vcpu
);
951 vcpu
->guest_debug
= dbg
->control
;
953 svm
->vmcb
->control
.intercept_exceptions
&=
954 ~((1 << DB_VECTOR
) | (1 << BP_VECTOR
));
955 if (vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
) {
956 if (vcpu
->guest_debug
&
957 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
))
958 svm
->vmcb
->control
.intercept_exceptions
|=
960 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_SW_BP
)
961 svm
->vmcb
->control
.intercept_exceptions
|=
964 vcpu
->guest_debug
= 0;
966 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
967 svm
->vmcb
->save
.dr7
= dbg
->arch
.debugreg
[7];
969 svm
->vmcb
->save
.dr7
= vcpu
->arch
.dr7
;
971 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
972 svm
->vmcb
->save
.rflags
|= X86_EFLAGS_TF
| X86_EFLAGS_RF
;
973 else if (old_debug
& KVM_GUESTDBG_SINGLESTEP
)
974 svm
->vmcb
->save
.rflags
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
979 static int svm_get_irq(struct kvm_vcpu
*vcpu
)
981 struct vcpu_svm
*svm
= to_svm(vcpu
);
982 u32 exit_int_info
= svm
->vmcb
->control
.exit_int_info
;
984 if (is_external_interrupt(exit_int_info
))
985 return exit_int_info
& SVM_EVTINJ_VEC_MASK
;
989 static void load_host_msrs(struct kvm_vcpu
*vcpu
)
992 wrmsrl(MSR_GS_BASE
, to_svm(vcpu
)->host_gs_base
);
996 static void save_host_msrs(struct kvm_vcpu
*vcpu
)
999 rdmsrl(MSR_GS_BASE
, to_svm(vcpu
)->host_gs_base
);
1003 static void new_asid(struct vcpu_svm
*svm
, struct svm_cpu_data
*svm_data
)
1005 if (svm_data
->next_asid
> svm_data
->max_asid
) {
1006 ++svm_data
->asid_generation
;
1007 svm_data
->next_asid
= 1;
1008 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ALL_ASID
;
1011 svm
->vcpu
.cpu
= svm_data
->cpu
;
1012 svm
->asid_generation
= svm_data
->asid_generation
;
1013 svm
->vmcb
->control
.asid
= svm_data
->next_asid
++;
1016 static unsigned long svm_get_dr(struct kvm_vcpu
*vcpu
, int dr
)
1018 struct vcpu_svm
*svm
= to_svm(vcpu
);
1023 val
= vcpu
->arch
.db
[dr
];
1026 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
1027 val
= vcpu
->arch
.dr6
;
1029 val
= svm
->vmcb
->save
.dr6
;
1032 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
1033 val
= vcpu
->arch
.dr7
;
1035 val
= svm
->vmcb
->save
.dr7
;
1041 KVMTRACE_2D(DR_READ
, vcpu
, (u32
)dr
, (u32
)val
, handler
);
1045 static void svm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long value
,
1048 struct vcpu_svm
*svm
= to_svm(vcpu
);
1050 KVMTRACE_2D(DR_WRITE
, vcpu
, (u32
)dr
, (u32
)value
, handler
);
1056 vcpu
->arch
.db
[dr
] = value
;
1057 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
))
1058 vcpu
->arch
.eff_db
[dr
] = value
;
1061 if (vcpu
->arch
.cr4
& X86_CR4_DE
)
1062 *exception
= UD_VECTOR
;
1065 if (value
& 0xffffffff00000000ULL
) {
1066 *exception
= GP_VECTOR
;
1069 vcpu
->arch
.dr6
= (value
& DR6_VOLATILE
) | DR6_FIXED_1
;
1072 if (value
& 0xffffffff00000000ULL
) {
1073 *exception
= GP_VECTOR
;
1076 vcpu
->arch
.dr7
= (value
& DR7_VOLATILE
) | DR7_FIXED_1
;
1077 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)) {
1078 svm
->vmcb
->save
.dr7
= vcpu
->arch
.dr7
;
1079 vcpu
->arch
.switch_db_regs
= (value
& DR7_BP_EN_MASK
);
1083 /* FIXME: Possible case? */
1084 printk(KERN_DEBUG
"%s: unexpected dr %u\n",
1086 *exception
= UD_VECTOR
;
1091 static int pf_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1093 u32 exit_int_info
= svm
->vmcb
->control
.exit_int_info
;
1094 struct kvm
*kvm
= svm
->vcpu
.kvm
;
1097 bool event_injection
= false;
1099 if (!irqchip_in_kernel(kvm
) &&
1100 is_external_interrupt(exit_int_info
)) {
1101 event_injection
= true;
1102 kvm_push_irq(&svm
->vcpu
, exit_int_info
& SVM_EVTINJ_VEC_MASK
);
1105 fault_address
= svm
->vmcb
->control
.exit_info_2
;
1106 error_code
= svm
->vmcb
->control
.exit_info_1
;
1109 KVMTRACE_3D(PAGE_FAULT
, &svm
->vcpu
, error_code
,
1110 (u32
)fault_address
, (u32
)(fault_address
>> 32),
1113 KVMTRACE_3D(TDP_FAULT
, &svm
->vcpu
, error_code
,
1114 (u32
)fault_address
, (u32
)(fault_address
>> 32),
1117 * FIXME: Tis shouldn't be necessary here, but there is a flush
1118 * missing in the MMU code. Until we find this bug, flush the
1119 * complete TLB here on an NPF
1122 svm_flush_tlb(&svm
->vcpu
);
1124 if (!npt_enabled
&& event_injection
)
1125 kvm_mmu_unprotect_page_virt(&svm
->vcpu
, fault_address
);
1126 return kvm_mmu_page_fault(&svm
->vcpu
, fault_address
, error_code
);
1129 static int db_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1131 if (!(svm
->vcpu
.guest_debug
&
1132 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
))) {
1133 kvm_queue_exception(&svm
->vcpu
, DB_VECTOR
);
1136 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
1137 kvm_run
->debug
.arch
.pc
= svm
->vmcb
->save
.cs
.base
+ svm
->vmcb
->save
.rip
;
1138 kvm_run
->debug
.arch
.exception
= DB_VECTOR
;
1142 static int bp_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1144 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
1145 kvm_run
->debug
.arch
.pc
= svm
->vmcb
->save
.cs
.base
+ svm
->vmcb
->save
.rip
;
1146 kvm_run
->debug
.arch
.exception
= BP_VECTOR
;
1150 static int ud_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1154 er
= emulate_instruction(&svm
->vcpu
, kvm_run
, 0, 0, EMULTYPE_TRAP_UD
);
1155 if (er
!= EMULATE_DONE
)
1156 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
1160 static int nm_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1162 svm
->vmcb
->control
.intercept_exceptions
&= ~(1 << NM_VECTOR
);
1163 if (!(svm
->vcpu
.arch
.cr0
& X86_CR0_TS
))
1164 svm
->vmcb
->save
.cr0
&= ~X86_CR0_TS
;
1165 svm
->vcpu
.fpu_active
= 1;
1170 static int mc_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1173 * On an #MC intercept the MCE handler is not called automatically in
1174 * the host. So do it by hand here.
1178 /* not sure if we ever come back to this point */
1183 static int shutdown_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1186 * VMCB is undefined after a SHUTDOWN intercept
1187 * so reinitialize it.
1189 clear_page(svm
->vmcb
);
1192 kvm_run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
1196 static int io_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1198 u32 io_info
= svm
->vmcb
->control
.exit_info_1
; /* address size bug? */
1199 int size
, in
, string
;
1202 ++svm
->vcpu
.stat
.io_exits
;
1204 svm
->next_rip
= svm
->vmcb
->control
.exit_info_2
;
1206 string
= (io_info
& SVM_IOIO_STR_MASK
) != 0;
1209 if (emulate_instruction(&svm
->vcpu
,
1210 kvm_run
, 0, 0, 0) == EMULATE_DO_MMIO
)
1215 in
= (io_info
& SVM_IOIO_TYPE_MASK
) != 0;
1216 port
= io_info
>> 16;
1217 size
= (io_info
& SVM_IOIO_SIZE_MASK
) >> SVM_IOIO_SIZE_SHIFT
;
1219 skip_emulated_instruction(&svm
->vcpu
);
1220 return kvm_emulate_pio(&svm
->vcpu
, kvm_run
, in
, size
, port
);
1223 static int nmi_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1225 KVMTRACE_0D(NMI
, &svm
->vcpu
, handler
);
1229 static int intr_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1231 ++svm
->vcpu
.stat
.irq_exits
;
1232 KVMTRACE_0D(INTR
, &svm
->vcpu
, handler
);
1236 static int nop_on_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1241 static int halt_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1243 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 1;
1244 skip_emulated_instruction(&svm
->vcpu
);
1245 return kvm_emulate_halt(&svm
->vcpu
);
1248 static int vmmcall_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1250 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1251 skip_emulated_instruction(&svm
->vcpu
);
1252 kvm_emulate_hypercall(&svm
->vcpu
);
1256 static int nested_svm_check_permissions(struct vcpu_svm
*svm
)
1258 if (!(svm
->vcpu
.arch
.shadow_efer
& EFER_SVME
)
1259 || !is_paging(&svm
->vcpu
)) {
1260 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
1264 if (svm
->vmcb
->save
.cpl
) {
1265 kvm_inject_gp(&svm
->vcpu
, 0);
1272 static int nested_svm_check_exception(struct vcpu_svm
*svm
, unsigned nr
,
1273 bool has_error_code
, u32 error_code
)
1275 if (is_nested(svm
)) {
1276 svm
->vmcb
->control
.exit_code
= SVM_EXIT_EXCP_BASE
+ nr
;
1277 svm
->vmcb
->control
.exit_code_hi
= 0;
1278 svm
->vmcb
->control
.exit_info_1
= error_code
;
1279 svm
->vmcb
->control
.exit_info_2
= svm
->vcpu
.arch
.cr2
;
1280 if (nested_svm_exit_handled(svm
, false)) {
1281 nsvm_printk("VMexit -> EXCP 0x%x\n", nr
);
1283 nested_svm_vmexit(svm
);
1291 static inline int nested_svm_intr(struct vcpu_svm
*svm
)
1293 if (is_nested(svm
)) {
1294 if (!(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
))
1297 if (!(svm
->vcpu
.arch
.hflags
& HF_HIF_MASK
))
1300 svm
->vmcb
->control
.exit_code
= SVM_EXIT_INTR
;
1302 if (nested_svm_exit_handled(svm
, false)) {
1303 nsvm_printk("VMexit -> INTR\n");
1304 nested_svm_vmexit(svm
);
1312 static struct page
*nested_svm_get_page(struct vcpu_svm
*svm
, u64 gpa
)
1316 down_read(¤t
->mm
->mmap_sem
);
1317 page
= gfn_to_page(svm
->vcpu
.kvm
, gpa
>> PAGE_SHIFT
);
1318 up_read(¤t
->mm
->mmap_sem
);
1320 if (is_error_page(page
)) {
1321 printk(KERN_INFO
"%s: could not find page at 0x%llx\n",
1323 kvm_release_page_clean(page
);
1324 kvm_inject_gp(&svm
->vcpu
, 0);
1330 static int nested_svm_do(struct vcpu_svm
*svm
,
1331 u64 arg1_gpa
, u64 arg2_gpa
, void *opaque
,
1332 int (*handler
)(struct vcpu_svm
*svm
,
1337 struct page
*arg1_page
;
1338 struct page
*arg2_page
= NULL
;
1343 arg1_page
= nested_svm_get_page(svm
, arg1_gpa
);
1344 if(arg1_page
== NULL
)
1348 arg2_page
= nested_svm_get_page(svm
, arg2_gpa
);
1349 if(arg2_page
== NULL
) {
1350 kvm_release_page_clean(arg1_page
);
1355 arg1
= kmap_atomic(arg1_page
, KM_USER0
);
1357 arg2
= kmap_atomic(arg2_page
, KM_USER1
);
1359 retval
= handler(svm
, arg1
, arg2
, opaque
);
1361 kunmap_atomic(arg1
, KM_USER0
);
1363 kunmap_atomic(arg2
, KM_USER1
);
1365 kvm_release_page_dirty(arg1_page
);
1367 kvm_release_page_dirty(arg2_page
);
1372 static int nested_svm_exit_handled_real(struct vcpu_svm
*svm
,
1377 struct vmcb
*nested_vmcb
= (struct vmcb
*)arg1
;
1378 bool kvm_overrides
= *(bool *)opaque
;
1379 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
1381 if (kvm_overrides
) {
1382 switch (exit_code
) {
1386 /* For now we are always handling NPFs when using them */
1391 /* When we're shadowing, trap PFs */
1392 case SVM_EXIT_EXCP_BASE
+ PF_VECTOR
:
1401 switch (exit_code
) {
1402 case SVM_EXIT_READ_CR0
... SVM_EXIT_READ_CR8
: {
1403 u32 cr_bits
= 1 << (exit_code
- SVM_EXIT_READ_CR0
);
1404 if (nested_vmcb
->control
.intercept_cr_read
& cr_bits
)
1408 case SVM_EXIT_WRITE_CR0
... SVM_EXIT_WRITE_CR8
: {
1409 u32 cr_bits
= 1 << (exit_code
- SVM_EXIT_WRITE_CR0
);
1410 if (nested_vmcb
->control
.intercept_cr_write
& cr_bits
)
1414 case SVM_EXIT_READ_DR0
... SVM_EXIT_READ_DR7
: {
1415 u32 dr_bits
= 1 << (exit_code
- SVM_EXIT_READ_DR0
);
1416 if (nested_vmcb
->control
.intercept_dr_read
& dr_bits
)
1420 case SVM_EXIT_WRITE_DR0
... SVM_EXIT_WRITE_DR7
: {
1421 u32 dr_bits
= 1 << (exit_code
- SVM_EXIT_WRITE_DR0
);
1422 if (nested_vmcb
->control
.intercept_dr_write
& dr_bits
)
1426 case SVM_EXIT_EXCP_BASE
... SVM_EXIT_EXCP_BASE
+ 0x1f: {
1427 u32 excp_bits
= 1 << (exit_code
- SVM_EXIT_EXCP_BASE
);
1428 if (nested_vmcb
->control
.intercept_exceptions
& excp_bits
)
1433 u64 exit_bits
= 1ULL << (exit_code
- SVM_EXIT_INTR
);
1434 nsvm_printk("exit code: 0x%x\n", exit_code
);
1435 if (nested_vmcb
->control
.intercept
& exit_bits
)
1443 static int nested_svm_exit_handled_msr(struct vcpu_svm
*svm
,
1444 void *arg1
, void *arg2
,
1447 struct vmcb
*nested_vmcb
= (struct vmcb
*)arg1
;
1448 u8
*msrpm
= (u8
*)arg2
;
1450 u32 msr
= svm
->vcpu
.arch
.regs
[VCPU_REGS_RCX
];
1451 u32 param
= svm
->vmcb
->control
.exit_info_1
& 1;
1453 if (!(nested_vmcb
->control
.intercept
& (1ULL << INTERCEPT_MSR_PROT
)))
1461 case 0xc0000000 ... 0xc0001fff:
1462 t0
= (8192 + msr
- 0xc0000000) * 2;
1466 case 0xc0010000 ... 0xc0011fff:
1467 t0
= (16384 + msr
- 0xc0010000) * 2;
1475 if (msrpm
[t1
] & ((1 << param
) << t0
))
1481 static int nested_svm_exit_handled(struct vcpu_svm
*svm
, bool kvm_override
)
1483 bool k
= kvm_override
;
1485 switch (svm
->vmcb
->control
.exit_code
) {
1487 return nested_svm_do(svm
, svm
->nested_vmcb
,
1488 svm
->nested_vmcb_msrpm
, NULL
,
1489 nested_svm_exit_handled_msr
);
1493 return nested_svm_do(svm
, svm
->nested_vmcb
, 0, &k
,
1494 nested_svm_exit_handled_real
);
1497 static int nested_svm_vmexit_real(struct vcpu_svm
*svm
, void *arg1
,
1498 void *arg2
, void *opaque
)
1500 struct vmcb
*nested_vmcb
= (struct vmcb
*)arg1
;
1501 struct vmcb
*hsave
= svm
->hsave
;
1502 u64 nested_save
[] = { nested_vmcb
->save
.cr0
,
1503 nested_vmcb
->save
.cr3
,
1504 nested_vmcb
->save
.cr4
,
1505 nested_vmcb
->save
.efer
,
1506 nested_vmcb
->control
.intercept_cr_read
,
1507 nested_vmcb
->control
.intercept_cr_write
,
1508 nested_vmcb
->control
.intercept_dr_read
,
1509 nested_vmcb
->control
.intercept_dr_write
,
1510 nested_vmcb
->control
.intercept_exceptions
,
1511 nested_vmcb
->control
.intercept
,
1512 nested_vmcb
->control
.msrpm_base_pa
,
1513 nested_vmcb
->control
.iopm_base_pa
,
1514 nested_vmcb
->control
.tsc_offset
};
1516 /* Give the current vmcb to the guest */
1517 memcpy(nested_vmcb
, svm
->vmcb
, sizeof(struct vmcb
));
1518 nested_vmcb
->save
.cr0
= nested_save
[0];
1520 nested_vmcb
->save
.cr3
= nested_save
[1];
1521 nested_vmcb
->save
.cr4
= nested_save
[2];
1522 nested_vmcb
->save
.efer
= nested_save
[3];
1523 nested_vmcb
->control
.intercept_cr_read
= nested_save
[4];
1524 nested_vmcb
->control
.intercept_cr_write
= nested_save
[5];
1525 nested_vmcb
->control
.intercept_dr_read
= nested_save
[6];
1526 nested_vmcb
->control
.intercept_dr_write
= nested_save
[7];
1527 nested_vmcb
->control
.intercept_exceptions
= nested_save
[8];
1528 nested_vmcb
->control
.intercept
= nested_save
[9];
1529 nested_vmcb
->control
.msrpm_base_pa
= nested_save
[10];
1530 nested_vmcb
->control
.iopm_base_pa
= nested_save
[11];
1531 nested_vmcb
->control
.tsc_offset
= nested_save
[12];
1533 /* We always set V_INTR_MASKING and remember the old value in hflags */
1534 if (!(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
))
1535 nested_vmcb
->control
.int_ctl
&= ~V_INTR_MASKING_MASK
;
1537 if ((nested_vmcb
->control
.int_ctl
& V_IRQ_MASK
) &&
1538 (nested_vmcb
->control
.int_vector
)) {
1539 nsvm_printk("WARNING: IRQ 0x%x still enabled on #VMEXIT\n",
1540 nested_vmcb
->control
.int_vector
);
1543 /* Restore the original control entries */
1544 svm
->vmcb
->control
= hsave
->control
;
1546 /* Kill any pending exceptions */
1547 if (svm
->vcpu
.arch
.exception
.pending
== true)
1548 nsvm_printk("WARNING: Pending Exception\n");
1549 svm
->vcpu
.arch
.exception
.pending
= false;
1551 /* Restore selected save entries */
1552 svm
->vmcb
->save
.es
= hsave
->save
.es
;
1553 svm
->vmcb
->save
.cs
= hsave
->save
.cs
;
1554 svm
->vmcb
->save
.ss
= hsave
->save
.ss
;
1555 svm
->vmcb
->save
.ds
= hsave
->save
.ds
;
1556 svm
->vmcb
->save
.gdtr
= hsave
->save
.gdtr
;
1557 svm
->vmcb
->save
.idtr
= hsave
->save
.idtr
;
1558 svm
->vmcb
->save
.rflags
= hsave
->save
.rflags
;
1559 svm_set_efer(&svm
->vcpu
, hsave
->save
.efer
);
1560 svm_set_cr0(&svm
->vcpu
, hsave
->save
.cr0
| X86_CR0_PE
);
1561 svm_set_cr4(&svm
->vcpu
, hsave
->save
.cr4
);
1563 svm
->vmcb
->save
.cr3
= hsave
->save
.cr3
;
1564 svm
->vcpu
.arch
.cr3
= hsave
->save
.cr3
;
1566 kvm_set_cr3(&svm
->vcpu
, hsave
->save
.cr3
);
1568 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
, hsave
->save
.rax
);
1569 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RSP
, hsave
->save
.rsp
);
1570 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RIP
, hsave
->save
.rip
);
1571 svm
->vmcb
->save
.dr7
= 0;
1572 svm
->vmcb
->save
.cpl
= 0;
1573 svm
->vmcb
->control
.exit_int_info
= 0;
1575 svm
->vcpu
.arch
.hflags
&= ~HF_GIF_MASK
;
1576 /* Exit nested SVM mode */
1577 svm
->nested_vmcb
= 0;
1582 static int nested_svm_vmexit(struct vcpu_svm
*svm
)
1584 nsvm_printk("VMexit\n");
1585 if (nested_svm_do(svm
, svm
->nested_vmcb
, 0,
1586 NULL
, nested_svm_vmexit_real
))
1589 kvm_mmu_reset_context(&svm
->vcpu
);
1590 kvm_mmu_load(&svm
->vcpu
);
1595 static int nested_svm_vmrun_msrpm(struct vcpu_svm
*svm
, void *arg1
,
1596 void *arg2
, void *opaque
)
1599 u32
*nested_msrpm
= (u32
*)arg1
;
1600 for (i
=0; i
< PAGE_SIZE
* (1 << MSRPM_ALLOC_ORDER
) / 4; i
++)
1601 svm
->nested_msrpm
[i
] = svm
->msrpm
[i
] | nested_msrpm
[i
];
1602 svm
->vmcb
->control
.msrpm_base_pa
= __pa(svm
->nested_msrpm
);
1607 static int nested_svm_vmrun(struct vcpu_svm
*svm
, void *arg1
,
1608 void *arg2
, void *opaque
)
1610 struct vmcb
*nested_vmcb
= (struct vmcb
*)arg1
;
1611 struct vmcb
*hsave
= svm
->hsave
;
1613 /* nested_vmcb is our indicator if nested SVM is activated */
1614 svm
->nested_vmcb
= svm
->vmcb
->save
.rax
;
1616 /* Clear internal status */
1617 svm
->vcpu
.arch
.exception
.pending
= false;
1619 /* Save the old vmcb, so we don't need to pick what we save, but
1620 can restore everything when a VMEXIT occurs */
1621 memcpy(hsave
, svm
->vmcb
, sizeof(struct vmcb
));
1622 /* We need to remember the original CR3 in the SPT case */
1624 hsave
->save
.cr3
= svm
->vcpu
.arch
.cr3
;
1625 hsave
->save
.cr4
= svm
->vcpu
.arch
.cr4
;
1626 hsave
->save
.rip
= svm
->next_rip
;
1628 if (svm
->vmcb
->save
.rflags
& X86_EFLAGS_IF
)
1629 svm
->vcpu
.arch
.hflags
|= HF_HIF_MASK
;
1631 svm
->vcpu
.arch
.hflags
&= ~HF_HIF_MASK
;
1633 /* Load the nested guest state */
1634 svm
->vmcb
->save
.es
= nested_vmcb
->save
.es
;
1635 svm
->vmcb
->save
.cs
= nested_vmcb
->save
.cs
;
1636 svm
->vmcb
->save
.ss
= nested_vmcb
->save
.ss
;
1637 svm
->vmcb
->save
.ds
= nested_vmcb
->save
.ds
;
1638 svm
->vmcb
->save
.gdtr
= nested_vmcb
->save
.gdtr
;
1639 svm
->vmcb
->save
.idtr
= nested_vmcb
->save
.idtr
;
1640 svm
->vmcb
->save
.rflags
= nested_vmcb
->save
.rflags
;
1641 svm_set_efer(&svm
->vcpu
, nested_vmcb
->save
.efer
);
1642 svm_set_cr0(&svm
->vcpu
, nested_vmcb
->save
.cr0
);
1643 svm_set_cr4(&svm
->vcpu
, nested_vmcb
->save
.cr4
);
1645 svm
->vmcb
->save
.cr3
= nested_vmcb
->save
.cr3
;
1646 svm
->vcpu
.arch
.cr3
= nested_vmcb
->save
.cr3
;
1648 kvm_set_cr3(&svm
->vcpu
, nested_vmcb
->save
.cr3
);
1649 kvm_mmu_reset_context(&svm
->vcpu
);
1651 svm
->vmcb
->save
.cr2
= nested_vmcb
->save
.cr2
;
1652 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
, nested_vmcb
->save
.rax
);
1653 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RSP
, nested_vmcb
->save
.rsp
);
1654 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RIP
, nested_vmcb
->save
.rip
);
1655 /* In case we don't even reach vcpu_run, the fields are not updated */
1656 svm
->vmcb
->save
.rax
= nested_vmcb
->save
.rax
;
1657 svm
->vmcb
->save
.rsp
= nested_vmcb
->save
.rsp
;
1658 svm
->vmcb
->save
.rip
= nested_vmcb
->save
.rip
;
1659 svm
->vmcb
->save
.dr7
= nested_vmcb
->save
.dr7
;
1660 svm
->vmcb
->save
.dr6
= nested_vmcb
->save
.dr6
;
1661 svm
->vmcb
->save
.cpl
= nested_vmcb
->save
.cpl
;
1663 /* We don't want a nested guest to be more powerful than the guest,
1664 so all intercepts are ORed */
1665 svm
->vmcb
->control
.intercept_cr_read
|=
1666 nested_vmcb
->control
.intercept_cr_read
;
1667 svm
->vmcb
->control
.intercept_cr_write
|=
1668 nested_vmcb
->control
.intercept_cr_write
;
1669 svm
->vmcb
->control
.intercept_dr_read
|=
1670 nested_vmcb
->control
.intercept_dr_read
;
1671 svm
->vmcb
->control
.intercept_dr_write
|=
1672 nested_vmcb
->control
.intercept_dr_write
;
1673 svm
->vmcb
->control
.intercept_exceptions
|=
1674 nested_vmcb
->control
.intercept_exceptions
;
1676 svm
->vmcb
->control
.intercept
|= nested_vmcb
->control
.intercept
;
1678 svm
->nested_vmcb_msrpm
= nested_vmcb
->control
.msrpm_base_pa
;
1680 force_new_asid(&svm
->vcpu
);
1681 svm
->vmcb
->control
.exit_int_info
= nested_vmcb
->control
.exit_int_info
;
1682 svm
->vmcb
->control
.exit_int_info_err
= nested_vmcb
->control
.exit_int_info_err
;
1683 svm
->vmcb
->control
.int_ctl
= nested_vmcb
->control
.int_ctl
| V_INTR_MASKING_MASK
;
1684 if (nested_vmcb
->control
.int_ctl
& V_IRQ_MASK
) {
1685 nsvm_printk("nSVM Injecting Interrupt: 0x%x\n",
1686 nested_vmcb
->control
.int_ctl
);
1688 if (nested_vmcb
->control
.int_ctl
& V_INTR_MASKING_MASK
)
1689 svm
->vcpu
.arch
.hflags
|= HF_VINTR_MASK
;
1691 svm
->vcpu
.arch
.hflags
&= ~HF_VINTR_MASK
;
1693 nsvm_printk("nSVM exit_int_info: 0x%x | int_state: 0x%x\n",
1694 nested_vmcb
->control
.exit_int_info
,
1695 nested_vmcb
->control
.int_state
);
1697 svm
->vmcb
->control
.int_vector
= nested_vmcb
->control
.int_vector
;
1698 svm
->vmcb
->control
.int_state
= nested_vmcb
->control
.int_state
;
1699 svm
->vmcb
->control
.tsc_offset
+= nested_vmcb
->control
.tsc_offset
;
1700 if (nested_vmcb
->control
.event_inj
& SVM_EVTINJ_VALID
)
1701 nsvm_printk("Injecting Event: 0x%x\n",
1702 nested_vmcb
->control
.event_inj
);
1703 svm
->vmcb
->control
.event_inj
= nested_vmcb
->control
.event_inj
;
1704 svm
->vmcb
->control
.event_inj_err
= nested_vmcb
->control
.event_inj_err
;
1706 svm
->vcpu
.arch
.hflags
|= HF_GIF_MASK
;
1711 static int nested_svm_vmloadsave(struct vmcb
*from_vmcb
, struct vmcb
*to_vmcb
)
1713 to_vmcb
->save
.fs
= from_vmcb
->save
.fs
;
1714 to_vmcb
->save
.gs
= from_vmcb
->save
.gs
;
1715 to_vmcb
->save
.tr
= from_vmcb
->save
.tr
;
1716 to_vmcb
->save
.ldtr
= from_vmcb
->save
.ldtr
;
1717 to_vmcb
->save
.kernel_gs_base
= from_vmcb
->save
.kernel_gs_base
;
1718 to_vmcb
->save
.star
= from_vmcb
->save
.star
;
1719 to_vmcb
->save
.lstar
= from_vmcb
->save
.lstar
;
1720 to_vmcb
->save
.cstar
= from_vmcb
->save
.cstar
;
1721 to_vmcb
->save
.sfmask
= from_vmcb
->save
.sfmask
;
1722 to_vmcb
->save
.sysenter_cs
= from_vmcb
->save
.sysenter_cs
;
1723 to_vmcb
->save
.sysenter_esp
= from_vmcb
->save
.sysenter_esp
;
1724 to_vmcb
->save
.sysenter_eip
= from_vmcb
->save
.sysenter_eip
;
1729 static int nested_svm_vmload(struct vcpu_svm
*svm
, void *nested_vmcb
,
1730 void *arg2
, void *opaque
)
1732 return nested_svm_vmloadsave((struct vmcb
*)nested_vmcb
, svm
->vmcb
);
1735 static int nested_svm_vmsave(struct vcpu_svm
*svm
, void *nested_vmcb
,
1736 void *arg2
, void *opaque
)
1738 return nested_svm_vmloadsave(svm
->vmcb
, (struct vmcb
*)nested_vmcb
);
1741 static int vmload_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1743 if (nested_svm_check_permissions(svm
))
1746 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1747 skip_emulated_instruction(&svm
->vcpu
);
1749 nested_svm_do(svm
, svm
->vmcb
->save
.rax
, 0, NULL
, nested_svm_vmload
);
1754 static int vmsave_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1756 if (nested_svm_check_permissions(svm
))
1759 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1760 skip_emulated_instruction(&svm
->vcpu
);
1762 nested_svm_do(svm
, svm
->vmcb
->save
.rax
, 0, NULL
, nested_svm_vmsave
);
1767 static int vmrun_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1769 nsvm_printk("VMrun\n");
1770 if (nested_svm_check_permissions(svm
))
1773 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1774 skip_emulated_instruction(&svm
->vcpu
);
1776 if (nested_svm_do(svm
, svm
->vmcb
->save
.rax
, 0,
1777 NULL
, nested_svm_vmrun
))
1780 if (nested_svm_do(svm
, svm
->nested_vmcb_msrpm
, 0,
1781 NULL
, nested_svm_vmrun_msrpm
))
1787 static int stgi_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1789 if (nested_svm_check_permissions(svm
))
1792 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1793 skip_emulated_instruction(&svm
->vcpu
);
1795 svm
->vcpu
.arch
.hflags
|= HF_GIF_MASK
;
1800 static int clgi_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1802 if (nested_svm_check_permissions(svm
))
1805 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1806 skip_emulated_instruction(&svm
->vcpu
);
1808 svm
->vcpu
.arch
.hflags
&= ~HF_GIF_MASK
;
1810 /* After a CLGI no interrupts should come */
1811 svm_clear_vintr(svm
);
1812 svm
->vmcb
->control
.int_ctl
&= ~V_IRQ_MASK
;
1817 static int invalid_op_interception(struct vcpu_svm
*svm
,
1818 struct kvm_run
*kvm_run
)
1820 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
1824 static int task_switch_interception(struct vcpu_svm
*svm
,
1825 struct kvm_run
*kvm_run
)
1829 tss_selector
= (u16
)svm
->vmcb
->control
.exit_info_1
;
1830 if (svm
->vmcb
->control
.exit_info_2
&
1831 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET
))
1832 return kvm_task_switch(&svm
->vcpu
, tss_selector
,
1834 if (svm
->vmcb
->control
.exit_info_2
&
1835 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP
))
1836 return kvm_task_switch(&svm
->vcpu
, tss_selector
,
1838 return kvm_task_switch(&svm
->vcpu
, tss_selector
, TASK_SWITCH_CALL
);
1841 static int cpuid_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1843 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
1844 kvm_emulate_cpuid(&svm
->vcpu
);
1848 static int invlpg_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1850 if (emulate_instruction(&svm
->vcpu
, kvm_run
, 0, 0, 0) != EMULATE_DONE
)
1851 pr_unimpl(&svm
->vcpu
, "%s: failed\n", __func__
);
1855 static int emulate_on_interception(struct vcpu_svm
*svm
,
1856 struct kvm_run
*kvm_run
)
1858 if (emulate_instruction(&svm
->vcpu
, NULL
, 0, 0, 0) != EMULATE_DONE
)
1859 pr_unimpl(&svm
->vcpu
, "%s: failed\n", __func__
);
1863 static int cr8_write_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1865 emulate_instruction(&svm
->vcpu
, NULL
, 0, 0, 0);
1866 if (irqchip_in_kernel(svm
->vcpu
.kvm
))
1868 kvm_run
->exit_reason
= KVM_EXIT_SET_TPR
;
1872 static int svm_get_msr(struct kvm_vcpu
*vcpu
, unsigned ecx
, u64
*data
)
1874 struct vcpu_svm
*svm
= to_svm(vcpu
);
1877 case MSR_IA32_TIME_STAMP_COUNTER
: {
1881 *data
= svm
->vmcb
->control
.tsc_offset
+ tsc
;
1885 *data
= svm
->vmcb
->save
.star
;
1887 #ifdef CONFIG_X86_64
1889 *data
= svm
->vmcb
->save
.lstar
;
1892 *data
= svm
->vmcb
->save
.cstar
;
1894 case MSR_KERNEL_GS_BASE
:
1895 *data
= svm
->vmcb
->save
.kernel_gs_base
;
1897 case MSR_SYSCALL_MASK
:
1898 *data
= svm
->vmcb
->save
.sfmask
;
1901 case MSR_IA32_SYSENTER_CS
:
1902 *data
= svm
->vmcb
->save
.sysenter_cs
;
1904 case MSR_IA32_SYSENTER_EIP
:
1905 *data
= svm
->vmcb
->save
.sysenter_eip
;
1907 case MSR_IA32_SYSENTER_ESP
:
1908 *data
= svm
->vmcb
->save
.sysenter_esp
;
1910 /* Nobody will change the following 5 values in the VMCB so
1911 we can safely return them on rdmsr. They will always be 0
1912 until LBRV is implemented. */
1913 case MSR_IA32_DEBUGCTLMSR
:
1914 *data
= svm
->vmcb
->save
.dbgctl
;
1916 case MSR_IA32_LASTBRANCHFROMIP
:
1917 *data
= svm
->vmcb
->save
.br_from
;
1919 case MSR_IA32_LASTBRANCHTOIP
:
1920 *data
= svm
->vmcb
->save
.br_to
;
1922 case MSR_IA32_LASTINTFROMIP
:
1923 *data
= svm
->vmcb
->save
.last_excp_from
;
1925 case MSR_IA32_LASTINTTOIP
:
1926 *data
= svm
->vmcb
->save
.last_excp_to
;
1928 case MSR_VM_HSAVE_PA
:
1929 *data
= svm
->hsave_msr
;
1934 case MSR_IA32_UCODE_REV
:
1938 return kvm_get_msr_common(vcpu
, ecx
, data
);
1943 static int rdmsr_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
1945 u32 ecx
= svm
->vcpu
.arch
.regs
[VCPU_REGS_RCX
];
1948 if (svm_get_msr(&svm
->vcpu
, ecx
, &data
))
1949 kvm_inject_gp(&svm
->vcpu
, 0);
1951 KVMTRACE_3D(MSR_READ
, &svm
->vcpu
, ecx
, (u32
)data
,
1952 (u32
)(data
>> 32), handler
);
1954 svm
->vcpu
.arch
.regs
[VCPU_REGS_RAX
] = data
& 0xffffffff;
1955 svm
->vcpu
.arch
.regs
[VCPU_REGS_RDX
] = data
>> 32;
1956 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
1957 skip_emulated_instruction(&svm
->vcpu
);
1962 static int svm_set_msr(struct kvm_vcpu
*vcpu
, unsigned ecx
, u64 data
)
1964 struct vcpu_svm
*svm
= to_svm(vcpu
);
1967 case MSR_IA32_TIME_STAMP_COUNTER
: {
1971 svm
->vmcb
->control
.tsc_offset
= data
- tsc
;
1975 svm
->vmcb
->save
.star
= data
;
1977 #ifdef CONFIG_X86_64
1979 svm
->vmcb
->save
.lstar
= data
;
1982 svm
->vmcb
->save
.cstar
= data
;
1984 case MSR_KERNEL_GS_BASE
:
1985 svm
->vmcb
->save
.kernel_gs_base
= data
;
1987 case MSR_SYSCALL_MASK
:
1988 svm
->vmcb
->save
.sfmask
= data
;
1991 case MSR_IA32_SYSENTER_CS
:
1992 svm
->vmcb
->save
.sysenter_cs
= data
;
1994 case MSR_IA32_SYSENTER_EIP
:
1995 svm
->vmcb
->save
.sysenter_eip
= data
;
1997 case MSR_IA32_SYSENTER_ESP
:
1998 svm
->vmcb
->save
.sysenter_esp
= data
;
2000 case MSR_IA32_DEBUGCTLMSR
:
2001 if (!svm_has(SVM_FEATURE_LBRV
)) {
2002 pr_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2006 if (data
& DEBUGCTL_RESERVED_BITS
)
2009 svm
->vmcb
->save
.dbgctl
= data
;
2010 if (data
& (1ULL<<0))
2011 svm_enable_lbrv(svm
);
2013 svm_disable_lbrv(svm
);
2015 case MSR_K7_EVNTSEL0
:
2016 case MSR_K7_EVNTSEL1
:
2017 case MSR_K7_EVNTSEL2
:
2018 case MSR_K7_EVNTSEL3
:
2019 case MSR_K7_PERFCTR0
:
2020 case MSR_K7_PERFCTR1
:
2021 case MSR_K7_PERFCTR2
:
2022 case MSR_K7_PERFCTR3
:
2024 * Just discard all writes to the performance counters; this
2025 * should keep both older linux and windows 64-bit guests
2028 pr_unimpl(vcpu
, "unimplemented perfctr wrmsr: 0x%x data 0x%llx\n", ecx
, data
);
2031 case MSR_VM_HSAVE_PA
:
2032 svm
->hsave_msr
= data
;
2035 return kvm_set_msr_common(vcpu
, ecx
, data
);
2040 static int wrmsr_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
2042 u32 ecx
= svm
->vcpu
.arch
.regs
[VCPU_REGS_RCX
];
2043 u64 data
= (svm
->vcpu
.arch
.regs
[VCPU_REGS_RAX
] & -1u)
2044 | ((u64
)(svm
->vcpu
.arch
.regs
[VCPU_REGS_RDX
] & -1u) << 32);
2046 KVMTRACE_3D(MSR_WRITE
, &svm
->vcpu
, ecx
, (u32
)data
, (u32
)(data
>> 32),
2049 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
2050 if (svm_set_msr(&svm
->vcpu
, ecx
, data
))
2051 kvm_inject_gp(&svm
->vcpu
, 0);
2053 skip_emulated_instruction(&svm
->vcpu
);
2057 static int msr_interception(struct vcpu_svm
*svm
, struct kvm_run
*kvm_run
)
2059 if (svm
->vmcb
->control
.exit_info_1
)
2060 return wrmsr_interception(svm
, kvm_run
);
2062 return rdmsr_interception(svm
, kvm_run
);
2065 static int interrupt_window_interception(struct vcpu_svm
*svm
,
2066 struct kvm_run
*kvm_run
)
2068 KVMTRACE_0D(PEND_INTR
, &svm
->vcpu
, handler
);
2070 svm_clear_vintr(svm
);
2071 svm
->vmcb
->control
.int_ctl
&= ~V_IRQ_MASK
;
2073 * If the user space waits to inject interrupts, exit as soon as
2076 if (kvm_run
->request_interrupt_window
&&
2077 !svm
->vcpu
.arch
.irq_summary
) {
2078 ++svm
->vcpu
.stat
.irq_window_exits
;
2079 kvm_run
->exit_reason
= KVM_EXIT_IRQ_WINDOW_OPEN
;
2086 static int (*svm_exit_handlers
[])(struct vcpu_svm
*svm
,
2087 struct kvm_run
*kvm_run
) = {
2088 [SVM_EXIT_READ_CR0
] = emulate_on_interception
,
2089 [SVM_EXIT_READ_CR3
] = emulate_on_interception
,
2090 [SVM_EXIT_READ_CR4
] = emulate_on_interception
,
2091 [SVM_EXIT_READ_CR8
] = emulate_on_interception
,
2093 [SVM_EXIT_WRITE_CR0
] = emulate_on_interception
,
2094 [SVM_EXIT_WRITE_CR3
] = emulate_on_interception
,
2095 [SVM_EXIT_WRITE_CR4
] = emulate_on_interception
,
2096 [SVM_EXIT_WRITE_CR8
] = cr8_write_interception
,
2097 [SVM_EXIT_READ_DR0
] = emulate_on_interception
,
2098 [SVM_EXIT_READ_DR1
] = emulate_on_interception
,
2099 [SVM_EXIT_READ_DR2
] = emulate_on_interception
,
2100 [SVM_EXIT_READ_DR3
] = emulate_on_interception
,
2101 [SVM_EXIT_WRITE_DR0
] = emulate_on_interception
,
2102 [SVM_EXIT_WRITE_DR1
] = emulate_on_interception
,
2103 [SVM_EXIT_WRITE_DR2
] = emulate_on_interception
,
2104 [SVM_EXIT_WRITE_DR3
] = emulate_on_interception
,
2105 [SVM_EXIT_WRITE_DR5
] = emulate_on_interception
,
2106 [SVM_EXIT_WRITE_DR7
] = emulate_on_interception
,
2107 [SVM_EXIT_EXCP_BASE
+ DB_VECTOR
] = db_interception
,
2108 [SVM_EXIT_EXCP_BASE
+ BP_VECTOR
] = bp_interception
,
2109 [SVM_EXIT_EXCP_BASE
+ UD_VECTOR
] = ud_interception
,
2110 [SVM_EXIT_EXCP_BASE
+ PF_VECTOR
] = pf_interception
,
2111 [SVM_EXIT_EXCP_BASE
+ NM_VECTOR
] = nm_interception
,
2112 [SVM_EXIT_EXCP_BASE
+ MC_VECTOR
] = mc_interception
,
2113 [SVM_EXIT_INTR
] = intr_interception
,
2114 [SVM_EXIT_NMI
] = nmi_interception
,
2115 [SVM_EXIT_SMI
] = nop_on_interception
,
2116 [SVM_EXIT_INIT
] = nop_on_interception
,
2117 [SVM_EXIT_VINTR
] = interrupt_window_interception
,
2118 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
2119 [SVM_EXIT_CPUID
] = cpuid_interception
,
2120 [SVM_EXIT_INVD
] = emulate_on_interception
,
2121 [SVM_EXIT_HLT
] = halt_interception
,
2122 [SVM_EXIT_INVLPG
] = invlpg_interception
,
2123 [SVM_EXIT_INVLPGA
] = invalid_op_interception
,
2124 [SVM_EXIT_IOIO
] = io_interception
,
2125 [SVM_EXIT_MSR
] = msr_interception
,
2126 [SVM_EXIT_TASK_SWITCH
] = task_switch_interception
,
2127 [SVM_EXIT_SHUTDOWN
] = shutdown_interception
,
2128 [SVM_EXIT_VMRUN
] = vmrun_interception
,
2129 [SVM_EXIT_VMMCALL
] = vmmcall_interception
,
2130 [SVM_EXIT_VMLOAD
] = vmload_interception
,
2131 [SVM_EXIT_VMSAVE
] = vmsave_interception
,
2132 [SVM_EXIT_STGI
] = stgi_interception
,
2133 [SVM_EXIT_CLGI
] = clgi_interception
,
2134 [SVM_EXIT_SKINIT
] = invalid_op_interception
,
2135 [SVM_EXIT_WBINVD
] = emulate_on_interception
,
2136 [SVM_EXIT_MONITOR
] = invalid_op_interception
,
2137 [SVM_EXIT_MWAIT
] = invalid_op_interception
,
2138 [SVM_EXIT_NPF
] = pf_interception
,
2141 static int handle_exit(struct kvm_run
*kvm_run
, struct kvm_vcpu
*vcpu
)
2143 struct vcpu_svm
*svm
= to_svm(vcpu
);
2144 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
2146 KVMTRACE_3D(VMEXIT
, vcpu
, exit_code
, (u32
)svm
->vmcb
->save
.rip
,
2147 (u32
)((u64
)svm
->vmcb
->save
.rip
>> 32), entryexit
);
2149 if (is_nested(svm
)) {
2150 nsvm_printk("nested handle_exit: 0x%x | 0x%lx | 0x%lx | 0x%lx\n",
2151 exit_code
, svm
->vmcb
->control
.exit_info_1
,
2152 svm
->vmcb
->control
.exit_info_2
, svm
->vmcb
->save
.rip
);
2153 if (nested_svm_exit_handled(svm
, true)) {
2154 nested_svm_vmexit(svm
);
2155 nsvm_printk("-> #VMEXIT\n");
2162 if ((vcpu
->arch
.cr0
^ svm
->vmcb
->save
.cr0
) & X86_CR0_PG
) {
2163 svm_set_cr0(vcpu
, svm
->vmcb
->save
.cr0
);
2166 vcpu
->arch
.cr0
= svm
->vmcb
->save
.cr0
;
2167 vcpu
->arch
.cr3
= svm
->vmcb
->save
.cr3
;
2168 if (is_paging(vcpu
) && is_pae(vcpu
) && !is_long_mode(vcpu
)) {
2169 if (!load_pdptrs(vcpu
, vcpu
->arch
.cr3
)) {
2170 kvm_inject_gp(vcpu
, 0);
2175 kvm_mmu_reset_context(vcpu
);
2182 if (svm
->vmcb
->control
.exit_code
== SVM_EXIT_ERR
) {
2183 kvm_run
->exit_reason
= KVM_EXIT_FAIL_ENTRY
;
2184 kvm_run
->fail_entry
.hardware_entry_failure_reason
2185 = svm
->vmcb
->control
.exit_code
;
2189 if (is_external_interrupt(svm
->vmcb
->control
.exit_int_info
) &&
2190 exit_code
!= SVM_EXIT_EXCP_BASE
+ PF_VECTOR
&&
2191 exit_code
!= SVM_EXIT_NPF
)
2192 printk(KERN_ERR
"%s: unexpected exit_ini_info 0x%x "
2194 __func__
, svm
->vmcb
->control
.exit_int_info
,
2197 if (exit_code
>= ARRAY_SIZE(svm_exit_handlers
)
2198 || !svm_exit_handlers
[exit_code
]) {
2199 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
2200 kvm_run
->hw
.hardware_exit_reason
= exit_code
;
2204 return svm_exit_handlers
[exit_code
](svm
, kvm_run
);
2207 static void reload_tss(struct kvm_vcpu
*vcpu
)
2209 int cpu
= raw_smp_processor_id();
2211 struct svm_cpu_data
*svm_data
= per_cpu(svm_data
, cpu
);
2212 svm_data
->tss_desc
->type
= 9; /* available 32/64-bit TSS */
2216 static void pre_svm_run(struct vcpu_svm
*svm
)
2218 int cpu
= raw_smp_processor_id();
2220 struct svm_cpu_data
*svm_data
= per_cpu(svm_data
, cpu
);
2222 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_DO_NOTHING
;
2223 if (svm
->vcpu
.cpu
!= cpu
||
2224 svm
->asid_generation
!= svm_data
->asid_generation
)
2225 new_asid(svm
, svm_data
);
2229 static inline void svm_inject_irq(struct vcpu_svm
*svm
, int irq
)
2231 struct vmcb_control_area
*control
;
2233 KVMTRACE_1D(INJ_VIRQ
, &svm
->vcpu
, (u32
)irq
, handler
);
2235 ++svm
->vcpu
.stat
.irq_injections
;
2236 control
= &svm
->vmcb
->control
;
2237 control
->int_vector
= irq
;
2238 control
->int_ctl
&= ~V_INTR_PRIO_MASK
;
2239 control
->int_ctl
|= V_IRQ_MASK
|
2240 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT
);
2243 static void svm_set_irq(struct kvm_vcpu
*vcpu
, int irq
)
2245 struct vcpu_svm
*svm
= to_svm(vcpu
);
2247 nested_svm_intr(svm
);
2249 svm_inject_irq(svm
, irq
);
2252 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
)
2254 struct vcpu_svm
*svm
= to_svm(vcpu
);
2255 struct vmcb
*vmcb
= svm
->vmcb
;
2258 if (!irqchip_in_kernel(vcpu
->kvm
) || vcpu
->arch
.apic
->vapic_addr
)
2261 vmcb
->control
.intercept_cr_write
&= ~INTERCEPT_CR8_MASK
;
2263 max_irr
= kvm_lapic_find_highest_irr(vcpu
);
2267 tpr
= kvm_lapic_get_cr8(vcpu
) << 4;
2269 if (tpr
>= (max_irr
& 0xf0))
2270 vmcb
->control
.intercept_cr_write
|= INTERCEPT_CR8_MASK
;
2273 static int svm_interrupt_allowed(struct kvm_vcpu
*vcpu
)
2275 struct vcpu_svm
*svm
= to_svm(vcpu
);
2276 struct vmcb
*vmcb
= svm
->vmcb
;
2277 return (vmcb
->save
.rflags
& X86_EFLAGS_IF
) &&
2278 !(vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
) &&
2279 (svm
->vcpu
.arch
.hflags
& HF_GIF_MASK
);
2282 static void svm_intr_assist(struct kvm_vcpu
*vcpu
)
2284 struct vcpu_svm
*svm
= to_svm(vcpu
);
2285 struct vmcb
*vmcb
= svm
->vmcb
;
2286 int intr_vector
= -1;
2288 if ((vmcb
->control
.exit_int_info
& SVM_EVTINJ_VALID
) &&
2289 ((vmcb
->control
.exit_int_info
& SVM_EVTINJ_TYPE_MASK
) == 0)) {
2290 intr_vector
= vmcb
->control
.exit_int_info
&
2291 SVM_EVTINJ_VEC_MASK
;
2292 vmcb
->control
.exit_int_info
= 0;
2293 svm_inject_irq(svm
, intr_vector
);
2297 if (vmcb
->control
.int_ctl
& V_IRQ_MASK
)
2300 if (!kvm_cpu_has_interrupt(vcpu
))
2303 if (nested_svm_intr(svm
))
2306 if (!(svm
->vcpu
.arch
.hflags
& HF_GIF_MASK
))
2309 if (!(vmcb
->save
.rflags
& X86_EFLAGS_IF
) ||
2310 (vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
) ||
2311 (vmcb
->control
.event_inj
& SVM_EVTINJ_VALID
)) {
2312 /* unable to deliver irq, set pending irq */
2314 svm_inject_irq(svm
, 0x0);
2317 /* Okay, we can deliver the interrupt: grab it and update PIC state. */
2318 intr_vector
= kvm_cpu_get_interrupt(vcpu
);
2319 svm_inject_irq(svm
, intr_vector
);
2321 update_cr8_intercept(vcpu
);
2324 static void kvm_reput_irq(struct vcpu_svm
*svm
)
2326 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
2328 if ((control
->int_ctl
& V_IRQ_MASK
)
2329 && !irqchip_in_kernel(svm
->vcpu
.kvm
)) {
2330 control
->int_ctl
&= ~V_IRQ_MASK
;
2331 kvm_push_irq(&svm
->vcpu
, control
->int_vector
);
2334 svm
->vcpu
.arch
.interrupt_window_open
=
2335 !(control
->int_state
& SVM_INTERRUPT_SHADOW_MASK
) &&
2336 (svm
->vcpu
.arch
.hflags
& HF_GIF_MASK
);
2339 static void svm_do_inject_vector(struct vcpu_svm
*svm
)
2341 svm_inject_irq(svm
, kvm_pop_irq(&svm
->vcpu
));
2344 static void do_interrupt_requests(struct kvm_vcpu
*vcpu
,
2345 struct kvm_run
*kvm_run
)
2347 struct vcpu_svm
*svm
= to_svm(vcpu
);
2348 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
2350 if (nested_svm_intr(svm
))
2353 svm
->vcpu
.arch
.interrupt_window_open
=
2354 (!(control
->int_state
& SVM_INTERRUPT_SHADOW_MASK
) &&
2355 (svm
->vmcb
->save
.rflags
& X86_EFLAGS_IF
) &&
2356 (svm
->vcpu
.arch
.hflags
& HF_GIF_MASK
));
2358 if (svm
->vcpu
.arch
.interrupt_window_open
&& svm
->vcpu
.arch
.irq_summary
)
2360 * If interrupts enabled, and not blocked by sti or mov ss. Good.
2362 svm_do_inject_vector(svm
);
2365 * Interrupts blocked. Wait for unblock.
2367 if (!svm
->vcpu
.arch
.interrupt_window_open
&&
2368 (svm
->vcpu
.arch
.irq_summary
|| kvm_run
->request_interrupt_window
))
2371 svm_clear_vintr(svm
);
2374 static int svm_set_tss_addr(struct kvm
*kvm
, unsigned int addr
)
2379 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
)
2381 force_new_asid(vcpu
);
2384 static void svm_prepare_guest_switch(struct kvm_vcpu
*vcpu
)
2388 static inline void sync_cr8_to_lapic(struct kvm_vcpu
*vcpu
)
2390 struct vcpu_svm
*svm
= to_svm(vcpu
);
2392 if (!(svm
->vmcb
->control
.intercept_cr_write
& INTERCEPT_CR8_MASK
)) {
2393 int cr8
= svm
->vmcb
->control
.int_ctl
& V_TPR_MASK
;
2394 kvm_lapic_set_tpr(vcpu
, cr8
);
2398 static inline void sync_lapic_to_cr8(struct kvm_vcpu
*vcpu
)
2400 struct vcpu_svm
*svm
= to_svm(vcpu
);
2403 if (!irqchip_in_kernel(vcpu
->kvm
))
2406 cr8
= kvm_get_cr8(vcpu
);
2407 svm
->vmcb
->control
.int_ctl
&= ~V_TPR_MASK
;
2408 svm
->vmcb
->control
.int_ctl
|= cr8
& V_TPR_MASK
;
2411 #ifdef CONFIG_X86_64
2417 static void svm_vcpu_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
2419 struct vcpu_svm
*svm
= to_svm(vcpu
);
2424 svm
->vmcb
->save
.rax
= vcpu
->arch
.regs
[VCPU_REGS_RAX
];
2425 svm
->vmcb
->save
.rsp
= vcpu
->arch
.regs
[VCPU_REGS_RSP
];
2426 svm
->vmcb
->save
.rip
= vcpu
->arch
.regs
[VCPU_REGS_RIP
];
2430 sync_lapic_to_cr8(vcpu
);
2432 save_host_msrs(vcpu
);
2433 fs_selector
= kvm_read_fs();
2434 gs_selector
= kvm_read_gs();
2435 ldt_selector
= kvm_read_ldt();
2436 svm
->host_cr2
= kvm_read_cr2();
2437 if (!is_nested(svm
))
2438 svm
->vmcb
->save
.cr2
= vcpu
->arch
.cr2
;
2439 /* required for live migration with NPT */
2441 svm
->vmcb
->save
.cr3
= vcpu
->arch
.cr3
;
2448 "push %%"R
"bp; \n\t"
2449 "mov %c[rbx](%[svm]), %%"R
"bx \n\t"
2450 "mov %c[rcx](%[svm]), %%"R
"cx \n\t"
2451 "mov %c[rdx](%[svm]), %%"R
"dx \n\t"
2452 "mov %c[rsi](%[svm]), %%"R
"si \n\t"
2453 "mov %c[rdi](%[svm]), %%"R
"di \n\t"
2454 "mov %c[rbp](%[svm]), %%"R
"bp \n\t"
2455 #ifdef CONFIG_X86_64
2456 "mov %c[r8](%[svm]), %%r8 \n\t"
2457 "mov %c[r9](%[svm]), %%r9 \n\t"
2458 "mov %c[r10](%[svm]), %%r10 \n\t"
2459 "mov %c[r11](%[svm]), %%r11 \n\t"
2460 "mov %c[r12](%[svm]), %%r12 \n\t"
2461 "mov %c[r13](%[svm]), %%r13 \n\t"
2462 "mov %c[r14](%[svm]), %%r14 \n\t"
2463 "mov %c[r15](%[svm]), %%r15 \n\t"
2466 /* Enter guest mode */
2468 "mov %c[vmcb](%[svm]), %%"R
"ax \n\t"
2469 __ex(SVM_VMLOAD
) "\n\t"
2470 __ex(SVM_VMRUN
) "\n\t"
2471 __ex(SVM_VMSAVE
) "\n\t"
2474 /* Save guest registers, load host registers */
2475 "mov %%"R
"bx, %c[rbx](%[svm]) \n\t"
2476 "mov %%"R
"cx, %c[rcx](%[svm]) \n\t"
2477 "mov %%"R
"dx, %c[rdx](%[svm]) \n\t"
2478 "mov %%"R
"si, %c[rsi](%[svm]) \n\t"
2479 "mov %%"R
"di, %c[rdi](%[svm]) \n\t"
2480 "mov %%"R
"bp, %c[rbp](%[svm]) \n\t"
2481 #ifdef CONFIG_X86_64
2482 "mov %%r8, %c[r8](%[svm]) \n\t"
2483 "mov %%r9, %c[r9](%[svm]) \n\t"
2484 "mov %%r10, %c[r10](%[svm]) \n\t"
2485 "mov %%r11, %c[r11](%[svm]) \n\t"
2486 "mov %%r12, %c[r12](%[svm]) \n\t"
2487 "mov %%r13, %c[r13](%[svm]) \n\t"
2488 "mov %%r14, %c[r14](%[svm]) \n\t"
2489 "mov %%r15, %c[r15](%[svm]) \n\t"
2494 [vmcb
]"i"(offsetof(struct vcpu_svm
, vmcb_pa
)),
2495 [rbx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RBX
])),
2496 [rcx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RCX
])),
2497 [rdx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RDX
])),
2498 [rsi
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RSI
])),
2499 [rdi
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RDI
])),
2500 [rbp
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RBP
]))
2501 #ifdef CONFIG_X86_64
2502 , [r8
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R8
])),
2503 [r9
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R9
])),
2504 [r10
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R10
])),
2505 [r11
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R11
])),
2506 [r12
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R12
])),
2507 [r13
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R13
])),
2508 [r14
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R14
])),
2509 [r15
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R15
]))
2512 , R
"bx", R
"cx", R
"dx", R
"si", R
"di"
2513 #ifdef CONFIG_X86_64
2514 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
2518 vcpu
->arch
.cr2
= svm
->vmcb
->save
.cr2
;
2519 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = svm
->vmcb
->save
.rax
;
2520 vcpu
->arch
.regs
[VCPU_REGS_RSP
] = svm
->vmcb
->save
.rsp
;
2521 vcpu
->arch
.regs
[VCPU_REGS_RIP
] = svm
->vmcb
->save
.rip
;
2523 kvm_write_cr2(svm
->host_cr2
);
2525 kvm_load_fs(fs_selector
);
2526 kvm_load_gs(gs_selector
);
2527 kvm_load_ldt(ldt_selector
);
2528 load_host_msrs(vcpu
);
2532 local_irq_disable();
2536 sync_cr8_to_lapic(vcpu
);
2543 static void svm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
2545 struct vcpu_svm
*svm
= to_svm(vcpu
);
2548 svm
->vmcb
->control
.nested_cr3
= root
;
2549 force_new_asid(vcpu
);
2553 svm
->vmcb
->save
.cr3
= root
;
2554 force_new_asid(vcpu
);
2556 if (vcpu
->fpu_active
) {
2557 svm
->vmcb
->control
.intercept_exceptions
|= (1 << NM_VECTOR
);
2558 svm
->vmcb
->save
.cr0
|= X86_CR0_TS
;
2559 vcpu
->fpu_active
= 0;
2563 static int is_disabled(void)
2567 rdmsrl(MSR_VM_CR
, vm_cr
);
2568 if (vm_cr
& (1 << SVM_VM_CR_SVM_DISABLE
))
2575 svm_patch_hypercall(struct kvm_vcpu
*vcpu
, unsigned char *hypercall
)
2578 * Patch in the VMMCALL instruction:
2580 hypercall
[0] = 0x0f;
2581 hypercall
[1] = 0x01;
2582 hypercall
[2] = 0xd9;
2585 static void svm_check_processor_compat(void *rtn
)
2590 static bool svm_cpu_has_accelerated_tpr(void)
2595 static int get_npt_level(void)
2597 #ifdef CONFIG_X86_64
2598 return PT64_ROOT_LEVEL
;
2600 return PT32E_ROOT_LEVEL
;
2604 static int svm_get_mt_mask_shift(void)
2609 static struct kvm_x86_ops svm_x86_ops
= {
2610 .cpu_has_kvm_support
= has_svm
,
2611 .disabled_by_bios
= is_disabled
,
2612 .hardware_setup
= svm_hardware_setup
,
2613 .hardware_unsetup
= svm_hardware_unsetup
,
2614 .check_processor_compatibility
= svm_check_processor_compat
,
2615 .hardware_enable
= svm_hardware_enable
,
2616 .hardware_disable
= svm_hardware_disable
,
2617 .cpu_has_accelerated_tpr
= svm_cpu_has_accelerated_tpr
,
2619 .vcpu_create
= svm_create_vcpu
,
2620 .vcpu_free
= svm_free_vcpu
,
2621 .vcpu_reset
= svm_vcpu_reset
,
2623 .prepare_guest_switch
= svm_prepare_guest_switch
,
2624 .vcpu_load
= svm_vcpu_load
,
2625 .vcpu_put
= svm_vcpu_put
,
2627 .set_guest_debug
= svm_guest_debug
,
2628 .get_msr
= svm_get_msr
,
2629 .set_msr
= svm_set_msr
,
2630 .get_segment_base
= svm_get_segment_base
,
2631 .get_segment
= svm_get_segment
,
2632 .set_segment
= svm_set_segment
,
2633 .get_cpl
= svm_get_cpl
,
2634 .get_cs_db_l_bits
= kvm_get_cs_db_l_bits
,
2635 .decache_cr4_guest_bits
= svm_decache_cr4_guest_bits
,
2636 .set_cr0
= svm_set_cr0
,
2637 .set_cr3
= svm_set_cr3
,
2638 .set_cr4
= svm_set_cr4
,
2639 .set_efer
= svm_set_efer
,
2640 .get_idt
= svm_get_idt
,
2641 .set_idt
= svm_set_idt
,
2642 .get_gdt
= svm_get_gdt
,
2643 .set_gdt
= svm_set_gdt
,
2644 .get_dr
= svm_get_dr
,
2645 .set_dr
= svm_set_dr
,
2646 .get_rflags
= svm_get_rflags
,
2647 .set_rflags
= svm_set_rflags
,
2649 .tlb_flush
= svm_flush_tlb
,
2651 .run
= svm_vcpu_run
,
2652 .handle_exit
= handle_exit
,
2653 .skip_emulated_instruction
= skip_emulated_instruction
,
2654 .patch_hypercall
= svm_patch_hypercall
,
2655 .get_irq
= svm_get_irq
,
2656 .set_irq
= svm_set_irq
,
2657 .queue_exception
= svm_queue_exception
,
2658 .exception_injected
= svm_exception_injected
,
2659 .inject_pending_irq
= svm_intr_assist
,
2660 .inject_pending_vectors
= do_interrupt_requests
,
2661 .interrupt_allowed
= svm_interrupt_allowed
,
2663 .set_tss_addr
= svm_set_tss_addr
,
2664 .get_tdp_level
= get_npt_level
,
2665 .get_mt_mask_shift
= svm_get_mt_mask_shift
,
2668 static int __init
svm_init(void)
2670 return kvm_init(&svm_x86_ops
, sizeof(struct vcpu_svm
),
2674 static void __exit
svm_exit(void)
2679 module_init(svm_init
)
2680 module_exit(svm_exit
)