2 * Kernel-based Virtual Machine driver for Linux
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10 * Yaniv Kamay <yaniv@qumranet.com>
11 * Avi Kivity <avi@qumranet.com>
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
17 #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>
29 #include <linux/ftrace_event.h>
30 #include <linux/slab.h>
32 #include <asm/tlbflush.h>
34 #include <asm/kvm_para.h>
36 #include <asm/virtext.h>
39 #define __ex(x) __kvm_handle_fault_on_reboot(x)
41 MODULE_AUTHOR("Qumranet");
42 MODULE_LICENSE("GPL");
44 #define IOPM_ALLOC_ORDER 2
45 #define MSRPM_ALLOC_ORDER 1
47 #define SEG_TYPE_LDT 2
48 #define SEG_TYPE_BUSY_TSS16 3
50 #define SVM_FEATURE_NPT (1 << 0)
51 #define SVM_FEATURE_LBRV (1 << 1)
52 #define SVM_FEATURE_SVML (1 << 2)
53 #define SVM_FEATURE_NRIP (1 << 3)
54 #define SVM_FEATURE_TSC_RATE (1 << 4)
55 #define SVM_FEATURE_VMCB_CLEAN (1 << 5)
56 #define SVM_FEATURE_FLUSH_ASID (1 << 6)
57 #define SVM_FEATURE_DECODE_ASSIST (1 << 7)
58 #define SVM_FEATURE_PAUSE_FILTER (1 << 10)
60 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
61 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
62 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
64 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
66 #define TSC_RATIO_RSVD 0xffffff0000000000ULL
67 #define TSC_RATIO_MIN 0x0000000000000001ULL
68 #define TSC_RATIO_MAX 0x000000ffffffffffULL
70 static bool erratum_383_found __read_mostly
;
72 static const u32 host_save_user_msrs
[] = {
74 MSR_STAR
, MSR_LSTAR
, MSR_CSTAR
, MSR_SYSCALL_MASK
, MSR_KERNEL_GS_BASE
,
77 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
80 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
90 /* These are the merged vectors */
93 /* gpa pointers to the real vectors */
97 /* A VMEXIT is required but not yet emulated */
100 /* cache for intercepts of the guest */
103 u32 intercept_exceptions
;
106 /* Nested Paging related state */
110 #define MSRPM_OFFSETS 16
111 static u32 msrpm_offsets
[MSRPM_OFFSETS
] __read_mostly
;
114 struct kvm_vcpu vcpu
;
116 unsigned long vmcb_pa
;
117 struct svm_cpu_data
*svm_data
;
118 uint64_t asid_generation
;
119 uint64_t sysenter_esp
;
120 uint64_t sysenter_eip
;
124 u64 host_user_msrs
[NR_HOST_SAVE_USER_MSRS
];
136 struct nested_state nested
;
140 unsigned int3_injected
;
141 unsigned long int3_rip
;
147 static DEFINE_PER_CPU(u64
, current_tsc_ratio
);
148 #define TSC_RATIO_DEFAULT 0x0100000000ULL
150 #define MSR_INVALID 0xffffffffU
152 static struct svm_direct_access_msrs
{
153 u32 index
; /* Index of the MSR */
154 bool always
; /* True if intercept is always on */
155 } direct_access_msrs
[] = {
156 { .index
= MSR_STAR
, .always
= true },
157 { .index
= MSR_IA32_SYSENTER_CS
, .always
= true },
159 { .index
= MSR_GS_BASE
, .always
= true },
160 { .index
= MSR_FS_BASE
, .always
= true },
161 { .index
= MSR_KERNEL_GS_BASE
, .always
= true },
162 { .index
= MSR_LSTAR
, .always
= true },
163 { .index
= MSR_CSTAR
, .always
= true },
164 { .index
= MSR_SYSCALL_MASK
, .always
= true },
166 { .index
= MSR_IA32_LASTBRANCHFROMIP
, .always
= false },
167 { .index
= MSR_IA32_LASTBRANCHTOIP
, .always
= false },
168 { .index
= MSR_IA32_LASTINTFROMIP
, .always
= false },
169 { .index
= MSR_IA32_LASTINTTOIP
, .always
= false },
170 { .index
= MSR_INVALID
, .always
= false },
173 /* enable NPT for AMD64 and X86 with PAE */
174 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
175 static bool npt_enabled
= true;
177 static bool npt_enabled
;
181 module_param(npt
, int, S_IRUGO
);
183 static int nested
= 1;
184 module_param(nested
, int, S_IRUGO
);
186 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
);
187 static void svm_complete_interrupts(struct vcpu_svm
*svm
);
189 static int nested_svm_exit_handled(struct vcpu_svm
*svm
);
190 static int nested_svm_intercept(struct vcpu_svm
*svm
);
191 static int nested_svm_vmexit(struct vcpu_svm
*svm
);
192 static int nested_svm_check_exception(struct vcpu_svm
*svm
, unsigned nr
,
193 bool has_error_code
, u32 error_code
);
194 static u64
__scale_tsc(u64 ratio
, u64 tsc
);
197 VMCB_INTERCEPTS
, /* Intercept vectors, TSC offset,
198 pause filter count */
199 VMCB_PERM_MAP
, /* IOPM Base and MSRPM Base */
200 VMCB_ASID
, /* ASID */
201 VMCB_INTR
, /* int_ctl, int_vector */
202 VMCB_NPT
, /* npt_en, nCR3, gPAT */
203 VMCB_CR
, /* CR0, CR3, CR4, EFER */
204 VMCB_DR
, /* DR6, DR7 */
205 VMCB_DT
, /* GDT, IDT */
206 VMCB_SEG
, /* CS, DS, SS, ES, CPL */
207 VMCB_CR2
, /* CR2 only */
208 VMCB_LBR
, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
212 /* TPR and CR2 are always written before VMRUN */
213 #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2))
215 static inline void mark_all_dirty(struct vmcb
*vmcb
)
217 vmcb
->control
.clean
= 0;
220 static inline void mark_all_clean(struct vmcb
*vmcb
)
222 vmcb
->control
.clean
= ((1 << VMCB_DIRTY_MAX
) - 1)
223 & ~VMCB_ALWAYS_DIRTY_MASK
;
226 static inline void mark_dirty(struct vmcb
*vmcb
, int bit
)
228 vmcb
->control
.clean
&= ~(1 << bit
);
231 static inline struct vcpu_svm
*to_svm(struct kvm_vcpu
*vcpu
)
233 return container_of(vcpu
, struct vcpu_svm
, vcpu
);
236 static void recalc_intercepts(struct vcpu_svm
*svm
)
238 struct vmcb_control_area
*c
, *h
;
239 struct nested_state
*g
;
241 mark_dirty(svm
->vmcb
, VMCB_INTERCEPTS
);
243 if (!is_guest_mode(&svm
->vcpu
))
246 c
= &svm
->vmcb
->control
;
247 h
= &svm
->nested
.hsave
->control
;
250 c
->intercept_cr
= h
->intercept_cr
| g
->intercept_cr
;
251 c
->intercept_dr
= h
->intercept_dr
| g
->intercept_dr
;
252 c
->intercept_exceptions
= h
->intercept_exceptions
| g
->intercept_exceptions
;
253 c
->intercept
= h
->intercept
| g
->intercept
;
256 static inline struct vmcb
*get_host_vmcb(struct vcpu_svm
*svm
)
258 if (is_guest_mode(&svm
->vcpu
))
259 return svm
->nested
.hsave
;
264 static inline void set_cr_intercept(struct vcpu_svm
*svm
, int bit
)
266 struct vmcb
*vmcb
= get_host_vmcb(svm
);
268 vmcb
->control
.intercept_cr
|= (1U << bit
);
270 recalc_intercepts(svm
);
273 static inline void clr_cr_intercept(struct vcpu_svm
*svm
, int bit
)
275 struct vmcb
*vmcb
= get_host_vmcb(svm
);
277 vmcb
->control
.intercept_cr
&= ~(1U << bit
);
279 recalc_intercepts(svm
);
282 static inline bool is_cr_intercept(struct vcpu_svm
*svm
, int bit
)
284 struct vmcb
*vmcb
= get_host_vmcb(svm
);
286 return vmcb
->control
.intercept_cr
& (1U << bit
);
289 static inline void set_dr_intercept(struct vcpu_svm
*svm
, int bit
)
291 struct vmcb
*vmcb
= get_host_vmcb(svm
);
293 vmcb
->control
.intercept_dr
|= (1U << bit
);
295 recalc_intercepts(svm
);
298 static inline void clr_dr_intercept(struct vcpu_svm
*svm
, int bit
)
300 struct vmcb
*vmcb
= get_host_vmcb(svm
);
302 vmcb
->control
.intercept_dr
&= ~(1U << bit
);
304 recalc_intercepts(svm
);
307 static inline void set_exception_intercept(struct vcpu_svm
*svm
, int bit
)
309 struct vmcb
*vmcb
= get_host_vmcb(svm
);
311 vmcb
->control
.intercept_exceptions
|= (1U << bit
);
313 recalc_intercepts(svm
);
316 static inline void clr_exception_intercept(struct vcpu_svm
*svm
, int bit
)
318 struct vmcb
*vmcb
= get_host_vmcb(svm
);
320 vmcb
->control
.intercept_exceptions
&= ~(1U << bit
);
322 recalc_intercepts(svm
);
325 static inline void set_intercept(struct vcpu_svm
*svm
, int bit
)
327 struct vmcb
*vmcb
= get_host_vmcb(svm
);
329 vmcb
->control
.intercept
|= (1ULL << bit
);
331 recalc_intercepts(svm
);
334 static inline void clr_intercept(struct vcpu_svm
*svm
, int bit
)
336 struct vmcb
*vmcb
= get_host_vmcb(svm
);
338 vmcb
->control
.intercept
&= ~(1ULL << bit
);
340 recalc_intercepts(svm
);
343 static inline void enable_gif(struct vcpu_svm
*svm
)
345 svm
->vcpu
.arch
.hflags
|= HF_GIF_MASK
;
348 static inline void disable_gif(struct vcpu_svm
*svm
)
350 svm
->vcpu
.arch
.hflags
&= ~HF_GIF_MASK
;
353 static inline bool gif_set(struct vcpu_svm
*svm
)
355 return !!(svm
->vcpu
.arch
.hflags
& HF_GIF_MASK
);
358 static unsigned long iopm_base
;
360 struct kvm_ldttss_desc
{
363 unsigned base1
:8, type
:5, dpl
:2, p
:1;
364 unsigned limit1
:4, zero0
:3, g
:1, base2
:8;
367 } __attribute__((packed
));
369 struct svm_cpu_data
{
375 struct kvm_ldttss_desc
*tss_desc
;
377 struct page
*save_area
;
380 static DEFINE_PER_CPU(struct svm_cpu_data
*, svm_data
);
382 struct svm_init_data
{
387 static u32 msrpm_ranges
[] = {0, 0xc0000000, 0xc0010000};
389 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
390 #define MSRS_RANGE_SIZE 2048
391 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
393 static u32
svm_msrpm_offset(u32 msr
)
398 for (i
= 0; i
< NUM_MSR_MAPS
; i
++) {
399 if (msr
< msrpm_ranges
[i
] ||
400 msr
>= msrpm_ranges
[i
] + MSRS_IN_RANGE
)
403 offset
= (msr
- msrpm_ranges
[i
]) / 4; /* 4 msrs per u8 */
404 offset
+= (i
* MSRS_RANGE_SIZE
); /* add range offset */
406 /* Now we have the u8 offset - but need the u32 offset */
410 /* MSR not in any range */
414 #define MAX_INST_SIZE 15
416 static inline void clgi(void)
418 asm volatile (__ex(SVM_CLGI
));
421 static inline void stgi(void)
423 asm volatile (__ex(SVM_STGI
));
426 static inline void invlpga(unsigned long addr
, u32 asid
)
428 asm volatile (__ex(SVM_INVLPGA
) : : "a"(addr
), "c"(asid
));
431 static int get_npt_level(void)
434 return PT64_ROOT_LEVEL
;
436 return PT32E_ROOT_LEVEL
;
440 static void svm_set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
442 vcpu
->arch
.efer
= efer
;
443 if (!npt_enabled
&& !(efer
& EFER_LMA
))
446 to_svm(vcpu
)->vmcb
->save
.efer
= efer
| EFER_SVME
;
447 mark_dirty(to_svm(vcpu
)->vmcb
, VMCB_CR
);
450 static int is_external_interrupt(u32 info
)
452 info
&= SVM_EVTINJ_TYPE_MASK
| SVM_EVTINJ_VALID
;
453 return info
== (SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
);
456 static u32
svm_get_interrupt_shadow(struct kvm_vcpu
*vcpu
, int mask
)
458 struct vcpu_svm
*svm
= to_svm(vcpu
);
461 if (svm
->vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
)
462 ret
|= KVM_X86_SHADOW_INT_STI
| KVM_X86_SHADOW_INT_MOV_SS
;
466 static void svm_set_interrupt_shadow(struct kvm_vcpu
*vcpu
, int mask
)
468 struct vcpu_svm
*svm
= to_svm(vcpu
);
471 svm
->vmcb
->control
.int_state
&= ~SVM_INTERRUPT_SHADOW_MASK
;
473 svm
->vmcb
->control
.int_state
|= SVM_INTERRUPT_SHADOW_MASK
;
477 static void skip_emulated_instruction(struct kvm_vcpu
*vcpu
)
479 struct vcpu_svm
*svm
= to_svm(vcpu
);
481 if (svm
->vmcb
->control
.next_rip
!= 0)
482 svm
->next_rip
= svm
->vmcb
->control
.next_rip
;
484 if (!svm
->next_rip
) {
485 if (emulate_instruction(vcpu
, EMULTYPE_SKIP
) !=
487 printk(KERN_DEBUG
"%s: NOP\n", __func__
);
490 if (svm
->next_rip
- kvm_rip_read(vcpu
) > MAX_INST_SIZE
)
491 printk(KERN_ERR
"%s: ip 0x%lx next 0x%llx\n",
492 __func__
, kvm_rip_read(vcpu
), svm
->next_rip
);
494 kvm_rip_write(vcpu
, svm
->next_rip
);
495 svm_set_interrupt_shadow(vcpu
, 0);
498 static void svm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
,
499 bool has_error_code
, u32 error_code
,
502 struct vcpu_svm
*svm
= to_svm(vcpu
);
505 * If we are within a nested VM we'd better #VMEXIT and let the guest
506 * handle the exception
509 nested_svm_check_exception(svm
, nr
, has_error_code
, error_code
))
512 if (nr
== BP_VECTOR
&& !static_cpu_has(X86_FEATURE_NRIPS
)) {
513 unsigned long rip
, old_rip
= kvm_rip_read(&svm
->vcpu
);
516 * For guest debugging where we have to reinject #BP if some
517 * INT3 is guest-owned:
518 * Emulate nRIP by moving RIP forward. Will fail if injection
519 * raises a fault that is not intercepted. Still better than
520 * failing in all cases.
522 skip_emulated_instruction(&svm
->vcpu
);
523 rip
= kvm_rip_read(&svm
->vcpu
);
524 svm
->int3_rip
= rip
+ svm
->vmcb
->save
.cs
.base
;
525 svm
->int3_injected
= rip
- old_rip
;
528 svm
->vmcb
->control
.event_inj
= nr
530 | (has_error_code
? SVM_EVTINJ_VALID_ERR
: 0)
531 | SVM_EVTINJ_TYPE_EXEPT
;
532 svm
->vmcb
->control
.event_inj_err
= error_code
;
535 static void svm_init_erratum_383(void)
541 if (!cpu_has_amd_erratum(amd_erratum_383
))
544 /* Use _safe variants to not break nested virtualization */
545 val
= native_read_msr_safe(MSR_AMD64_DC_CFG
, &err
);
551 low
= lower_32_bits(val
);
552 high
= upper_32_bits(val
);
554 native_write_msr_safe(MSR_AMD64_DC_CFG
, low
, high
);
556 erratum_383_found
= true;
559 static int has_svm(void)
563 if (!cpu_has_svm(&msg
)) {
564 printk(KERN_INFO
"has_svm: %s\n", msg
);
571 static void svm_hardware_disable(void *garbage
)
573 /* Make sure we clean up behind us */
574 if (static_cpu_has(X86_FEATURE_TSCRATEMSR
))
575 wrmsrl(MSR_AMD64_TSC_RATIO
, TSC_RATIO_DEFAULT
);
580 static int svm_hardware_enable(void *garbage
)
583 struct svm_cpu_data
*sd
;
585 struct desc_ptr gdt_descr
;
586 struct desc_struct
*gdt
;
587 int me
= raw_smp_processor_id();
589 rdmsrl(MSR_EFER
, efer
);
590 if (efer
& EFER_SVME
)
594 printk(KERN_ERR
"svm_hardware_enable: err EOPNOTSUPP on %d\n",
598 sd
= per_cpu(svm_data
, me
);
601 printk(KERN_ERR
"svm_hardware_enable: svm_data is NULL on %d\n",
606 sd
->asid_generation
= 1;
607 sd
->max_asid
= cpuid_ebx(SVM_CPUID_FUNC
) - 1;
608 sd
->next_asid
= sd
->max_asid
+ 1;
610 native_store_gdt(&gdt_descr
);
611 gdt
= (struct desc_struct
*)gdt_descr
.address
;
612 sd
->tss_desc
= (struct kvm_ldttss_desc
*)(gdt
+ GDT_ENTRY_TSS
);
614 wrmsrl(MSR_EFER
, efer
| EFER_SVME
);
616 wrmsrl(MSR_VM_HSAVE_PA
, page_to_pfn(sd
->save_area
) << PAGE_SHIFT
);
618 if (static_cpu_has(X86_FEATURE_TSCRATEMSR
)) {
619 wrmsrl(MSR_AMD64_TSC_RATIO
, TSC_RATIO_DEFAULT
);
620 __get_cpu_var(current_tsc_ratio
) = TSC_RATIO_DEFAULT
;
623 svm_init_erratum_383();
628 static void svm_cpu_uninit(int cpu
)
630 struct svm_cpu_data
*sd
= per_cpu(svm_data
, raw_smp_processor_id());
635 per_cpu(svm_data
, raw_smp_processor_id()) = NULL
;
636 __free_page(sd
->save_area
);
640 static int svm_cpu_init(int cpu
)
642 struct svm_cpu_data
*sd
;
645 sd
= kzalloc(sizeof(struct svm_cpu_data
), GFP_KERNEL
);
649 sd
->save_area
= alloc_page(GFP_KERNEL
);
654 per_cpu(svm_data
, cpu
) = sd
;
664 static bool valid_msr_intercept(u32 index
)
668 for (i
= 0; direct_access_msrs
[i
].index
!= MSR_INVALID
; i
++)
669 if (direct_access_msrs
[i
].index
== index
)
675 static void set_msr_interception(u32
*msrpm
, unsigned msr
,
678 u8 bit_read
, bit_write
;
683 * If this warning triggers extend the direct_access_msrs list at the
684 * beginning of the file
686 WARN_ON(!valid_msr_intercept(msr
));
688 offset
= svm_msrpm_offset(msr
);
689 bit_read
= 2 * (msr
& 0x0f);
690 bit_write
= 2 * (msr
& 0x0f) + 1;
693 BUG_ON(offset
== MSR_INVALID
);
695 read
? clear_bit(bit_read
, &tmp
) : set_bit(bit_read
, &tmp
);
696 write
? clear_bit(bit_write
, &tmp
) : set_bit(bit_write
, &tmp
);
701 static void svm_vcpu_init_msrpm(u32
*msrpm
)
705 memset(msrpm
, 0xff, PAGE_SIZE
* (1 << MSRPM_ALLOC_ORDER
));
707 for (i
= 0; direct_access_msrs
[i
].index
!= MSR_INVALID
; i
++) {
708 if (!direct_access_msrs
[i
].always
)
711 set_msr_interception(msrpm
, direct_access_msrs
[i
].index
, 1, 1);
715 static void add_msr_offset(u32 offset
)
719 for (i
= 0; i
< MSRPM_OFFSETS
; ++i
) {
721 /* Offset already in list? */
722 if (msrpm_offsets
[i
] == offset
)
725 /* Slot used by another offset? */
726 if (msrpm_offsets
[i
] != MSR_INVALID
)
729 /* Add offset to list */
730 msrpm_offsets
[i
] = offset
;
736 * If this BUG triggers the msrpm_offsets table has an overflow. Just
737 * increase MSRPM_OFFSETS in this case.
742 static void init_msrpm_offsets(void)
746 memset(msrpm_offsets
, 0xff, sizeof(msrpm_offsets
));
748 for (i
= 0; direct_access_msrs
[i
].index
!= MSR_INVALID
; i
++) {
751 offset
= svm_msrpm_offset(direct_access_msrs
[i
].index
);
752 BUG_ON(offset
== MSR_INVALID
);
754 add_msr_offset(offset
);
758 static void svm_enable_lbrv(struct vcpu_svm
*svm
)
760 u32
*msrpm
= svm
->msrpm
;
762 svm
->vmcb
->control
.lbr_ctl
= 1;
763 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHFROMIP
, 1, 1);
764 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHTOIP
, 1, 1);
765 set_msr_interception(msrpm
, MSR_IA32_LASTINTFROMIP
, 1, 1);
766 set_msr_interception(msrpm
, MSR_IA32_LASTINTTOIP
, 1, 1);
769 static void svm_disable_lbrv(struct vcpu_svm
*svm
)
771 u32
*msrpm
= svm
->msrpm
;
773 svm
->vmcb
->control
.lbr_ctl
= 0;
774 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHFROMIP
, 0, 0);
775 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHTOIP
, 0, 0);
776 set_msr_interception(msrpm
, MSR_IA32_LASTINTFROMIP
, 0, 0);
777 set_msr_interception(msrpm
, MSR_IA32_LASTINTTOIP
, 0, 0);
780 static __init
int svm_hardware_setup(void)
783 struct page
*iopm_pages
;
787 iopm_pages
= alloc_pages(GFP_KERNEL
, IOPM_ALLOC_ORDER
);
792 iopm_va
= page_address(iopm_pages
);
793 memset(iopm_va
, 0xff, PAGE_SIZE
* (1 << IOPM_ALLOC_ORDER
));
794 iopm_base
= page_to_pfn(iopm_pages
) << PAGE_SHIFT
;
796 init_msrpm_offsets();
798 if (boot_cpu_has(X86_FEATURE_NX
))
799 kvm_enable_efer_bits(EFER_NX
);
801 if (boot_cpu_has(X86_FEATURE_FXSR_OPT
))
802 kvm_enable_efer_bits(EFER_FFXSR
);
804 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR
)) {
807 kvm_has_tsc_control
= true;
810 * Make sure the user can only configure tsc_khz values that
811 * fit into a signed integer.
812 * A min value is not calculated needed because it will always
813 * be 1 on all machines and a value of 0 is used to disable
814 * tsc-scaling for the vcpu.
816 max
= min(0x7fffffffULL
, __scale_tsc(tsc_khz
, TSC_RATIO_MAX
));
818 kvm_max_guest_tsc_khz
= max
;
822 printk(KERN_INFO
"kvm: Nested Virtualization enabled\n");
823 kvm_enable_efer_bits(EFER_SVME
| EFER_LMSLE
);
826 for_each_possible_cpu(cpu
) {
827 r
= svm_cpu_init(cpu
);
832 if (!boot_cpu_has(X86_FEATURE_NPT
))
835 if (npt_enabled
&& !npt
) {
836 printk(KERN_INFO
"kvm: Nested Paging disabled\n");
841 printk(KERN_INFO
"kvm: Nested Paging enabled\n");
849 __free_pages(iopm_pages
, IOPM_ALLOC_ORDER
);
854 static __exit
void svm_hardware_unsetup(void)
858 for_each_possible_cpu(cpu
)
861 __free_pages(pfn_to_page(iopm_base
>> PAGE_SHIFT
), IOPM_ALLOC_ORDER
);
865 static void init_seg(struct vmcb_seg
*seg
)
868 seg
->attrib
= SVM_SELECTOR_P_MASK
| SVM_SELECTOR_S_MASK
|
869 SVM_SELECTOR_WRITE_MASK
; /* Read/Write Data Segment */
874 static void init_sys_seg(struct vmcb_seg
*seg
, uint32_t type
)
877 seg
->attrib
= SVM_SELECTOR_P_MASK
| type
;
882 static u64
__scale_tsc(u64 ratio
, u64 tsc
)
884 u64 mult
, frac
, _tsc
;
887 frac
= ratio
& ((1ULL << 32) - 1);
891 _tsc
+= (tsc
>> 32) * frac
;
892 _tsc
+= ((tsc
& ((1ULL << 32) - 1)) * frac
) >> 32;
897 static u64
svm_scale_tsc(struct kvm_vcpu
*vcpu
, u64 tsc
)
899 struct vcpu_svm
*svm
= to_svm(vcpu
);
902 if (svm
->tsc_ratio
!= TSC_RATIO_DEFAULT
)
903 _tsc
= __scale_tsc(svm
->tsc_ratio
, tsc
);
908 static void svm_set_tsc_khz(struct kvm_vcpu
*vcpu
, u32 user_tsc_khz
)
910 struct vcpu_svm
*svm
= to_svm(vcpu
);
914 /* TSC scaling supported? */
915 if (!boot_cpu_has(X86_FEATURE_TSCRATEMSR
))
918 /* TSC-Scaling disabled or guest TSC same frequency as host TSC? */
919 if (user_tsc_khz
== 0) {
920 vcpu
->arch
.virtual_tsc_khz
= 0;
921 svm
->tsc_ratio
= TSC_RATIO_DEFAULT
;
927 /* TSC scaling required - calculate ratio */
929 do_div(ratio
, tsc_khz
);
931 if (ratio
== 0 || ratio
& TSC_RATIO_RSVD
) {
932 WARN_ONCE(1, "Invalid TSC ratio - virtual-tsc-khz=%u\n",
936 vcpu
->arch
.virtual_tsc_khz
= user_tsc_khz
;
937 svm
->tsc_ratio
= ratio
;
940 static void svm_write_tsc_offset(struct kvm_vcpu
*vcpu
, u64 offset
)
942 struct vcpu_svm
*svm
= to_svm(vcpu
);
943 u64 g_tsc_offset
= 0;
945 if (is_guest_mode(vcpu
)) {
946 g_tsc_offset
= svm
->vmcb
->control
.tsc_offset
-
947 svm
->nested
.hsave
->control
.tsc_offset
;
948 svm
->nested
.hsave
->control
.tsc_offset
= offset
;
951 svm
->vmcb
->control
.tsc_offset
= offset
+ g_tsc_offset
;
953 mark_dirty(svm
->vmcb
, VMCB_INTERCEPTS
);
956 static void svm_adjust_tsc_offset(struct kvm_vcpu
*vcpu
, s64 adjustment
)
958 struct vcpu_svm
*svm
= to_svm(vcpu
);
960 svm
->vmcb
->control
.tsc_offset
+= adjustment
;
961 if (is_guest_mode(vcpu
))
962 svm
->nested
.hsave
->control
.tsc_offset
+= adjustment
;
963 mark_dirty(svm
->vmcb
, VMCB_INTERCEPTS
);
966 static u64
svm_compute_tsc_offset(struct kvm_vcpu
*vcpu
, u64 target_tsc
)
970 tsc
= svm_scale_tsc(vcpu
, native_read_tsc());
972 return target_tsc
- tsc
;
975 static void init_vmcb(struct vcpu_svm
*svm
)
977 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
978 struct vmcb_save_area
*save
= &svm
->vmcb
->save
;
980 svm
->vcpu
.fpu_active
= 1;
981 svm
->vcpu
.arch
.hflags
= 0;
983 set_cr_intercept(svm
, INTERCEPT_CR0_READ
);
984 set_cr_intercept(svm
, INTERCEPT_CR3_READ
);
985 set_cr_intercept(svm
, INTERCEPT_CR4_READ
);
986 set_cr_intercept(svm
, INTERCEPT_CR0_WRITE
);
987 set_cr_intercept(svm
, INTERCEPT_CR3_WRITE
);
988 set_cr_intercept(svm
, INTERCEPT_CR4_WRITE
);
989 set_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
991 set_dr_intercept(svm
, INTERCEPT_DR0_READ
);
992 set_dr_intercept(svm
, INTERCEPT_DR1_READ
);
993 set_dr_intercept(svm
, INTERCEPT_DR2_READ
);
994 set_dr_intercept(svm
, INTERCEPT_DR3_READ
);
995 set_dr_intercept(svm
, INTERCEPT_DR4_READ
);
996 set_dr_intercept(svm
, INTERCEPT_DR5_READ
);
997 set_dr_intercept(svm
, INTERCEPT_DR6_READ
);
998 set_dr_intercept(svm
, INTERCEPT_DR7_READ
);
1000 set_dr_intercept(svm
, INTERCEPT_DR0_WRITE
);
1001 set_dr_intercept(svm
, INTERCEPT_DR1_WRITE
);
1002 set_dr_intercept(svm
, INTERCEPT_DR2_WRITE
);
1003 set_dr_intercept(svm
, INTERCEPT_DR3_WRITE
);
1004 set_dr_intercept(svm
, INTERCEPT_DR4_WRITE
);
1005 set_dr_intercept(svm
, INTERCEPT_DR5_WRITE
);
1006 set_dr_intercept(svm
, INTERCEPT_DR6_WRITE
);
1007 set_dr_intercept(svm
, INTERCEPT_DR7_WRITE
);
1009 set_exception_intercept(svm
, PF_VECTOR
);
1010 set_exception_intercept(svm
, UD_VECTOR
);
1011 set_exception_intercept(svm
, MC_VECTOR
);
1013 set_intercept(svm
, INTERCEPT_INTR
);
1014 set_intercept(svm
, INTERCEPT_NMI
);
1015 set_intercept(svm
, INTERCEPT_SMI
);
1016 set_intercept(svm
, INTERCEPT_SELECTIVE_CR0
);
1017 set_intercept(svm
, INTERCEPT_CPUID
);
1018 set_intercept(svm
, INTERCEPT_INVD
);
1019 set_intercept(svm
, INTERCEPT_HLT
);
1020 set_intercept(svm
, INTERCEPT_INVLPG
);
1021 set_intercept(svm
, INTERCEPT_INVLPGA
);
1022 set_intercept(svm
, INTERCEPT_IOIO_PROT
);
1023 set_intercept(svm
, INTERCEPT_MSR_PROT
);
1024 set_intercept(svm
, INTERCEPT_TASK_SWITCH
);
1025 set_intercept(svm
, INTERCEPT_SHUTDOWN
);
1026 set_intercept(svm
, INTERCEPT_VMRUN
);
1027 set_intercept(svm
, INTERCEPT_VMMCALL
);
1028 set_intercept(svm
, INTERCEPT_VMLOAD
);
1029 set_intercept(svm
, INTERCEPT_VMSAVE
);
1030 set_intercept(svm
, INTERCEPT_STGI
);
1031 set_intercept(svm
, INTERCEPT_CLGI
);
1032 set_intercept(svm
, INTERCEPT_SKINIT
);
1033 set_intercept(svm
, INTERCEPT_WBINVD
);
1034 set_intercept(svm
, INTERCEPT_MONITOR
);
1035 set_intercept(svm
, INTERCEPT_MWAIT
);
1036 set_intercept(svm
, INTERCEPT_XSETBV
);
1038 control
->iopm_base_pa
= iopm_base
;
1039 control
->msrpm_base_pa
= __pa(svm
->msrpm
);
1040 control
->int_ctl
= V_INTR_MASKING_MASK
;
1042 init_seg(&save
->es
);
1043 init_seg(&save
->ss
);
1044 init_seg(&save
->ds
);
1045 init_seg(&save
->fs
);
1046 init_seg(&save
->gs
);
1048 save
->cs
.selector
= 0xf000;
1049 /* Executable/Readable Code Segment */
1050 save
->cs
.attrib
= SVM_SELECTOR_READ_MASK
| SVM_SELECTOR_P_MASK
|
1051 SVM_SELECTOR_S_MASK
| SVM_SELECTOR_CODE_MASK
;
1052 save
->cs
.limit
= 0xffff;
1054 * cs.base should really be 0xffff0000, but vmx can't handle that, so
1055 * be consistent with it.
1057 * Replace when we have real mode working for vmx.
1059 save
->cs
.base
= 0xf0000;
1061 save
->gdtr
.limit
= 0xffff;
1062 save
->idtr
.limit
= 0xffff;
1064 init_sys_seg(&save
->ldtr
, SEG_TYPE_LDT
);
1065 init_sys_seg(&save
->tr
, SEG_TYPE_BUSY_TSS16
);
1067 svm_set_efer(&svm
->vcpu
, 0);
1068 save
->dr6
= 0xffff0ff0;
1070 kvm_set_rflags(&svm
->vcpu
, 2);
1071 save
->rip
= 0x0000fff0;
1072 svm
->vcpu
.arch
.regs
[VCPU_REGS_RIP
] = save
->rip
;
1075 * This is the guest-visible cr0 value.
1076 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1078 svm
->vcpu
.arch
.cr0
= 0;
1079 (void)kvm_set_cr0(&svm
->vcpu
, X86_CR0_NW
| X86_CR0_CD
| X86_CR0_ET
);
1081 save
->cr4
= X86_CR4_PAE
;
1085 /* Setup VMCB for Nested Paging */
1086 control
->nested_ctl
= 1;
1087 clr_intercept(svm
, INTERCEPT_TASK_SWITCH
);
1088 clr_intercept(svm
, INTERCEPT_INVLPG
);
1089 clr_exception_intercept(svm
, PF_VECTOR
);
1090 clr_cr_intercept(svm
, INTERCEPT_CR3_READ
);
1091 clr_cr_intercept(svm
, INTERCEPT_CR3_WRITE
);
1092 save
->g_pat
= 0x0007040600070406ULL
;
1096 svm
->asid_generation
= 0;
1098 svm
->nested
.vmcb
= 0;
1099 svm
->vcpu
.arch
.hflags
= 0;
1101 if (boot_cpu_has(X86_FEATURE_PAUSEFILTER
)) {
1102 control
->pause_filter_count
= 3000;
1103 set_intercept(svm
, INTERCEPT_PAUSE
);
1106 mark_all_dirty(svm
->vmcb
);
1111 static int svm_vcpu_reset(struct kvm_vcpu
*vcpu
)
1113 struct vcpu_svm
*svm
= to_svm(vcpu
);
1117 if (!kvm_vcpu_is_bsp(vcpu
)) {
1118 kvm_rip_write(vcpu
, 0);
1119 svm
->vmcb
->save
.cs
.base
= svm
->vcpu
.arch
.sipi_vector
<< 12;
1120 svm
->vmcb
->save
.cs
.selector
= svm
->vcpu
.arch
.sipi_vector
<< 8;
1122 vcpu
->arch
.regs_avail
= ~0;
1123 vcpu
->arch
.regs_dirty
= ~0;
1128 static struct kvm_vcpu
*svm_create_vcpu(struct kvm
*kvm
, unsigned int id
)
1130 struct vcpu_svm
*svm
;
1132 struct page
*msrpm_pages
;
1133 struct page
*hsave_page
;
1134 struct page
*nested_msrpm_pages
;
1137 svm
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
1143 svm
->tsc_ratio
= TSC_RATIO_DEFAULT
;
1145 err
= kvm_vcpu_init(&svm
->vcpu
, kvm
, id
);
1150 page
= alloc_page(GFP_KERNEL
);
1154 msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
1158 nested_msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
1159 if (!nested_msrpm_pages
)
1162 hsave_page
= alloc_page(GFP_KERNEL
);
1166 svm
->nested
.hsave
= page_address(hsave_page
);
1168 svm
->msrpm
= page_address(msrpm_pages
);
1169 svm_vcpu_init_msrpm(svm
->msrpm
);
1171 svm
->nested
.msrpm
= page_address(nested_msrpm_pages
);
1172 svm_vcpu_init_msrpm(svm
->nested
.msrpm
);
1174 svm
->vmcb
= page_address(page
);
1175 clear_page(svm
->vmcb
);
1176 svm
->vmcb_pa
= page_to_pfn(page
) << PAGE_SHIFT
;
1177 svm
->asid_generation
= 0;
1179 kvm_write_tsc(&svm
->vcpu
, 0);
1181 err
= fx_init(&svm
->vcpu
);
1185 svm
->vcpu
.arch
.apic_base
= 0xfee00000 | MSR_IA32_APICBASE_ENABLE
;
1186 if (kvm_vcpu_is_bsp(&svm
->vcpu
))
1187 svm
->vcpu
.arch
.apic_base
|= MSR_IA32_APICBASE_BSP
;
1192 __free_page(hsave_page
);
1194 __free_pages(nested_msrpm_pages
, MSRPM_ALLOC_ORDER
);
1196 __free_pages(msrpm_pages
, MSRPM_ALLOC_ORDER
);
1200 kvm_vcpu_uninit(&svm
->vcpu
);
1202 kmem_cache_free(kvm_vcpu_cache
, svm
);
1204 return ERR_PTR(err
);
1207 static void svm_free_vcpu(struct kvm_vcpu
*vcpu
)
1209 struct vcpu_svm
*svm
= to_svm(vcpu
);
1211 __free_page(pfn_to_page(svm
->vmcb_pa
>> PAGE_SHIFT
));
1212 __free_pages(virt_to_page(svm
->msrpm
), MSRPM_ALLOC_ORDER
);
1213 __free_page(virt_to_page(svm
->nested
.hsave
));
1214 __free_pages(virt_to_page(svm
->nested
.msrpm
), MSRPM_ALLOC_ORDER
);
1215 kvm_vcpu_uninit(vcpu
);
1216 kmem_cache_free(kvm_vcpu_cache
, svm
);
1219 static void svm_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
1221 struct vcpu_svm
*svm
= to_svm(vcpu
);
1224 if (unlikely(cpu
!= vcpu
->cpu
)) {
1225 svm
->asid_generation
= 0;
1226 mark_all_dirty(svm
->vmcb
);
1229 #ifdef CONFIG_X86_64
1230 rdmsrl(MSR_GS_BASE
, to_svm(vcpu
)->host
.gs_base
);
1232 savesegment(fs
, svm
->host
.fs
);
1233 savesegment(gs
, svm
->host
.gs
);
1234 svm
->host
.ldt
= kvm_read_ldt();
1236 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
1237 rdmsrl(host_save_user_msrs
[i
], svm
->host_user_msrs
[i
]);
1239 if (static_cpu_has(X86_FEATURE_TSCRATEMSR
) &&
1240 svm
->tsc_ratio
!= __get_cpu_var(current_tsc_ratio
)) {
1241 __get_cpu_var(current_tsc_ratio
) = svm
->tsc_ratio
;
1242 wrmsrl(MSR_AMD64_TSC_RATIO
, svm
->tsc_ratio
);
1246 static void svm_vcpu_put(struct kvm_vcpu
*vcpu
)
1248 struct vcpu_svm
*svm
= to_svm(vcpu
);
1251 ++vcpu
->stat
.host_state_reload
;
1252 kvm_load_ldt(svm
->host
.ldt
);
1253 #ifdef CONFIG_X86_64
1254 loadsegment(fs
, svm
->host
.fs
);
1255 wrmsrl(MSR_KERNEL_GS_BASE
, current
->thread
.gs
);
1256 load_gs_index(svm
->host
.gs
);
1258 #ifdef CONFIG_X86_32_LAZY_GS
1259 loadsegment(gs
, svm
->host
.gs
);
1262 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
1263 wrmsrl(host_save_user_msrs
[i
], svm
->host_user_msrs
[i
]);
1266 static unsigned long svm_get_rflags(struct kvm_vcpu
*vcpu
)
1268 return to_svm(vcpu
)->vmcb
->save
.rflags
;
1271 static void svm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
1273 to_svm(vcpu
)->vmcb
->save
.rflags
= rflags
;
1276 static void svm_cache_reg(struct kvm_vcpu
*vcpu
, enum kvm_reg reg
)
1279 case VCPU_EXREG_PDPTR
:
1280 BUG_ON(!npt_enabled
);
1281 load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, kvm_read_cr3(vcpu
));
1288 static void svm_set_vintr(struct vcpu_svm
*svm
)
1290 set_intercept(svm
, INTERCEPT_VINTR
);
1293 static void svm_clear_vintr(struct vcpu_svm
*svm
)
1295 clr_intercept(svm
, INTERCEPT_VINTR
);
1298 static struct vmcb_seg
*svm_seg(struct kvm_vcpu
*vcpu
, int seg
)
1300 struct vmcb_save_area
*save
= &to_svm(vcpu
)->vmcb
->save
;
1303 case VCPU_SREG_CS
: return &save
->cs
;
1304 case VCPU_SREG_DS
: return &save
->ds
;
1305 case VCPU_SREG_ES
: return &save
->es
;
1306 case VCPU_SREG_FS
: return &save
->fs
;
1307 case VCPU_SREG_GS
: return &save
->gs
;
1308 case VCPU_SREG_SS
: return &save
->ss
;
1309 case VCPU_SREG_TR
: return &save
->tr
;
1310 case VCPU_SREG_LDTR
: return &save
->ldtr
;
1316 static u64
svm_get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
1318 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
1323 static void svm_get_segment(struct kvm_vcpu
*vcpu
,
1324 struct kvm_segment
*var
, int seg
)
1326 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
1328 var
->base
= s
->base
;
1329 var
->limit
= s
->limit
;
1330 var
->selector
= s
->selector
;
1331 var
->type
= s
->attrib
& SVM_SELECTOR_TYPE_MASK
;
1332 var
->s
= (s
->attrib
>> SVM_SELECTOR_S_SHIFT
) & 1;
1333 var
->dpl
= (s
->attrib
>> SVM_SELECTOR_DPL_SHIFT
) & 3;
1334 var
->present
= (s
->attrib
>> SVM_SELECTOR_P_SHIFT
) & 1;
1335 var
->avl
= (s
->attrib
>> SVM_SELECTOR_AVL_SHIFT
) & 1;
1336 var
->l
= (s
->attrib
>> SVM_SELECTOR_L_SHIFT
) & 1;
1337 var
->db
= (s
->attrib
>> SVM_SELECTOR_DB_SHIFT
) & 1;
1338 var
->g
= (s
->attrib
>> SVM_SELECTOR_G_SHIFT
) & 1;
1341 * AMD's VMCB does not have an explicit unusable field, so emulate it
1342 * for cross vendor migration purposes by "not present"
1344 var
->unusable
= !var
->present
|| (var
->type
== 0);
1349 * SVM always stores 0 for the 'G' bit in the CS selector in
1350 * the VMCB on a VMEXIT. This hurts cross-vendor migration:
1351 * Intel's VMENTRY has a check on the 'G' bit.
1353 var
->g
= s
->limit
> 0xfffff;
1357 * Work around a bug where the busy flag in the tr selector
1367 * The accessed bit must always be set in the segment
1368 * descriptor cache, although it can be cleared in the
1369 * descriptor, the cached bit always remains at 1. Since
1370 * Intel has a check on this, set it here to support
1371 * cross-vendor migration.
1378 * On AMD CPUs sometimes the DB bit in the segment
1379 * descriptor is left as 1, although the whole segment has
1380 * been made unusable. Clear it here to pass an Intel VMX
1381 * entry check when cross vendor migrating.
1389 static int svm_get_cpl(struct kvm_vcpu
*vcpu
)
1391 struct vmcb_save_area
*save
= &to_svm(vcpu
)->vmcb
->save
;
1396 static void svm_get_idt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1398 struct vcpu_svm
*svm
= to_svm(vcpu
);
1400 dt
->size
= svm
->vmcb
->save
.idtr
.limit
;
1401 dt
->address
= svm
->vmcb
->save
.idtr
.base
;
1404 static void svm_set_idt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1406 struct vcpu_svm
*svm
= to_svm(vcpu
);
1408 svm
->vmcb
->save
.idtr
.limit
= dt
->size
;
1409 svm
->vmcb
->save
.idtr
.base
= dt
->address
;
1410 mark_dirty(svm
->vmcb
, VMCB_DT
);
1413 static void svm_get_gdt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1415 struct vcpu_svm
*svm
= to_svm(vcpu
);
1417 dt
->size
= svm
->vmcb
->save
.gdtr
.limit
;
1418 dt
->address
= svm
->vmcb
->save
.gdtr
.base
;
1421 static void svm_set_gdt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1423 struct vcpu_svm
*svm
= to_svm(vcpu
);
1425 svm
->vmcb
->save
.gdtr
.limit
= dt
->size
;
1426 svm
->vmcb
->save
.gdtr
.base
= dt
->address
;
1427 mark_dirty(svm
->vmcb
, VMCB_DT
);
1430 static void svm_decache_cr0_guest_bits(struct kvm_vcpu
*vcpu
)
1434 static void svm_decache_cr3(struct kvm_vcpu
*vcpu
)
1438 static void svm_decache_cr4_guest_bits(struct kvm_vcpu
*vcpu
)
1442 static void update_cr0_intercept(struct vcpu_svm
*svm
)
1444 ulong gcr0
= svm
->vcpu
.arch
.cr0
;
1445 u64
*hcr0
= &svm
->vmcb
->save
.cr0
;
1447 if (!svm
->vcpu
.fpu_active
)
1448 *hcr0
|= SVM_CR0_SELECTIVE_MASK
;
1450 *hcr0
= (*hcr0
& ~SVM_CR0_SELECTIVE_MASK
)
1451 | (gcr0
& SVM_CR0_SELECTIVE_MASK
);
1453 mark_dirty(svm
->vmcb
, VMCB_CR
);
1455 if (gcr0
== *hcr0
&& svm
->vcpu
.fpu_active
) {
1456 clr_cr_intercept(svm
, INTERCEPT_CR0_READ
);
1457 clr_cr_intercept(svm
, INTERCEPT_CR0_WRITE
);
1459 set_cr_intercept(svm
, INTERCEPT_CR0_READ
);
1460 set_cr_intercept(svm
, INTERCEPT_CR0_WRITE
);
1464 static void svm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
1466 struct vcpu_svm
*svm
= to_svm(vcpu
);
1468 #ifdef CONFIG_X86_64
1469 if (vcpu
->arch
.efer
& EFER_LME
) {
1470 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
1471 vcpu
->arch
.efer
|= EFER_LMA
;
1472 svm
->vmcb
->save
.efer
|= EFER_LMA
| EFER_LME
;
1475 if (is_paging(vcpu
) && !(cr0
& X86_CR0_PG
)) {
1476 vcpu
->arch
.efer
&= ~EFER_LMA
;
1477 svm
->vmcb
->save
.efer
&= ~(EFER_LMA
| EFER_LME
);
1481 vcpu
->arch
.cr0
= cr0
;
1484 cr0
|= X86_CR0_PG
| X86_CR0_WP
;
1486 if (!vcpu
->fpu_active
)
1489 * re-enable caching here because the QEMU bios
1490 * does not do it - this results in some delay at
1493 cr0
&= ~(X86_CR0_CD
| X86_CR0_NW
);
1494 svm
->vmcb
->save
.cr0
= cr0
;
1495 mark_dirty(svm
->vmcb
, VMCB_CR
);
1496 update_cr0_intercept(svm
);
1499 static void svm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
1501 unsigned long host_cr4_mce
= read_cr4() & X86_CR4_MCE
;
1502 unsigned long old_cr4
= to_svm(vcpu
)->vmcb
->save
.cr4
;
1504 if (npt_enabled
&& ((old_cr4
^ cr4
) & X86_CR4_PGE
))
1505 svm_flush_tlb(vcpu
);
1507 vcpu
->arch
.cr4
= cr4
;
1510 cr4
|= host_cr4_mce
;
1511 to_svm(vcpu
)->vmcb
->save
.cr4
= cr4
;
1512 mark_dirty(to_svm(vcpu
)->vmcb
, VMCB_CR
);
1515 static void svm_set_segment(struct kvm_vcpu
*vcpu
,
1516 struct kvm_segment
*var
, int seg
)
1518 struct vcpu_svm
*svm
= to_svm(vcpu
);
1519 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
1521 s
->base
= var
->base
;
1522 s
->limit
= var
->limit
;
1523 s
->selector
= var
->selector
;
1527 s
->attrib
= (var
->type
& SVM_SELECTOR_TYPE_MASK
);
1528 s
->attrib
|= (var
->s
& 1) << SVM_SELECTOR_S_SHIFT
;
1529 s
->attrib
|= (var
->dpl
& 3) << SVM_SELECTOR_DPL_SHIFT
;
1530 s
->attrib
|= (var
->present
& 1) << SVM_SELECTOR_P_SHIFT
;
1531 s
->attrib
|= (var
->avl
& 1) << SVM_SELECTOR_AVL_SHIFT
;
1532 s
->attrib
|= (var
->l
& 1) << SVM_SELECTOR_L_SHIFT
;
1533 s
->attrib
|= (var
->db
& 1) << SVM_SELECTOR_DB_SHIFT
;
1534 s
->attrib
|= (var
->g
& 1) << SVM_SELECTOR_G_SHIFT
;
1536 if (seg
== VCPU_SREG_CS
)
1538 = (svm
->vmcb
->save
.cs
.attrib
1539 >> SVM_SELECTOR_DPL_SHIFT
) & 3;
1541 mark_dirty(svm
->vmcb
, VMCB_SEG
);
1544 static void update_db_intercept(struct kvm_vcpu
*vcpu
)
1546 struct vcpu_svm
*svm
= to_svm(vcpu
);
1548 clr_exception_intercept(svm
, DB_VECTOR
);
1549 clr_exception_intercept(svm
, BP_VECTOR
);
1551 if (svm
->nmi_singlestep
)
1552 set_exception_intercept(svm
, DB_VECTOR
);
1554 if (vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
) {
1555 if (vcpu
->guest_debug
&
1556 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
))
1557 set_exception_intercept(svm
, DB_VECTOR
);
1558 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_SW_BP
)
1559 set_exception_intercept(svm
, BP_VECTOR
);
1561 vcpu
->guest_debug
= 0;
1564 static void svm_guest_debug(struct kvm_vcpu
*vcpu
, struct kvm_guest_debug
*dbg
)
1566 struct vcpu_svm
*svm
= to_svm(vcpu
);
1568 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
1569 svm
->vmcb
->save
.dr7
= dbg
->arch
.debugreg
[7];
1571 svm
->vmcb
->save
.dr7
= vcpu
->arch
.dr7
;
1573 mark_dirty(svm
->vmcb
, VMCB_DR
);
1575 update_db_intercept(vcpu
);
1578 static void new_asid(struct vcpu_svm
*svm
, struct svm_cpu_data
*sd
)
1580 if (sd
->next_asid
> sd
->max_asid
) {
1581 ++sd
->asid_generation
;
1583 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ALL_ASID
;
1586 svm
->asid_generation
= sd
->asid_generation
;
1587 svm
->vmcb
->control
.asid
= sd
->next_asid
++;
1589 mark_dirty(svm
->vmcb
, VMCB_ASID
);
1592 static void svm_set_dr7(struct kvm_vcpu
*vcpu
, unsigned long value
)
1594 struct vcpu_svm
*svm
= to_svm(vcpu
);
1596 svm
->vmcb
->save
.dr7
= value
;
1597 mark_dirty(svm
->vmcb
, VMCB_DR
);
1600 static int pf_interception(struct vcpu_svm
*svm
)
1602 u64 fault_address
= svm
->vmcb
->control
.exit_info_2
;
1606 switch (svm
->apf_reason
) {
1608 error_code
= svm
->vmcb
->control
.exit_info_1
;
1610 trace_kvm_page_fault(fault_address
, error_code
);
1611 if (!npt_enabled
&& kvm_event_needs_reinjection(&svm
->vcpu
))
1612 kvm_mmu_unprotect_page_virt(&svm
->vcpu
, fault_address
);
1613 r
= kvm_mmu_page_fault(&svm
->vcpu
, fault_address
, error_code
,
1614 svm
->vmcb
->control
.insn_bytes
,
1615 svm
->vmcb
->control
.insn_len
);
1617 case KVM_PV_REASON_PAGE_NOT_PRESENT
:
1618 svm
->apf_reason
= 0;
1619 local_irq_disable();
1620 kvm_async_pf_task_wait(fault_address
);
1623 case KVM_PV_REASON_PAGE_READY
:
1624 svm
->apf_reason
= 0;
1625 local_irq_disable();
1626 kvm_async_pf_task_wake(fault_address
);
1633 static int db_interception(struct vcpu_svm
*svm
)
1635 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
1637 if (!(svm
->vcpu
.guest_debug
&
1638 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
)) &&
1639 !svm
->nmi_singlestep
) {
1640 kvm_queue_exception(&svm
->vcpu
, DB_VECTOR
);
1644 if (svm
->nmi_singlestep
) {
1645 svm
->nmi_singlestep
= false;
1646 if (!(svm
->vcpu
.guest_debug
& KVM_GUESTDBG_SINGLESTEP
))
1647 svm
->vmcb
->save
.rflags
&=
1648 ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
1649 update_db_intercept(&svm
->vcpu
);
1652 if (svm
->vcpu
.guest_debug
&
1653 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
)) {
1654 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
1655 kvm_run
->debug
.arch
.pc
=
1656 svm
->vmcb
->save
.cs
.base
+ svm
->vmcb
->save
.rip
;
1657 kvm_run
->debug
.arch
.exception
= DB_VECTOR
;
1664 static int bp_interception(struct vcpu_svm
*svm
)
1666 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
1668 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
1669 kvm_run
->debug
.arch
.pc
= svm
->vmcb
->save
.cs
.base
+ svm
->vmcb
->save
.rip
;
1670 kvm_run
->debug
.arch
.exception
= BP_VECTOR
;
1674 static int ud_interception(struct vcpu_svm
*svm
)
1678 er
= emulate_instruction(&svm
->vcpu
, EMULTYPE_TRAP_UD
);
1679 if (er
!= EMULATE_DONE
)
1680 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
1684 static void svm_fpu_activate(struct kvm_vcpu
*vcpu
)
1686 struct vcpu_svm
*svm
= to_svm(vcpu
);
1688 clr_exception_intercept(svm
, NM_VECTOR
);
1690 svm
->vcpu
.fpu_active
= 1;
1691 update_cr0_intercept(svm
);
1694 static int nm_interception(struct vcpu_svm
*svm
)
1696 svm_fpu_activate(&svm
->vcpu
);
1700 static bool is_erratum_383(void)
1705 if (!erratum_383_found
)
1708 value
= native_read_msr_safe(MSR_IA32_MC0_STATUS
, &err
);
1712 /* Bit 62 may or may not be set for this mce */
1713 value
&= ~(1ULL << 62);
1715 if (value
!= 0xb600000000010015ULL
)
1718 /* Clear MCi_STATUS registers */
1719 for (i
= 0; i
< 6; ++i
)
1720 native_write_msr_safe(MSR_IA32_MCx_STATUS(i
), 0, 0);
1722 value
= native_read_msr_safe(MSR_IA32_MCG_STATUS
, &err
);
1726 value
&= ~(1ULL << 2);
1727 low
= lower_32_bits(value
);
1728 high
= upper_32_bits(value
);
1730 native_write_msr_safe(MSR_IA32_MCG_STATUS
, low
, high
);
1733 /* Flush tlb to evict multi-match entries */
1739 static void svm_handle_mce(struct vcpu_svm
*svm
)
1741 if (is_erratum_383()) {
1743 * Erratum 383 triggered. Guest state is corrupt so kill the
1746 pr_err("KVM: Guest triggered AMD Erratum 383\n");
1748 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, &svm
->vcpu
);
1754 * On an #MC intercept the MCE handler is not called automatically in
1755 * the host. So do it by hand here.
1759 /* not sure if we ever come back to this point */
1764 static int mc_interception(struct vcpu_svm
*svm
)
1769 static int shutdown_interception(struct vcpu_svm
*svm
)
1771 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
1774 * VMCB is undefined after a SHUTDOWN intercept
1775 * so reinitialize it.
1777 clear_page(svm
->vmcb
);
1780 kvm_run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
1784 static int io_interception(struct vcpu_svm
*svm
)
1786 struct kvm_vcpu
*vcpu
= &svm
->vcpu
;
1787 u32 io_info
= svm
->vmcb
->control
.exit_info_1
; /* address size bug? */
1788 int size
, in
, string
;
1791 ++svm
->vcpu
.stat
.io_exits
;
1792 string
= (io_info
& SVM_IOIO_STR_MASK
) != 0;
1793 in
= (io_info
& SVM_IOIO_TYPE_MASK
) != 0;
1795 return emulate_instruction(vcpu
, 0) == EMULATE_DONE
;
1797 port
= io_info
>> 16;
1798 size
= (io_info
& SVM_IOIO_SIZE_MASK
) >> SVM_IOIO_SIZE_SHIFT
;
1799 svm
->next_rip
= svm
->vmcb
->control
.exit_info_2
;
1800 skip_emulated_instruction(&svm
->vcpu
);
1802 return kvm_fast_pio_out(vcpu
, size
, port
);
1805 static int nmi_interception(struct vcpu_svm
*svm
)
1810 static int intr_interception(struct vcpu_svm
*svm
)
1812 ++svm
->vcpu
.stat
.irq_exits
;
1816 static int nop_on_interception(struct vcpu_svm
*svm
)
1821 static int halt_interception(struct vcpu_svm
*svm
)
1823 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 1;
1824 skip_emulated_instruction(&svm
->vcpu
);
1825 return kvm_emulate_halt(&svm
->vcpu
);
1828 static int vmmcall_interception(struct vcpu_svm
*svm
)
1830 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1831 skip_emulated_instruction(&svm
->vcpu
);
1832 kvm_emulate_hypercall(&svm
->vcpu
);
1836 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu
*vcpu
)
1838 struct vcpu_svm
*svm
= to_svm(vcpu
);
1840 return svm
->nested
.nested_cr3
;
1843 static void nested_svm_set_tdp_cr3(struct kvm_vcpu
*vcpu
,
1846 struct vcpu_svm
*svm
= to_svm(vcpu
);
1848 svm
->vmcb
->control
.nested_cr3
= root
;
1849 mark_dirty(svm
->vmcb
, VMCB_NPT
);
1850 svm_flush_tlb(vcpu
);
1853 static void nested_svm_inject_npf_exit(struct kvm_vcpu
*vcpu
,
1854 struct x86_exception
*fault
)
1856 struct vcpu_svm
*svm
= to_svm(vcpu
);
1858 svm
->vmcb
->control
.exit_code
= SVM_EXIT_NPF
;
1859 svm
->vmcb
->control
.exit_code_hi
= 0;
1860 svm
->vmcb
->control
.exit_info_1
= fault
->error_code
;
1861 svm
->vmcb
->control
.exit_info_2
= fault
->address
;
1863 nested_svm_vmexit(svm
);
1866 static int nested_svm_init_mmu_context(struct kvm_vcpu
*vcpu
)
1870 r
= kvm_init_shadow_mmu(vcpu
, &vcpu
->arch
.mmu
);
1872 vcpu
->arch
.mmu
.set_cr3
= nested_svm_set_tdp_cr3
;
1873 vcpu
->arch
.mmu
.get_cr3
= nested_svm_get_tdp_cr3
;
1874 vcpu
->arch
.mmu
.inject_page_fault
= nested_svm_inject_npf_exit
;
1875 vcpu
->arch
.mmu
.shadow_root_level
= get_npt_level();
1876 vcpu
->arch
.walk_mmu
= &vcpu
->arch
.nested_mmu
;
1881 static void nested_svm_uninit_mmu_context(struct kvm_vcpu
*vcpu
)
1883 vcpu
->arch
.walk_mmu
= &vcpu
->arch
.mmu
;
1886 static int nested_svm_check_permissions(struct vcpu_svm
*svm
)
1888 if (!(svm
->vcpu
.arch
.efer
& EFER_SVME
)
1889 || !is_paging(&svm
->vcpu
)) {
1890 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
1894 if (svm
->vmcb
->save
.cpl
) {
1895 kvm_inject_gp(&svm
->vcpu
, 0);
1902 static int nested_svm_check_exception(struct vcpu_svm
*svm
, unsigned nr
,
1903 bool has_error_code
, u32 error_code
)
1907 if (!is_guest_mode(&svm
->vcpu
))
1910 svm
->vmcb
->control
.exit_code
= SVM_EXIT_EXCP_BASE
+ nr
;
1911 svm
->vmcb
->control
.exit_code_hi
= 0;
1912 svm
->vmcb
->control
.exit_info_1
= error_code
;
1913 svm
->vmcb
->control
.exit_info_2
= svm
->vcpu
.arch
.cr2
;
1915 vmexit
= nested_svm_intercept(svm
);
1916 if (vmexit
== NESTED_EXIT_DONE
)
1917 svm
->nested
.exit_required
= true;
1922 /* This function returns true if it is save to enable the irq window */
1923 static inline bool nested_svm_intr(struct vcpu_svm
*svm
)
1925 if (!is_guest_mode(&svm
->vcpu
))
1928 if (!(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
))
1931 if (!(svm
->vcpu
.arch
.hflags
& HF_HIF_MASK
))
1935 * if vmexit was already requested (by intercepted exception
1936 * for instance) do not overwrite it with "external interrupt"
1939 if (svm
->nested
.exit_required
)
1942 svm
->vmcb
->control
.exit_code
= SVM_EXIT_INTR
;
1943 svm
->vmcb
->control
.exit_info_1
= 0;
1944 svm
->vmcb
->control
.exit_info_2
= 0;
1946 if (svm
->nested
.intercept
& 1ULL) {
1948 * The #vmexit can't be emulated here directly because this
1949 * code path runs with irqs and preemtion disabled. A
1950 * #vmexit emulation might sleep. Only signal request for
1953 svm
->nested
.exit_required
= true;
1954 trace_kvm_nested_intr_vmexit(svm
->vmcb
->save
.rip
);
1961 /* This function returns true if it is save to enable the nmi window */
1962 static inline bool nested_svm_nmi(struct vcpu_svm
*svm
)
1964 if (!is_guest_mode(&svm
->vcpu
))
1967 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_NMI
)))
1970 svm
->vmcb
->control
.exit_code
= SVM_EXIT_NMI
;
1971 svm
->nested
.exit_required
= true;
1976 static void *nested_svm_map(struct vcpu_svm
*svm
, u64 gpa
, struct page
**_page
)
1982 page
= gfn_to_page(svm
->vcpu
.kvm
, gpa
>> PAGE_SHIFT
);
1983 if (is_error_page(page
))
1991 kvm_release_page_clean(page
);
1992 kvm_inject_gp(&svm
->vcpu
, 0);
1997 static void nested_svm_unmap(struct page
*page
)
2000 kvm_release_page_dirty(page
);
2003 static int nested_svm_intercept_ioio(struct vcpu_svm
*svm
)
2009 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_IOIO_PROT
)))
2010 return NESTED_EXIT_HOST
;
2012 port
= svm
->vmcb
->control
.exit_info_1
>> 16;
2013 gpa
= svm
->nested
.vmcb_iopm
+ (port
/ 8);
2017 if (kvm_read_guest(svm
->vcpu
.kvm
, gpa
, &val
, 1))
2020 return val
? NESTED_EXIT_DONE
: NESTED_EXIT_HOST
;
2023 static int nested_svm_exit_handled_msr(struct vcpu_svm
*svm
)
2025 u32 offset
, msr
, value
;
2028 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_MSR_PROT
)))
2029 return NESTED_EXIT_HOST
;
2031 msr
= svm
->vcpu
.arch
.regs
[VCPU_REGS_RCX
];
2032 offset
= svm_msrpm_offset(msr
);
2033 write
= svm
->vmcb
->control
.exit_info_1
& 1;
2034 mask
= 1 << ((2 * (msr
& 0xf)) + write
);
2036 if (offset
== MSR_INVALID
)
2037 return NESTED_EXIT_DONE
;
2039 /* Offset is in 32 bit units but need in 8 bit units */
2042 if (kvm_read_guest(svm
->vcpu
.kvm
, svm
->nested
.vmcb_msrpm
+ offset
, &value
, 4))
2043 return NESTED_EXIT_DONE
;
2045 return (value
& mask
) ? NESTED_EXIT_DONE
: NESTED_EXIT_HOST
;
2048 static int nested_svm_exit_special(struct vcpu_svm
*svm
)
2050 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
2052 switch (exit_code
) {
2055 case SVM_EXIT_EXCP_BASE
+ MC_VECTOR
:
2056 return NESTED_EXIT_HOST
;
2058 /* For now we are always handling NPFs when using them */
2060 return NESTED_EXIT_HOST
;
2062 case SVM_EXIT_EXCP_BASE
+ PF_VECTOR
:
2063 /* When we're shadowing, trap PFs, but not async PF */
2064 if (!npt_enabled
&& svm
->apf_reason
== 0)
2065 return NESTED_EXIT_HOST
;
2067 case SVM_EXIT_EXCP_BASE
+ NM_VECTOR
:
2068 nm_interception(svm
);
2074 return NESTED_EXIT_CONTINUE
;
2078 * If this function returns true, this #vmexit was already handled
2080 static int nested_svm_intercept(struct vcpu_svm
*svm
)
2082 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
2083 int vmexit
= NESTED_EXIT_HOST
;
2085 switch (exit_code
) {
2087 vmexit
= nested_svm_exit_handled_msr(svm
);
2090 vmexit
= nested_svm_intercept_ioio(svm
);
2092 case SVM_EXIT_READ_CR0
... SVM_EXIT_WRITE_CR8
: {
2093 u32 bit
= 1U << (exit_code
- SVM_EXIT_READ_CR0
);
2094 if (svm
->nested
.intercept_cr
& bit
)
2095 vmexit
= NESTED_EXIT_DONE
;
2098 case SVM_EXIT_READ_DR0
... SVM_EXIT_WRITE_DR7
: {
2099 u32 bit
= 1U << (exit_code
- SVM_EXIT_READ_DR0
);
2100 if (svm
->nested
.intercept_dr
& bit
)
2101 vmexit
= NESTED_EXIT_DONE
;
2104 case SVM_EXIT_EXCP_BASE
... SVM_EXIT_EXCP_BASE
+ 0x1f: {
2105 u32 excp_bits
= 1 << (exit_code
- SVM_EXIT_EXCP_BASE
);
2106 if (svm
->nested
.intercept_exceptions
& excp_bits
)
2107 vmexit
= NESTED_EXIT_DONE
;
2108 /* async page fault always cause vmexit */
2109 else if ((exit_code
== SVM_EXIT_EXCP_BASE
+ PF_VECTOR
) &&
2110 svm
->apf_reason
!= 0)
2111 vmexit
= NESTED_EXIT_DONE
;
2114 case SVM_EXIT_ERR
: {
2115 vmexit
= NESTED_EXIT_DONE
;
2119 u64 exit_bits
= 1ULL << (exit_code
- SVM_EXIT_INTR
);
2120 if (svm
->nested
.intercept
& exit_bits
)
2121 vmexit
= NESTED_EXIT_DONE
;
2128 static int nested_svm_exit_handled(struct vcpu_svm
*svm
)
2132 vmexit
= nested_svm_intercept(svm
);
2134 if (vmexit
== NESTED_EXIT_DONE
)
2135 nested_svm_vmexit(svm
);
2140 static inline void copy_vmcb_control_area(struct vmcb
*dst_vmcb
, struct vmcb
*from_vmcb
)
2142 struct vmcb_control_area
*dst
= &dst_vmcb
->control
;
2143 struct vmcb_control_area
*from
= &from_vmcb
->control
;
2145 dst
->intercept_cr
= from
->intercept_cr
;
2146 dst
->intercept_dr
= from
->intercept_dr
;
2147 dst
->intercept_exceptions
= from
->intercept_exceptions
;
2148 dst
->intercept
= from
->intercept
;
2149 dst
->iopm_base_pa
= from
->iopm_base_pa
;
2150 dst
->msrpm_base_pa
= from
->msrpm_base_pa
;
2151 dst
->tsc_offset
= from
->tsc_offset
;
2152 dst
->asid
= from
->asid
;
2153 dst
->tlb_ctl
= from
->tlb_ctl
;
2154 dst
->int_ctl
= from
->int_ctl
;
2155 dst
->int_vector
= from
->int_vector
;
2156 dst
->int_state
= from
->int_state
;
2157 dst
->exit_code
= from
->exit_code
;
2158 dst
->exit_code_hi
= from
->exit_code_hi
;
2159 dst
->exit_info_1
= from
->exit_info_1
;
2160 dst
->exit_info_2
= from
->exit_info_2
;
2161 dst
->exit_int_info
= from
->exit_int_info
;
2162 dst
->exit_int_info_err
= from
->exit_int_info_err
;
2163 dst
->nested_ctl
= from
->nested_ctl
;
2164 dst
->event_inj
= from
->event_inj
;
2165 dst
->event_inj_err
= from
->event_inj_err
;
2166 dst
->nested_cr3
= from
->nested_cr3
;
2167 dst
->lbr_ctl
= from
->lbr_ctl
;
2170 static int nested_svm_vmexit(struct vcpu_svm
*svm
)
2172 struct vmcb
*nested_vmcb
;
2173 struct vmcb
*hsave
= svm
->nested
.hsave
;
2174 struct vmcb
*vmcb
= svm
->vmcb
;
2177 trace_kvm_nested_vmexit_inject(vmcb
->control
.exit_code
,
2178 vmcb
->control
.exit_info_1
,
2179 vmcb
->control
.exit_info_2
,
2180 vmcb
->control
.exit_int_info
,
2181 vmcb
->control
.exit_int_info_err
);
2183 nested_vmcb
= nested_svm_map(svm
, svm
->nested
.vmcb
, &page
);
2187 /* Exit Guest-Mode */
2188 leave_guest_mode(&svm
->vcpu
);
2189 svm
->nested
.vmcb
= 0;
2191 /* Give the current vmcb to the guest */
2194 nested_vmcb
->save
.es
= vmcb
->save
.es
;
2195 nested_vmcb
->save
.cs
= vmcb
->save
.cs
;
2196 nested_vmcb
->save
.ss
= vmcb
->save
.ss
;
2197 nested_vmcb
->save
.ds
= vmcb
->save
.ds
;
2198 nested_vmcb
->save
.gdtr
= vmcb
->save
.gdtr
;
2199 nested_vmcb
->save
.idtr
= vmcb
->save
.idtr
;
2200 nested_vmcb
->save
.efer
= svm
->vcpu
.arch
.efer
;
2201 nested_vmcb
->save
.cr0
= kvm_read_cr0(&svm
->vcpu
);
2202 nested_vmcb
->save
.cr3
= kvm_read_cr3(&svm
->vcpu
);
2203 nested_vmcb
->save
.cr2
= vmcb
->save
.cr2
;
2204 nested_vmcb
->save
.cr4
= svm
->vcpu
.arch
.cr4
;
2205 nested_vmcb
->save
.rflags
= kvm_get_rflags(&svm
->vcpu
);
2206 nested_vmcb
->save
.rip
= vmcb
->save
.rip
;
2207 nested_vmcb
->save
.rsp
= vmcb
->save
.rsp
;
2208 nested_vmcb
->save
.rax
= vmcb
->save
.rax
;
2209 nested_vmcb
->save
.dr7
= vmcb
->save
.dr7
;
2210 nested_vmcb
->save
.dr6
= vmcb
->save
.dr6
;
2211 nested_vmcb
->save
.cpl
= vmcb
->save
.cpl
;
2213 nested_vmcb
->control
.int_ctl
= vmcb
->control
.int_ctl
;
2214 nested_vmcb
->control
.int_vector
= vmcb
->control
.int_vector
;
2215 nested_vmcb
->control
.int_state
= vmcb
->control
.int_state
;
2216 nested_vmcb
->control
.exit_code
= vmcb
->control
.exit_code
;
2217 nested_vmcb
->control
.exit_code_hi
= vmcb
->control
.exit_code_hi
;
2218 nested_vmcb
->control
.exit_info_1
= vmcb
->control
.exit_info_1
;
2219 nested_vmcb
->control
.exit_info_2
= vmcb
->control
.exit_info_2
;
2220 nested_vmcb
->control
.exit_int_info
= vmcb
->control
.exit_int_info
;
2221 nested_vmcb
->control
.exit_int_info_err
= vmcb
->control
.exit_int_info_err
;
2222 nested_vmcb
->control
.next_rip
= vmcb
->control
.next_rip
;
2225 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2226 * to make sure that we do not lose injected events. So check event_inj
2227 * here and copy it to exit_int_info if it is valid.
2228 * Exit_int_info and event_inj can't be both valid because the case
2229 * below only happens on a VMRUN instruction intercept which has
2230 * no valid exit_int_info set.
2232 if (vmcb
->control
.event_inj
& SVM_EVTINJ_VALID
) {
2233 struct vmcb_control_area
*nc
= &nested_vmcb
->control
;
2235 nc
->exit_int_info
= vmcb
->control
.event_inj
;
2236 nc
->exit_int_info_err
= vmcb
->control
.event_inj_err
;
2239 nested_vmcb
->control
.tlb_ctl
= 0;
2240 nested_vmcb
->control
.event_inj
= 0;
2241 nested_vmcb
->control
.event_inj_err
= 0;
2243 /* We always set V_INTR_MASKING and remember the old value in hflags */
2244 if (!(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
))
2245 nested_vmcb
->control
.int_ctl
&= ~V_INTR_MASKING_MASK
;
2247 /* Restore the original control entries */
2248 copy_vmcb_control_area(vmcb
, hsave
);
2250 kvm_clear_exception_queue(&svm
->vcpu
);
2251 kvm_clear_interrupt_queue(&svm
->vcpu
);
2253 svm
->nested
.nested_cr3
= 0;
2255 /* Restore selected save entries */
2256 svm
->vmcb
->save
.es
= hsave
->save
.es
;
2257 svm
->vmcb
->save
.cs
= hsave
->save
.cs
;
2258 svm
->vmcb
->save
.ss
= hsave
->save
.ss
;
2259 svm
->vmcb
->save
.ds
= hsave
->save
.ds
;
2260 svm
->vmcb
->save
.gdtr
= hsave
->save
.gdtr
;
2261 svm
->vmcb
->save
.idtr
= hsave
->save
.idtr
;
2262 kvm_set_rflags(&svm
->vcpu
, hsave
->save
.rflags
);
2263 svm_set_efer(&svm
->vcpu
, hsave
->save
.efer
);
2264 svm_set_cr0(&svm
->vcpu
, hsave
->save
.cr0
| X86_CR0_PE
);
2265 svm_set_cr4(&svm
->vcpu
, hsave
->save
.cr4
);
2267 svm
->vmcb
->save
.cr3
= hsave
->save
.cr3
;
2268 svm
->vcpu
.arch
.cr3
= hsave
->save
.cr3
;
2270 (void)kvm_set_cr3(&svm
->vcpu
, hsave
->save
.cr3
);
2272 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
, hsave
->save
.rax
);
2273 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RSP
, hsave
->save
.rsp
);
2274 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RIP
, hsave
->save
.rip
);
2275 svm
->vmcb
->save
.dr7
= 0;
2276 svm
->vmcb
->save
.cpl
= 0;
2277 svm
->vmcb
->control
.exit_int_info
= 0;
2279 mark_all_dirty(svm
->vmcb
);
2281 nested_svm_unmap(page
);
2283 nested_svm_uninit_mmu_context(&svm
->vcpu
);
2284 kvm_mmu_reset_context(&svm
->vcpu
);
2285 kvm_mmu_load(&svm
->vcpu
);
2290 static bool nested_svm_vmrun_msrpm(struct vcpu_svm
*svm
)
2293 * This function merges the msr permission bitmaps of kvm and the
2294 * nested vmcb. It is omptimized in that it only merges the parts where
2295 * the kvm msr permission bitmap may contain zero bits
2299 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_MSR_PROT
)))
2302 for (i
= 0; i
< MSRPM_OFFSETS
; i
++) {
2306 if (msrpm_offsets
[i
] == 0xffffffff)
2309 p
= msrpm_offsets
[i
];
2310 offset
= svm
->nested
.vmcb_msrpm
+ (p
* 4);
2312 if (kvm_read_guest(svm
->vcpu
.kvm
, offset
, &value
, 4))
2315 svm
->nested
.msrpm
[p
] = svm
->msrpm
[p
] | value
;
2318 svm
->vmcb
->control
.msrpm_base_pa
= __pa(svm
->nested
.msrpm
);
2323 static bool nested_vmcb_checks(struct vmcb
*vmcb
)
2325 if ((vmcb
->control
.intercept
& (1ULL << INTERCEPT_VMRUN
)) == 0)
2328 if (vmcb
->control
.asid
== 0)
2331 if (vmcb
->control
.nested_ctl
&& !npt_enabled
)
2337 static bool nested_svm_vmrun(struct vcpu_svm
*svm
)
2339 struct vmcb
*nested_vmcb
;
2340 struct vmcb
*hsave
= svm
->nested
.hsave
;
2341 struct vmcb
*vmcb
= svm
->vmcb
;
2345 vmcb_gpa
= svm
->vmcb
->save
.rax
;
2347 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, &page
);
2351 if (!nested_vmcb_checks(nested_vmcb
)) {
2352 nested_vmcb
->control
.exit_code
= SVM_EXIT_ERR
;
2353 nested_vmcb
->control
.exit_code_hi
= 0;
2354 nested_vmcb
->control
.exit_info_1
= 0;
2355 nested_vmcb
->control
.exit_info_2
= 0;
2357 nested_svm_unmap(page
);
2362 trace_kvm_nested_vmrun(svm
->vmcb
->save
.rip
, vmcb_gpa
,
2363 nested_vmcb
->save
.rip
,
2364 nested_vmcb
->control
.int_ctl
,
2365 nested_vmcb
->control
.event_inj
,
2366 nested_vmcb
->control
.nested_ctl
);
2368 trace_kvm_nested_intercepts(nested_vmcb
->control
.intercept_cr
& 0xffff,
2369 nested_vmcb
->control
.intercept_cr
>> 16,
2370 nested_vmcb
->control
.intercept_exceptions
,
2371 nested_vmcb
->control
.intercept
);
2373 /* Clear internal status */
2374 kvm_clear_exception_queue(&svm
->vcpu
);
2375 kvm_clear_interrupt_queue(&svm
->vcpu
);
2378 * Save the old vmcb, so we don't need to pick what we save, but can
2379 * restore everything when a VMEXIT occurs
2381 hsave
->save
.es
= vmcb
->save
.es
;
2382 hsave
->save
.cs
= vmcb
->save
.cs
;
2383 hsave
->save
.ss
= vmcb
->save
.ss
;
2384 hsave
->save
.ds
= vmcb
->save
.ds
;
2385 hsave
->save
.gdtr
= vmcb
->save
.gdtr
;
2386 hsave
->save
.idtr
= vmcb
->save
.idtr
;
2387 hsave
->save
.efer
= svm
->vcpu
.arch
.efer
;
2388 hsave
->save
.cr0
= kvm_read_cr0(&svm
->vcpu
);
2389 hsave
->save
.cr4
= svm
->vcpu
.arch
.cr4
;
2390 hsave
->save
.rflags
= kvm_get_rflags(&svm
->vcpu
);
2391 hsave
->save
.rip
= kvm_rip_read(&svm
->vcpu
);
2392 hsave
->save
.rsp
= vmcb
->save
.rsp
;
2393 hsave
->save
.rax
= vmcb
->save
.rax
;
2395 hsave
->save
.cr3
= vmcb
->save
.cr3
;
2397 hsave
->save
.cr3
= kvm_read_cr3(&svm
->vcpu
);
2399 copy_vmcb_control_area(hsave
, vmcb
);
2401 if (kvm_get_rflags(&svm
->vcpu
) & X86_EFLAGS_IF
)
2402 svm
->vcpu
.arch
.hflags
|= HF_HIF_MASK
;
2404 svm
->vcpu
.arch
.hflags
&= ~HF_HIF_MASK
;
2406 if (nested_vmcb
->control
.nested_ctl
) {
2407 kvm_mmu_unload(&svm
->vcpu
);
2408 svm
->nested
.nested_cr3
= nested_vmcb
->control
.nested_cr3
;
2409 nested_svm_init_mmu_context(&svm
->vcpu
);
2412 /* Load the nested guest state */
2413 svm
->vmcb
->save
.es
= nested_vmcb
->save
.es
;
2414 svm
->vmcb
->save
.cs
= nested_vmcb
->save
.cs
;
2415 svm
->vmcb
->save
.ss
= nested_vmcb
->save
.ss
;
2416 svm
->vmcb
->save
.ds
= nested_vmcb
->save
.ds
;
2417 svm
->vmcb
->save
.gdtr
= nested_vmcb
->save
.gdtr
;
2418 svm
->vmcb
->save
.idtr
= nested_vmcb
->save
.idtr
;
2419 kvm_set_rflags(&svm
->vcpu
, nested_vmcb
->save
.rflags
);
2420 svm_set_efer(&svm
->vcpu
, nested_vmcb
->save
.efer
);
2421 svm_set_cr0(&svm
->vcpu
, nested_vmcb
->save
.cr0
);
2422 svm_set_cr4(&svm
->vcpu
, nested_vmcb
->save
.cr4
);
2424 svm
->vmcb
->save
.cr3
= nested_vmcb
->save
.cr3
;
2425 svm
->vcpu
.arch
.cr3
= nested_vmcb
->save
.cr3
;
2427 (void)kvm_set_cr3(&svm
->vcpu
, nested_vmcb
->save
.cr3
);
2429 /* Guest paging mode is active - reset mmu */
2430 kvm_mmu_reset_context(&svm
->vcpu
);
2432 svm
->vmcb
->save
.cr2
= svm
->vcpu
.arch
.cr2
= nested_vmcb
->save
.cr2
;
2433 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
, nested_vmcb
->save
.rax
);
2434 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RSP
, nested_vmcb
->save
.rsp
);
2435 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RIP
, nested_vmcb
->save
.rip
);
2437 /* In case we don't even reach vcpu_run, the fields are not updated */
2438 svm
->vmcb
->save
.rax
= nested_vmcb
->save
.rax
;
2439 svm
->vmcb
->save
.rsp
= nested_vmcb
->save
.rsp
;
2440 svm
->vmcb
->save
.rip
= nested_vmcb
->save
.rip
;
2441 svm
->vmcb
->save
.dr7
= nested_vmcb
->save
.dr7
;
2442 svm
->vmcb
->save
.dr6
= nested_vmcb
->save
.dr6
;
2443 svm
->vmcb
->save
.cpl
= nested_vmcb
->save
.cpl
;
2445 svm
->nested
.vmcb_msrpm
= nested_vmcb
->control
.msrpm_base_pa
& ~0x0fffULL
;
2446 svm
->nested
.vmcb_iopm
= nested_vmcb
->control
.iopm_base_pa
& ~0x0fffULL
;
2448 /* cache intercepts */
2449 svm
->nested
.intercept_cr
= nested_vmcb
->control
.intercept_cr
;
2450 svm
->nested
.intercept_dr
= nested_vmcb
->control
.intercept_dr
;
2451 svm
->nested
.intercept_exceptions
= nested_vmcb
->control
.intercept_exceptions
;
2452 svm
->nested
.intercept
= nested_vmcb
->control
.intercept
;
2454 svm_flush_tlb(&svm
->vcpu
);
2455 svm
->vmcb
->control
.int_ctl
= nested_vmcb
->control
.int_ctl
| V_INTR_MASKING_MASK
;
2456 if (nested_vmcb
->control
.int_ctl
& V_INTR_MASKING_MASK
)
2457 svm
->vcpu
.arch
.hflags
|= HF_VINTR_MASK
;
2459 svm
->vcpu
.arch
.hflags
&= ~HF_VINTR_MASK
;
2461 if (svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
) {
2462 /* We only want the cr8 intercept bits of the guest */
2463 clr_cr_intercept(svm
, INTERCEPT_CR8_READ
);
2464 clr_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
2467 /* We don't want to see VMMCALLs from a nested guest */
2468 clr_intercept(svm
, INTERCEPT_VMMCALL
);
2470 svm
->vmcb
->control
.lbr_ctl
= nested_vmcb
->control
.lbr_ctl
;
2471 svm
->vmcb
->control
.int_vector
= nested_vmcb
->control
.int_vector
;
2472 svm
->vmcb
->control
.int_state
= nested_vmcb
->control
.int_state
;
2473 svm
->vmcb
->control
.tsc_offset
+= nested_vmcb
->control
.tsc_offset
;
2474 svm
->vmcb
->control
.event_inj
= nested_vmcb
->control
.event_inj
;
2475 svm
->vmcb
->control
.event_inj_err
= nested_vmcb
->control
.event_inj_err
;
2477 nested_svm_unmap(page
);
2479 /* Enter Guest-Mode */
2480 enter_guest_mode(&svm
->vcpu
);
2483 * Merge guest and host intercepts - must be called with vcpu in
2484 * guest-mode to take affect here
2486 recalc_intercepts(svm
);
2488 svm
->nested
.vmcb
= vmcb_gpa
;
2492 mark_all_dirty(svm
->vmcb
);
2497 static void nested_svm_vmloadsave(struct vmcb
*from_vmcb
, struct vmcb
*to_vmcb
)
2499 to_vmcb
->save
.fs
= from_vmcb
->save
.fs
;
2500 to_vmcb
->save
.gs
= from_vmcb
->save
.gs
;
2501 to_vmcb
->save
.tr
= from_vmcb
->save
.tr
;
2502 to_vmcb
->save
.ldtr
= from_vmcb
->save
.ldtr
;
2503 to_vmcb
->save
.kernel_gs_base
= from_vmcb
->save
.kernel_gs_base
;
2504 to_vmcb
->save
.star
= from_vmcb
->save
.star
;
2505 to_vmcb
->save
.lstar
= from_vmcb
->save
.lstar
;
2506 to_vmcb
->save
.cstar
= from_vmcb
->save
.cstar
;
2507 to_vmcb
->save
.sfmask
= from_vmcb
->save
.sfmask
;
2508 to_vmcb
->save
.sysenter_cs
= from_vmcb
->save
.sysenter_cs
;
2509 to_vmcb
->save
.sysenter_esp
= from_vmcb
->save
.sysenter_esp
;
2510 to_vmcb
->save
.sysenter_eip
= from_vmcb
->save
.sysenter_eip
;
2513 static int vmload_interception(struct vcpu_svm
*svm
)
2515 struct vmcb
*nested_vmcb
;
2518 if (nested_svm_check_permissions(svm
))
2521 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, &page
);
2525 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2526 skip_emulated_instruction(&svm
->vcpu
);
2528 nested_svm_vmloadsave(nested_vmcb
, svm
->vmcb
);
2529 nested_svm_unmap(page
);
2534 static int vmsave_interception(struct vcpu_svm
*svm
)
2536 struct vmcb
*nested_vmcb
;
2539 if (nested_svm_check_permissions(svm
))
2542 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, &page
);
2546 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2547 skip_emulated_instruction(&svm
->vcpu
);
2549 nested_svm_vmloadsave(svm
->vmcb
, nested_vmcb
);
2550 nested_svm_unmap(page
);
2555 static int vmrun_interception(struct vcpu_svm
*svm
)
2557 if (nested_svm_check_permissions(svm
))
2560 /* Save rip after vmrun instruction */
2561 kvm_rip_write(&svm
->vcpu
, kvm_rip_read(&svm
->vcpu
) + 3);
2563 if (!nested_svm_vmrun(svm
))
2566 if (!nested_svm_vmrun_msrpm(svm
))
2573 svm
->vmcb
->control
.exit_code
= SVM_EXIT_ERR
;
2574 svm
->vmcb
->control
.exit_code_hi
= 0;
2575 svm
->vmcb
->control
.exit_info_1
= 0;
2576 svm
->vmcb
->control
.exit_info_2
= 0;
2578 nested_svm_vmexit(svm
);
2583 static int stgi_interception(struct vcpu_svm
*svm
)
2585 if (nested_svm_check_permissions(svm
))
2588 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2589 skip_emulated_instruction(&svm
->vcpu
);
2590 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
2597 static int clgi_interception(struct vcpu_svm
*svm
)
2599 if (nested_svm_check_permissions(svm
))
2602 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2603 skip_emulated_instruction(&svm
->vcpu
);
2607 /* After a CLGI no interrupts should come */
2608 svm_clear_vintr(svm
);
2609 svm
->vmcb
->control
.int_ctl
&= ~V_IRQ_MASK
;
2611 mark_dirty(svm
->vmcb
, VMCB_INTR
);
2616 static int invlpga_interception(struct vcpu_svm
*svm
)
2618 struct kvm_vcpu
*vcpu
= &svm
->vcpu
;
2620 trace_kvm_invlpga(svm
->vmcb
->save
.rip
, vcpu
->arch
.regs
[VCPU_REGS_RCX
],
2621 vcpu
->arch
.regs
[VCPU_REGS_RAX
]);
2623 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2624 kvm_mmu_invlpg(vcpu
, vcpu
->arch
.regs
[VCPU_REGS_RAX
]);
2626 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2627 skip_emulated_instruction(&svm
->vcpu
);
2631 static int skinit_interception(struct vcpu_svm
*svm
)
2633 trace_kvm_skinit(svm
->vmcb
->save
.rip
, svm
->vcpu
.arch
.regs
[VCPU_REGS_RAX
]);
2635 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
2639 static int xsetbv_interception(struct vcpu_svm
*svm
)
2641 u64 new_bv
= kvm_read_edx_eax(&svm
->vcpu
);
2642 u32 index
= kvm_register_read(&svm
->vcpu
, VCPU_REGS_RCX
);
2644 if (kvm_set_xcr(&svm
->vcpu
, index
, new_bv
) == 0) {
2645 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2646 skip_emulated_instruction(&svm
->vcpu
);
2652 static int invalid_op_interception(struct vcpu_svm
*svm
)
2654 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
2658 static int task_switch_interception(struct vcpu_svm
*svm
)
2662 int int_type
= svm
->vmcb
->control
.exit_int_info
&
2663 SVM_EXITINTINFO_TYPE_MASK
;
2664 int int_vec
= svm
->vmcb
->control
.exit_int_info
& SVM_EVTINJ_VEC_MASK
;
2666 svm
->vmcb
->control
.exit_int_info
& SVM_EXITINTINFO_TYPE_MASK
;
2668 svm
->vmcb
->control
.exit_int_info
& SVM_EXITINTINFO_VALID
;
2669 bool has_error_code
= false;
2672 tss_selector
= (u16
)svm
->vmcb
->control
.exit_info_1
;
2674 if (svm
->vmcb
->control
.exit_info_2
&
2675 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET
))
2676 reason
= TASK_SWITCH_IRET
;
2677 else if (svm
->vmcb
->control
.exit_info_2
&
2678 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP
))
2679 reason
= TASK_SWITCH_JMP
;
2681 reason
= TASK_SWITCH_GATE
;
2683 reason
= TASK_SWITCH_CALL
;
2685 if (reason
== TASK_SWITCH_GATE
) {
2687 case SVM_EXITINTINFO_TYPE_NMI
:
2688 svm
->vcpu
.arch
.nmi_injected
= false;
2690 case SVM_EXITINTINFO_TYPE_EXEPT
:
2691 if (svm
->vmcb
->control
.exit_info_2
&
2692 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE
)) {
2693 has_error_code
= true;
2695 (u32
)svm
->vmcb
->control
.exit_info_2
;
2697 kvm_clear_exception_queue(&svm
->vcpu
);
2699 case SVM_EXITINTINFO_TYPE_INTR
:
2700 kvm_clear_interrupt_queue(&svm
->vcpu
);
2707 if (reason
!= TASK_SWITCH_GATE
||
2708 int_type
== SVM_EXITINTINFO_TYPE_SOFT
||
2709 (int_type
== SVM_EXITINTINFO_TYPE_EXEPT
&&
2710 (int_vec
== OF_VECTOR
|| int_vec
== BP_VECTOR
)))
2711 skip_emulated_instruction(&svm
->vcpu
);
2713 if (kvm_task_switch(&svm
->vcpu
, tss_selector
, reason
,
2714 has_error_code
, error_code
) == EMULATE_FAIL
) {
2715 svm
->vcpu
.run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
2716 svm
->vcpu
.run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
2717 svm
->vcpu
.run
->internal
.ndata
= 0;
2723 static int cpuid_interception(struct vcpu_svm
*svm
)
2725 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
2726 kvm_emulate_cpuid(&svm
->vcpu
);
2730 static int iret_interception(struct vcpu_svm
*svm
)
2732 ++svm
->vcpu
.stat
.nmi_window_exits
;
2733 clr_intercept(svm
, INTERCEPT_IRET
);
2734 svm
->vcpu
.arch
.hflags
|= HF_IRET_MASK
;
2735 svm
->nmi_iret_rip
= kvm_rip_read(&svm
->vcpu
);
2739 static int invlpg_interception(struct vcpu_svm
*svm
)
2741 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS
))
2742 return emulate_instruction(&svm
->vcpu
, 0) == EMULATE_DONE
;
2744 kvm_mmu_invlpg(&svm
->vcpu
, svm
->vmcb
->control
.exit_info_1
);
2745 skip_emulated_instruction(&svm
->vcpu
);
2749 static int emulate_on_interception(struct vcpu_svm
*svm
)
2751 return emulate_instruction(&svm
->vcpu
, 0) == EMULATE_DONE
;
2754 bool check_selective_cr0_intercepted(struct vcpu_svm
*svm
, unsigned long val
)
2756 unsigned long cr0
= svm
->vcpu
.arch
.cr0
;
2760 intercept
= svm
->nested
.intercept
;
2762 if (!is_guest_mode(&svm
->vcpu
) ||
2763 (!(intercept
& (1ULL << INTERCEPT_SELECTIVE_CR0
))))
2766 cr0
&= ~SVM_CR0_SELECTIVE_MASK
;
2767 val
&= ~SVM_CR0_SELECTIVE_MASK
;
2770 svm
->vmcb
->control
.exit_code
= SVM_EXIT_CR0_SEL_WRITE
;
2771 ret
= (nested_svm_exit_handled(svm
) == NESTED_EXIT_DONE
);
2777 #define CR_VALID (1ULL << 63)
2779 static int cr_interception(struct vcpu_svm
*svm
)
2785 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS
))
2786 return emulate_on_interception(svm
);
2788 if (unlikely((svm
->vmcb
->control
.exit_info_1
& CR_VALID
) == 0))
2789 return emulate_on_interception(svm
);
2791 reg
= svm
->vmcb
->control
.exit_info_1
& SVM_EXITINFO_REG_MASK
;
2792 cr
= svm
->vmcb
->control
.exit_code
- SVM_EXIT_READ_CR0
;
2795 if (cr
>= 16) { /* mov to cr */
2797 val
= kvm_register_read(&svm
->vcpu
, reg
);
2800 if (!check_selective_cr0_intercepted(svm
, val
))
2801 err
= kvm_set_cr0(&svm
->vcpu
, val
);
2807 err
= kvm_set_cr3(&svm
->vcpu
, val
);
2810 err
= kvm_set_cr4(&svm
->vcpu
, val
);
2813 err
= kvm_set_cr8(&svm
->vcpu
, val
);
2816 WARN(1, "unhandled write to CR%d", cr
);
2817 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
2820 } else { /* mov from cr */
2823 val
= kvm_read_cr0(&svm
->vcpu
);
2826 val
= svm
->vcpu
.arch
.cr2
;
2829 val
= kvm_read_cr3(&svm
->vcpu
);
2832 val
= kvm_read_cr4(&svm
->vcpu
);
2835 val
= kvm_get_cr8(&svm
->vcpu
);
2838 WARN(1, "unhandled read from CR%d", cr
);
2839 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
2842 kvm_register_write(&svm
->vcpu
, reg
, val
);
2844 kvm_complete_insn_gp(&svm
->vcpu
, err
);
2849 static int dr_interception(struct vcpu_svm
*svm
)
2855 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS
))
2856 return emulate_on_interception(svm
);
2858 reg
= svm
->vmcb
->control
.exit_info_1
& SVM_EXITINFO_REG_MASK
;
2859 dr
= svm
->vmcb
->control
.exit_code
- SVM_EXIT_READ_DR0
;
2861 if (dr
>= 16) { /* mov to DRn */
2862 val
= kvm_register_read(&svm
->vcpu
, reg
);
2863 kvm_set_dr(&svm
->vcpu
, dr
- 16, val
);
2865 err
= kvm_get_dr(&svm
->vcpu
, dr
, &val
);
2867 kvm_register_write(&svm
->vcpu
, reg
, val
);
2870 skip_emulated_instruction(&svm
->vcpu
);
2875 static int cr8_write_interception(struct vcpu_svm
*svm
)
2877 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
2880 u8 cr8_prev
= kvm_get_cr8(&svm
->vcpu
);
2881 /* instruction emulation calls kvm_set_cr8() */
2882 r
= cr_interception(svm
);
2883 if (irqchip_in_kernel(svm
->vcpu
.kvm
)) {
2884 clr_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
2887 if (cr8_prev
<= kvm_get_cr8(&svm
->vcpu
))
2889 kvm_run
->exit_reason
= KVM_EXIT_SET_TPR
;
2893 static int svm_get_msr(struct kvm_vcpu
*vcpu
, unsigned ecx
, u64
*data
)
2895 struct vcpu_svm
*svm
= to_svm(vcpu
);
2898 case MSR_IA32_TSC
: {
2899 struct vmcb
*vmcb
= get_host_vmcb(svm
);
2901 *data
= vmcb
->control
.tsc_offset
+
2902 svm_scale_tsc(vcpu
, native_read_tsc());
2907 *data
= svm
->vmcb
->save
.star
;
2909 #ifdef CONFIG_X86_64
2911 *data
= svm
->vmcb
->save
.lstar
;
2914 *data
= svm
->vmcb
->save
.cstar
;
2916 case MSR_KERNEL_GS_BASE
:
2917 *data
= svm
->vmcb
->save
.kernel_gs_base
;
2919 case MSR_SYSCALL_MASK
:
2920 *data
= svm
->vmcb
->save
.sfmask
;
2923 case MSR_IA32_SYSENTER_CS
:
2924 *data
= svm
->vmcb
->save
.sysenter_cs
;
2926 case MSR_IA32_SYSENTER_EIP
:
2927 *data
= svm
->sysenter_eip
;
2929 case MSR_IA32_SYSENTER_ESP
:
2930 *data
= svm
->sysenter_esp
;
2933 * Nobody will change the following 5 values in the VMCB so we can
2934 * safely return them on rdmsr. They will always be 0 until LBRV is
2937 case MSR_IA32_DEBUGCTLMSR
:
2938 *data
= svm
->vmcb
->save
.dbgctl
;
2940 case MSR_IA32_LASTBRANCHFROMIP
:
2941 *data
= svm
->vmcb
->save
.br_from
;
2943 case MSR_IA32_LASTBRANCHTOIP
:
2944 *data
= svm
->vmcb
->save
.br_to
;
2946 case MSR_IA32_LASTINTFROMIP
:
2947 *data
= svm
->vmcb
->save
.last_excp_from
;
2949 case MSR_IA32_LASTINTTOIP
:
2950 *data
= svm
->vmcb
->save
.last_excp_to
;
2952 case MSR_VM_HSAVE_PA
:
2953 *data
= svm
->nested
.hsave_msr
;
2956 *data
= svm
->nested
.vm_cr_msr
;
2958 case MSR_IA32_UCODE_REV
:
2962 return kvm_get_msr_common(vcpu
, ecx
, data
);
2967 static int rdmsr_interception(struct vcpu_svm
*svm
)
2969 u32 ecx
= svm
->vcpu
.arch
.regs
[VCPU_REGS_RCX
];
2972 if (svm_get_msr(&svm
->vcpu
, ecx
, &data
)) {
2973 trace_kvm_msr_read_ex(ecx
);
2974 kvm_inject_gp(&svm
->vcpu
, 0);
2976 trace_kvm_msr_read(ecx
, data
);
2978 svm
->vcpu
.arch
.regs
[VCPU_REGS_RAX
] = data
& 0xffffffff;
2979 svm
->vcpu
.arch
.regs
[VCPU_REGS_RDX
] = data
>> 32;
2980 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
2981 skip_emulated_instruction(&svm
->vcpu
);
2986 static int svm_set_vm_cr(struct kvm_vcpu
*vcpu
, u64 data
)
2988 struct vcpu_svm
*svm
= to_svm(vcpu
);
2989 int svm_dis
, chg_mask
;
2991 if (data
& ~SVM_VM_CR_VALID_MASK
)
2994 chg_mask
= SVM_VM_CR_VALID_MASK
;
2996 if (svm
->nested
.vm_cr_msr
& SVM_VM_CR_SVM_DIS_MASK
)
2997 chg_mask
&= ~(SVM_VM_CR_SVM_LOCK_MASK
| SVM_VM_CR_SVM_DIS_MASK
);
2999 svm
->nested
.vm_cr_msr
&= ~chg_mask
;
3000 svm
->nested
.vm_cr_msr
|= (data
& chg_mask
);
3002 svm_dis
= svm
->nested
.vm_cr_msr
& SVM_VM_CR_SVM_DIS_MASK
;
3004 /* check for svm_disable while efer.svme is set */
3005 if (svm_dis
&& (vcpu
->arch
.efer
& EFER_SVME
))
3011 static int svm_set_msr(struct kvm_vcpu
*vcpu
, unsigned ecx
, u64 data
)
3013 struct vcpu_svm
*svm
= to_svm(vcpu
);
3017 kvm_write_tsc(vcpu
, data
);
3020 svm
->vmcb
->save
.star
= data
;
3022 #ifdef CONFIG_X86_64
3024 svm
->vmcb
->save
.lstar
= data
;
3027 svm
->vmcb
->save
.cstar
= data
;
3029 case MSR_KERNEL_GS_BASE
:
3030 svm
->vmcb
->save
.kernel_gs_base
= data
;
3032 case MSR_SYSCALL_MASK
:
3033 svm
->vmcb
->save
.sfmask
= data
;
3036 case MSR_IA32_SYSENTER_CS
:
3037 svm
->vmcb
->save
.sysenter_cs
= data
;
3039 case MSR_IA32_SYSENTER_EIP
:
3040 svm
->sysenter_eip
= data
;
3041 svm
->vmcb
->save
.sysenter_eip
= data
;
3043 case MSR_IA32_SYSENTER_ESP
:
3044 svm
->sysenter_esp
= data
;
3045 svm
->vmcb
->save
.sysenter_esp
= data
;
3047 case MSR_IA32_DEBUGCTLMSR
:
3048 if (!boot_cpu_has(X86_FEATURE_LBRV
)) {
3049 pr_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3053 if (data
& DEBUGCTL_RESERVED_BITS
)
3056 svm
->vmcb
->save
.dbgctl
= data
;
3057 mark_dirty(svm
->vmcb
, VMCB_LBR
);
3058 if (data
& (1ULL<<0))
3059 svm_enable_lbrv(svm
);
3061 svm_disable_lbrv(svm
);
3063 case MSR_VM_HSAVE_PA
:
3064 svm
->nested
.hsave_msr
= data
;
3067 return svm_set_vm_cr(vcpu
, data
);
3069 pr_unimpl(vcpu
, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx
, data
);
3072 return kvm_set_msr_common(vcpu
, ecx
, data
);
3077 static int wrmsr_interception(struct vcpu_svm
*svm
)
3079 u32 ecx
= svm
->vcpu
.arch
.regs
[VCPU_REGS_RCX
];
3080 u64 data
= (svm
->vcpu
.arch
.regs
[VCPU_REGS_RAX
] & -1u)
3081 | ((u64
)(svm
->vcpu
.arch
.regs
[VCPU_REGS_RDX
] & -1u) << 32);
3084 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
3085 if (svm_set_msr(&svm
->vcpu
, ecx
, data
)) {
3086 trace_kvm_msr_write_ex(ecx
, data
);
3087 kvm_inject_gp(&svm
->vcpu
, 0);
3089 trace_kvm_msr_write(ecx
, data
);
3090 skip_emulated_instruction(&svm
->vcpu
);
3095 static int msr_interception(struct vcpu_svm
*svm
)
3097 if (svm
->vmcb
->control
.exit_info_1
)
3098 return wrmsr_interception(svm
);
3100 return rdmsr_interception(svm
);
3103 static int interrupt_window_interception(struct vcpu_svm
*svm
)
3105 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
3107 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
3108 svm_clear_vintr(svm
);
3109 svm
->vmcb
->control
.int_ctl
&= ~V_IRQ_MASK
;
3110 mark_dirty(svm
->vmcb
, VMCB_INTR
);
3112 * If the user space waits to inject interrupts, exit as soon as
3115 if (!irqchip_in_kernel(svm
->vcpu
.kvm
) &&
3116 kvm_run
->request_interrupt_window
&&
3117 !kvm_cpu_has_interrupt(&svm
->vcpu
)) {
3118 ++svm
->vcpu
.stat
.irq_window_exits
;
3119 kvm_run
->exit_reason
= KVM_EXIT_IRQ_WINDOW_OPEN
;
3126 static int pause_interception(struct vcpu_svm
*svm
)
3128 kvm_vcpu_on_spin(&(svm
->vcpu
));
3132 static int (*svm_exit_handlers
[])(struct vcpu_svm
*svm
) = {
3133 [SVM_EXIT_READ_CR0
] = cr_interception
,
3134 [SVM_EXIT_READ_CR3
] = cr_interception
,
3135 [SVM_EXIT_READ_CR4
] = cr_interception
,
3136 [SVM_EXIT_READ_CR8
] = cr_interception
,
3137 [SVM_EXIT_CR0_SEL_WRITE
] = emulate_on_interception
,
3138 [SVM_EXIT_WRITE_CR0
] = cr_interception
,
3139 [SVM_EXIT_WRITE_CR3
] = cr_interception
,
3140 [SVM_EXIT_WRITE_CR4
] = cr_interception
,
3141 [SVM_EXIT_WRITE_CR8
] = cr8_write_interception
,
3142 [SVM_EXIT_READ_DR0
] = dr_interception
,
3143 [SVM_EXIT_READ_DR1
] = dr_interception
,
3144 [SVM_EXIT_READ_DR2
] = dr_interception
,
3145 [SVM_EXIT_READ_DR3
] = dr_interception
,
3146 [SVM_EXIT_READ_DR4
] = dr_interception
,
3147 [SVM_EXIT_READ_DR5
] = dr_interception
,
3148 [SVM_EXIT_READ_DR6
] = dr_interception
,
3149 [SVM_EXIT_READ_DR7
] = dr_interception
,
3150 [SVM_EXIT_WRITE_DR0
] = dr_interception
,
3151 [SVM_EXIT_WRITE_DR1
] = dr_interception
,
3152 [SVM_EXIT_WRITE_DR2
] = dr_interception
,
3153 [SVM_EXIT_WRITE_DR3
] = dr_interception
,
3154 [SVM_EXIT_WRITE_DR4
] = dr_interception
,
3155 [SVM_EXIT_WRITE_DR5
] = dr_interception
,
3156 [SVM_EXIT_WRITE_DR6
] = dr_interception
,
3157 [SVM_EXIT_WRITE_DR7
] = dr_interception
,
3158 [SVM_EXIT_EXCP_BASE
+ DB_VECTOR
] = db_interception
,
3159 [SVM_EXIT_EXCP_BASE
+ BP_VECTOR
] = bp_interception
,
3160 [SVM_EXIT_EXCP_BASE
+ UD_VECTOR
] = ud_interception
,
3161 [SVM_EXIT_EXCP_BASE
+ PF_VECTOR
] = pf_interception
,
3162 [SVM_EXIT_EXCP_BASE
+ NM_VECTOR
] = nm_interception
,
3163 [SVM_EXIT_EXCP_BASE
+ MC_VECTOR
] = mc_interception
,
3164 [SVM_EXIT_INTR
] = intr_interception
,
3165 [SVM_EXIT_NMI
] = nmi_interception
,
3166 [SVM_EXIT_SMI
] = nop_on_interception
,
3167 [SVM_EXIT_INIT
] = nop_on_interception
,
3168 [SVM_EXIT_VINTR
] = interrupt_window_interception
,
3169 [SVM_EXIT_CPUID
] = cpuid_interception
,
3170 [SVM_EXIT_IRET
] = iret_interception
,
3171 [SVM_EXIT_INVD
] = emulate_on_interception
,
3172 [SVM_EXIT_PAUSE
] = pause_interception
,
3173 [SVM_EXIT_HLT
] = halt_interception
,
3174 [SVM_EXIT_INVLPG
] = invlpg_interception
,
3175 [SVM_EXIT_INVLPGA
] = invlpga_interception
,
3176 [SVM_EXIT_IOIO
] = io_interception
,
3177 [SVM_EXIT_MSR
] = msr_interception
,
3178 [SVM_EXIT_TASK_SWITCH
] = task_switch_interception
,
3179 [SVM_EXIT_SHUTDOWN
] = shutdown_interception
,
3180 [SVM_EXIT_VMRUN
] = vmrun_interception
,
3181 [SVM_EXIT_VMMCALL
] = vmmcall_interception
,
3182 [SVM_EXIT_VMLOAD
] = vmload_interception
,
3183 [SVM_EXIT_VMSAVE
] = vmsave_interception
,
3184 [SVM_EXIT_STGI
] = stgi_interception
,
3185 [SVM_EXIT_CLGI
] = clgi_interception
,
3186 [SVM_EXIT_SKINIT
] = skinit_interception
,
3187 [SVM_EXIT_WBINVD
] = emulate_on_interception
,
3188 [SVM_EXIT_MONITOR
] = invalid_op_interception
,
3189 [SVM_EXIT_MWAIT
] = invalid_op_interception
,
3190 [SVM_EXIT_XSETBV
] = xsetbv_interception
,
3191 [SVM_EXIT_NPF
] = pf_interception
,
3194 static void dump_vmcb(struct kvm_vcpu
*vcpu
)
3196 struct vcpu_svm
*svm
= to_svm(vcpu
);
3197 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
3198 struct vmcb_save_area
*save
= &svm
->vmcb
->save
;
3200 pr_err("VMCB Control Area:\n");
3201 pr_err("%-20s%04x\n", "cr_read:", control
->intercept_cr
& 0xffff);
3202 pr_err("%-20s%04x\n", "cr_write:", control
->intercept_cr
>> 16);
3203 pr_err("%-20s%04x\n", "dr_read:", control
->intercept_dr
& 0xffff);
3204 pr_err("%-20s%04x\n", "dr_write:", control
->intercept_dr
>> 16);
3205 pr_err("%-20s%08x\n", "exceptions:", control
->intercept_exceptions
);
3206 pr_err("%-20s%016llx\n", "intercepts:", control
->intercept
);
3207 pr_err("%-20s%d\n", "pause filter count:", control
->pause_filter_count
);
3208 pr_err("%-20s%016llx\n", "iopm_base_pa:", control
->iopm_base_pa
);
3209 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control
->msrpm_base_pa
);
3210 pr_err("%-20s%016llx\n", "tsc_offset:", control
->tsc_offset
);
3211 pr_err("%-20s%d\n", "asid:", control
->asid
);
3212 pr_err("%-20s%d\n", "tlb_ctl:", control
->tlb_ctl
);
3213 pr_err("%-20s%08x\n", "int_ctl:", control
->int_ctl
);
3214 pr_err("%-20s%08x\n", "int_vector:", control
->int_vector
);
3215 pr_err("%-20s%08x\n", "int_state:", control
->int_state
);
3216 pr_err("%-20s%08x\n", "exit_code:", control
->exit_code
);
3217 pr_err("%-20s%016llx\n", "exit_info1:", control
->exit_info_1
);
3218 pr_err("%-20s%016llx\n", "exit_info2:", control
->exit_info_2
);
3219 pr_err("%-20s%08x\n", "exit_int_info:", control
->exit_int_info
);
3220 pr_err("%-20s%08x\n", "exit_int_info_err:", control
->exit_int_info_err
);
3221 pr_err("%-20s%lld\n", "nested_ctl:", control
->nested_ctl
);
3222 pr_err("%-20s%016llx\n", "nested_cr3:", control
->nested_cr3
);
3223 pr_err("%-20s%08x\n", "event_inj:", control
->event_inj
);
3224 pr_err("%-20s%08x\n", "event_inj_err:", control
->event_inj_err
);
3225 pr_err("%-20s%lld\n", "lbr_ctl:", control
->lbr_ctl
);
3226 pr_err("%-20s%016llx\n", "next_rip:", control
->next_rip
);
3227 pr_err("VMCB State Save Area:\n");
3228 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3230 save
->es
.selector
, save
->es
.attrib
,
3231 save
->es
.limit
, save
->es
.base
);
3232 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3234 save
->cs
.selector
, save
->cs
.attrib
,
3235 save
->cs
.limit
, save
->cs
.base
);
3236 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3238 save
->ss
.selector
, save
->ss
.attrib
,
3239 save
->ss
.limit
, save
->ss
.base
);
3240 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3242 save
->ds
.selector
, save
->ds
.attrib
,
3243 save
->ds
.limit
, save
->ds
.base
);
3244 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3246 save
->fs
.selector
, save
->fs
.attrib
,
3247 save
->fs
.limit
, save
->fs
.base
);
3248 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3250 save
->gs
.selector
, save
->gs
.attrib
,
3251 save
->gs
.limit
, save
->gs
.base
);
3252 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3254 save
->gdtr
.selector
, save
->gdtr
.attrib
,
3255 save
->gdtr
.limit
, save
->gdtr
.base
);
3256 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3258 save
->ldtr
.selector
, save
->ldtr
.attrib
,
3259 save
->ldtr
.limit
, save
->ldtr
.base
);
3260 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3262 save
->idtr
.selector
, save
->idtr
.attrib
,
3263 save
->idtr
.limit
, save
->idtr
.base
);
3264 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3266 save
->tr
.selector
, save
->tr
.attrib
,
3267 save
->tr
.limit
, save
->tr
.base
);
3268 pr_err("cpl: %d efer: %016llx\n",
3269 save
->cpl
, save
->efer
);
3270 pr_err("%-15s %016llx %-13s %016llx\n",
3271 "cr0:", save
->cr0
, "cr2:", save
->cr2
);
3272 pr_err("%-15s %016llx %-13s %016llx\n",
3273 "cr3:", save
->cr3
, "cr4:", save
->cr4
);
3274 pr_err("%-15s %016llx %-13s %016llx\n",
3275 "dr6:", save
->dr6
, "dr7:", save
->dr7
);
3276 pr_err("%-15s %016llx %-13s %016llx\n",
3277 "rip:", save
->rip
, "rflags:", save
->rflags
);
3278 pr_err("%-15s %016llx %-13s %016llx\n",
3279 "rsp:", save
->rsp
, "rax:", save
->rax
);
3280 pr_err("%-15s %016llx %-13s %016llx\n",
3281 "star:", save
->star
, "lstar:", save
->lstar
);
3282 pr_err("%-15s %016llx %-13s %016llx\n",
3283 "cstar:", save
->cstar
, "sfmask:", save
->sfmask
);
3284 pr_err("%-15s %016llx %-13s %016llx\n",
3285 "kernel_gs_base:", save
->kernel_gs_base
,
3286 "sysenter_cs:", save
->sysenter_cs
);
3287 pr_err("%-15s %016llx %-13s %016llx\n",
3288 "sysenter_esp:", save
->sysenter_esp
,
3289 "sysenter_eip:", save
->sysenter_eip
);
3290 pr_err("%-15s %016llx %-13s %016llx\n",
3291 "gpat:", save
->g_pat
, "dbgctl:", save
->dbgctl
);
3292 pr_err("%-15s %016llx %-13s %016llx\n",
3293 "br_from:", save
->br_from
, "br_to:", save
->br_to
);
3294 pr_err("%-15s %016llx %-13s %016llx\n",
3295 "excp_from:", save
->last_excp_from
,
3296 "excp_to:", save
->last_excp_to
);
3299 static void svm_get_exit_info(struct kvm_vcpu
*vcpu
, u64
*info1
, u64
*info2
)
3301 struct vmcb_control_area
*control
= &to_svm(vcpu
)->vmcb
->control
;
3303 *info1
= control
->exit_info_1
;
3304 *info2
= control
->exit_info_2
;
3307 static int handle_exit(struct kvm_vcpu
*vcpu
)
3309 struct vcpu_svm
*svm
= to_svm(vcpu
);
3310 struct kvm_run
*kvm_run
= vcpu
->run
;
3311 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
3313 trace_kvm_exit(exit_code
, vcpu
, KVM_ISA_SVM
);
3315 if (!is_cr_intercept(svm
, INTERCEPT_CR0_WRITE
))
3316 vcpu
->arch
.cr0
= svm
->vmcb
->save
.cr0
;
3318 vcpu
->arch
.cr3
= svm
->vmcb
->save
.cr3
;
3320 if (unlikely(svm
->nested
.exit_required
)) {
3321 nested_svm_vmexit(svm
);
3322 svm
->nested
.exit_required
= false;
3327 if (is_guest_mode(vcpu
)) {
3330 trace_kvm_nested_vmexit(svm
->vmcb
->save
.rip
, exit_code
,
3331 svm
->vmcb
->control
.exit_info_1
,
3332 svm
->vmcb
->control
.exit_info_2
,
3333 svm
->vmcb
->control
.exit_int_info
,
3334 svm
->vmcb
->control
.exit_int_info_err
);
3336 vmexit
= nested_svm_exit_special(svm
);
3338 if (vmexit
== NESTED_EXIT_CONTINUE
)
3339 vmexit
= nested_svm_exit_handled(svm
);
3341 if (vmexit
== NESTED_EXIT_DONE
)
3345 svm_complete_interrupts(svm
);
3347 if (svm
->vmcb
->control
.exit_code
== SVM_EXIT_ERR
) {
3348 kvm_run
->exit_reason
= KVM_EXIT_FAIL_ENTRY
;
3349 kvm_run
->fail_entry
.hardware_entry_failure_reason
3350 = svm
->vmcb
->control
.exit_code
;
3351 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
3356 if (is_external_interrupt(svm
->vmcb
->control
.exit_int_info
) &&
3357 exit_code
!= SVM_EXIT_EXCP_BASE
+ PF_VECTOR
&&
3358 exit_code
!= SVM_EXIT_NPF
&& exit_code
!= SVM_EXIT_TASK_SWITCH
&&
3359 exit_code
!= SVM_EXIT_INTR
&& exit_code
!= SVM_EXIT_NMI
)
3360 printk(KERN_ERR
"%s: unexpected exit_ini_info 0x%x "
3362 __func__
, svm
->vmcb
->control
.exit_int_info
,
3365 if (exit_code
>= ARRAY_SIZE(svm_exit_handlers
)
3366 || !svm_exit_handlers
[exit_code
]) {
3367 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
3368 kvm_run
->hw
.hardware_exit_reason
= exit_code
;
3372 return svm_exit_handlers
[exit_code
](svm
);
3375 static void reload_tss(struct kvm_vcpu
*vcpu
)
3377 int cpu
= raw_smp_processor_id();
3379 struct svm_cpu_data
*sd
= per_cpu(svm_data
, cpu
);
3380 sd
->tss_desc
->type
= 9; /* available 32/64-bit TSS */
3384 static void pre_svm_run(struct vcpu_svm
*svm
)
3386 int cpu
= raw_smp_processor_id();
3388 struct svm_cpu_data
*sd
= per_cpu(svm_data
, cpu
);
3390 /* FIXME: handle wraparound of asid_generation */
3391 if (svm
->asid_generation
!= sd
->asid_generation
)
3395 static void svm_inject_nmi(struct kvm_vcpu
*vcpu
)
3397 struct vcpu_svm
*svm
= to_svm(vcpu
);
3399 svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_NMI
;
3400 vcpu
->arch
.hflags
|= HF_NMI_MASK
;
3401 set_intercept(svm
, INTERCEPT_IRET
);
3402 ++vcpu
->stat
.nmi_injections
;
3405 static inline void svm_inject_irq(struct vcpu_svm
*svm
, int irq
)
3407 struct vmcb_control_area
*control
;
3409 control
= &svm
->vmcb
->control
;
3410 control
->int_vector
= irq
;
3411 control
->int_ctl
&= ~V_INTR_PRIO_MASK
;
3412 control
->int_ctl
|= V_IRQ_MASK
|
3413 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT
);
3414 mark_dirty(svm
->vmcb
, VMCB_INTR
);
3417 static void svm_set_irq(struct kvm_vcpu
*vcpu
)
3419 struct vcpu_svm
*svm
= to_svm(vcpu
);
3421 BUG_ON(!(gif_set(svm
)));
3423 trace_kvm_inj_virq(vcpu
->arch
.interrupt
.nr
);
3424 ++vcpu
->stat
.irq_injections
;
3426 svm
->vmcb
->control
.event_inj
= vcpu
->arch
.interrupt
.nr
|
3427 SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
;
3430 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
, int tpr
, int irr
)
3432 struct vcpu_svm
*svm
= to_svm(vcpu
);
3434 if (is_guest_mode(vcpu
) && (vcpu
->arch
.hflags
& HF_VINTR_MASK
))
3441 set_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
3444 static int svm_nmi_allowed(struct kvm_vcpu
*vcpu
)
3446 struct vcpu_svm
*svm
= to_svm(vcpu
);
3447 struct vmcb
*vmcb
= svm
->vmcb
;
3449 ret
= !(vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
) &&
3450 !(svm
->vcpu
.arch
.hflags
& HF_NMI_MASK
);
3451 ret
= ret
&& gif_set(svm
) && nested_svm_nmi(svm
);
3456 static bool svm_get_nmi_mask(struct kvm_vcpu
*vcpu
)
3458 struct vcpu_svm
*svm
= to_svm(vcpu
);
3460 return !!(svm
->vcpu
.arch
.hflags
& HF_NMI_MASK
);
3463 static void svm_set_nmi_mask(struct kvm_vcpu
*vcpu
, bool masked
)
3465 struct vcpu_svm
*svm
= to_svm(vcpu
);
3468 svm
->vcpu
.arch
.hflags
|= HF_NMI_MASK
;
3469 set_intercept(svm
, INTERCEPT_IRET
);
3471 svm
->vcpu
.arch
.hflags
&= ~HF_NMI_MASK
;
3472 clr_intercept(svm
, INTERCEPT_IRET
);
3476 static int svm_interrupt_allowed(struct kvm_vcpu
*vcpu
)
3478 struct vcpu_svm
*svm
= to_svm(vcpu
);
3479 struct vmcb
*vmcb
= svm
->vmcb
;
3482 if (!gif_set(svm
) ||
3483 (vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
))
3486 ret
= !!(kvm_get_rflags(vcpu
) & X86_EFLAGS_IF
);
3488 if (is_guest_mode(vcpu
))
3489 return ret
&& !(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
);
3494 static void enable_irq_window(struct kvm_vcpu
*vcpu
)
3496 struct vcpu_svm
*svm
= to_svm(vcpu
);
3499 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3500 * 1, because that's a separate STGI/VMRUN intercept. The next time we
3501 * get that intercept, this function will be called again though and
3502 * we'll get the vintr intercept.
3504 if (gif_set(svm
) && nested_svm_intr(svm
)) {
3506 svm_inject_irq(svm
, 0x0);
3510 static void enable_nmi_window(struct kvm_vcpu
*vcpu
)
3512 struct vcpu_svm
*svm
= to_svm(vcpu
);
3514 if ((svm
->vcpu
.arch
.hflags
& (HF_NMI_MASK
| HF_IRET_MASK
))
3516 return; /* IRET will cause a vm exit */
3519 * Something prevents NMI from been injected. Single step over possible
3520 * problem (IRET or exception injection or interrupt shadow)
3522 svm
->nmi_singlestep
= true;
3523 svm
->vmcb
->save
.rflags
|= (X86_EFLAGS_TF
| X86_EFLAGS_RF
);
3524 update_db_intercept(vcpu
);
3527 static int svm_set_tss_addr(struct kvm
*kvm
, unsigned int addr
)
3532 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
)
3534 struct vcpu_svm
*svm
= to_svm(vcpu
);
3536 if (static_cpu_has(X86_FEATURE_FLUSHBYASID
))
3537 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ASID
;
3539 svm
->asid_generation
--;
3542 static void svm_prepare_guest_switch(struct kvm_vcpu
*vcpu
)
3546 static inline void sync_cr8_to_lapic(struct kvm_vcpu
*vcpu
)
3548 struct vcpu_svm
*svm
= to_svm(vcpu
);
3550 if (is_guest_mode(vcpu
) && (vcpu
->arch
.hflags
& HF_VINTR_MASK
))
3553 if (!is_cr_intercept(svm
, INTERCEPT_CR8_WRITE
)) {
3554 int cr8
= svm
->vmcb
->control
.int_ctl
& V_TPR_MASK
;
3555 kvm_set_cr8(vcpu
, cr8
);
3559 static inline void sync_lapic_to_cr8(struct kvm_vcpu
*vcpu
)
3561 struct vcpu_svm
*svm
= to_svm(vcpu
);
3564 if (is_guest_mode(vcpu
) && (vcpu
->arch
.hflags
& HF_VINTR_MASK
))
3567 cr8
= kvm_get_cr8(vcpu
);
3568 svm
->vmcb
->control
.int_ctl
&= ~V_TPR_MASK
;
3569 svm
->vmcb
->control
.int_ctl
|= cr8
& V_TPR_MASK
;
3572 static void svm_complete_interrupts(struct vcpu_svm
*svm
)
3576 u32 exitintinfo
= svm
->vmcb
->control
.exit_int_info
;
3577 unsigned int3_injected
= svm
->int3_injected
;
3579 svm
->int3_injected
= 0;
3582 * If we've made progress since setting HF_IRET_MASK, we've
3583 * executed an IRET and can allow NMI injection.
3585 if ((svm
->vcpu
.arch
.hflags
& HF_IRET_MASK
)
3586 && kvm_rip_read(&svm
->vcpu
) != svm
->nmi_iret_rip
) {
3587 svm
->vcpu
.arch
.hflags
&= ~(HF_NMI_MASK
| HF_IRET_MASK
);
3588 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
3591 svm
->vcpu
.arch
.nmi_injected
= false;
3592 kvm_clear_exception_queue(&svm
->vcpu
);
3593 kvm_clear_interrupt_queue(&svm
->vcpu
);
3595 if (!(exitintinfo
& SVM_EXITINTINFO_VALID
))
3598 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
3600 vector
= exitintinfo
& SVM_EXITINTINFO_VEC_MASK
;
3601 type
= exitintinfo
& SVM_EXITINTINFO_TYPE_MASK
;
3604 case SVM_EXITINTINFO_TYPE_NMI
:
3605 svm
->vcpu
.arch
.nmi_injected
= true;
3607 case SVM_EXITINTINFO_TYPE_EXEPT
:
3609 * In case of software exceptions, do not reinject the vector,
3610 * but re-execute the instruction instead. Rewind RIP first
3611 * if we emulated INT3 before.
3613 if (kvm_exception_is_soft(vector
)) {
3614 if (vector
== BP_VECTOR
&& int3_injected
&&
3615 kvm_is_linear_rip(&svm
->vcpu
, svm
->int3_rip
))
3616 kvm_rip_write(&svm
->vcpu
,
3617 kvm_rip_read(&svm
->vcpu
) -
3621 if (exitintinfo
& SVM_EXITINTINFO_VALID_ERR
) {
3622 u32 err
= svm
->vmcb
->control
.exit_int_info_err
;
3623 kvm_requeue_exception_e(&svm
->vcpu
, vector
, err
);
3626 kvm_requeue_exception(&svm
->vcpu
, vector
);
3628 case SVM_EXITINTINFO_TYPE_INTR
:
3629 kvm_queue_interrupt(&svm
->vcpu
, vector
, false);
3636 static void svm_cancel_injection(struct kvm_vcpu
*vcpu
)
3638 struct vcpu_svm
*svm
= to_svm(vcpu
);
3639 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
3641 control
->exit_int_info
= control
->event_inj
;
3642 control
->exit_int_info_err
= control
->event_inj_err
;
3643 control
->event_inj
= 0;
3644 svm_complete_interrupts(svm
);
3647 #ifdef CONFIG_X86_64
3653 static void svm_vcpu_run(struct kvm_vcpu
*vcpu
)
3655 struct vcpu_svm
*svm
= to_svm(vcpu
);
3657 svm
->vmcb
->save
.rax
= vcpu
->arch
.regs
[VCPU_REGS_RAX
];
3658 svm
->vmcb
->save
.rsp
= vcpu
->arch
.regs
[VCPU_REGS_RSP
];
3659 svm
->vmcb
->save
.rip
= vcpu
->arch
.regs
[VCPU_REGS_RIP
];
3662 * A vmexit emulation is required before the vcpu can be executed
3665 if (unlikely(svm
->nested
.exit_required
))
3670 sync_lapic_to_cr8(vcpu
);
3672 svm
->vmcb
->save
.cr2
= vcpu
->arch
.cr2
;
3679 "push %%"R
"bp; \n\t"
3680 "mov %c[rbx](%[svm]), %%"R
"bx \n\t"
3681 "mov %c[rcx](%[svm]), %%"R
"cx \n\t"
3682 "mov %c[rdx](%[svm]), %%"R
"dx \n\t"
3683 "mov %c[rsi](%[svm]), %%"R
"si \n\t"
3684 "mov %c[rdi](%[svm]), %%"R
"di \n\t"
3685 "mov %c[rbp](%[svm]), %%"R
"bp \n\t"
3686 #ifdef CONFIG_X86_64
3687 "mov %c[r8](%[svm]), %%r8 \n\t"
3688 "mov %c[r9](%[svm]), %%r9 \n\t"
3689 "mov %c[r10](%[svm]), %%r10 \n\t"
3690 "mov %c[r11](%[svm]), %%r11 \n\t"
3691 "mov %c[r12](%[svm]), %%r12 \n\t"
3692 "mov %c[r13](%[svm]), %%r13 \n\t"
3693 "mov %c[r14](%[svm]), %%r14 \n\t"
3694 "mov %c[r15](%[svm]), %%r15 \n\t"
3697 /* Enter guest mode */
3699 "mov %c[vmcb](%[svm]), %%"R
"ax \n\t"
3700 __ex(SVM_VMLOAD
) "\n\t"
3701 __ex(SVM_VMRUN
) "\n\t"
3702 __ex(SVM_VMSAVE
) "\n\t"
3705 /* Save guest registers, load host registers */
3706 "mov %%"R
"bx, %c[rbx](%[svm]) \n\t"
3707 "mov %%"R
"cx, %c[rcx](%[svm]) \n\t"
3708 "mov %%"R
"dx, %c[rdx](%[svm]) \n\t"
3709 "mov %%"R
"si, %c[rsi](%[svm]) \n\t"
3710 "mov %%"R
"di, %c[rdi](%[svm]) \n\t"
3711 "mov %%"R
"bp, %c[rbp](%[svm]) \n\t"
3712 #ifdef CONFIG_X86_64
3713 "mov %%r8, %c[r8](%[svm]) \n\t"
3714 "mov %%r9, %c[r9](%[svm]) \n\t"
3715 "mov %%r10, %c[r10](%[svm]) \n\t"
3716 "mov %%r11, %c[r11](%[svm]) \n\t"
3717 "mov %%r12, %c[r12](%[svm]) \n\t"
3718 "mov %%r13, %c[r13](%[svm]) \n\t"
3719 "mov %%r14, %c[r14](%[svm]) \n\t"
3720 "mov %%r15, %c[r15](%[svm]) \n\t"
3725 [vmcb
]"i"(offsetof(struct vcpu_svm
, vmcb_pa
)),
3726 [rbx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RBX
])),
3727 [rcx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RCX
])),
3728 [rdx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RDX
])),
3729 [rsi
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RSI
])),
3730 [rdi
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RDI
])),
3731 [rbp
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RBP
]))
3732 #ifdef CONFIG_X86_64
3733 , [r8
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R8
])),
3734 [r9
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R9
])),
3735 [r10
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R10
])),
3736 [r11
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R11
])),
3737 [r12
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R12
])),
3738 [r13
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R13
])),
3739 [r14
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R14
])),
3740 [r15
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R15
]))
3743 , R
"bx", R
"cx", R
"dx", R
"si", R
"di"
3744 #ifdef CONFIG_X86_64
3745 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
3749 #ifdef CONFIG_X86_64
3750 wrmsrl(MSR_GS_BASE
, svm
->host
.gs_base
);
3752 loadsegment(fs
, svm
->host
.fs
);
3753 #ifndef CONFIG_X86_32_LAZY_GS
3754 loadsegment(gs
, svm
->host
.gs
);
3760 local_irq_disable();
3762 vcpu
->arch
.cr2
= svm
->vmcb
->save
.cr2
;
3763 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = svm
->vmcb
->save
.rax
;
3764 vcpu
->arch
.regs
[VCPU_REGS_RSP
] = svm
->vmcb
->save
.rsp
;
3765 vcpu
->arch
.regs
[VCPU_REGS_RIP
] = svm
->vmcb
->save
.rip
;
3767 if (unlikely(svm
->vmcb
->control
.exit_code
== SVM_EXIT_NMI
))
3768 kvm_before_handle_nmi(&svm
->vcpu
);
3772 /* Any pending NMI will happen here */
3774 if (unlikely(svm
->vmcb
->control
.exit_code
== SVM_EXIT_NMI
))
3775 kvm_after_handle_nmi(&svm
->vcpu
);
3777 sync_cr8_to_lapic(vcpu
);
3781 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_DO_NOTHING
;
3783 /* if exit due to PF check for async PF */
3784 if (svm
->vmcb
->control
.exit_code
== SVM_EXIT_EXCP_BASE
+ PF_VECTOR
)
3785 svm
->apf_reason
= kvm_read_and_reset_pf_reason();
3788 vcpu
->arch
.regs_avail
&= ~(1 << VCPU_EXREG_PDPTR
);
3789 vcpu
->arch
.regs_dirty
&= ~(1 << VCPU_EXREG_PDPTR
);
3793 * We need to handle MC intercepts here before the vcpu has a chance to
3794 * change the physical cpu
3796 if (unlikely(svm
->vmcb
->control
.exit_code
==
3797 SVM_EXIT_EXCP_BASE
+ MC_VECTOR
))
3798 svm_handle_mce(svm
);
3800 mark_all_clean(svm
->vmcb
);
3805 static void svm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
3807 struct vcpu_svm
*svm
= to_svm(vcpu
);
3809 svm
->vmcb
->save
.cr3
= root
;
3810 mark_dirty(svm
->vmcb
, VMCB_CR
);
3811 svm_flush_tlb(vcpu
);
3814 static void set_tdp_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
3816 struct vcpu_svm
*svm
= to_svm(vcpu
);
3818 svm
->vmcb
->control
.nested_cr3
= root
;
3819 mark_dirty(svm
->vmcb
, VMCB_NPT
);
3821 /* Also sync guest cr3 here in case we live migrate */
3822 svm
->vmcb
->save
.cr3
= kvm_read_cr3(vcpu
);
3823 mark_dirty(svm
->vmcb
, VMCB_CR
);
3825 svm_flush_tlb(vcpu
);
3828 static int is_disabled(void)
3832 rdmsrl(MSR_VM_CR
, vm_cr
);
3833 if (vm_cr
& (1 << SVM_VM_CR_SVM_DISABLE
))
3840 svm_patch_hypercall(struct kvm_vcpu
*vcpu
, unsigned char *hypercall
)
3843 * Patch in the VMMCALL instruction:
3845 hypercall
[0] = 0x0f;
3846 hypercall
[1] = 0x01;
3847 hypercall
[2] = 0xd9;
3850 static void svm_check_processor_compat(void *rtn
)
3855 static bool svm_cpu_has_accelerated_tpr(void)
3860 static u64
svm_get_mt_mask(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool is_mmio
)
3865 static void svm_cpuid_update(struct kvm_vcpu
*vcpu
)
3869 static void svm_set_supported_cpuid(u32 func
, struct kvm_cpuid_entry2
*entry
)
3874 entry
->ecx
|= (1 << 2); /* Set SVM bit */
3877 entry
->eax
= 1; /* SVM revision 1 */
3878 entry
->ebx
= 8; /* Lets support 8 ASIDs in case we add proper
3879 ASID emulation to nested SVM */
3880 entry
->ecx
= 0; /* Reserved */
3881 entry
->edx
= 0; /* Per default do not support any
3882 additional features */
3884 /* Support next_rip if host supports it */
3885 if (boot_cpu_has(X86_FEATURE_NRIPS
))
3886 entry
->edx
|= SVM_FEATURE_NRIP
;
3888 /* Support NPT for the guest if enabled */
3890 entry
->edx
|= SVM_FEATURE_NPT
;
3896 static const struct trace_print_flags svm_exit_reasons_str
[] = {
3897 { SVM_EXIT_READ_CR0
, "read_cr0" },
3898 { SVM_EXIT_READ_CR3
, "read_cr3" },
3899 { SVM_EXIT_READ_CR4
, "read_cr4" },
3900 { SVM_EXIT_READ_CR8
, "read_cr8" },
3901 { SVM_EXIT_WRITE_CR0
, "write_cr0" },
3902 { SVM_EXIT_WRITE_CR3
, "write_cr3" },
3903 { SVM_EXIT_WRITE_CR4
, "write_cr4" },
3904 { SVM_EXIT_WRITE_CR8
, "write_cr8" },
3905 { SVM_EXIT_READ_DR0
, "read_dr0" },
3906 { SVM_EXIT_READ_DR1
, "read_dr1" },
3907 { SVM_EXIT_READ_DR2
, "read_dr2" },
3908 { SVM_EXIT_READ_DR3
, "read_dr3" },
3909 { SVM_EXIT_WRITE_DR0
, "write_dr0" },
3910 { SVM_EXIT_WRITE_DR1
, "write_dr1" },
3911 { SVM_EXIT_WRITE_DR2
, "write_dr2" },
3912 { SVM_EXIT_WRITE_DR3
, "write_dr3" },
3913 { SVM_EXIT_WRITE_DR5
, "write_dr5" },
3914 { SVM_EXIT_WRITE_DR7
, "write_dr7" },
3915 { SVM_EXIT_EXCP_BASE
+ DB_VECTOR
, "DB excp" },
3916 { SVM_EXIT_EXCP_BASE
+ BP_VECTOR
, "BP excp" },
3917 { SVM_EXIT_EXCP_BASE
+ UD_VECTOR
, "UD excp" },
3918 { SVM_EXIT_EXCP_BASE
+ PF_VECTOR
, "PF excp" },
3919 { SVM_EXIT_EXCP_BASE
+ NM_VECTOR
, "NM excp" },
3920 { SVM_EXIT_EXCP_BASE
+ MC_VECTOR
, "MC excp" },
3921 { SVM_EXIT_INTR
, "interrupt" },
3922 { SVM_EXIT_NMI
, "nmi" },
3923 { SVM_EXIT_SMI
, "smi" },
3924 { SVM_EXIT_INIT
, "init" },
3925 { SVM_EXIT_VINTR
, "vintr" },
3926 { SVM_EXIT_CPUID
, "cpuid" },
3927 { SVM_EXIT_INVD
, "invd" },
3928 { SVM_EXIT_HLT
, "hlt" },
3929 { SVM_EXIT_INVLPG
, "invlpg" },
3930 { SVM_EXIT_INVLPGA
, "invlpga" },
3931 { SVM_EXIT_IOIO
, "io" },
3932 { SVM_EXIT_MSR
, "msr" },
3933 { SVM_EXIT_TASK_SWITCH
, "task_switch" },
3934 { SVM_EXIT_SHUTDOWN
, "shutdown" },
3935 { SVM_EXIT_VMRUN
, "vmrun" },
3936 { SVM_EXIT_VMMCALL
, "hypercall" },
3937 { SVM_EXIT_VMLOAD
, "vmload" },
3938 { SVM_EXIT_VMSAVE
, "vmsave" },
3939 { SVM_EXIT_STGI
, "stgi" },
3940 { SVM_EXIT_CLGI
, "clgi" },
3941 { SVM_EXIT_SKINIT
, "skinit" },
3942 { SVM_EXIT_WBINVD
, "wbinvd" },
3943 { SVM_EXIT_MONITOR
, "monitor" },
3944 { SVM_EXIT_MWAIT
, "mwait" },
3945 { SVM_EXIT_XSETBV
, "xsetbv" },
3946 { SVM_EXIT_NPF
, "npf" },
3950 static int svm_get_lpage_level(void)
3952 return PT_PDPE_LEVEL
;
3955 static bool svm_rdtscp_supported(void)
3960 static bool svm_has_wbinvd_exit(void)
3965 static void svm_fpu_deactivate(struct kvm_vcpu
*vcpu
)
3967 struct vcpu_svm
*svm
= to_svm(vcpu
);
3969 set_exception_intercept(svm
, NM_VECTOR
);
3970 update_cr0_intercept(svm
);
3973 #define PRE_EX(exit) { .exit_code = (exit), \
3974 .stage = X86_ICPT_PRE_EXCEPT, }
3975 #define POST_EX(exit) { .exit_code = (exit), \
3976 .stage = X86_ICPT_POST_EXCEPT, }
3977 #define POST_MEM(exit) { .exit_code = (exit), \
3978 .stage = X86_ICPT_POST_MEMACCESS, }
3980 static struct __x86_intercept
{
3982 enum x86_intercept_stage stage
;
3983 } x86_intercept_map
[] = {
3984 [x86_intercept_cr_read
] = POST_EX(SVM_EXIT_READ_CR0
),
3985 [x86_intercept_cr_write
] = POST_EX(SVM_EXIT_WRITE_CR0
),
3986 [x86_intercept_clts
] = POST_EX(SVM_EXIT_WRITE_CR0
),
3987 [x86_intercept_lmsw
] = POST_EX(SVM_EXIT_WRITE_CR0
),
3988 [x86_intercept_smsw
] = POST_EX(SVM_EXIT_READ_CR0
),
3989 [x86_intercept_dr_read
] = POST_EX(SVM_EXIT_READ_DR0
),
3990 [x86_intercept_dr_write
] = POST_EX(SVM_EXIT_WRITE_DR0
),
3991 [x86_intercept_sldt
] = POST_EX(SVM_EXIT_LDTR_READ
),
3992 [x86_intercept_str
] = POST_EX(SVM_EXIT_TR_READ
),
3993 [x86_intercept_lldt
] = POST_EX(SVM_EXIT_LDTR_WRITE
),
3994 [x86_intercept_ltr
] = POST_EX(SVM_EXIT_TR_WRITE
),
3995 [x86_intercept_sgdt
] = POST_EX(SVM_EXIT_GDTR_READ
),
3996 [x86_intercept_sidt
] = POST_EX(SVM_EXIT_IDTR_READ
),
3997 [x86_intercept_lgdt
] = POST_EX(SVM_EXIT_GDTR_WRITE
),
3998 [x86_intercept_lidt
] = POST_EX(SVM_EXIT_IDTR_WRITE
),
3999 [x86_intercept_vmrun
] = POST_EX(SVM_EXIT_VMRUN
),
4000 [x86_intercept_vmmcall
] = POST_EX(SVM_EXIT_VMMCALL
),
4001 [x86_intercept_vmload
] = POST_EX(SVM_EXIT_VMLOAD
),
4002 [x86_intercept_vmsave
] = POST_EX(SVM_EXIT_VMSAVE
),
4003 [x86_intercept_stgi
] = POST_EX(SVM_EXIT_STGI
),
4004 [x86_intercept_clgi
] = POST_EX(SVM_EXIT_CLGI
),
4005 [x86_intercept_skinit
] = POST_EX(SVM_EXIT_SKINIT
),
4006 [x86_intercept_invlpga
] = POST_EX(SVM_EXIT_INVLPGA
),
4007 [x86_intercept_rdtscp
] = POST_EX(SVM_EXIT_RDTSCP
),
4008 [x86_intercept_monitor
] = POST_MEM(SVM_EXIT_MONITOR
),
4009 [x86_intercept_mwait
] = POST_EX(SVM_EXIT_MWAIT
),
4010 [x86_intercept_invlpg
] = POST_EX(SVM_EXIT_INVLPG
),
4011 [x86_intercept_invd
] = POST_EX(SVM_EXIT_INVD
),
4012 [x86_intercept_wbinvd
] = POST_EX(SVM_EXIT_WBINVD
),
4013 [x86_intercept_wrmsr
] = POST_EX(SVM_EXIT_MSR
),
4014 [x86_intercept_rdtsc
] = POST_EX(SVM_EXIT_RDTSC
),
4015 [x86_intercept_rdmsr
] = POST_EX(SVM_EXIT_MSR
),
4016 [x86_intercept_rdpmc
] = POST_EX(SVM_EXIT_RDPMC
),
4017 [x86_intercept_cpuid
] = PRE_EX(SVM_EXIT_CPUID
),
4018 [x86_intercept_rsm
] = PRE_EX(SVM_EXIT_RSM
),
4019 [x86_intercept_pause
] = PRE_EX(SVM_EXIT_PAUSE
),
4020 [x86_intercept_pushf
] = PRE_EX(SVM_EXIT_PUSHF
),
4021 [x86_intercept_popf
] = PRE_EX(SVM_EXIT_POPF
),
4022 [x86_intercept_intn
] = PRE_EX(SVM_EXIT_SWINT
),
4023 [x86_intercept_iret
] = PRE_EX(SVM_EXIT_IRET
),
4024 [x86_intercept_icebp
] = PRE_EX(SVM_EXIT_ICEBP
),
4025 [x86_intercept_hlt
] = POST_EX(SVM_EXIT_HLT
),
4026 [x86_intercept_in
] = POST_EX(SVM_EXIT_IOIO
),
4027 [x86_intercept_ins
] = POST_EX(SVM_EXIT_IOIO
),
4028 [x86_intercept_out
] = POST_EX(SVM_EXIT_IOIO
),
4029 [x86_intercept_outs
] = POST_EX(SVM_EXIT_IOIO
),
4036 static int svm_check_intercept(struct kvm_vcpu
*vcpu
,
4037 struct x86_instruction_info
*info
,
4038 enum x86_intercept_stage stage
)
4040 struct vcpu_svm
*svm
= to_svm(vcpu
);
4041 int vmexit
, ret
= X86EMUL_CONTINUE
;
4042 struct __x86_intercept icpt_info
;
4043 struct vmcb
*vmcb
= svm
->vmcb
;
4045 if (info
->intercept
>= ARRAY_SIZE(x86_intercept_map
))
4048 icpt_info
= x86_intercept_map
[info
->intercept
];
4050 if (stage
!= icpt_info
.stage
)
4053 switch (icpt_info
.exit_code
) {
4054 case SVM_EXIT_READ_CR0
:
4055 if (info
->intercept
== x86_intercept_cr_read
)
4056 icpt_info
.exit_code
+= info
->modrm_reg
;
4058 case SVM_EXIT_WRITE_CR0
: {
4059 unsigned long cr0
, val
;
4062 if (info
->intercept
== x86_intercept_cr_write
)
4063 icpt_info
.exit_code
+= info
->modrm_reg
;
4065 if (icpt_info
.exit_code
!= SVM_EXIT_WRITE_CR0
)
4068 intercept
= svm
->nested
.intercept
;
4070 if (!(intercept
& (1ULL << INTERCEPT_SELECTIVE_CR0
)))
4073 cr0
= vcpu
->arch
.cr0
& ~SVM_CR0_SELECTIVE_MASK
;
4074 val
= info
->src_val
& ~SVM_CR0_SELECTIVE_MASK
;
4076 if (info
->intercept
== x86_intercept_lmsw
) {
4079 /* lmsw can't clear PE - catch this here */
4080 if (cr0
& X86_CR0_PE
)
4085 icpt_info
.exit_code
= SVM_EXIT_CR0_SEL_WRITE
;
4089 case SVM_EXIT_READ_DR0
:
4090 case SVM_EXIT_WRITE_DR0
:
4091 icpt_info
.exit_code
+= info
->modrm_reg
;
4094 if (info
->intercept
== x86_intercept_wrmsr
)
4095 vmcb
->control
.exit_info_1
= 1;
4097 vmcb
->control
.exit_info_1
= 0;
4099 case SVM_EXIT_PAUSE
:
4101 * We get this for NOP only, but pause
4102 * is rep not, check this here
4104 if (info
->rep_prefix
!= REPE_PREFIX
)
4106 case SVM_EXIT_IOIO
: {
4110 exit_info
= (vcpu
->arch
.regs
[VCPU_REGS_RDX
] & 0xffff) << 16;
4112 if (info
->intercept
== x86_intercept_in
||
4113 info
->intercept
== x86_intercept_ins
) {
4114 exit_info
|= SVM_IOIO_TYPE_MASK
;
4115 bytes
= info
->src_bytes
;
4117 bytes
= info
->dst_bytes
;
4120 if (info
->intercept
== x86_intercept_outs
||
4121 info
->intercept
== x86_intercept_ins
)
4122 exit_info
|= SVM_IOIO_STR_MASK
;
4124 if (info
->rep_prefix
)
4125 exit_info
|= SVM_IOIO_REP_MASK
;
4127 bytes
= min(bytes
, 4u);
4129 exit_info
|= bytes
<< SVM_IOIO_SIZE_SHIFT
;
4131 exit_info
|= (u32
)info
->ad_bytes
<< (SVM_IOIO_ASIZE_SHIFT
- 1);
4133 vmcb
->control
.exit_info_1
= exit_info
;
4134 vmcb
->control
.exit_info_2
= info
->next_rip
;
4142 vmcb
->control
.next_rip
= info
->next_rip
;
4143 vmcb
->control
.exit_code
= icpt_info
.exit_code
;
4144 vmexit
= nested_svm_exit_handled(svm
);
4146 ret
= (vmexit
== NESTED_EXIT_DONE
) ? X86EMUL_INTERCEPTED
4153 static struct kvm_x86_ops svm_x86_ops
= {
4154 .cpu_has_kvm_support
= has_svm
,
4155 .disabled_by_bios
= is_disabled
,
4156 .hardware_setup
= svm_hardware_setup
,
4157 .hardware_unsetup
= svm_hardware_unsetup
,
4158 .check_processor_compatibility
= svm_check_processor_compat
,
4159 .hardware_enable
= svm_hardware_enable
,
4160 .hardware_disable
= svm_hardware_disable
,
4161 .cpu_has_accelerated_tpr
= svm_cpu_has_accelerated_tpr
,
4163 .vcpu_create
= svm_create_vcpu
,
4164 .vcpu_free
= svm_free_vcpu
,
4165 .vcpu_reset
= svm_vcpu_reset
,
4167 .prepare_guest_switch
= svm_prepare_guest_switch
,
4168 .vcpu_load
= svm_vcpu_load
,
4169 .vcpu_put
= svm_vcpu_put
,
4171 .set_guest_debug
= svm_guest_debug
,
4172 .get_msr
= svm_get_msr
,
4173 .set_msr
= svm_set_msr
,
4174 .get_segment_base
= svm_get_segment_base
,
4175 .get_segment
= svm_get_segment
,
4176 .set_segment
= svm_set_segment
,
4177 .get_cpl
= svm_get_cpl
,
4178 .get_cs_db_l_bits
= kvm_get_cs_db_l_bits
,
4179 .decache_cr0_guest_bits
= svm_decache_cr0_guest_bits
,
4180 .decache_cr3
= svm_decache_cr3
,
4181 .decache_cr4_guest_bits
= svm_decache_cr4_guest_bits
,
4182 .set_cr0
= svm_set_cr0
,
4183 .set_cr3
= svm_set_cr3
,
4184 .set_cr4
= svm_set_cr4
,
4185 .set_efer
= svm_set_efer
,
4186 .get_idt
= svm_get_idt
,
4187 .set_idt
= svm_set_idt
,
4188 .get_gdt
= svm_get_gdt
,
4189 .set_gdt
= svm_set_gdt
,
4190 .set_dr7
= svm_set_dr7
,
4191 .cache_reg
= svm_cache_reg
,
4192 .get_rflags
= svm_get_rflags
,
4193 .set_rflags
= svm_set_rflags
,
4194 .fpu_activate
= svm_fpu_activate
,
4195 .fpu_deactivate
= svm_fpu_deactivate
,
4197 .tlb_flush
= svm_flush_tlb
,
4199 .run
= svm_vcpu_run
,
4200 .handle_exit
= handle_exit
,
4201 .skip_emulated_instruction
= skip_emulated_instruction
,
4202 .set_interrupt_shadow
= svm_set_interrupt_shadow
,
4203 .get_interrupt_shadow
= svm_get_interrupt_shadow
,
4204 .patch_hypercall
= svm_patch_hypercall
,
4205 .set_irq
= svm_set_irq
,
4206 .set_nmi
= svm_inject_nmi
,
4207 .queue_exception
= svm_queue_exception
,
4208 .cancel_injection
= svm_cancel_injection
,
4209 .interrupt_allowed
= svm_interrupt_allowed
,
4210 .nmi_allowed
= svm_nmi_allowed
,
4211 .get_nmi_mask
= svm_get_nmi_mask
,
4212 .set_nmi_mask
= svm_set_nmi_mask
,
4213 .enable_nmi_window
= enable_nmi_window
,
4214 .enable_irq_window
= enable_irq_window
,
4215 .update_cr8_intercept
= update_cr8_intercept
,
4217 .set_tss_addr
= svm_set_tss_addr
,
4218 .get_tdp_level
= get_npt_level
,
4219 .get_mt_mask
= svm_get_mt_mask
,
4221 .get_exit_info
= svm_get_exit_info
,
4222 .exit_reasons_str
= svm_exit_reasons_str
,
4224 .get_lpage_level
= svm_get_lpage_level
,
4226 .cpuid_update
= svm_cpuid_update
,
4228 .rdtscp_supported
= svm_rdtscp_supported
,
4230 .set_supported_cpuid
= svm_set_supported_cpuid
,
4232 .has_wbinvd_exit
= svm_has_wbinvd_exit
,
4234 .set_tsc_khz
= svm_set_tsc_khz
,
4235 .write_tsc_offset
= svm_write_tsc_offset
,
4236 .adjust_tsc_offset
= svm_adjust_tsc_offset
,
4237 .compute_tsc_offset
= svm_compute_tsc_offset
,
4239 .set_tdp_cr3
= set_tdp_cr3
,
4241 .check_intercept
= svm_check_intercept
,
4244 static int __init
svm_init(void)
4246 return kvm_init(&svm_x86_ops
, sizeof(struct vcpu_svm
),
4247 __alignof__(struct vcpu_svm
), THIS_MODULE
);
4250 static void __exit
svm_exit(void)
4255 module_init(svm_init
)
4256 module_exit(svm_exit
)