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 * Set osvw_len to higher value when updated Revision Guides
115 * are published and we know what the new status bits are
117 static uint64_t osvw_len
= 4, osvw_status
;
120 struct kvm_vcpu vcpu
;
122 unsigned long vmcb_pa
;
123 struct svm_cpu_data
*svm_data
;
124 uint64_t asid_generation
;
125 uint64_t sysenter_esp
;
126 uint64_t sysenter_eip
;
130 u64 host_user_msrs
[NR_HOST_SAVE_USER_MSRS
];
142 struct nested_state nested
;
146 unsigned int3_injected
;
147 unsigned long int3_rip
;
153 static DEFINE_PER_CPU(u64
, current_tsc_ratio
);
154 #define TSC_RATIO_DEFAULT 0x0100000000ULL
156 #define MSR_INVALID 0xffffffffU
158 static struct svm_direct_access_msrs
{
159 u32 index
; /* Index of the MSR */
160 bool always
; /* True if intercept is always on */
161 } direct_access_msrs
[] = {
162 { .index
= MSR_STAR
, .always
= true },
163 { .index
= MSR_IA32_SYSENTER_CS
, .always
= true },
165 { .index
= MSR_GS_BASE
, .always
= true },
166 { .index
= MSR_FS_BASE
, .always
= true },
167 { .index
= MSR_KERNEL_GS_BASE
, .always
= true },
168 { .index
= MSR_LSTAR
, .always
= true },
169 { .index
= MSR_CSTAR
, .always
= true },
170 { .index
= MSR_SYSCALL_MASK
, .always
= true },
172 { .index
= MSR_IA32_LASTBRANCHFROMIP
, .always
= false },
173 { .index
= MSR_IA32_LASTBRANCHTOIP
, .always
= false },
174 { .index
= MSR_IA32_LASTINTFROMIP
, .always
= false },
175 { .index
= MSR_IA32_LASTINTTOIP
, .always
= false },
176 { .index
= MSR_INVALID
, .always
= false },
179 /* enable NPT for AMD64 and X86 with PAE */
180 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
181 static bool npt_enabled
= true;
183 static bool npt_enabled
;
186 /* allow nested paging (virtualized MMU) for all guests */
187 static int npt
= true;
188 module_param(npt
, int, S_IRUGO
);
190 /* allow nested virtualization in KVM/SVM */
191 static int nested
= true;
192 module_param(nested
, int, S_IRUGO
);
194 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
);
195 static void svm_complete_interrupts(struct vcpu_svm
*svm
);
197 static int nested_svm_exit_handled(struct vcpu_svm
*svm
);
198 static int nested_svm_intercept(struct vcpu_svm
*svm
);
199 static int nested_svm_vmexit(struct vcpu_svm
*svm
);
200 static int nested_svm_check_exception(struct vcpu_svm
*svm
, unsigned nr
,
201 bool has_error_code
, u32 error_code
);
202 static u64
__scale_tsc(u64 ratio
, u64 tsc
);
205 VMCB_INTERCEPTS
, /* Intercept vectors, TSC offset,
206 pause filter count */
207 VMCB_PERM_MAP
, /* IOPM Base and MSRPM Base */
208 VMCB_ASID
, /* ASID */
209 VMCB_INTR
, /* int_ctl, int_vector */
210 VMCB_NPT
, /* npt_en, nCR3, gPAT */
211 VMCB_CR
, /* CR0, CR3, CR4, EFER */
212 VMCB_DR
, /* DR6, DR7 */
213 VMCB_DT
, /* GDT, IDT */
214 VMCB_SEG
, /* CS, DS, SS, ES, CPL */
215 VMCB_CR2
, /* CR2 only */
216 VMCB_LBR
, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
220 /* TPR and CR2 are always written before VMRUN */
221 #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2))
223 static inline void mark_all_dirty(struct vmcb
*vmcb
)
225 vmcb
->control
.clean
= 0;
228 static inline void mark_all_clean(struct vmcb
*vmcb
)
230 vmcb
->control
.clean
= ((1 << VMCB_DIRTY_MAX
) - 1)
231 & ~VMCB_ALWAYS_DIRTY_MASK
;
234 static inline void mark_dirty(struct vmcb
*vmcb
, int bit
)
236 vmcb
->control
.clean
&= ~(1 << bit
);
239 static inline struct vcpu_svm
*to_svm(struct kvm_vcpu
*vcpu
)
241 return container_of(vcpu
, struct vcpu_svm
, vcpu
);
244 static void recalc_intercepts(struct vcpu_svm
*svm
)
246 struct vmcb_control_area
*c
, *h
;
247 struct nested_state
*g
;
249 mark_dirty(svm
->vmcb
, VMCB_INTERCEPTS
);
251 if (!is_guest_mode(&svm
->vcpu
))
254 c
= &svm
->vmcb
->control
;
255 h
= &svm
->nested
.hsave
->control
;
258 c
->intercept_cr
= h
->intercept_cr
| g
->intercept_cr
;
259 c
->intercept_dr
= h
->intercept_dr
| g
->intercept_dr
;
260 c
->intercept_exceptions
= h
->intercept_exceptions
| g
->intercept_exceptions
;
261 c
->intercept
= h
->intercept
| g
->intercept
;
264 static inline struct vmcb
*get_host_vmcb(struct vcpu_svm
*svm
)
266 if (is_guest_mode(&svm
->vcpu
))
267 return svm
->nested
.hsave
;
272 static inline void set_cr_intercept(struct vcpu_svm
*svm
, int bit
)
274 struct vmcb
*vmcb
= get_host_vmcb(svm
);
276 vmcb
->control
.intercept_cr
|= (1U << bit
);
278 recalc_intercepts(svm
);
281 static inline void clr_cr_intercept(struct vcpu_svm
*svm
, int bit
)
283 struct vmcb
*vmcb
= get_host_vmcb(svm
);
285 vmcb
->control
.intercept_cr
&= ~(1U << bit
);
287 recalc_intercepts(svm
);
290 static inline bool is_cr_intercept(struct vcpu_svm
*svm
, int bit
)
292 struct vmcb
*vmcb
= get_host_vmcb(svm
);
294 return vmcb
->control
.intercept_cr
& (1U << bit
);
297 static inline void set_dr_intercept(struct vcpu_svm
*svm
, int bit
)
299 struct vmcb
*vmcb
= get_host_vmcb(svm
);
301 vmcb
->control
.intercept_dr
|= (1U << bit
);
303 recalc_intercepts(svm
);
306 static inline void clr_dr_intercept(struct vcpu_svm
*svm
, int bit
)
308 struct vmcb
*vmcb
= get_host_vmcb(svm
);
310 vmcb
->control
.intercept_dr
&= ~(1U << bit
);
312 recalc_intercepts(svm
);
315 static inline void set_exception_intercept(struct vcpu_svm
*svm
, int bit
)
317 struct vmcb
*vmcb
= get_host_vmcb(svm
);
319 vmcb
->control
.intercept_exceptions
|= (1U << bit
);
321 recalc_intercepts(svm
);
324 static inline void clr_exception_intercept(struct vcpu_svm
*svm
, int bit
)
326 struct vmcb
*vmcb
= get_host_vmcb(svm
);
328 vmcb
->control
.intercept_exceptions
&= ~(1U << bit
);
330 recalc_intercepts(svm
);
333 static inline void set_intercept(struct vcpu_svm
*svm
, int bit
)
335 struct vmcb
*vmcb
= get_host_vmcb(svm
);
337 vmcb
->control
.intercept
|= (1ULL << bit
);
339 recalc_intercepts(svm
);
342 static inline void clr_intercept(struct vcpu_svm
*svm
, int bit
)
344 struct vmcb
*vmcb
= get_host_vmcb(svm
);
346 vmcb
->control
.intercept
&= ~(1ULL << bit
);
348 recalc_intercepts(svm
);
351 static inline void enable_gif(struct vcpu_svm
*svm
)
353 svm
->vcpu
.arch
.hflags
|= HF_GIF_MASK
;
356 static inline void disable_gif(struct vcpu_svm
*svm
)
358 svm
->vcpu
.arch
.hflags
&= ~HF_GIF_MASK
;
361 static inline bool gif_set(struct vcpu_svm
*svm
)
363 return !!(svm
->vcpu
.arch
.hflags
& HF_GIF_MASK
);
366 static unsigned long iopm_base
;
368 struct kvm_ldttss_desc
{
371 unsigned base1
:8, type
:5, dpl
:2, p
:1;
372 unsigned limit1
:4, zero0
:3, g
:1, base2
:8;
375 } __attribute__((packed
));
377 struct svm_cpu_data
{
383 struct kvm_ldttss_desc
*tss_desc
;
385 struct page
*save_area
;
388 static DEFINE_PER_CPU(struct svm_cpu_data
*, svm_data
);
390 struct svm_init_data
{
395 static u32 msrpm_ranges
[] = {0, 0xc0000000, 0xc0010000};
397 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
398 #define MSRS_RANGE_SIZE 2048
399 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
401 static u32
svm_msrpm_offset(u32 msr
)
406 for (i
= 0; i
< NUM_MSR_MAPS
; i
++) {
407 if (msr
< msrpm_ranges
[i
] ||
408 msr
>= msrpm_ranges
[i
] + MSRS_IN_RANGE
)
411 offset
= (msr
- msrpm_ranges
[i
]) / 4; /* 4 msrs per u8 */
412 offset
+= (i
* MSRS_RANGE_SIZE
); /* add range offset */
414 /* Now we have the u8 offset - but need the u32 offset */
418 /* MSR not in any range */
422 #define MAX_INST_SIZE 15
424 static inline void clgi(void)
426 asm volatile (__ex(SVM_CLGI
));
429 static inline void stgi(void)
431 asm volatile (__ex(SVM_STGI
));
434 static inline void invlpga(unsigned long addr
, u32 asid
)
436 asm volatile (__ex(SVM_INVLPGA
) : : "a"(addr
), "c"(asid
));
439 static int get_npt_level(void)
442 return PT64_ROOT_LEVEL
;
444 return PT32E_ROOT_LEVEL
;
448 static void svm_set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
450 vcpu
->arch
.efer
= efer
;
451 if (!npt_enabled
&& !(efer
& EFER_LMA
))
454 to_svm(vcpu
)->vmcb
->save
.efer
= efer
| EFER_SVME
;
455 mark_dirty(to_svm(vcpu
)->vmcb
, VMCB_CR
);
458 static int is_external_interrupt(u32 info
)
460 info
&= SVM_EVTINJ_TYPE_MASK
| SVM_EVTINJ_VALID
;
461 return info
== (SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
);
464 static u32
svm_get_interrupt_shadow(struct kvm_vcpu
*vcpu
, int mask
)
466 struct vcpu_svm
*svm
= to_svm(vcpu
);
469 if (svm
->vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
)
470 ret
|= KVM_X86_SHADOW_INT_STI
| KVM_X86_SHADOW_INT_MOV_SS
;
474 static void svm_set_interrupt_shadow(struct kvm_vcpu
*vcpu
, int mask
)
476 struct vcpu_svm
*svm
= to_svm(vcpu
);
479 svm
->vmcb
->control
.int_state
&= ~SVM_INTERRUPT_SHADOW_MASK
;
481 svm
->vmcb
->control
.int_state
|= SVM_INTERRUPT_SHADOW_MASK
;
485 static void skip_emulated_instruction(struct kvm_vcpu
*vcpu
)
487 struct vcpu_svm
*svm
= to_svm(vcpu
);
489 if (svm
->vmcb
->control
.next_rip
!= 0)
490 svm
->next_rip
= svm
->vmcb
->control
.next_rip
;
492 if (!svm
->next_rip
) {
493 if (emulate_instruction(vcpu
, EMULTYPE_SKIP
) !=
495 printk(KERN_DEBUG
"%s: NOP\n", __func__
);
498 if (svm
->next_rip
- kvm_rip_read(vcpu
) > MAX_INST_SIZE
)
499 printk(KERN_ERR
"%s: ip 0x%lx next 0x%llx\n",
500 __func__
, kvm_rip_read(vcpu
), svm
->next_rip
);
502 kvm_rip_write(vcpu
, svm
->next_rip
);
503 svm_set_interrupt_shadow(vcpu
, 0);
506 static void svm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
,
507 bool has_error_code
, u32 error_code
,
510 struct vcpu_svm
*svm
= to_svm(vcpu
);
513 * If we are within a nested VM we'd better #VMEXIT and let the guest
514 * handle the exception
517 nested_svm_check_exception(svm
, nr
, has_error_code
, error_code
))
520 if (nr
== BP_VECTOR
&& !static_cpu_has(X86_FEATURE_NRIPS
)) {
521 unsigned long rip
, old_rip
= kvm_rip_read(&svm
->vcpu
);
524 * For guest debugging where we have to reinject #BP if some
525 * INT3 is guest-owned:
526 * Emulate nRIP by moving RIP forward. Will fail if injection
527 * raises a fault that is not intercepted. Still better than
528 * failing in all cases.
530 skip_emulated_instruction(&svm
->vcpu
);
531 rip
= kvm_rip_read(&svm
->vcpu
);
532 svm
->int3_rip
= rip
+ svm
->vmcb
->save
.cs
.base
;
533 svm
->int3_injected
= rip
- old_rip
;
536 svm
->vmcb
->control
.event_inj
= nr
538 | (has_error_code
? SVM_EVTINJ_VALID_ERR
: 0)
539 | SVM_EVTINJ_TYPE_EXEPT
;
540 svm
->vmcb
->control
.event_inj_err
= error_code
;
543 static void svm_init_erratum_383(void)
549 if (!cpu_has_amd_erratum(amd_erratum_383
))
552 /* Use _safe variants to not break nested virtualization */
553 val
= native_read_msr_safe(MSR_AMD64_DC_CFG
, &err
);
559 low
= lower_32_bits(val
);
560 high
= upper_32_bits(val
);
562 native_write_msr_safe(MSR_AMD64_DC_CFG
, low
, high
);
564 erratum_383_found
= true;
567 static void svm_init_osvw(struct kvm_vcpu
*vcpu
)
570 * Guests should see errata 400 and 415 as fixed (assuming that
571 * HLT and IO instructions are intercepted).
573 vcpu
->arch
.osvw
.length
= (osvw_len
>= 3) ? (osvw_len
) : 3;
574 vcpu
->arch
.osvw
.status
= osvw_status
& ~(6ULL);
577 * By increasing VCPU's osvw.length to 3 we are telling the guest that
578 * all osvw.status bits inside that length, including bit 0 (which is
579 * reserved for erratum 298), are valid. However, if host processor's
580 * osvw_len is 0 then osvw_status[0] carries no information. We need to
581 * be conservative here and therefore we tell the guest that erratum 298
582 * is present (because we really don't know).
584 if (osvw_len
== 0 && boot_cpu_data
.x86
== 0x10)
585 vcpu
->arch
.osvw
.status
|= 1;
588 static int has_svm(void)
592 if (!cpu_has_svm(&msg
)) {
593 printk(KERN_INFO
"has_svm: %s\n", msg
);
600 static void svm_hardware_disable(void *garbage
)
602 /* Make sure we clean up behind us */
603 if (static_cpu_has(X86_FEATURE_TSCRATEMSR
))
604 wrmsrl(MSR_AMD64_TSC_RATIO
, TSC_RATIO_DEFAULT
);
609 static int svm_hardware_enable(void *garbage
)
612 struct svm_cpu_data
*sd
;
614 struct desc_ptr gdt_descr
;
615 struct desc_struct
*gdt
;
616 int me
= raw_smp_processor_id();
618 rdmsrl(MSR_EFER
, efer
);
619 if (efer
& EFER_SVME
)
623 printk(KERN_ERR
"svm_hardware_enable: err EOPNOTSUPP on %d\n",
627 sd
= per_cpu(svm_data
, me
);
630 printk(KERN_ERR
"svm_hardware_enable: svm_data is NULL on %d\n",
635 sd
->asid_generation
= 1;
636 sd
->max_asid
= cpuid_ebx(SVM_CPUID_FUNC
) - 1;
637 sd
->next_asid
= sd
->max_asid
+ 1;
639 native_store_gdt(&gdt_descr
);
640 gdt
= (struct desc_struct
*)gdt_descr
.address
;
641 sd
->tss_desc
= (struct kvm_ldttss_desc
*)(gdt
+ GDT_ENTRY_TSS
);
643 wrmsrl(MSR_EFER
, efer
| EFER_SVME
);
645 wrmsrl(MSR_VM_HSAVE_PA
, page_to_pfn(sd
->save_area
) << PAGE_SHIFT
);
647 if (static_cpu_has(X86_FEATURE_TSCRATEMSR
)) {
648 wrmsrl(MSR_AMD64_TSC_RATIO
, TSC_RATIO_DEFAULT
);
649 __get_cpu_var(current_tsc_ratio
) = TSC_RATIO_DEFAULT
;
656 * Note that it is possible to have a system with mixed processor
657 * revisions and therefore different OSVW bits. If bits are not the same
658 * on different processors then choose the worst case (i.e. if erratum
659 * is present on one processor and not on another then assume that the
660 * erratum is present everywhere).
662 if (cpu_has(&boot_cpu_data
, X86_FEATURE_OSVW
)) {
663 uint64_t len
, status
= 0;
666 len
= native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH
, &err
);
668 status
= native_read_msr_safe(MSR_AMD64_OSVW_STATUS
,
672 osvw_status
= osvw_len
= 0;
676 osvw_status
|= status
;
677 osvw_status
&= (1ULL << osvw_len
) - 1;
680 osvw_status
= osvw_len
= 0;
682 svm_init_erratum_383();
687 static void svm_cpu_uninit(int cpu
)
689 struct svm_cpu_data
*sd
= per_cpu(svm_data
, raw_smp_processor_id());
694 per_cpu(svm_data
, raw_smp_processor_id()) = NULL
;
695 __free_page(sd
->save_area
);
699 static int svm_cpu_init(int cpu
)
701 struct svm_cpu_data
*sd
;
704 sd
= kzalloc(sizeof(struct svm_cpu_data
), GFP_KERNEL
);
708 sd
->save_area
= alloc_page(GFP_KERNEL
);
713 per_cpu(svm_data
, cpu
) = sd
;
723 static bool valid_msr_intercept(u32 index
)
727 for (i
= 0; direct_access_msrs
[i
].index
!= MSR_INVALID
; i
++)
728 if (direct_access_msrs
[i
].index
== index
)
734 static void set_msr_interception(u32
*msrpm
, unsigned msr
,
737 u8 bit_read
, bit_write
;
742 * If this warning triggers extend the direct_access_msrs list at the
743 * beginning of the file
745 WARN_ON(!valid_msr_intercept(msr
));
747 offset
= svm_msrpm_offset(msr
);
748 bit_read
= 2 * (msr
& 0x0f);
749 bit_write
= 2 * (msr
& 0x0f) + 1;
752 BUG_ON(offset
== MSR_INVALID
);
754 read
? clear_bit(bit_read
, &tmp
) : set_bit(bit_read
, &tmp
);
755 write
? clear_bit(bit_write
, &tmp
) : set_bit(bit_write
, &tmp
);
760 static void svm_vcpu_init_msrpm(u32
*msrpm
)
764 memset(msrpm
, 0xff, PAGE_SIZE
* (1 << MSRPM_ALLOC_ORDER
));
766 for (i
= 0; direct_access_msrs
[i
].index
!= MSR_INVALID
; i
++) {
767 if (!direct_access_msrs
[i
].always
)
770 set_msr_interception(msrpm
, direct_access_msrs
[i
].index
, 1, 1);
774 static void add_msr_offset(u32 offset
)
778 for (i
= 0; i
< MSRPM_OFFSETS
; ++i
) {
780 /* Offset already in list? */
781 if (msrpm_offsets
[i
] == offset
)
784 /* Slot used by another offset? */
785 if (msrpm_offsets
[i
] != MSR_INVALID
)
788 /* Add offset to list */
789 msrpm_offsets
[i
] = offset
;
795 * If this BUG triggers the msrpm_offsets table has an overflow. Just
796 * increase MSRPM_OFFSETS in this case.
801 static void init_msrpm_offsets(void)
805 memset(msrpm_offsets
, 0xff, sizeof(msrpm_offsets
));
807 for (i
= 0; direct_access_msrs
[i
].index
!= MSR_INVALID
; i
++) {
810 offset
= svm_msrpm_offset(direct_access_msrs
[i
].index
);
811 BUG_ON(offset
== MSR_INVALID
);
813 add_msr_offset(offset
);
817 static void svm_enable_lbrv(struct vcpu_svm
*svm
)
819 u32
*msrpm
= svm
->msrpm
;
821 svm
->vmcb
->control
.lbr_ctl
= 1;
822 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHFROMIP
, 1, 1);
823 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHTOIP
, 1, 1);
824 set_msr_interception(msrpm
, MSR_IA32_LASTINTFROMIP
, 1, 1);
825 set_msr_interception(msrpm
, MSR_IA32_LASTINTTOIP
, 1, 1);
828 static void svm_disable_lbrv(struct vcpu_svm
*svm
)
830 u32
*msrpm
= svm
->msrpm
;
832 svm
->vmcb
->control
.lbr_ctl
= 0;
833 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHFROMIP
, 0, 0);
834 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHTOIP
, 0, 0);
835 set_msr_interception(msrpm
, MSR_IA32_LASTINTFROMIP
, 0, 0);
836 set_msr_interception(msrpm
, MSR_IA32_LASTINTTOIP
, 0, 0);
839 static __init
int svm_hardware_setup(void)
842 struct page
*iopm_pages
;
846 iopm_pages
= alloc_pages(GFP_KERNEL
, IOPM_ALLOC_ORDER
);
851 iopm_va
= page_address(iopm_pages
);
852 memset(iopm_va
, 0xff, PAGE_SIZE
* (1 << IOPM_ALLOC_ORDER
));
853 iopm_base
= page_to_pfn(iopm_pages
) << PAGE_SHIFT
;
855 init_msrpm_offsets();
857 if (boot_cpu_has(X86_FEATURE_NX
))
858 kvm_enable_efer_bits(EFER_NX
);
860 if (boot_cpu_has(X86_FEATURE_FXSR_OPT
))
861 kvm_enable_efer_bits(EFER_FFXSR
);
863 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR
)) {
866 kvm_has_tsc_control
= true;
869 * Make sure the user can only configure tsc_khz values that
870 * fit into a signed integer.
871 * A min value is not calculated needed because it will always
872 * be 1 on all machines and a value of 0 is used to disable
873 * tsc-scaling for the vcpu.
875 max
= min(0x7fffffffULL
, __scale_tsc(tsc_khz
, TSC_RATIO_MAX
));
877 kvm_max_guest_tsc_khz
= max
;
881 printk(KERN_INFO
"kvm: Nested Virtualization enabled\n");
882 kvm_enable_efer_bits(EFER_SVME
| EFER_LMSLE
);
885 for_each_possible_cpu(cpu
) {
886 r
= svm_cpu_init(cpu
);
891 if (!boot_cpu_has(X86_FEATURE_NPT
))
894 if (npt_enabled
&& !npt
) {
895 printk(KERN_INFO
"kvm: Nested Paging disabled\n");
900 printk(KERN_INFO
"kvm: Nested Paging enabled\n");
908 __free_pages(iopm_pages
, IOPM_ALLOC_ORDER
);
913 static __exit
void svm_hardware_unsetup(void)
917 for_each_possible_cpu(cpu
)
920 __free_pages(pfn_to_page(iopm_base
>> PAGE_SHIFT
), IOPM_ALLOC_ORDER
);
924 static void init_seg(struct vmcb_seg
*seg
)
927 seg
->attrib
= SVM_SELECTOR_P_MASK
| SVM_SELECTOR_S_MASK
|
928 SVM_SELECTOR_WRITE_MASK
; /* Read/Write Data Segment */
933 static void init_sys_seg(struct vmcb_seg
*seg
, uint32_t type
)
936 seg
->attrib
= SVM_SELECTOR_P_MASK
| type
;
941 static u64
__scale_tsc(u64 ratio
, u64 tsc
)
943 u64 mult
, frac
, _tsc
;
946 frac
= ratio
& ((1ULL << 32) - 1);
950 _tsc
+= (tsc
>> 32) * frac
;
951 _tsc
+= ((tsc
& ((1ULL << 32) - 1)) * frac
) >> 32;
956 static u64
svm_scale_tsc(struct kvm_vcpu
*vcpu
, u64 tsc
)
958 struct vcpu_svm
*svm
= to_svm(vcpu
);
961 if (svm
->tsc_ratio
!= TSC_RATIO_DEFAULT
)
962 _tsc
= __scale_tsc(svm
->tsc_ratio
, tsc
);
967 static void svm_set_tsc_khz(struct kvm_vcpu
*vcpu
, u32 user_tsc_khz
, bool scale
)
969 struct vcpu_svm
*svm
= to_svm(vcpu
);
973 /* Guest TSC same frequency as host TSC? */
975 svm
->tsc_ratio
= TSC_RATIO_DEFAULT
;
979 /* TSC scaling supported? */
980 if (!boot_cpu_has(X86_FEATURE_TSCRATEMSR
)) {
981 if (user_tsc_khz
> tsc_khz
) {
982 vcpu
->arch
.tsc_catchup
= 1;
983 vcpu
->arch
.tsc_always_catchup
= 1;
985 WARN(1, "user requested TSC rate below hardware speed\n");
991 /* TSC scaling required - calculate ratio */
993 do_div(ratio
, tsc_khz
);
995 if (ratio
== 0 || ratio
& TSC_RATIO_RSVD
) {
996 WARN_ONCE(1, "Invalid TSC ratio - virtual-tsc-khz=%u\n",
1000 svm
->tsc_ratio
= ratio
;
1003 static void svm_write_tsc_offset(struct kvm_vcpu
*vcpu
, u64 offset
)
1005 struct vcpu_svm
*svm
= to_svm(vcpu
);
1006 u64 g_tsc_offset
= 0;
1008 if (is_guest_mode(vcpu
)) {
1009 g_tsc_offset
= svm
->vmcb
->control
.tsc_offset
-
1010 svm
->nested
.hsave
->control
.tsc_offset
;
1011 svm
->nested
.hsave
->control
.tsc_offset
= offset
;
1014 svm
->vmcb
->control
.tsc_offset
= offset
+ g_tsc_offset
;
1016 mark_dirty(svm
->vmcb
, VMCB_INTERCEPTS
);
1019 static void svm_adjust_tsc_offset(struct kvm_vcpu
*vcpu
, s64 adjustment
, bool host
)
1021 struct vcpu_svm
*svm
= to_svm(vcpu
);
1023 WARN_ON(adjustment
< 0);
1025 adjustment
= svm_scale_tsc(vcpu
, adjustment
);
1027 svm
->vmcb
->control
.tsc_offset
+= adjustment
;
1028 if (is_guest_mode(vcpu
))
1029 svm
->nested
.hsave
->control
.tsc_offset
+= adjustment
;
1030 mark_dirty(svm
->vmcb
, VMCB_INTERCEPTS
);
1033 static u64
svm_compute_tsc_offset(struct kvm_vcpu
*vcpu
, u64 target_tsc
)
1037 tsc
= svm_scale_tsc(vcpu
, native_read_tsc());
1039 return target_tsc
- tsc
;
1042 static void init_vmcb(struct vcpu_svm
*svm
)
1044 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
1045 struct vmcb_save_area
*save
= &svm
->vmcb
->save
;
1047 svm
->vcpu
.fpu_active
= 1;
1048 svm
->vcpu
.arch
.hflags
= 0;
1050 set_cr_intercept(svm
, INTERCEPT_CR0_READ
);
1051 set_cr_intercept(svm
, INTERCEPT_CR3_READ
);
1052 set_cr_intercept(svm
, INTERCEPT_CR4_READ
);
1053 set_cr_intercept(svm
, INTERCEPT_CR0_WRITE
);
1054 set_cr_intercept(svm
, INTERCEPT_CR3_WRITE
);
1055 set_cr_intercept(svm
, INTERCEPT_CR4_WRITE
);
1056 set_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
1058 set_dr_intercept(svm
, INTERCEPT_DR0_READ
);
1059 set_dr_intercept(svm
, INTERCEPT_DR1_READ
);
1060 set_dr_intercept(svm
, INTERCEPT_DR2_READ
);
1061 set_dr_intercept(svm
, INTERCEPT_DR3_READ
);
1062 set_dr_intercept(svm
, INTERCEPT_DR4_READ
);
1063 set_dr_intercept(svm
, INTERCEPT_DR5_READ
);
1064 set_dr_intercept(svm
, INTERCEPT_DR6_READ
);
1065 set_dr_intercept(svm
, INTERCEPT_DR7_READ
);
1067 set_dr_intercept(svm
, INTERCEPT_DR0_WRITE
);
1068 set_dr_intercept(svm
, INTERCEPT_DR1_WRITE
);
1069 set_dr_intercept(svm
, INTERCEPT_DR2_WRITE
);
1070 set_dr_intercept(svm
, INTERCEPT_DR3_WRITE
);
1071 set_dr_intercept(svm
, INTERCEPT_DR4_WRITE
);
1072 set_dr_intercept(svm
, INTERCEPT_DR5_WRITE
);
1073 set_dr_intercept(svm
, INTERCEPT_DR6_WRITE
);
1074 set_dr_intercept(svm
, INTERCEPT_DR7_WRITE
);
1076 set_exception_intercept(svm
, PF_VECTOR
);
1077 set_exception_intercept(svm
, UD_VECTOR
);
1078 set_exception_intercept(svm
, MC_VECTOR
);
1080 set_intercept(svm
, INTERCEPT_INTR
);
1081 set_intercept(svm
, INTERCEPT_NMI
);
1082 set_intercept(svm
, INTERCEPT_SMI
);
1083 set_intercept(svm
, INTERCEPT_SELECTIVE_CR0
);
1084 set_intercept(svm
, INTERCEPT_RDPMC
);
1085 set_intercept(svm
, INTERCEPT_CPUID
);
1086 set_intercept(svm
, INTERCEPT_INVD
);
1087 set_intercept(svm
, INTERCEPT_HLT
);
1088 set_intercept(svm
, INTERCEPT_INVLPG
);
1089 set_intercept(svm
, INTERCEPT_INVLPGA
);
1090 set_intercept(svm
, INTERCEPT_IOIO_PROT
);
1091 set_intercept(svm
, INTERCEPT_MSR_PROT
);
1092 set_intercept(svm
, INTERCEPT_TASK_SWITCH
);
1093 set_intercept(svm
, INTERCEPT_SHUTDOWN
);
1094 set_intercept(svm
, INTERCEPT_VMRUN
);
1095 set_intercept(svm
, INTERCEPT_VMMCALL
);
1096 set_intercept(svm
, INTERCEPT_VMLOAD
);
1097 set_intercept(svm
, INTERCEPT_VMSAVE
);
1098 set_intercept(svm
, INTERCEPT_STGI
);
1099 set_intercept(svm
, INTERCEPT_CLGI
);
1100 set_intercept(svm
, INTERCEPT_SKINIT
);
1101 set_intercept(svm
, INTERCEPT_WBINVD
);
1102 set_intercept(svm
, INTERCEPT_MONITOR
);
1103 set_intercept(svm
, INTERCEPT_MWAIT
);
1104 set_intercept(svm
, INTERCEPT_XSETBV
);
1106 control
->iopm_base_pa
= iopm_base
;
1107 control
->msrpm_base_pa
= __pa(svm
->msrpm
);
1108 control
->int_ctl
= V_INTR_MASKING_MASK
;
1110 init_seg(&save
->es
);
1111 init_seg(&save
->ss
);
1112 init_seg(&save
->ds
);
1113 init_seg(&save
->fs
);
1114 init_seg(&save
->gs
);
1116 save
->cs
.selector
= 0xf000;
1117 /* Executable/Readable Code Segment */
1118 save
->cs
.attrib
= SVM_SELECTOR_READ_MASK
| SVM_SELECTOR_P_MASK
|
1119 SVM_SELECTOR_S_MASK
| SVM_SELECTOR_CODE_MASK
;
1120 save
->cs
.limit
= 0xffff;
1122 * cs.base should really be 0xffff0000, but vmx can't handle that, so
1123 * be consistent with it.
1125 * Replace when we have real mode working for vmx.
1127 save
->cs
.base
= 0xf0000;
1129 save
->gdtr
.limit
= 0xffff;
1130 save
->idtr
.limit
= 0xffff;
1132 init_sys_seg(&save
->ldtr
, SEG_TYPE_LDT
);
1133 init_sys_seg(&save
->tr
, SEG_TYPE_BUSY_TSS16
);
1135 svm_set_efer(&svm
->vcpu
, 0);
1136 save
->dr6
= 0xffff0ff0;
1138 kvm_set_rflags(&svm
->vcpu
, 2);
1139 save
->rip
= 0x0000fff0;
1140 svm
->vcpu
.arch
.regs
[VCPU_REGS_RIP
] = save
->rip
;
1143 * This is the guest-visible cr0 value.
1144 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1146 svm
->vcpu
.arch
.cr0
= 0;
1147 (void)kvm_set_cr0(&svm
->vcpu
, X86_CR0_NW
| X86_CR0_CD
| X86_CR0_ET
);
1149 save
->cr4
= X86_CR4_PAE
;
1153 /* Setup VMCB for Nested Paging */
1154 control
->nested_ctl
= 1;
1155 clr_intercept(svm
, INTERCEPT_INVLPG
);
1156 clr_exception_intercept(svm
, PF_VECTOR
);
1157 clr_cr_intercept(svm
, INTERCEPT_CR3_READ
);
1158 clr_cr_intercept(svm
, INTERCEPT_CR3_WRITE
);
1159 save
->g_pat
= 0x0007040600070406ULL
;
1163 svm
->asid_generation
= 0;
1165 svm
->nested
.vmcb
= 0;
1166 svm
->vcpu
.arch
.hflags
= 0;
1168 if (boot_cpu_has(X86_FEATURE_PAUSEFILTER
)) {
1169 control
->pause_filter_count
= 3000;
1170 set_intercept(svm
, INTERCEPT_PAUSE
);
1173 mark_all_dirty(svm
->vmcb
);
1178 static int svm_vcpu_reset(struct kvm_vcpu
*vcpu
)
1180 struct vcpu_svm
*svm
= to_svm(vcpu
);
1184 if (!kvm_vcpu_is_bsp(vcpu
)) {
1185 kvm_rip_write(vcpu
, 0);
1186 svm
->vmcb
->save
.cs
.base
= svm
->vcpu
.arch
.sipi_vector
<< 12;
1187 svm
->vmcb
->save
.cs
.selector
= svm
->vcpu
.arch
.sipi_vector
<< 8;
1189 vcpu
->arch
.regs_avail
= ~0;
1190 vcpu
->arch
.regs_dirty
= ~0;
1195 static struct kvm_vcpu
*svm_create_vcpu(struct kvm
*kvm
, unsigned int id
)
1197 struct vcpu_svm
*svm
;
1199 struct page
*msrpm_pages
;
1200 struct page
*hsave_page
;
1201 struct page
*nested_msrpm_pages
;
1204 svm
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
1210 svm
->tsc_ratio
= TSC_RATIO_DEFAULT
;
1212 err
= kvm_vcpu_init(&svm
->vcpu
, kvm
, id
);
1217 page
= alloc_page(GFP_KERNEL
);
1221 msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
1225 nested_msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
1226 if (!nested_msrpm_pages
)
1229 hsave_page
= alloc_page(GFP_KERNEL
);
1233 svm
->nested
.hsave
= page_address(hsave_page
);
1235 svm
->msrpm
= page_address(msrpm_pages
);
1236 svm_vcpu_init_msrpm(svm
->msrpm
);
1238 svm
->nested
.msrpm
= page_address(nested_msrpm_pages
);
1239 svm_vcpu_init_msrpm(svm
->nested
.msrpm
);
1241 svm
->vmcb
= page_address(page
);
1242 clear_page(svm
->vmcb
);
1243 svm
->vmcb_pa
= page_to_pfn(page
) << PAGE_SHIFT
;
1244 svm
->asid_generation
= 0;
1246 kvm_write_tsc(&svm
->vcpu
, 0);
1248 err
= fx_init(&svm
->vcpu
);
1252 svm
->vcpu
.arch
.apic_base
= 0xfee00000 | MSR_IA32_APICBASE_ENABLE
;
1253 if (kvm_vcpu_is_bsp(&svm
->vcpu
))
1254 svm
->vcpu
.arch
.apic_base
|= MSR_IA32_APICBASE_BSP
;
1256 svm_init_osvw(&svm
->vcpu
);
1261 __free_page(hsave_page
);
1263 __free_pages(nested_msrpm_pages
, MSRPM_ALLOC_ORDER
);
1265 __free_pages(msrpm_pages
, MSRPM_ALLOC_ORDER
);
1269 kvm_vcpu_uninit(&svm
->vcpu
);
1271 kmem_cache_free(kvm_vcpu_cache
, svm
);
1273 return ERR_PTR(err
);
1276 static void svm_free_vcpu(struct kvm_vcpu
*vcpu
)
1278 struct vcpu_svm
*svm
= to_svm(vcpu
);
1280 __free_page(pfn_to_page(svm
->vmcb_pa
>> PAGE_SHIFT
));
1281 __free_pages(virt_to_page(svm
->msrpm
), MSRPM_ALLOC_ORDER
);
1282 __free_page(virt_to_page(svm
->nested
.hsave
));
1283 __free_pages(virt_to_page(svm
->nested
.msrpm
), MSRPM_ALLOC_ORDER
);
1284 kvm_vcpu_uninit(vcpu
);
1285 kmem_cache_free(kvm_vcpu_cache
, svm
);
1288 static void svm_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
1290 struct vcpu_svm
*svm
= to_svm(vcpu
);
1293 if (unlikely(cpu
!= vcpu
->cpu
)) {
1294 svm
->asid_generation
= 0;
1295 mark_all_dirty(svm
->vmcb
);
1298 #ifdef CONFIG_X86_64
1299 rdmsrl(MSR_GS_BASE
, to_svm(vcpu
)->host
.gs_base
);
1301 savesegment(fs
, svm
->host
.fs
);
1302 savesegment(gs
, svm
->host
.gs
);
1303 svm
->host
.ldt
= kvm_read_ldt();
1305 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
1306 rdmsrl(host_save_user_msrs
[i
], svm
->host_user_msrs
[i
]);
1308 if (static_cpu_has(X86_FEATURE_TSCRATEMSR
) &&
1309 svm
->tsc_ratio
!= __get_cpu_var(current_tsc_ratio
)) {
1310 __get_cpu_var(current_tsc_ratio
) = svm
->tsc_ratio
;
1311 wrmsrl(MSR_AMD64_TSC_RATIO
, svm
->tsc_ratio
);
1315 static void svm_vcpu_put(struct kvm_vcpu
*vcpu
)
1317 struct vcpu_svm
*svm
= to_svm(vcpu
);
1320 ++vcpu
->stat
.host_state_reload
;
1321 kvm_load_ldt(svm
->host
.ldt
);
1322 #ifdef CONFIG_X86_64
1323 loadsegment(fs
, svm
->host
.fs
);
1324 wrmsrl(MSR_KERNEL_GS_BASE
, current
->thread
.gs
);
1325 load_gs_index(svm
->host
.gs
);
1327 #ifdef CONFIG_X86_32_LAZY_GS
1328 loadsegment(gs
, svm
->host
.gs
);
1331 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
1332 wrmsrl(host_save_user_msrs
[i
], svm
->host_user_msrs
[i
]);
1335 static void svm_update_cpl(struct kvm_vcpu
*vcpu
)
1337 struct vcpu_svm
*svm
= to_svm(vcpu
);
1340 if (!is_protmode(vcpu
))
1342 else if (svm
->vmcb
->save
.rflags
& X86_EFLAGS_VM
)
1345 cpl
= svm
->vmcb
->save
.cs
.selector
& 0x3;
1347 svm
->vmcb
->save
.cpl
= cpl
;
1350 static unsigned long svm_get_rflags(struct kvm_vcpu
*vcpu
)
1352 return to_svm(vcpu
)->vmcb
->save
.rflags
;
1355 static void svm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
1357 unsigned long old_rflags
= to_svm(vcpu
)->vmcb
->save
.rflags
;
1359 to_svm(vcpu
)->vmcb
->save
.rflags
= rflags
;
1360 if ((old_rflags
^ rflags
) & X86_EFLAGS_VM
)
1361 svm_update_cpl(vcpu
);
1364 static void svm_cache_reg(struct kvm_vcpu
*vcpu
, enum kvm_reg reg
)
1367 case VCPU_EXREG_PDPTR
:
1368 BUG_ON(!npt_enabled
);
1369 load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, kvm_read_cr3(vcpu
));
1376 static void svm_set_vintr(struct vcpu_svm
*svm
)
1378 set_intercept(svm
, INTERCEPT_VINTR
);
1381 static void svm_clear_vintr(struct vcpu_svm
*svm
)
1383 clr_intercept(svm
, INTERCEPT_VINTR
);
1386 static struct vmcb_seg
*svm_seg(struct kvm_vcpu
*vcpu
, int seg
)
1388 struct vmcb_save_area
*save
= &to_svm(vcpu
)->vmcb
->save
;
1391 case VCPU_SREG_CS
: return &save
->cs
;
1392 case VCPU_SREG_DS
: return &save
->ds
;
1393 case VCPU_SREG_ES
: return &save
->es
;
1394 case VCPU_SREG_FS
: return &save
->fs
;
1395 case VCPU_SREG_GS
: return &save
->gs
;
1396 case VCPU_SREG_SS
: return &save
->ss
;
1397 case VCPU_SREG_TR
: return &save
->tr
;
1398 case VCPU_SREG_LDTR
: return &save
->ldtr
;
1404 static u64
svm_get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
1406 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
1411 static void svm_get_segment(struct kvm_vcpu
*vcpu
,
1412 struct kvm_segment
*var
, int seg
)
1414 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
1416 var
->base
= s
->base
;
1417 var
->limit
= s
->limit
;
1418 var
->selector
= s
->selector
;
1419 var
->type
= s
->attrib
& SVM_SELECTOR_TYPE_MASK
;
1420 var
->s
= (s
->attrib
>> SVM_SELECTOR_S_SHIFT
) & 1;
1421 var
->dpl
= (s
->attrib
>> SVM_SELECTOR_DPL_SHIFT
) & 3;
1422 var
->present
= (s
->attrib
>> SVM_SELECTOR_P_SHIFT
) & 1;
1423 var
->avl
= (s
->attrib
>> SVM_SELECTOR_AVL_SHIFT
) & 1;
1424 var
->l
= (s
->attrib
>> SVM_SELECTOR_L_SHIFT
) & 1;
1425 var
->db
= (s
->attrib
>> SVM_SELECTOR_DB_SHIFT
) & 1;
1426 var
->g
= (s
->attrib
>> SVM_SELECTOR_G_SHIFT
) & 1;
1429 * AMD's VMCB does not have an explicit unusable field, so emulate it
1430 * for cross vendor migration purposes by "not present"
1432 var
->unusable
= !var
->present
|| (var
->type
== 0);
1437 * SVM always stores 0 for the 'G' bit in the CS selector in
1438 * the VMCB on a VMEXIT. This hurts cross-vendor migration:
1439 * Intel's VMENTRY has a check on the 'G' bit.
1441 var
->g
= s
->limit
> 0xfffff;
1445 * Work around a bug where the busy flag in the tr selector
1455 * The accessed bit must always be set in the segment
1456 * descriptor cache, although it can be cleared in the
1457 * descriptor, the cached bit always remains at 1. Since
1458 * Intel has a check on this, set it here to support
1459 * cross-vendor migration.
1466 * On AMD CPUs sometimes the DB bit in the segment
1467 * descriptor is left as 1, although the whole segment has
1468 * been made unusable. Clear it here to pass an Intel VMX
1469 * entry check when cross vendor migrating.
1477 static int svm_get_cpl(struct kvm_vcpu
*vcpu
)
1479 struct vmcb_save_area
*save
= &to_svm(vcpu
)->vmcb
->save
;
1484 static void svm_get_idt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1486 struct vcpu_svm
*svm
= to_svm(vcpu
);
1488 dt
->size
= svm
->vmcb
->save
.idtr
.limit
;
1489 dt
->address
= svm
->vmcb
->save
.idtr
.base
;
1492 static void svm_set_idt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1494 struct vcpu_svm
*svm
= to_svm(vcpu
);
1496 svm
->vmcb
->save
.idtr
.limit
= dt
->size
;
1497 svm
->vmcb
->save
.idtr
.base
= dt
->address
;
1498 mark_dirty(svm
->vmcb
, VMCB_DT
);
1501 static void svm_get_gdt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1503 struct vcpu_svm
*svm
= to_svm(vcpu
);
1505 dt
->size
= svm
->vmcb
->save
.gdtr
.limit
;
1506 dt
->address
= svm
->vmcb
->save
.gdtr
.base
;
1509 static void svm_set_gdt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1511 struct vcpu_svm
*svm
= to_svm(vcpu
);
1513 svm
->vmcb
->save
.gdtr
.limit
= dt
->size
;
1514 svm
->vmcb
->save
.gdtr
.base
= dt
->address
;
1515 mark_dirty(svm
->vmcb
, VMCB_DT
);
1518 static void svm_decache_cr0_guest_bits(struct kvm_vcpu
*vcpu
)
1522 static void svm_decache_cr3(struct kvm_vcpu
*vcpu
)
1526 static void svm_decache_cr4_guest_bits(struct kvm_vcpu
*vcpu
)
1530 static void update_cr0_intercept(struct vcpu_svm
*svm
)
1532 ulong gcr0
= svm
->vcpu
.arch
.cr0
;
1533 u64
*hcr0
= &svm
->vmcb
->save
.cr0
;
1535 if (!svm
->vcpu
.fpu_active
)
1536 *hcr0
|= SVM_CR0_SELECTIVE_MASK
;
1538 *hcr0
= (*hcr0
& ~SVM_CR0_SELECTIVE_MASK
)
1539 | (gcr0
& SVM_CR0_SELECTIVE_MASK
);
1541 mark_dirty(svm
->vmcb
, VMCB_CR
);
1543 if (gcr0
== *hcr0
&& svm
->vcpu
.fpu_active
) {
1544 clr_cr_intercept(svm
, INTERCEPT_CR0_READ
);
1545 clr_cr_intercept(svm
, INTERCEPT_CR0_WRITE
);
1547 set_cr_intercept(svm
, INTERCEPT_CR0_READ
);
1548 set_cr_intercept(svm
, INTERCEPT_CR0_WRITE
);
1552 static void svm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
1554 struct vcpu_svm
*svm
= to_svm(vcpu
);
1556 #ifdef CONFIG_X86_64
1557 if (vcpu
->arch
.efer
& EFER_LME
) {
1558 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
1559 vcpu
->arch
.efer
|= EFER_LMA
;
1560 svm
->vmcb
->save
.efer
|= EFER_LMA
| EFER_LME
;
1563 if (is_paging(vcpu
) && !(cr0
& X86_CR0_PG
)) {
1564 vcpu
->arch
.efer
&= ~EFER_LMA
;
1565 svm
->vmcb
->save
.efer
&= ~(EFER_LMA
| EFER_LME
);
1569 vcpu
->arch
.cr0
= cr0
;
1572 cr0
|= X86_CR0_PG
| X86_CR0_WP
;
1574 if (!vcpu
->fpu_active
)
1577 * re-enable caching here because the QEMU bios
1578 * does not do it - this results in some delay at
1581 cr0
&= ~(X86_CR0_CD
| X86_CR0_NW
);
1582 svm
->vmcb
->save
.cr0
= cr0
;
1583 mark_dirty(svm
->vmcb
, VMCB_CR
);
1584 update_cr0_intercept(svm
);
1587 static int svm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
1589 unsigned long host_cr4_mce
= read_cr4() & X86_CR4_MCE
;
1590 unsigned long old_cr4
= to_svm(vcpu
)->vmcb
->save
.cr4
;
1592 if (cr4
& X86_CR4_VMXE
)
1595 if (npt_enabled
&& ((old_cr4
^ cr4
) & X86_CR4_PGE
))
1596 svm_flush_tlb(vcpu
);
1598 vcpu
->arch
.cr4
= cr4
;
1601 cr4
|= host_cr4_mce
;
1602 to_svm(vcpu
)->vmcb
->save
.cr4
= cr4
;
1603 mark_dirty(to_svm(vcpu
)->vmcb
, VMCB_CR
);
1607 static void svm_set_segment(struct kvm_vcpu
*vcpu
,
1608 struct kvm_segment
*var
, int seg
)
1610 struct vcpu_svm
*svm
= to_svm(vcpu
);
1611 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
1613 s
->base
= var
->base
;
1614 s
->limit
= var
->limit
;
1615 s
->selector
= var
->selector
;
1619 s
->attrib
= (var
->type
& SVM_SELECTOR_TYPE_MASK
);
1620 s
->attrib
|= (var
->s
& 1) << SVM_SELECTOR_S_SHIFT
;
1621 s
->attrib
|= (var
->dpl
& 3) << SVM_SELECTOR_DPL_SHIFT
;
1622 s
->attrib
|= (var
->present
& 1) << SVM_SELECTOR_P_SHIFT
;
1623 s
->attrib
|= (var
->avl
& 1) << SVM_SELECTOR_AVL_SHIFT
;
1624 s
->attrib
|= (var
->l
& 1) << SVM_SELECTOR_L_SHIFT
;
1625 s
->attrib
|= (var
->db
& 1) << SVM_SELECTOR_DB_SHIFT
;
1626 s
->attrib
|= (var
->g
& 1) << SVM_SELECTOR_G_SHIFT
;
1628 if (seg
== VCPU_SREG_CS
)
1629 svm_update_cpl(vcpu
);
1631 mark_dirty(svm
->vmcb
, VMCB_SEG
);
1634 static void update_db_intercept(struct kvm_vcpu
*vcpu
)
1636 struct vcpu_svm
*svm
= to_svm(vcpu
);
1638 clr_exception_intercept(svm
, DB_VECTOR
);
1639 clr_exception_intercept(svm
, BP_VECTOR
);
1641 if (svm
->nmi_singlestep
)
1642 set_exception_intercept(svm
, DB_VECTOR
);
1644 if (vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
) {
1645 if (vcpu
->guest_debug
&
1646 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
))
1647 set_exception_intercept(svm
, DB_VECTOR
);
1648 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_SW_BP
)
1649 set_exception_intercept(svm
, BP_VECTOR
);
1651 vcpu
->guest_debug
= 0;
1654 static void svm_guest_debug(struct kvm_vcpu
*vcpu
, struct kvm_guest_debug
*dbg
)
1656 struct vcpu_svm
*svm
= to_svm(vcpu
);
1658 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
1659 svm
->vmcb
->save
.dr7
= dbg
->arch
.debugreg
[7];
1661 svm
->vmcb
->save
.dr7
= vcpu
->arch
.dr7
;
1663 mark_dirty(svm
->vmcb
, VMCB_DR
);
1665 update_db_intercept(vcpu
);
1668 static void new_asid(struct vcpu_svm
*svm
, struct svm_cpu_data
*sd
)
1670 if (sd
->next_asid
> sd
->max_asid
) {
1671 ++sd
->asid_generation
;
1673 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ALL_ASID
;
1676 svm
->asid_generation
= sd
->asid_generation
;
1677 svm
->vmcb
->control
.asid
= sd
->next_asid
++;
1679 mark_dirty(svm
->vmcb
, VMCB_ASID
);
1682 static void svm_set_dr7(struct kvm_vcpu
*vcpu
, unsigned long value
)
1684 struct vcpu_svm
*svm
= to_svm(vcpu
);
1686 svm
->vmcb
->save
.dr7
= value
;
1687 mark_dirty(svm
->vmcb
, VMCB_DR
);
1690 static int pf_interception(struct vcpu_svm
*svm
)
1692 u64 fault_address
= svm
->vmcb
->control
.exit_info_2
;
1696 switch (svm
->apf_reason
) {
1698 error_code
= svm
->vmcb
->control
.exit_info_1
;
1700 trace_kvm_page_fault(fault_address
, error_code
);
1701 if (!npt_enabled
&& kvm_event_needs_reinjection(&svm
->vcpu
))
1702 kvm_mmu_unprotect_page_virt(&svm
->vcpu
, fault_address
);
1703 r
= kvm_mmu_page_fault(&svm
->vcpu
, fault_address
, error_code
,
1704 svm
->vmcb
->control
.insn_bytes
,
1705 svm
->vmcb
->control
.insn_len
);
1707 case KVM_PV_REASON_PAGE_NOT_PRESENT
:
1708 svm
->apf_reason
= 0;
1709 local_irq_disable();
1710 kvm_async_pf_task_wait(fault_address
);
1713 case KVM_PV_REASON_PAGE_READY
:
1714 svm
->apf_reason
= 0;
1715 local_irq_disable();
1716 kvm_async_pf_task_wake(fault_address
);
1723 static int db_interception(struct vcpu_svm
*svm
)
1725 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
1727 if (!(svm
->vcpu
.guest_debug
&
1728 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
)) &&
1729 !svm
->nmi_singlestep
) {
1730 kvm_queue_exception(&svm
->vcpu
, DB_VECTOR
);
1734 if (svm
->nmi_singlestep
) {
1735 svm
->nmi_singlestep
= false;
1736 if (!(svm
->vcpu
.guest_debug
& KVM_GUESTDBG_SINGLESTEP
))
1737 svm
->vmcb
->save
.rflags
&=
1738 ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
1739 update_db_intercept(&svm
->vcpu
);
1742 if (svm
->vcpu
.guest_debug
&
1743 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
)) {
1744 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
1745 kvm_run
->debug
.arch
.pc
=
1746 svm
->vmcb
->save
.cs
.base
+ svm
->vmcb
->save
.rip
;
1747 kvm_run
->debug
.arch
.exception
= DB_VECTOR
;
1754 static int bp_interception(struct vcpu_svm
*svm
)
1756 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
1758 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
1759 kvm_run
->debug
.arch
.pc
= svm
->vmcb
->save
.cs
.base
+ svm
->vmcb
->save
.rip
;
1760 kvm_run
->debug
.arch
.exception
= BP_VECTOR
;
1764 static int ud_interception(struct vcpu_svm
*svm
)
1768 er
= emulate_instruction(&svm
->vcpu
, EMULTYPE_TRAP_UD
);
1769 if (er
!= EMULATE_DONE
)
1770 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
1774 static void svm_fpu_activate(struct kvm_vcpu
*vcpu
)
1776 struct vcpu_svm
*svm
= to_svm(vcpu
);
1778 clr_exception_intercept(svm
, NM_VECTOR
);
1780 svm
->vcpu
.fpu_active
= 1;
1781 update_cr0_intercept(svm
);
1784 static int nm_interception(struct vcpu_svm
*svm
)
1786 svm_fpu_activate(&svm
->vcpu
);
1790 static bool is_erratum_383(void)
1795 if (!erratum_383_found
)
1798 value
= native_read_msr_safe(MSR_IA32_MC0_STATUS
, &err
);
1802 /* Bit 62 may or may not be set for this mce */
1803 value
&= ~(1ULL << 62);
1805 if (value
!= 0xb600000000010015ULL
)
1808 /* Clear MCi_STATUS registers */
1809 for (i
= 0; i
< 6; ++i
)
1810 native_write_msr_safe(MSR_IA32_MCx_STATUS(i
), 0, 0);
1812 value
= native_read_msr_safe(MSR_IA32_MCG_STATUS
, &err
);
1816 value
&= ~(1ULL << 2);
1817 low
= lower_32_bits(value
);
1818 high
= upper_32_bits(value
);
1820 native_write_msr_safe(MSR_IA32_MCG_STATUS
, low
, high
);
1823 /* Flush tlb to evict multi-match entries */
1829 static void svm_handle_mce(struct vcpu_svm
*svm
)
1831 if (is_erratum_383()) {
1833 * Erratum 383 triggered. Guest state is corrupt so kill the
1836 pr_err("KVM: Guest triggered AMD Erratum 383\n");
1838 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, &svm
->vcpu
);
1844 * On an #MC intercept the MCE handler is not called automatically in
1845 * the host. So do it by hand here.
1849 /* not sure if we ever come back to this point */
1854 static int mc_interception(struct vcpu_svm
*svm
)
1859 static int shutdown_interception(struct vcpu_svm
*svm
)
1861 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
1864 * VMCB is undefined after a SHUTDOWN intercept
1865 * so reinitialize it.
1867 clear_page(svm
->vmcb
);
1870 kvm_run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
1874 static int io_interception(struct vcpu_svm
*svm
)
1876 struct kvm_vcpu
*vcpu
= &svm
->vcpu
;
1877 u32 io_info
= svm
->vmcb
->control
.exit_info_1
; /* address size bug? */
1878 int size
, in
, string
;
1881 ++svm
->vcpu
.stat
.io_exits
;
1882 string
= (io_info
& SVM_IOIO_STR_MASK
) != 0;
1883 in
= (io_info
& SVM_IOIO_TYPE_MASK
) != 0;
1885 return emulate_instruction(vcpu
, 0) == EMULATE_DONE
;
1887 port
= io_info
>> 16;
1888 size
= (io_info
& SVM_IOIO_SIZE_MASK
) >> SVM_IOIO_SIZE_SHIFT
;
1889 svm
->next_rip
= svm
->vmcb
->control
.exit_info_2
;
1890 skip_emulated_instruction(&svm
->vcpu
);
1892 return kvm_fast_pio_out(vcpu
, size
, port
);
1895 static int nmi_interception(struct vcpu_svm
*svm
)
1900 static int intr_interception(struct vcpu_svm
*svm
)
1902 ++svm
->vcpu
.stat
.irq_exits
;
1906 static int nop_on_interception(struct vcpu_svm
*svm
)
1911 static int halt_interception(struct vcpu_svm
*svm
)
1913 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 1;
1914 skip_emulated_instruction(&svm
->vcpu
);
1915 return kvm_emulate_halt(&svm
->vcpu
);
1918 static int vmmcall_interception(struct vcpu_svm
*svm
)
1920 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1921 skip_emulated_instruction(&svm
->vcpu
);
1922 kvm_emulate_hypercall(&svm
->vcpu
);
1926 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu
*vcpu
)
1928 struct vcpu_svm
*svm
= to_svm(vcpu
);
1930 return svm
->nested
.nested_cr3
;
1933 static u64
nested_svm_get_tdp_pdptr(struct kvm_vcpu
*vcpu
, int index
)
1935 struct vcpu_svm
*svm
= to_svm(vcpu
);
1936 u64 cr3
= svm
->nested
.nested_cr3
;
1940 ret
= kvm_read_guest_page(vcpu
->kvm
, gpa_to_gfn(cr3
), &pdpte
,
1941 offset_in_page(cr3
) + index
* 8, 8);
1947 static void nested_svm_set_tdp_cr3(struct kvm_vcpu
*vcpu
,
1950 struct vcpu_svm
*svm
= to_svm(vcpu
);
1952 svm
->vmcb
->control
.nested_cr3
= root
;
1953 mark_dirty(svm
->vmcb
, VMCB_NPT
);
1954 svm_flush_tlb(vcpu
);
1957 static void nested_svm_inject_npf_exit(struct kvm_vcpu
*vcpu
,
1958 struct x86_exception
*fault
)
1960 struct vcpu_svm
*svm
= to_svm(vcpu
);
1962 svm
->vmcb
->control
.exit_code
= SVM_EXIT_NPF
;
1963 svm
->vmcb
->control
.exit_code_hi
= 0;
1964 svm
->vmcb
->control
.exit_info_1
= fault
->error_code
;
1965 svm
->vmcb
->control
.exit_info_2
= fault
->address
;
1967 nested_svm_vmexit(svm
);
1970 static int nested_svm_init_mmu_context(struct kvm_vcpu
*vcpu
)
1974 r
= kvm_init_shadow_mmu(vcpu
, &vcpu
->arch
.mmu
);
1976 vcpu
->arch
.mmu
.set_cr3
= nested_svm_set_tdp_cr3
;
1977 vcpu
->arch
.mmu
.get_cr3
= nested_svm_get_tdp_cr3
;
1978 vcpu
->arch
.mmu
.get_pdptr
= nested_svm_get_tdp_pdptr
;
1979 vcpu
->arch
.mmu
.inject_page_fault
= nested_svm_inject_npf_exit
;
1980 vcpu
->arch
.mmu
.shadow_root_level
= get_npt_level();
1981 vcpu
->arch
.walk_mmu
= &vcpu
->arch
.nested_mmu
;
1986 static void nested_svm_uninit_mmu_context(struct kvm_vcpu
*vcpu
)
1988 vcpu
->arch
.walk_mmu
= &vcpu
->arch
.mmu
;
1991 static int nested_svm_check_permissions(struct vcpu_svm
*svm
)
1993 if (!(svm
->vcpu
.arch
.efer
& EFER_SVME
)
1994 || !is_paging(&svm
->vcpu
)) {
1995 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
1999 if (svm
->vmcb
->save
.cpl
) {
2000 kvm_inject_gp(&svm
->vcpu
, 0);
2007 static int nested_svm_check_exception(struct vcpu_svm
*svm
, unsigned nr
,
2008 bool has_error_code
, u32 error_code
)
2012 if (!is_guest_mode(&svm
->vcpu
))
2015 svm
->vmcb
->control
.exit_code
= SVM_EXIT_EXCP_BASE
+ nr
;
2016 svm
->vmcb
->control
.exit_code_hi
= 0;
2017 svm
->vmcb
->control
.exit_info_1
= error_code
;
2018 svm
->vmcb
->control
.exit_info_2
= svm
->vcpu
.arch
.cr2
;
2020 vmexit
= nested_svm_intercept(svm
);
2021 if (vmexit
== NESTED_EXIT_DONE
)
2022 svm
->nested
.exit_required
= true;
2027 /* This function returns true if it is save to enable the irq window */
2028 static inline bool nested_svm_intr(struct vcpu_svm
*svm
)
2030 if (!is_guest_mode(&svm
->vcpu
))
2033 if (!(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
))
2036 if (!(svm
->vcpu
.arch
.hflags
& HF_HIF_MASK
))
2040 * if vmexit was already requested (by intercepted exception
2041 * for instance) do not overwrite it with "external interrupt"
2044 if (svm
->nested
.exit_required
)
2047 svm
->vmcb
->control
.exit_code
= SVM_EXIT_INTR
;
2048 svm
->vmcb
->control
.exit_info_1
= 0;
2049 svm
->vmcb
->control
.exit_info_2
= 0;
2051 if (svm
->nested
.intercept
& 1ULL) {
2053 * The #vmexit can't be emulated here directly because this
2054 * code path runs with irqs and preemtion disabled. A
2055 * #vmexit emulation might sleep. Only signal request for
2058 svm
->nested
.exit_required
= true;
2059 trace_kvm_nested_intr_vmexit(svm
->vmcb
->save
.rip
);
2066 /* This function returns true if it is save to enable the nmi window */
2067 static inline bool nested_svm_nmi(struct vcpu_svm
*svm
)
2069 if (!is_guest_mode(&svm
->vcpu
))
2072 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_NMI
)))
2075 svm
->vmcb
->control
.exit_code
= SVM_EXIT_NMI
;
2076 svm
->nested
.exit_required
= true;
2081 static void *nested_svm_map(struct vcpu_svm
*svm
, u64 gpa
, struct page
**_page
)
2087 page
= gfn_to_page(svm
->vcpu
.kvm
, gpa
>> PAGE_SHIFT
);
2088 if (is_error_page(page
))
2096 kvm_release_page_clean(page
);
2097 kvm_inject_gp(&svm
->vcpu
, 0);
2102 static void nested_svm_unmap(struct page
*page
)
2105 kvm_release_page_dirty(page
);
2108 static int nested_svm_intercept_ioio(struct vcpu_svm
*svm
)
2114 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_IOIO_PROT
)))
2115 return NESTED_EXIT_HOST
;
2117 port
= svm
->vmcb
->control
.exit_info_1
>> 16;
2118 gpa
= svm
->nested
.vmcb_iopm
+ (port
/ 8);
2122 if (kvm_read_guest(svm
->vcpu
.kvm
, gpa
, &val
, 1))
2125 return val
? NESTED_EXIT_DONE
: NESTED_EXIT_HOST
;
2128 static int nested_svm_exit_handled_msr(struct vcpu_svm
*svm
)
2130 u32 offset
, msr
, value
;
2133 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_MSR_PROT
)))
2134 return NESTED_EXIT_HOST
;
2136 msr
= svm
->vcpu
.arch
.regs
[VCPU_REGS_RCX
];
2137 offset
= svm_msrpm_offset(msr
);
2138 write
= svm
->vmcb
->control
.exit_info_1
& 1;
2139 mask
= 1 << ((2 * (msr
& 0xf)) + write
);
2141 if (offset
== MSR_INVALID
)
2142 return NESTED_EXIT_DONE
;
2144 /* Offset is in 32 bit units but need in 8 bit units */
2147 if (kvm_read_guest(svm
->vcpu
.kvm
, svm
->nested
.vmcb_msrpm
+ offset
, &value
, 4))
2148 return NESTED_EXIT_DONE
;
2150 return (value
& mask
) ? NESTED_EXIT_DONE
: NESTED_EXIT_HOST
;
2153 static int nested_svm_exit_special(struct vcpu_svm
*svm
)
2155 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
2157 switch (exit_code
) {
2160 case SVM_EXIT_EXCP_BASE
+ MC_VECTOR
:
2161 return NESTED_EXIT_HOST
;
2163 /* For now we are always handling NPFs when using them */
2165 return NESTED_EXIT_HOST
;
2167 case SVM_EXIT_EXCP_BASE
+ PF_VECTOR
:
2168 /* When we're shadowing, trap PFs, but not async PF */
2169 if (!npt_enabled
&& svm
->apf_reason
== 0)
2170 return NESTED_EXIT_HOST
;
2172 case SVM_EXIT_EXCP_BASE
+ NM_VECTOR
:
2173 nm_interception(svm
);
2179 return NESTED_EXIT_CONTINUE
;
2183 * If this function returns true, this #vmexit was already handled
2185 static int nested_svm_intercept(struct vcpu_svm
*svm
)
2187 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
2188 int vmexit
= NESTED_EXIT_HOST
;
2190 switch (exit_code
) {
2192 vmexit
= nested_svm_exit_handled_msr(svm
);
2195 vmexit
= nested_svm_intercept_ioio(svm
);
2197 case SVM_EXIT_READ_CR0
... SVM_EXIT_WRITE_CR8
: {
2198 u32 bit
= 1U << (exit_code
- SVM_EXIT_READ_CR0
);
2199 if (svm
->nested
.intercept_cr
& bit
)
2200 vmexit
= NESTED_EXIT_DONE
;
2203 case SVM_EXIT_READ_DR0
... SVM_EXIT_WRITE_DR7
: {
2204 u32 bit
= 1U << (exit_code
- SVM_EXIT_READ_DR0
);
2205 if (svm
->nested
.intercept_dr
& bit
)
2206 vmexit
= NESTED_EXIT_DONE
;
2209 case SVM_EXIT_EXCP_BASE
... SVM_EXIT_EXCP_BASE
+ 0x1f: {
2210 u32 excp_bits
= 1 << (exit_code
- SVM_EXIT_EXCP_BASE
);
2211 if (svm
->nested
.intercept_exceptions
& excp_bits
)
2212 vmexit
= NESTED_EXIT_DONE
;
2213 /* async page fault always cause vmexit */
2214 else if ((exit_code
== SVM_EXIT_EXCP_BASE
+ PF_VECTOR
) &&
2215 svm
->apf_reason
!= 0)
2216 vmexit
= NESTED_EXIT_DONE
;
2219 case SVM_EXIT_ERR
: {
2220 vmexit
= NESTED_EXIT_DONE
;
2224 u64 exit_bits
= 1ULL << (exit_code
- SVM_EXIT_INTR
);
2225 if (svm
->nested
.intercept
& exit_bits
)
2226 vmexit
= NESTED_EXIT_DONE
;
2233 static int nested_svm_exit_handled(struct vcpu_svm
*svm
)
2237 vmexit
= nested_svm_intercept(svm
);
2239 if (vmexit
== NESTED_EXIT_DONE
)
2240 nested_svm_vmexit(svm
);
2245 static inline void copy_vmcb_control_area(struct vmcb
*dst_vmcb
, struct vmcb
*from_vmcb
)
2247 struct vmcb_control_area
*dst
= &dst_vmcb
->control
;
2248 struct vmcb_control_area
*from
= &from_vmcb
->control
;
2250 dst
->intercept_cr
= from
->intercept_cr
;
2251 dst
->intercept_dr
= from
->intercept_dr
;
2252 dst
->intercept_exceptions
= from
->intercept_exceptions
;
2253 dst
->intercept
= from
->intercept
;
2254 dst
->iopm_base_pa
= from
->iopm_base_pa
;
2255 dst
->msrpm_base_pa
= from
->msrpm_base_pa
;
2256 dst
->tsc_offset
= from
->tsc_offset
;
2257 dst
->asid
= from
->asid
;
2258 dst
->tlb_ctl
= from
->tlb_ctl
;
2259 dst
->int_ctl
= from
->int_ctl
;
2260 dst
->int_vector
= from
->int_vector
;
2261 dst
->int_state
= from
->int_state
;
2262 dst
->exit_code
= from
->exit_code
;
2263 dst
->exit_code_hi
= from
->exit_code_hi
;
2264 dst
->exit_info_1
= from
->exit_info_1
;
2265 dst
->exit_info_2
= from
->exit_info_2
;
2266 dst
->exit_int_info
= from
->exit_int_info
;
2267 dst
->exit_int_info_err
= from
->exit_int_info_err
;
2268 dst
->nested_ctl
= from
->nested_ctl
;
2269 dst
->event_inj
= from
->event_inj
;
2270 dst
->event_inj_err
= from
->event_inj_err
;
2271 dst
->nested_cr3
= from
->nested_cr3
;
2272 dst
->lbr_ctl
= from
->lbr_ctl
;
2275 static int nested_svm_vmexit(struct vcpu_svm
*svm
)
2277 struct vmcb
*nested_vmcb
;
2278 struct vmcb
*hsave
= svm
->nested
.hsave
;
2279 struct vmcb
*vmcb
= svm
->vmcb
;
2282 trace_kvm_nested_vmexit_inject(vmcb
->control
.exit_code
,
2283 vmcb
->control
.exit_info_1
,
2284 vmcb
->control
.exit_info_2
,
2285 vmcb
->control
.exit_int_info
,
2286 vmcb
->control
.exit_int_info_err
,
2289 nested_vmcb
= nested_svm_map(svm
, svm
->nested
.vmcb
, &page
);
2293 /* Exit Guest-Mode */
2294 leave_guest_mode(&svm
->vcpu
);
2295 svm
->nested
.vmcb
= 0;
2297 /* Give the current vmcb to the guest */
2300 nested_vmcb
->save
.es
= vmcb
->save
.es
;
2301 nested_vmcb
->save
.cs
= vmcb
->save
.cs
;
2302 nested_vmcb
->save
.ss
= vmcb
->save
.ss
;
2303 nested_vmcb
->save
.ds
= vmcb
->save
.ds
;
2304 nested_vmcb
->save
.gdtr
= vmcb
->save
.gdtr
;
2305 nested_vmcb
->save
.idtr
= vmcb
->save
.idtr
;
2306 nested_vmcb
->save
.efer
= svm
->vcpu
.arch
.efer
;
2307 nested_vmcb
->save
.cr0
= kvm_read_cr0(&svm
->vcpu
);
2308 nested_vmcb
->save
.cr3
= kvm_read_cr3(&svm
->vcpu
);
2309 nested_vmcb
->save
.cr2
= vmcb
->save
.cr2
;
2310 nested_vmcb
->save
.cr4
= svm
->vcpu
.arch
.cr4
;
2311 nested_vmcb
->save
.rflags
= kvm_get_rflags(&svm
->vcpu
);
2312 nested_vmcb
->save
.rip
= vmcb
->save
.rip
;
2313 nested_vmcb
->save
.rsp
= vmcb
->save
.rsp
;
2314 nested_vmcb
->save
.rax
= vmcb
->save
.rax
;
2315 nested_vmcb
->save
.dr7
= vmcb
->save
.dr7
;
2316 nested_vmcb
->save
.dr6
= vmcb
->save
.dr6
;
2317 nested_vmcb
->save
.cpl
= vmcb
->save
.cpl
;
2319 nested_vmcb
->control
.int_ctl
= vmcb
->control
.int_ctl
;
2320 nested_vmcb
->control
.int_vector
= vmcb
->control
.int_vector
;
2321 nested_vmcb
->control
.int_state
= vmcb
->control
.int_state
;
2322 nested_vmcb
->control
.exit_code
= vmcb
->control
.exit_code
;
2323 nested_vmcb
->control
.exit_code_hi
= vmcb
->control
.exit_code_hi
;
2324 nested_vmcb
->control
.exit_info_1
= vmcb
->control
.exit_info_1
;
2325 nested_vmcb
->control
.exit_info_2
= vmcb
->control
.exit_info_2
;
2326 nested_vmcb
->control
.exit_int_info
= vmcb
->control
.exit_int_info
;
2327 nested_vmcb
->control
.exit_int_info_err
= vmcb
->control
.exit_int_info_err
;
2328 nested_vmcb
->control
.next_rip
= vmcb
->control
.next_rip
;
2331 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2332 * to make sure that we do not lose injected events. So check event_inj
2333 * here and copy it to exit_int_info if it is valid.
2334 * Exit_int_info and event_inj can't be both valid because the case
2335 * below only happens on a VMRUN instruction intercept which has
2336 * no valid exit_int_info set.
2338 if (vmcb
->control
.event_inj
& SVM_EVTINJ_VALID
) {
2339 struct vmcb_control_area
*nc
= &nested_vmcb
->control
;
2341 nc
->exit_int_info
= vmcb
->control
.event_inj
;
2342 nc
->exit_int_info_err
= vmcb
->control
.event_inj_err
;
2345 nested_vmcb
->control
.tlb_ctl
= 0;
2346 nested_vmcb
->control
.event_inj
= 0;
2347 nested_vmcb
->control
.event_inj_err
= 0;
2349 /* We always set V_INTR_MASKING and remember the old value in hflags */
2350 if (!(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
))
2351 nested_vmcb
->control
.int_ctl
&= ~V_INTR_MASKING_MASK
;
2353 /* Restore the original control entries */
2354 copy_vmcb_control_area(vmcb
, hsave
);
2356 kvm_clear_exception_queue(&svm
->vcpu
);
2357 kvm_clear_interrupt_queue(&svm
->vcpu
);
2359 svm
->nested
.nested_cr3
= 0;
2361 /* Restore selected save entries */
2362 svm
->vmcb
->save
.es
= hsave
->save
.es
;
2363 svm
->vmcb
->save
.cs
= hsave
->save
.cs
;
2364 svm
->vmcb
->save
.ss
= hsave
->save
.ss
;
2365 svm
->vmcb
->save
.ds
= hsave
->save
.ds
;
2366 svm
->vmcb
->save
.gdtr
= hsave
->save
.gdtr
;
2367 svm
->vmcb
->save
.idtr
= hsave
->save
.idtr
;
2368 kvm_set_rflags(&svm
->vcpu
, hsave
->save
.rflags
);
2369 svm_set_efer(&svm
->vcpu
, hsave
->save
.efer
);
2370 svm_set_cr0(&svm
->vcpu
, hsave
->save
.cr0
| X86_CR0_PE
);
2371 svm_set_cr4(&svm
->vcpu
, hsave
->save
.cr4
);
2373 svm
->vmcb
->save
.cr3
= hsave
->save
.cr3
;
2374 svm
->vcpu
.arch
.cr3
= hsave
->save
.cr3
;
2376 (void)kvm_set_cr3(&svm
->vcpu
, hsave
->save
.cr3
);
2378 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
, hsave
->save
.rax
);
2379 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RSP
, hsave
->save
.rsp
);
2380 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RIP
, hsave
->save
.rip
);
2381 svm
->vmcb
->save
.dr7
= 0;
2382 svm
->vmcb
->save
.cpl
= 0;
2383 svm
->vmcb
->control
.exit_int_info
= 0;
2385 mark_all_dirty(svm
->vmcb
);
2387 nested_svm_unmap(page
);
2389 nested_svm_uninit_mmu_context(&svm
->vcpu
);
2390 kvm_mmu_reset_context(&svm
->vcpu
);
2391 kvm_mmu_load(&svm
->vcpu
);
2396 static bool nested_svm_vmrun_msrpm(struct vcpu_svm
*svm
)
2399 * This function merges the msr permission bitmaps of kvm and the
2400 * nested vmcb. It is omptimized in that it only merges the parts where
2401 * the kvm msr permission bitmap may contain zero bits
2405 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_MSR_PROT
)))
2408 for (i
= 0; i
< MSRPM_OFFSETS
; i
++) {
2412 if (msrpm_offsets
[i
] == 0xffffffff)
2415 p
= msrpm_offsets
[i
];
2416 offset
= svm
->nested
.vmcb_msrpm
+ (p
* 4);
2418 if (kvm_read_guest(svm
->vcpu
.kvm
, offset
, &value
, 4))
2421 svm
->nested
.msrpm
[p
] = svm
->msrpm
[p
] | value
;
2424 svm
->vmcb
->control
.msrpm_base_pa
= __pa(svm
->nested
.msrpm
);
2429 static bool nested_vmcb_checks(struct vmcb
*vmcb
)
2431 if ((vmcb
->control
.intercept
& (1ULL << INTERCEPT_VMRUN
)) == 0)
2434 if (vmcb
->control
.asid
== 0)
2437 if (vmcb
->control
.nested_ctl
&& !npt_enabled
)
2443 static bool nested_svm_vmrun(struct vcpu_svm
*svm
)
2445 struct vmcb
*nested_vmcb
;
2446 struct vmcb
*hsave
= svm
->nested
.hsave
;
2447 struct vmcb
*vmcb
= svm
->vmcb
;
2451 vmcb_gpa
= svm
->vmcb
->save
.rax
;
2453 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, &page
);
2457 if (!nested_vmcb_checks(nested_vmcb
)) {
2458 nested_vmcb
->control
.exit_code
= SVM_EXIT_ERR
;
2459 nested_vmcb
->control
.exit_code_hi
= 0;
2460 nested_vmcb
->control
.exit_info_1
= 0;
2461 nested_vmcb
->control
.exit_info_2
= 0;
2463 nested_svm_unmap(page
);
2468 trace_kvm_nested_vmrun(svm
->vmcb
->save
.rip
, vmcb_gpa
,
2469 nested_vmcb
->save
.rip
,
2470 nested_vmcb
->control
.int_ctl
,
2471 nested_vmcb
->control
.event_inj
,
2472 nested_vmcb
->control
.nested_ctl
);
2474 trace_kvm_nested_intercepts(nested_vmcb
->control
.intercept_cr
& 0xffff,
2475 nested_vmcb
->control
.intercept_cr
>> 16,
2476 nested_vmcb
->control
.intercept_exceptions
,
2477 nested_vmcb
->control
.intercept
);
2479 /* Clear internal status */
2480 kvm_clear_exception_queue(&svm
->vcpu
);
2481 kvm_clear_interrupt_queue(&svm
->vcpu
);
2484 * Save the old vmcb, so we don't need to pick what we save, but can
2485 * restore everything when a VMEXIT occurs
2487 hsave
->save
.es
= vmcb
->save
.es
;
2488 hsave
->save
.cs
= vmcb
->save
.cs
;
2489 hsave
->save
.ss
= vmcb
->save
.ss
;
2490 hsave
->save
.ds
= vmcb
->save
.ds
;
2491 hsave
->save
.gdtr
= vmcb
->save
.gdtr
;
2492 hsave
->save
.idtr
= vmcb
->save
.idtr
;
2493 hsave
->save
.efer
= svm
->vcpu
.arch
.efer
;
2494 hsave
->save
.cr0
= kvm_read_cr0(&svm
->vcpu
);
2495 hsave
->save
.cr4
= svm
->vcpu
.arch
.cr4
;
2496 hsave
->save
.rflags
= kvm_get_rflags(&svm
->vcpu
);
2497 hsave
->save
.rip
= kvm_rip_read(&svm
->vcpu
);
2498 hsave
->save
.rsp
= vmcb
->save
.rsp
;
2499 hsave
->save
.rax
= vmcb
->save
.rax
;
2501 hsave
->save
.cr3
= vmcb
->save
.cr3
;
2503 hsave
->save
.cr3
= kvm_read_cr3(&svm
->vcpu
);
2505 copy_vmcb_control_area(hsave
, vmcb
);
2507 if (kvm_get_rflags(&svm
->vcpu
) & X86_EFLAGS_IF
)
2508 svm
->vcpu
.arch
.hflags
|= HF_HIF_MASK
;
2510 svm
->vcpu
.arch
.hflags
&= ~HF_HIF_MASK
;
2512 if (nested_vmcb
->control
.nested_ctl
) {
2513 kvm_mmu_unload(&svm
->vcpu
);
2514 svm
->nested
.nested_cr3
= nested_vmcb
->control
.nested_cr3
;
2515 nested_svm_init_mmu_context(&svm
->vcpu
);
2518 /* Load the nested guest state */
2519 svm
->vmcb
->save
.es
= nested_vmcb
->save
.es
;
2520 svm
->vmcb
->save
.cs
= nested_vmcb
->save
.cs
;
2521 svm
->vmcb
->save
.ss
= nested_vmcb
->save
.ss
;
2522 svm
->vmcb
->save
.ds
= nested_vmcb
->save
.ds
;
2523 svm
->vmcb
->save
.gdtr
= nested_vmcb
->save
.gdtr
;
2524 svm
->vmcb
->save
.idtr
= nested_vmcb
->save
.idtr
;
2525 kvm_set_rflags(&svm
->vcpu
, nested_vmcb
->save
.rflags
);
2526 svm_set_efer(&svm
->vcpu
, nested_vmcb
->save
.efer
);
2527 svm_set_cr0(&svm
->vcpu
, nested_vmcb
->save
.cr0
);
2528 svm_set_cr4(&svm
->vcpu
, nested_vmcb
->save
.cr4
);
2530 svm
->vmcb
->save
.cr3
= nested_vmcb
->save
.cr3
;
2531 svm
->vcpu
.arch
.cr3
= nested_vmcb
->save
.cr3
;
2533 (void)kvm_set_cr3(&svm
->vcpu
, nested_vmcb
->save
.cr3
);
2535 /* Guest paging mode is active - reset mmu */
2536 kvm_mmu_reset_context(&svm
->vcpu
);
2538 svm
->vmcb
->save
.cr2
= svm
->vcpu
.arch
.cr2
= nested_vmcb
->save
.cr2
;
2539 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
, nested_vmcb
->save
.rax
);
2540 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RSP
, nested_vmcb
->save
.rsp
);
2541 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RIP
, nested_vmcb
->save
.rip
);
2543 /* In case we don't even reach vcpu_run, the fields are not updated */
2544 svm
->vmcb
->save
.rax
= nested_vmcb
->save
.rax
;
2545 svm
->vmcb
->save
.rsp
= nested_vmcb
->save
.rsp
;
2546 svm
->vmcb
->save
.rip
= nested_vmcb
->save
.rip
;
2547 svm
->vmcb
->save
.dr7
= nested_vmcb
->save
.dr7
;
2548 svm
->vmcb
->save
.dr6
= nested_vmcb
->save
.dr6
;
2549 svm
->vmcb
->save
.cpl
= nested_vmcb
->save
.cpl
;
2551 svm
->nested
.vmcb_msrpm
= nested_vmcb
->control
.msrpm_base_pa
& ~0x0fffULL
;
2552 svm
->nested
.vmcb_iopm
= nested_vmcb
->control
.iopm_base_pa
& ~0x0fffULL
;
2554 /* cache intercepts */
2555 svm
->nested
.intercept_cr
= nested_vmcb
->control
.intercept_cr
;
2556 svm
->nested
.intercept_dr
= nested_vmcb
->control
.intercept_dr
;
2557 svm
->nested
.intercept_exceptions
= nested_vmcb
->control
.intercept_exceptions
;
2558 svm
->nested
.intercept
= nested_vmcb
->control
.intercept
;
2560 svm_flush_tlb(&svm
->vcpu
);
2561 svm
->vmcb
->control
.int_ctl
= nested_vmcb
->control
.int_ctl
| V_INTR_MASKING_MASK
;
2562 if (nested_vmcb
->control
.int_ctl
& V_INTR_MASKING_MASK
)
2563 svm
->vcpu
.arch
.hflags
|= HF_VINTR_MASK
;
2565 svm
->vcpu
.arch
.hflags
&= ~HF_VINTR_MASK
;
2567 if (svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
) {
2568 /* We only want the cr8 intercept bits of the guest */
2569 clr_cr_intercept(svm
, INTERCEPT_CR8_READ
);
2570 clr_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
2573 /* We don't want to see VMMCALLs from a nested guest */
2574 clr_intercept(svm
, INTERCEPT_VMMCALL
);
2576 svm
->vmcb
->control
.lbr_ctl
= nested_vmcb
->control
.lbr_ctl
;
2577 svm
->vmcb
->control
.int_vector
= nested_vmcb
->control
.int_vector
;
2578 svm
->vmcb
->control
.int_state
= nested_vmcb
->control
.int_state
;
2579 svm
->vmcb
->control
.tsc_offset
+= nested_vmcb
->control
.tsc_offset
;
2580 svm
->vmcb
->control
.event_inj
= nested_vmcb
->control
.event_inj
;
2581 svm
->vmcb
->control
.event_inj_err
= nested_vmcb
->control
.event_inj_err
;
2583 nested_svm_unmap(page
);
2585 /* Enter Guest-Mode */
2586 enter_guest_mode(&svm
->vcpu
);
2589 * Merge guest and host intercepts - must be called with vcpu in
2590 * guest-mode to take affect here
2592 recalc_intercepts(svm
);
2594 svm
->nested
.vmcb
= vmcb_gpa
;
2598 mark_all_dirty(svm
->vmcb
);
2603 static void nested_svm_vmloadsave(struct vmcb
*from_vmcb
, struct vmcb
*to_vmcb
)
2605 to_vmcb
->save
.fs
= from_vmcb
->save
.fs
;
2606 to_vmcb
->save
.gs
= from_vmcb
->save
.gs
;
2607 to_vmcb
->save
.tr
= from_vmcb
->save
.tr
;
2608 to_vmcb
->save
.ldtr
= from_vmcb
->save
.ldtr
;
2609 to_vmcb
->save
.kernel_gs_base
= from_vmcb
->save
.kernel_gs_base
;
2610 to_vmcb
->save
.star
= from_vmcb
->save
.star
;
2611 to_vmcb
->save
.lstar
= from_vmcb
->save
.lstar
;
2612 to_vmcb
->save
.cstar
= from_vmcb
->save
.cstar
;
2613 to_vmcb
->save
.sfmask
= from_vmcb
->save
.sfmask
;
2614 to_vmcb
->save
.sysenter_cs
= from_vmcb
->save
.sysenter_cs
;
2615 to_vmcb
->save
.sysenter_esp
= from_vmcb
->save
.sysenter_esp
;
2616 to_vmcb
->save
.sysenter_eip
= from_vmcb
->save
.sysenter_eip
;
2619 static int vmload_interception(struct vcpu_svm
*svm
)
2621 struct vmcb
*nested_vmcb
;
2624 if (nested_svm_check_permissions(svm
))
2627 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, &page
);
2631 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2632 skip_emulated_instruction(&svm
->vcpu
);
2634 nested_svm_vmloadsave(nested_vmcb
, svm
->vmcb
);
2635 nested_svm_unmap(page
);
2640 static int vmsave_interception(struct vcpu_svm
*svm
)
2642 struct vmcb
*nested_vmcb
;
2645 if (nested_svm_check_permissions(svm
))
2648 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, &page
);
2652 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2653 skip_emulated_instruction(&svm
->vcpu
);
2655 nested_svm_vmloadsave(svm
->vmcb
, nested_vmcb
);
2656 nested_svm_unmap(page
);
2661 static int vmrun_interception(struct vcpu_svm
*svm
)
2663 if (nested_svm_check_permissions(svm
))
2666 /* Save rip after vmrun instruction */
2667 kvm_rip_write(&svm
->vcpu
, kvm_rip_read(&svm
->vcpu
) + 3);
2669 if (!nested_svm_vmrun(svm
))
2672 if (!nested_svm_vmrun_msrpm(svm
))
2679 svm
->vmcb
->control
.exit_code
= SVM_EXIT_ERR
;
2680 svm
->vmcb
->control
.exit_code_hi
= 0;
2681 svm
->vmcb
->control
.exit_info_1
= 0;
2682 svm
->vmcb
->control
.exit_info_2
= 0;
2684 nested_svm_vmexit(svm
);
2689 static int stgi_interception(struct vcpu_svm
*svm
)
2691 if (nested_svm_check_permissions(svm
))
2694 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2695 skip_emulated_instruction(&svm
->vcpu
);
2696 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
2703 static int clgi_interception(struct vcpu_svm
*svm
)
2705 if (nested_svm_check_permissions(svm
))
2708 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2709 skip_emulated_instruction(&svm
->vcpu
);
2713 /* After a CLGI no interrupts should come */
2714 svm_clear_vintr(svm
);
2715 svm
->vmcb
->control
.int_ctl
&= ~V_IRQ_MASK
;
2717 mark_dirty(svm
->vmcb
, VMCB_INTR
);
2722 static int invlpga_interception(struct vcpu_svm
*svm
)
2724 struct kvm_vcpu
*vcpu
= &svm
->vcpu
;
2726 trace_kvm_invlpga(svm
->vmcb
->save
.rip
, vcpu
->arch
.regs
[VCPU_REGS_RCX
],
2727 vcpu
->arch
.regs
[VCPU_REGS_RAX
]);
2729 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2730 kvm_mmu_invlpg(vcpu
, vcpu
->arch
.regs
[VCPU_REGS_RAX
]);
2732 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2733 skip_emulated_instruction(&svm
->vcpu
);
2737 static int skinit_interception(struct vcpu_svm
*svm
)
2739 trace_kvm_skinit(svm
->vmcb
->save
.rip
, svm
->vcpu
.arch
.regs
[VCPU_REGS_RAX
]);
2741 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
2745 static int xsetbv_interception(struct vcpu_svm
*svm
)
2747 u64 new_bv
= kvm_read_edx_eax(&svm
->vcpu
);
2748 u32 index
= kvm_register_read(&svm
->vcpu
, VCPU_REGS_RCX
);
2750 if (kvm_set_xcr(&svm
->vcpu
, index
, new_bv
) == 0) {
2751 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2752 skip_emulated_instruction(&svm
->vcpu
);
2758 static int invalid_op_interception(struct vcpu_svm
*svm
)
2760 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
2764 static int task_switch_interception(struct vcpu_svm
*svm
)
2768 int int_type
= svm
->vmcb
->control
.exit_int_info
&
2769 SVM_EXITINTINFO_TYPE_MASK
;
2770 int int_vec
= svm
->vmcb
->control
.exit_int_info
& SVM_EVTINJ_VEC_MASK
;
2772 svm
->vmcb
->control
.exit_int_info
& SVM_EXITINTINFO_TYPE_MASK
;
2774 svm
->vmcb
->control
.exit_int_info
& SVM_EXITINTINFO_VALID
;
2775 bool has_error_code
= false;
2778 tss_selector
= (u16
)svm
->vmcb
->control
.exit_info_1
;
2780 if (svm
->vmcb
->control
.exit_info_2
&
2781 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET
))
2782 reason
= TASK_SWITCH_IRET
;
2783 else if (svm
->vmcb
->control
.exit_info_2
&
2784 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP
))
2785 reason
= TASK_SWITCH_JMP
;
2787 reason
= TASK_SWITCH_GATE
;
2789 reason
= TASK_SWITCH_CALL
;
2791 if (reason
== TASK_SWITCH_GATE
) {
2793 case SVM_EXITINTINFO_TYPE_NMI
:
2794 svm
->vcpu
.arch
.nmi_injected
= false;
2796 case SVM_EXITINTINFO_TYPE_EXEPT
:
2797 if (svm
->vmcb
->control
.exit_info_2
&
2798 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE
)) {
2799 has_error_code
= true;
2801 (u32
)svm
->vmcb
->control
.exit_info_2
;
2803 kvm_clear_exception_queue(&svm
->vcpu
);
2805 case SVM_EXITINTINFO_TYPE_INTR
:
2806 kvm_clear_interrupt_queue(&svm
->vcpu
);
2813 if (reason
!= TASK_SWITCH_GATE
||
2814 int_type
== SVM_EXITINTINFO_TYPE_SOFT
||
2815 (int_type
== SVM_EXITINTINFO_TYPE_EXEPT
&&
2816 (int_vec
== OF_VECTOR
|| int_vec
== BP_VECTOR
)))
2817 skip_emulated_instruction(&svm
->vcpu
);
2819 if (int_type
!= SVM_EXITINTINFO_TYPE_SOFT
)
2822 if (kvm_task_switch(&svm
->vcpu
, tss_selector
, int_vec
, reason
,
2823 has_error_code
, error_code
) == EMULATE_FAIL
) {
2824 svm
->vcpu
.run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
2825 svm
->vcpu
.run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
2826 svm
->vcpu
.run
->internal
.ndata
= 0;
2832 static int cpuid_interception(struct vcpu_svm
*svm
)
2834 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
2835 kvm_emulate_cpuid(&svm
->vcpu
);
2839 static int iret_interception(struct vcpu_svm
*svm
)
2841 ++svm
->vcpu
.stat
.nmi_window_exits
;
2842 clr_intercept(svm
, INTERCEPT_IRET
);
2843 svm
->vcpu
.arch
.hflags
|= HF_IRET_MASK
;
2844 svm
->nmi_iret_rip
= kvm_rip_read(&svm
->vcpu
);
2848 static int invlpg_interception(struct vcpu_svm
*svm
)
2850 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS
))
2851 return emulate_instruction(&svm
->vcpu
, 0) == EMULATE_DONE
;
2853 kvm_mmu_invlpg(&svm
->vcpu
, svm
->vmcb
->control
.exit_info_1
);
2854 skip_emulated_instruction(&svm
->vcpu
);
2858 static int emulate_on_interception(struct vcpu_svm
*svm
)
2860 return emulate_instruction(&svm
->vcpu
, 0) == EMULATE_DONE
;
2863 static int rdpmc_interception(struct vcpu_svm
*svm
)
2867 if (!static_cpu_has(X86_FEATURE_NRIPS
))
2868 return emulate_on_interception(svm
);
2870 err
= kvm_rdpmc(&svm
->vcpu
);
2871 kvm_complete_insn_gp(&svm
->vcpu
, err
);
2876 bool check_selective_cr0_intercepted(struct vcpu_svm
*svm
, unsigned long val
)
2878 unsigned long cr0
= svm
->vcpu
.arch
.cr0
;
2882 intercept
= svm
->nested
.intercept
;
2884 if (!is_guest_mode(&svm
->vcpu
) ||
2885 (!(intercept
& (1ULL << INTERCEPT_SELECTIVE_CR0
))))
2888 cr0
&= ~SVM_CR0_SELECTIVE_MASK
;
2889 val
&= ~SVM_CR0_SELECTIVE_MASK
;
2892 svm
->vmcb
->control
.exit_code
= SVM_EXIT_CR0_SEL_WRITE
;
2893 ret
= (nested_svm_exit_handled(svm
) == NESTED_EXIT_DONE
);
2899 #define CR_VALID (1ULL << 63)
2901 static int cr_interception(struct vcpu_svm
*svm
)
2907 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS
))
2908 return emulate_on_interception(svm
);
2910 if (unlikely((svm
->vmcb
->control
.exit_info_1
& CR_VALID
) == 0))
2911 return emulate_on_interception(svm
);
2913 reg
= svm
->vmcb
->control
.exit_info_1
& SVM_EXITINFO_REG_MASK
;
2914 cr
= svm
->vmcb
->control
.exit_code
- SVM_EXIT_READ_CR0
;
2917 if (cr
>= 16) { /* mov to cr */
2919 val
= kvm_register_read(&svm
->vcpu
, reg
);
2922 if (!check_selective_cr0_intercepted(svm
, val
))
2923 err
= kvm_set_cr0(&svm
->vcpu
, val
);
2929 err
= kvm_set_cr3(&svm
->vcpu
, val
);
2932 err
= kvm_set_cr4(&svm
->vcpu
, val
);
2935 err
= kvm_set_cr8(&svm
->vcpu
, val
);
2938 WARN(1, "unhandled write to CR%d", cr
);
2939 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
2942 } else { /* mov from cr */
2945 val
= kvm_read_cr0(&svm
->vcpu
);
2948 val
= svm
->vcpu
.arch
.cr2
;
2951 val
= kvm_read_cr3(&svm
->vcpu
);
2954 val
= kvm_read_cr4(&svm
->vcpu
);
2957 val
= kvm_get_cr8(&svm
->vcpu
);
2960 WARN(1, "unhandled read from CR%d", cr
);
2961 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
2964 kvm_register_write(&svm
->vcpu
, reg
, val
);
2966 kvm_complete_insn_gp(&svm
->vcpu
, err
);
2971 static int dr_interception(struct vcpu_svm
*svm
)
2977 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS
))
2978 return emulate_on_interception(svm
);
2980 reg
= svm
->vmcb
->control
.exit_info_1
& SVM_EXITINFO_REG_MASK
;
2981 dr
= svm
->vmcb
->control
.exit_code
- SVM_EXIT_READ_DR0
;
2983 if (dr
>= 16) { /* mov to DRn */
2984 val
= kvm_register_read(&svm
->vcpu
, reg
);
2985 kvm_set_dr(&svm
->vcpu
, dr
- 16, val
);
2987 err
= kvm_get_dr(&svm
->vcpu
, dr
, &val
);
2989 kvm_register_write(&svm
->vcpu
, reg
, val
);
2992 skip_emulated_instruction(&svm
->vcpu
);
2997 static int cr8_write_interception(struct vcpu_svm
*svm
)
2999 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
3002 u8 cr8_prev
= kvm_get_cr8(&svm
->vcpu
);
3003 /* instruction emulation calls kvm_set_cr8() */
3004 r
= cr_interception(svm
);
3005 if (irqchip_in_kernel(svm
->vcpu
.kvm
)) {
3006 clr_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
3009 if (cr8_prev
<= kvm_get_cr8(&svm
->vcpu
))
3011 kvm_run
->exit_reason
= KVM_EXIT_SET_TPR
;
3015 u64
svm_read_l1_tsc(struct kvm_vcpu
*vcpu
)
3017 struct vmcb
*vmcb
= get_host_vmcb(to_svm(vcpu
));
3018 return vmcb
->control
.tsc_offset
+
3019 svm_scale_tsc(vcpu
, native_read_tsc());
3022 static int svm_get_msr(struct kvm_vcpu
*vcpu
, unsigned ecx
, u64
*data
)
3024 struct vcpu_svm
*svm
= to_svm(vcpu
);
3027 case MSR_IA32_TSC
: {
3028 *data
= svm
->vmcb
->control
.tsc_offset
+
3029 svm_scale_tsc(vcpu
, native_read_tsc());
3034 *data
= svm
->vmcb
->save
.star
;
3036 #ifdef CONFIG_X86_64
3038 *data
= svm
->vmcb
->save
.lstar
;
3041 *data
= svm
->vmcb
->save
.cstar
;
3043 case MSR_KERNEL_GS_BASE
:
3044 *data
= svm
->vmcb
->save
.kernel_gs_base
;
3046 case MSR_SYSCALL_MASK
:
3047 *data
= svm
->vmcb
->save
.sfmask
;
3050 case MSR_IA32_SYSENTER_CS
:
3051 *data
= svm
->vmcb
->save
.sysenter_cs
;
3053 case MSR_IA32_SYSENTER_EIP
:
3054 *data
= svm
->sysenter_eip
;
3056 case MSR_IA32_SYSENTER_ESP
:
3057 *data
= svm
->sysenter_esp
;
3060 * Nobody will change the following 5 values in the VMCB so we can
3061 * safely return them on rdmsr. They will always be 0 until LBRV is
3064 case MSR_IA32_DEBUGCTLMSR
:
3065 *data
= svm
->vmcb
->save
.dbgctl
;
3067 case MSR_IA32_LASTBRANCHFROMIP
:
3068 *data
= svm
->vmcb
->save
.br_from
;
3070 case MSR_IA32_LASTBRANCHTOIP
:
3071 *data
= svm
->vmcb
->save
.br_to
;
3073 case MSR_IA32_LASTINTFROMIP
:
3074 *data
= svm
->vmcb
->save
.last_excp_from
;
3076 case MSR_IA32_LASTINTTOIP
:
3077 *data
= svm
->vmcb
->save
.last_excp_to
;
3079 case MSR_VM_HSAVE_PA
:
3080 *data
= svm
->nested
.hsave_msr
;
3083 *data
= svm
->nested
.vm_cr_msr
;
3085 case MSR_IA32_UCODE_REV
:
3089 return kvm_get_msr_common(vcpu
, ecx
, data
);
3094 static int rdmsr_interception(struct vcpu_svm
*svm
)
3096 u32 ecx
= svm
->vcpu
.arch
.regs
[VCPU_REGS_RCX
];
3099 if (svm_get_msr(&svm
->vcpu
, ecx
, &data
)) {
3100 trace_kvm_msr_read_ex(ecx
);
3101 kvm_inject_gp(&svm
->vcpu
, 0);
3103 trace_kvm_msr_read(ecx
, data
);
3105 svm
->vcpu
.arch
.regs
[VCPU_REGS_RAX
] = data
& 0xffffffff;
3106 svm
->vcpu
.arch
.regs
[VCPU_REGS_RDX
] = data
>> 32;
3107 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
3108 skip_emulated_instruction(&svm
->vcpu
);
3113 static int svm_set_vm_cr(struct kvm_vcpu
*vcpu
, u64 data
)
3115 struct vcpu_svm
*svm
= to_svm(vcpu
);
3116 int svm_dis
, chg_mask
;
3118 if (data
& ~SVM_VM_CR_VALID_MASK
)
3121 chg_mask
= SVM_VM_CR_VALID_MASK
;
3123 if (svm
->nested
.vm_cr_msr
& SVM_VM_CR_SVM_DIS_MASK
)
3124 chg_mask
&= ~(SVM_VM_CR_SVM_LOCK_MASK
| SVM_VM_CR_SVM_DIS_MASK
);
3126 svm
->nested
.vm_cr_msr
&= ~chg_mask
;
3127 svm
->nested
.vm_cr_msr
|= (data
& chg_mask
);
3129 svm_dis
= svm
->nested
.vm_cr_msr
& SVM_VM_CR_SVM_DIS_MASK
;
3131 /* check for svm_disable while efer.svme is set */
3132 if (svm_dis
&& (vcpu
->arch
.efer
& EFER_SVME
))
3138 static int svm_set_msr(struct kvm_vcpu
*vcpu
, unsigned ecx
, u64 data
)
3140 struct vcpu_svm
*svm
= to_svm(vcpu
);
3144 kvm_write_tsc(vcpu
, data
);
3147 svm
->vmcb
->save
.star
= data
;
3149 #ifdef CONFIG_X86_64
3151 svm
->vmcb
->save
.lstar
= data
;
3154 svm
->vmcb
->save
.cstar
= data
;
3156 case MSR_KERNEL_GS_BASE
:
3157 svm
->vmcb
->save
.kernel_gs_base
= data
;
3159 case MSR_SYSCALL_MASK
:
3160 svm
->vmcb
->save
.sfmask
= data
;
3163 case MSR_IA32_SYSENTER_CS
:
3164 svm
->vmcb
->save
.sysenter_cs
= data
;
3166 case MSR_IA32_SYSENTER_EIP
:
3167 svm
->sysenter_eip
= data
;
3168 svm
->vmcb
->save
.sysenter_eip
= data
;
3170 case MSR_IA32_SYSENTER_ESP
:
3171 svm
->sysenter_esp
= data
;
3172 svm
->vmcb
->save
.sysenter_esp
= data
;
3174 case MSR_IA32_DEBUGCTLMSR
:
3175 if (!boot_cpu_has(X86_FEATURE_LBRV
)) {
3176 pr_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3180 if (data
& DEBUGCTL_RESERVED_BITS
)
3183 svm
->vmcb
->save
.dbgctl
= data
;
3184 mark_dirty(svm
->vmcb
, VMCB_LBR
);
3185 if (data
& (1ULL<<0))
3186 svm_enable_lbrv(svm
);
3188 svm_disable_lbrv(svm
);
3190 case MSR_VM_HSAVE_PA
:
3191 svm
->nested
.hsave_msr
= data
;
3194 return svm_set_vm_cr(vcpu
, data
);
3196 pr_unimpl(vcpu
, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx
, data
);
3199 return kvm_set_msr_common(vcpu
, ecx
, data
);
3204 static int wrmsr_interception(struct vcpu_svm
*svm
)
3206 u32 ecx
= svm
->vcpu
.arch
.regs
[VCPU_REGS_RCX
];
3207 u64 data
= (svm
->vcpu
.arch
.regs
[VCPU_REGS_RAX
] & -1u)
3208 | ((u64
)(svm
->vcpu
.arch
.regs
[VCPU_REGS_RDX
] & -1u) << 32);
3211 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
3212 if (svm_set_msr(&svm
->vcpu
, ecx
, data
)) {
3213 trace_kvm_msr_write_ex(ecx
, data
);
3214 kvm_inject_gp(&svm
->vcpu
, 0);
3216 trace_kvm_msr_write(ecx
, data
);
3217 skip_emulated_instruction(&svm
->vcpu
);
3222 static int msr_interception(struct vcpu_svm
*svm
)
3224 if (svm
->vmcb
->control
.exit_info_1
)
3225 return wrmsr_interception(svm
);
3227 return rdmsr_interception(svm
);
3230 static int interrupt_window_interception(struct vcpu_svm
*svm
)
3232 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
3234 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
3235 svm_clear_vintr(svm
);
3236 svm
->vmcb
->control
.int_ctl
&= ~V_IRQ_MASK
;
3237 mark_dirty(svm
->vmcb
, VMCB_INTR
);
3239 * If the user space waits to inject interrupts, exit as soon as
3242 if (!irqchip_in_kernel(svm
->vcpu
.kvm
) &&
3243 kvm_run
->request_interrupt_window
&&
3244 !kvm_cpu_has_interrupt(&svm
->vcpu
)) {
3245 ++svm
->vcpu
.stat
.irq_window_exits
;
3246 kvm_run
->exit_reason
= KVM_EXIT_IRQ_WINDOW_OPEN
;
3253 static int pause_interception(struct vcpu_svm
*svm
)
3255 kvm_vcpu_on_spin(&(svm
->vcpu
));
3259 static int (*svm_exit_handlers
[])(struct vcpu_svm
*svm
) = {
3260 [SVM_EXIT_READ_CR0
] = cr_interception
,
3261 [SVM_EXIT_READ_CR3
] = cr_interception
,
3262 [SVM_EXIT_READ_CR4
] = cr_interception
,
3263 [SVM_EXIT_READ_CR8
] = cr_interception
,
3264 [SVM_EXIT_CR0_SEL_WRITE
] = emulate_on_interception
,
3265 [SVM_EXIT_WRITE_CR0
] = cr_interception
,
3266 [SVM_EXIT_WRITE_CR3
] = cr_interception
,
3267 [SVM_EXIT_WRITE_CR4
] = cr_interception
,
3268 [SVM_EXIT_WRITE_CR8
] = cr8_write_interception
,
3269 [SVM_EXIT_READ_DR0
] = dr_interception
,
3270 [SVM_EXIT_READ_DR1
] = dr_interception
,
3271 [SVM_EXIT_READ_DR2
] = dr_interception
,
3272 [SVM_EXIT_READ_DR3
] = dr_interception
,
3273 [SVM_EXIT_READ_DR4
] = dr_interception
,
3274 [SVM_EXIT_READ_DR5
] = dr_interception
,
3275 [SVM_EXIT_READ_DR6
] = dr_interception
,
3276 [SVM_EXIT_READ_DR7
] = dr_interception
,
3277 [SVM_EXIT_WRITE_DR0
] = dr_interception
,
3278 [SVM_EXIT_WRITE_DR1
] = dr_interception
,
3279 [SVM_EXIT_WRITE_DR2
] = dr_interception
,
3280 [SVM_EXIT_WRITE_DR3
] = dr_interception
,
3281 [SVM_EXIT_WRITE_DR4
] = dr_interception
,
3282 [SVM_EXIT_WRITE_DR5
] = dr_interception
,
3283 [SVM_EXIT_WRITE_DR6
] = dr_interception
,
3284 [SVM_EXIT_WRITE_DR7
] = dr_interception
,
3285 [SVM_EXIT_EXCP_BASE
+ DB_VECTOR
] = db_interception
,
3286 [SVM_EXIT_EXCP_BASE
+ BP_VECTOR
] = bp_interception
,
3287 [SVM_EXIT_EXCP_BASE
+ UD_VECTOR
] = ud_interception
,
3288 [SVM_EXIT_EXCP_BASE
+ PF_VECTOR
] = pf_interception
,
3289 [SVM_EXIT_EXCP_BASE
+ NM_VECTOR
] = nm_interception
,
3290 [SVM_EXIT_EXCP_BASE
+ MC_VECTOR
] = mc_interception
,
3291 [SVM_EXIT_INTR
] = intr_interception
,
3292 [SVM_EXIT_NMI
] = nmi_interception
,
3293 [SVM_EXIT_SMI
] = nop_on_interception
,
3294 [SVM_EXIT_INIT
] = nop_on_interception
,
3295 [SVM_EXIT_VINTR
] = interrupt_window_interception
,
3296 [SVM_EXIT_RDPMC
] = rdpmc_interception
,
3297 [SVM_EXIT_CPUID
] = cpuid_interception
,
3298 [SVM_EXIT_IRET
] = iret_interception
,
3299 [SVM_EXIT_INVD
] = emulate_on_interception
,
3300 [SVM_EXIT_PAUSE
] = pause_interception
,
3301 [SVM_EXIT_HLT
] = halt_interception
,
3302 [SVM_EXIT_INVLPG
] = invlpg_interception
,
3303 [SVM_EXIT_INVLPGA
] = invlpga_interception
,
3304 [SVM_EXIT_IOIO
] = io_interception
,
3305 [SVM_EXIT_MSR
] = msr_interception
,
3306 [SVM_EXIT_TASK_SWITCH
] = task_switch_interception
,
3307 [SVM_EXIT_SHUTDOWN
] = shutdown_interception
,
3308 [SVM_EXIT_VMRUN
] = vmrun_interception
,
3309 [SVM_EXIT_VMMCALL
] = vmmcall_interception
,
3310 [SVM_EXIT_VMLOAD
] = vmload_interception
,
3311 [SVM_EXIT_VMSAVE
] = vmsave_interception
,
3312 [SVM_EXIT_STGI
] = stgi_interception
,
3313 [SVM_EXIT_CLGI
] = clgi_interception
,
3314 [SVM_EXIT_SKINIT
] = skinit_interception
,
3315 [SVM_EXIT_WBINVD
] = emulate_on_interception
,
3316 [SVM_EXIT_MONITOR
] = invalid_op_interception
,
3317 [SVM_EXIT_MWAIT
] = invalid_op_interception
,
3318 [SVM_EXIT_XSETBV
] = xsetbv_interception
,
3319 [SVM_EXIT_NPF
] = pf_interception
,
3322 static void dump_vmcb(struct kvm_vcpu
*vcpu
)
3324 struct vcpu_svm
*svm
= to_svm(vcpu
);
3325 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
3326 struct vmcb_save_area
*save
= &svm
->vmcb
->save
;
3328 pr_err("VMCB Control Area:\n");
3329 pr_err("%-20s%04x\n", "cr_read:", control
->intercept_cr
& 0xffff);
3330 pr_err("%-20s%04x\n", "cr_write:", control
->intercept_cr
>> 16);
3331 pr_err("%-20s%04x\n", "dr_read:", control
->intercept_dr
& 0xffff);
3332 pr_err("%-20s%04x\n", "dr_write:", control
->intercept_dr
>> 16);
3333 pr_err("%-20s%08x\n", "exceptions:", control
->intercept_exceptions
);
3334 pr_err("%-20s%016llx\n", "intercepts:", control
->intercept
);
3335 pr_err("%-20s%d\n", "pause filter count:", control
->pause_filter_count
);
3336 pr_err("%-20s%016llx\n", "iopm_base_pa:", control
->iopm_base_pa
);
3337 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control
->msrpm_base_pa
);
3338 pr_err("%-20s%016llx\n", "tsc_offset:", control
->tsc_offset
);
3339 pr_err("%-20s%d\n", "asid:", control
->asid
);
3340 pr_err("%-20s%d\n", "tlb_ctl:", control
->tlb_ctl
);
3341 pr_err("%-20s%08x\n", "int_ctl:", control
->int_ctl
);
3342 pr_err("%-20s%08x\n", "int_vector:", control
->int_vector
);
3343 pr_err("%-20s%08x\n", "int_state:", control
->int_state
);
3344 pr_err("%-20s%08x\n", "exit_code:", control
->exit_code
);
3345 pr_err("%-20s%016llx\n", "exit_info1:", control
->exit_info_1
);
3346 pr_err("%-20s%016llx\n", "exit_info2:", control
->exit_info_2
);
3347 pr_err("%-20s%08x\n", "exit_int_info:", control
->exit_int_info
);
3348 pr_err("%-20s%08x\n", "exit_int_info_err:", control
->exit_int_info_err
);
3349 pr_err("%-20s%lld\n", "nested_ctl:", control
->nested_ctl
);
3350 pr_err("%-20s%016llx\n", "nested_cr3:", control
->nested_cr3
);
3351 pr_err("%-20s%08x\n", "event_inj:", control
->event_inj
);
3352 pr_err("%-20s%08x\n", "event_inj_err:", control
->event_inj_err
);
3353 pr_err("%-20s%lld\n", "lbr_ctl:", control
->lbr_ctl
);
3354 pr_err("%-20s%016llx\n", "next_rip:", control
->next_rip
);
3355 pr_err("VMCB State Save Area:\n");
3356 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3358 save
->es
.selector
, save
->es
.attrib
,
3359 save
->es
.limit
, save
->es
.base
);
3360 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3362 save
->cs
.selector
, save
->cs
.attrib
,
3363 save
->cs
.limit
, save
->cs
.base
);
3364 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3366 save
->ss
.selector
, save
->ss
.attrib
,
3367 save
->ss
.limit
, save
->ss
.base
);
3368 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3370 save
->ds
.selector
, save
->ds
.attrib
,
3371 save
->ds
.limit
, save
->ds
.base
);
3372 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3374 save
->fs
.selector
, save
->fs
.attrib
,
3375 save
->fs
.limit
, save
->fs
.base
);
3376 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3378 save
->gs
.selector
, save
->gs
.attrib
,
3379 save
->gs
.limit
, save
->gs
.base
);
3380 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3382 save
->gdtr
.selector
, save
->gdtr
.attrib
,
3383 save
->gdtr
.limit
, save
->gdtr
.base
);
3384 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3386 save
->ldtr
.selector
, save
->ldtr
.attrib
,
3387 save
->ldtr
.limit
, save
->ldtr
.base
);
3388 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3390 save
->idtr
.selector
, save
->idtr
.attrib
,
3391 save
->idtr
.limit
, save
->idtr
.base
);
3392 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3394 save
->tr
.selector
, save
->tr
.attrib
,
3395 save
->tr
.limit
, save
->tr
.base
);
3396 pr_err("cpl: %d efer: %016llx\n",
3397 save
->cpl
, save
->efer
);
3398 pr_err("%-15s %016llx %-13s %016llx\n",
3399 "cr0:", save
->cr0
, "cr2:", save
->cr2
);
3400 pr_err("%-15s %016llx %-13s %016llx\n",
3401 "cr3:", save
->cr3
, "cr4:", save
->cr4
);
3402 pr_err("%-15s %016llx %-13s %016llx\n",
3403 "dr6:", save
->dr6
, "dr7:", save
->dr7
);
3404 pr_err("%-15s %016llx %-13s %016llx\n",
3405 "rip:", save
->rip
, "rflags:", save
->rflags
);
3406 pr_err("%-15s %016llx %-13s %016llx\n",
3407 "rsp:", save
->rsp
, "rax:", save
->rax
);
3408 pr_err("%-15s %016llx %-13s %016llx\n",
3409 "star:", save
->star
, "lstar:", save
->lstar
);
3410 pr_err("%-15s %016llx %-13s %016llx\n",
3411 "cstar:", save
->cstar
, "sfmask:", save
->sfmask
);
3412 pr_err("%-15s %016llx %-13s %016llx\n",
3413 "kernel_gs_base:", save
->kernel_gs_base
,
3414 "sysenter_cs:", save
->sysenter_cs
);
3415 pr_err("%-15s %016llx %-13s %016llx\n",
3416 "sysenter_esp:", save
->sysenter_esp
,
3417 "sysenter_eip:", save
->sysenter_eip
);
3418 pr_err("%-15s %016llx %-13s %016llx\n",
3419 "gpat:", save
->g_pat
, "dbgctl:", save
->dbgctl
);
3420 pr_err("%-15s %016llx %-13s %016llx\n",
3421 "br_from:", save
->br_from
, "br_to:", save
->br_to
);
3422 pr_err("%-15s %016llx %-13s %016llx\n",
3423 "excp_from:", save
->last_excp_from
,
3424 "excp_to:", save
->last_excp_to
);
3427 static void svm_get_exit_info(struct kvm_vcpu
*vcpu
, u64
*info1
, u64
*info2
)
3429 struct vmcb_control_area
*control
= &to_svm(vcpu
)->vmcb
->control
;
3431 *info1
= control
->exit_info_1
;
3432 *info2
= control
->exit_info_2
;
3435 static int handle_exit(struct kvm_vcpu
*vcpu
)
3437 struct vcpu_svm
*svm
= to_svm(vcpu
);
3438 struct kvm_run
*kvm_run
= vcpu
->run
;
3439 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
3441 if (!is_cr_intercept(svm
, INTERCEPT_CR0_WRITE
))
3442 vcpu
->arch
.cr0
= svm
->vmcb
->save
.cr0
;
3444 vcpu
->arch
.cr3
= svm
->vmcb
->save
.cr3
;
3446 if (unlikely(svm
->nested
.exit_required
)) {
3447 nested_svm_vmexit(svm
);
3448 svm
->nested
.exit_required
= false;
3453 if (is_guest_mode(vcpu
)) {
3456 trace_kvm_nested_vmexit(svm
->vmcb
->save
.rip
, exit_code
,
3457 svm
->vmcb
->control
.exit_info_1
,
3458 svm
->vmcb
->control
.exit_info_2
,
3459 svm
->vmcb
->control
.exit_int_info
,
3460 svm
->vmcb
->control
.exit_int_info_err
,
3463 vmexit
= nested_svm_exit_special(svm
);
3465 if (vmexit
== NESTED_EXIT_CONTINUE
)
3466 vmexit
= nested_svm_exit_handled(svm
);
3468 if (vmexit
== NESTED_EXIT_DONE
)
3472 svm_complete_interrupts(svm
);
3474 if (svm
->vmcb
->control
.exit_code
== SVM_EXIT_ERR
) {
3475 kvm_run
->exit_reason
= KVM_EXIT_FAIL_ENTRY
;
3476 kvm_run
->fail_entry
.hardware_entry_failure_reason
3477 = svm
->vmcb
->control
.exit_code
;
3478 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
3483 if (is_external_interrupt(svm
->vmcb
->control
.exit_int_info
) &&
3484 exit_code
!= SVM_EXIT_EXCP_BASE
+ PF_VECTOR
&&
3485 exit_code
!= SVM_EXIT_NPF
&& exit_code
!= SVM_EXIT_TASK_SWITCH
&&
3486 exit_code
!= SVM_EXIT_INTR
&& exit_code
!= SVM_EXIT_NMI
)
3487 printk(KERN_ERR
"%s: unexpected exit_ini_info 0x%x "
3489 __func__
, svm
->vmcb
->control
.exit_int_info
,
3492 if (exit_code
>= ARRAY_SIZE(svm_exit_handlers
)
3493 || !svm_exit_handlers
[exit_code
]) {
3494 kvm_run
->exit_reason
= KVM_EXIT_UNKNOWN
;
3495 kvm_run
->hw
.hardware_exit_reason
= exit_code
;
3499 return svm_exit_handlers
[exit_code
](svm
);
3502 static void reload_tss(struct kvm_vcpu
*vcpu
)
3504 int cpu
= raw_smp_processor_id();
3506 struct svm_cpu_data
*sd
= per_cpu(svm_data
, cpu
);
3507 sd
->tss_desc
->type
= 9; /* available 32/64-bit TSS */
3511 static void pre_svm_run(struct vcpu_svm
*svm
)
3513 int cpu
= raw_smp_processor_id();
3515 struct svm_cpu_data
*sd
= per_cpu(svm_data
, cpu
);
3517 /* FIXME: handle wraparound of asid_generation */
3518 if (svm
->asid_generation
!= sd
->asid_generation
)
3522 static void svm_inject_nmi(struct kvm_vcpu
*vcpu
)
3524 struct vcpu_svm
*svm
= to_svm(vcpu
);
3526 svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_NMI
;
3527 vcpu
->arch
.hflags
|= HF_NMI_MASK
;
3528 set_intercept(svm
, INTERCEPT_IRET
);
3529 ++vcpu
->stat
.nmi_injections
;
3532 static inline void svm_inject_irq(struct vcpu_svm
*svm
, int irq
)
3534 struct vmcb_control_area
*control
;
3536 control
= &svm
->vmcb
->control
;
3537 control
->int_vector
= irq
;
3538 control
->int_ctl
&= ~V_INTR_PRIO_MASK
;
3539 control
->int_ctl
|= V_IRQ_MASK
|
3540 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT
);
3541 mark_dirty(svm
->vmcb
, VMCB_INTR
);
3544 static void svm_set_irq(struct kvm_vcpu
*vcpu
)
3546 struct vcpu_svm
*svm
= to_svm(vcpu
);
3548 BUG_ON(!(gif_set(svm
)));
3550 trace_kvm_inj_virq(vcpu
->arch
.interrupt
.nr
);
3551 ++vcpu
->stat
.irq_injections
;
3553 svm
->vmcb
->control
.event_inj
= vcpu
->arch
.interrupt
.nr
|
3554 SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
;
3557 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
, int tpr
, int irr
)
3559 struct vcpu_svm
*svm
= to_svm(vcpu
);
3561 if (is_guest_mode(vcpu
) && (vcpu
->arch
.hflags
& HF_VINTR_MASK
))
3568 set_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
3571 static int svm_nmi_allowed(struct kvm_vcpu
*vcpu
)
3573 struct vcpu_svm
*svm
= to_svm(vcpu
);
3574 struct vmcb
*vmcb
= svm
->vmcb
;
3576 ret
= !(vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
) &&
3577 !(svm
->vcpu
.arch
.hflags
& HF_NMI_MASK
);
3578 ret
= ret
&& gif_set(svm
) && nested_svm_nmi(svm
);
3583 static bool svm_get_nmi_mask(struct kvm_vcpu
*vcpu
)
3585 struct vcpu_svm
*svm
= to_svm(vcpu
);
3587 return !!(svm
->vcpu
.arch
.hflags
& HF_NMI_MASK
);
3590 static void svm_set_nmi_mask(struct kvm_vcpu
*vcpu
, bool masked
)
3592 struct vcpu_svm
*svm
= to_svm(vcpu
);
3595 svm
->vcpu
.arch
.hflags
|= HF_NMI_MASK
;
3596 set_intercept(svm
, INTERCEPT_IRET
);
3598 svm
->vcpu
.arch
.hflags
&= ~HF_NMI_MASK
;
3599 clr_intercept(svm
, INTERCEPT_IRET
);
3603 static int svm_interrupt_allowed(struct kvm_vcpu
*vcpu
)
3605 struct vcpu_svm
*svm
= to_svm(vcpu
);
3606 struct vmcb
*vmcb
= svm
->vmcb
;
3609 if (!gif_set(svm
) ||
3610 (vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
))
3613 ret
= !!(kvm_get_rflags(vcpu
) & X86_EFLAGS_IF
);
3615 if (is_guest_mode(vcpu
))
3616 return ret
&& !(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
);
3621 static void enable_irq_window(struct kvm_vcpu
*vcpu
)
3623 struct vcpu_svm
*svm
= to_svm(vcpu
);
3626 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3627 * 1, because that's a separate STGI/VMRUN intercept. The next time we
3628 * get that intercept, this function will be called again though and
3629 * we'll get the vintr intercept.
3631 if (gif_set(svm
) && nested_svm_intr(svm
)) {
3633 svm_inject_irq(svm
, 0x0);
3637 static void enable_nmi_window(struct kvm_vcpu
*vcpu
)
3639 struct vcpu_svm
*svm
= to_svm(vcpu
);
3641 if ((svm
->vcpu
.arch
.hflags
& (HF_NMI_MASK
| HF_IRET_MASK
))
3643 return; /* IRET will cause a vm exit */
3646 * Something prevents NMI from been injected. Single step over possible
3647 * problem (IRET or exception injection or interrupt shadow)
3649 svm
->nmi_singlestep
= true;
3650 svm
->vmcb
->save
.rflags
|= (X86_EFLAGS_TF
| X86_EFLAGS_RF
);
3651 update_db_intercept(vcpu
);
3654 static int svm_set_tss_addr(struct kvm
*kvm
, unsigned int addr
)
3659 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
)
3661 struct vcpu_svm
*svm
= to_svm(vcpu
);
3663 if (static_cpu_has(X86_FEATURE_FLUSHBYASID
))
3664 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ASID
;
3666 svm
->asid_generation
--;
3669 static void svm_prepare_guest_switch(struct kvm_vcpu
*vcpu
)
3673 static inline void sync_cr8_to_lapic(struct kvm_vcpu
*vcpu
)
3675 struct vcpu_svm
*svm
= to_svm(vcpu
);
3677 if (is_guest_mode(vcpu
) && (vcpu
->arch
.hflags
& HF_VINTR_MASK
))
3680 if (!is_cr_intercept(svm
, INTERCEPT_CR8_WRITE
)) {
3681 int cr8
= svm
->vmcb
->control
.int_ctl
& V_TPR_MASK
;
3682 kvm_set_cr8(vcpu
, cr8
);
3686 static inline void sync_lapic_to_cr8(struct kvm_vcpu
*vcpu
)
3688 struct vcpu_svm
*svm
= to_svm(vcpu
);
3691 if (is_guest_mode(vcpu
) && (vcpu
->arch
.hflags
& HF_VINTR_MASK
))
3694 cr8
= kvm_get_cr8(vcpu
);
3695 svm
->vmcb
->control
.int_ctl
&= ~V_TPR_MASK
;
3696 svm
->vmcb
->control
.int_ctl
|= cr8
& V_TPR_MASK
;
3699 static void svm_complete_interrupts(struct vcpu_svm
*svm
)
3703 u32 exitintinfo
= svm
->vmcb
->control
.exit_int_info
;
3704 unsigned int3_injected
= svm
->int3_injected
;
3706 svm
->int3_injected
= 0;
3709 * If we've made progress since setting HF_IRET_MASK, we've
3710 * executed an IRET and can allow NMI injection.
3712 if ((svm
->vcpu
.arch
.hflags
& HF_IRET_MASK
)
3713 && kvm_rip_read(&svm
->vcpu
) != svm
->nmi_iret_rip
) {
3714 svm
->vcpu
.arch
.hflags
&= ~(HF_NMI_MASK
| HF_IRET_MASK
);
3715 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
3718 svm
->vcpu
.arch
.nmi_injected
= false;
3719 kvm_clear_exception_queue(&svm
->vcpu
);
3720 kvm_clear_interrupt_queue(&svm
->vcpu
);
3722 if (!(exitintinfo
& SVM_EXITINTINFO_VALID
))
3725 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
3727 vector
= exitintinfo
& SVM_EXITINTINFO_VEC_MASK
;
3728 type
= exitintinfo
& SVM_EXITINTINFO_TYPE_MASK
;
3731 case SVM_EXITINTINFO_TYPE_NMI
:
3732 svm
->vcpu
.arch
.nmi_injected
= true;
3734 case SVM_EXITINTINFO_TYPE_EXEPT
:
3736 * In case of software exceptions, do not reinject the vector,
3737 * but re-execute the instruction instead. Rewind RIP first
3738 * if we emulated INT3 before.
3740 if (kvm_exception_is_soft(vector
)) {
3741 if (vector
== BP_VECTOR
&& int3_injected
&&
3742 kvm_is_linear_rip(&svm
->vcpu
, svm
->int3_rip
))
3743 kvm_rip_write(&svm
->vcpu
,
3744 kvm_rip_read(&svm
->vcpu
) -
3748 if (exitintinfo
& SVM_EXITINTINFO_VALID_ERR
) {
3749 u32 err
= svm
->vmcb
->control
.exit_int_info_err
;
3750 kvm_requeue_exception_e(&svm
->vcpu
, vector
, err
);
3753 kvm_requeue_exception(&svm
->vcpu
, vector
);
3755 case SVM_EXITINTINFO_TYPE_INTR
:
3756 kvm_queue_interrupt(&svm
->vcpu
, vector
, false);
3763 static void svm_cancel_injection(struct kvm_vcpu
*vcpu
)
3765 struct vcpu_svm
*svm
= to_svm(vcpu
);
3766 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
3768 control
->exit_int_info
= control
->event_inj
;
3769 control
->exit_int_info_err
= control
->event_inj_err
;
3770 control
->event_inj
= 0;
3771 svm_complete_interrupts(svm
);
3774 #ifdef CONFIG_X86_64
3780 static void svm_vcpu_run(struct kvm_vcpu
*vcpu
)
3782 struct vcpu_svm
*svm
= to_svm(vcpu
);
3784 svm
->vmcb
->save
.rax
= vcpu
->arch
.regs
[VCPU_REGS_RAX
];
3785 svm
->vmcb
->save
.rsp
= vcpu
->arch
.regs
[VCPU_REGS_RSP
];
3786 svm
->vmcb
->save
.rip
= vcpu
->arch
.regs
[VCPU_REGS_RIP
];
3789 * A vmexit emulation is required before the vcpu can be executed
3792 if (unlikely(svm
->nested
.exit_required
))
3797 sync_lapic_to_cr8(vcpu
);
3799 svm
->vmcb
->save
.cr2
= vcpu
->arch
.cr2
;
3806 "push %%"R
"bp; \n\t"
3807 "mov %c[rbx](%[svm]), %%"R
"bx \n\t"
3808 "mov %c[rcx](%[svm]), %%"R
"cx \n\t"
3809 "mov %c[rdx](%[svm]), %%"R
"dx \n\t"
3810 "mov %c[rsi](%[svm]), %%"R
"si \n\t"
3811 "mov %c[rdi](%[svm]), %%"R
"di \n\t"
3812 "mov %c[rbp](%[svm]), %%"R
"bp \n\t"
3813 #ifdef CONFIG_X86_64
3814 "mov %c[r8](%[svm]), %%r8 \n\t"
3815 "mov %c[r9](%[svm]), %%r9 \n\t"
3816 "mov %c[r10](%[svm]), %%r10 \n\t"
3817 "mov %c[r11](%[svm]), %%r11 \n\t"
3818 "mov %c[r12](%[svm]), %%r12 \n\t"
3819 "mov %c[r13](%[svm]), %%r13 \n\t"
3820 "mov %c[r14](%[svm]), %%r14 \n\t"
3821 "mov %c[r15](%[svm]), %%r15 \n\t"
3824 /* Enter guest mode */
3826 "mov %c[vmcb](%[svm]), %%"R
"ax \n\t"
3827 __ex(SVM_VMLOAD
) "\n\t"
3828 __ex(SVM_VMRUN
) "\n\t"
3829 __ex(SVM_VMSAVE
) "\n\t"
3832 /* Save guest registers, load host registers */
3833 "mov %%"R
"bx, %c[rbx](%[svm]) \n\t"
3834 "mov %%"R
"cx, %c[rcx](%[svm]) \n\t"
3835 "mov %%"R
"dx, %c[rdx](%[svm]) \n\t"
3836 "mov %%"R
"si, %c[rsi](%[svm]) \n\t"
3837 "mov %%"R
"di, %c[rdi](%[svm]) \n\t"
3838 "mov %%"R
"bp, %c[rbp](%[svm]) \n\t"
3839 #ifdef CONFIG_X86_64
3840 "mov %%r8, %c[r8](%[svm]) \n\t"
3841 "mov %%r9, %c[r9](%[svm]) \n\t"
3842 "mov %%r10, %c[r10](%[svm]) \n\t"
3843 "mov %%r11, %c[r11](%[svm]) \n\t"
3844 "mov %%r12, %c[r12](%[svm]) \n\t"
3845 "mov %%r13, %c[r13](%[svm]) \n\t"
3846 "mov %%r14, %c[r14](%[svm]) \n\t"
3847 "mov %%r15, %c[r15](%[svm]) \n\t"
3852 [vmcb
]"i"(offsetof(struct vcpu_svm
, vmcb_pa
)),
3853 [rbx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RBX
])),
3854 [rcx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RCX
])),
3855 [rdx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RDX
])),
3856 [rsi
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RSI
])),
3857 [rdi
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RDI
])),
3858 [rbp
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RBP
]))
3859 #ifdef CONFIG_X86_64
3860 , [r8
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R8
])),
3861 [r9
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R9
])),
3862 [r10
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R10
])),
3863 [r11
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R11
])),
3864 [r12
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R12
])),
3865 [r13
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R13
])),
3866 [r14
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R14
])),
3867 [r15
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R15
]))
3870 , R
"bx", R
"cx", R
"dx", R
"si", R
"di"
3871 #ifdef CONFIG_X86_64
3872 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
3876 #ifdef CONFIG_X86_64
3877 wrmsrl(MSR_GS_BASE
, svm
->host
.gs_base
);
3879 loadsegment(fs
, svm
->host
.fs
);
3880 #ifndef CONFIG_X86_32_LAZY_GS
3881 loadsegment(gs
, svm
->host
.gs
);
3887 local_irq_disable();
3889 vcpu
->arch
.cr2
= svm
->vmcb
->save
.cr2
;
3890 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = svm
->vmcb
->save
.rax
;
3891 vcpu
->arch
.regs
[VCPU_REGS_RSP
] = svm
->vmcb
->save
.rsp
;
3892 vcpu
->arch
.regs
[VCPU_REGS_RIP
] = svm
->vmcb
->save
.rip
;
3894 trace_kvm_exit(svm
->vmcb
->control
.exit_code
, vcpu
, KVM_ISA_SVM
);
3896 if (unlikely(svm
->vmcb
->control
.exit_code
== SVM_EXIT_NMI
))
3897 kvm_before_handle_nmi(&svm
->vcpu
);
3901 /* Any pending NMI will happen here */
3903 if (unlikely(svm
->vmcb
->control
.exit_code
== SVM_EXIT_NMI
))
3904 kvm_after_handle_nmi(&svm
->vcpu
);
3906 sync_cr8_to_lapic(vcpu
);
3910 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_DO_NOTHING
;
3912 /* if exit due to PF check for async PF */
3913 if (svm
->vmcb
->control
.exit_code
== SVM_EXIT_EXCP_BASE
+ PF_VECTOR
)
3914 svm
->apf_reason
= kvm_read_and_reset_pf_reason();
3917 vcpu
->arch
.regs_avail
&= ~(1 << VCPU_EXREG_PDPTR
);
3918 vcpu
->arch
.regs_dirty
&= ~(1 << VCPU_EXREG_PDPTR
);
3922 * We need to handle MC intercepts here before the vcpu has a chance to
3923 * change the physical cpu
3925 if (unlikely(svm
->vmcb
->control
.exit_code
==
3926 SVM_EXIT_EXCP_BASE
+ MC_VECTOR
))
3927 svm_handle_mce(svm
);
3929 mark_all_clean(svm
->vmcb
);
3934 static void svm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
3936 struct vcpu_svm
*svm
= to_svm(vcpu
);
3938 svm
->vmcb
->save
.cr3
= root
;
3939 mark_dirty(svm
->vmcb
, VMCB_CR
);
3940 svm_flush_tlb(vcpu
);
3943 static void set_tdp_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
3945 struct vcpu_svm
*svm
= to_svm(vcpu
);
3947 svm
->vmcb
->control
.nested_cr3
= root
;
3948 mark_dirty(svm
->vmcb
, VMCB_NPT
);
3950 /* Also sync guest cr3 here in case we live migrate */
3951 svm
->vmcb
->save
.cr3
= kvm_read_cr3(vcpu
);
3952 mark_dirty(svm
->vmcb
, VMCB_CR
);
3954 svm_flush_tlb(vcpu
);
3957 static int is_disabled(void)
3961 rdmsrl(MSR_VM_CR
, vm_cr
);
3962 if (vm_cr
& (1 << SVM_VM_CR_SVM_DISABLE
))
3969 svm_patch_hypercall(struct kvm_vcpu
*vcpu
, unsigned char *hypercall
)
3972 * Patch in the VMMCALL instruction:
3974 hypercall
[0] = 0x0f;
3975 hypercall
[1] = 0x01;
3976 hypercall
[2] = 0xd9;
3979 static void svm_check_processor_compat(void *rtn
)
3984 static bool svm_cpu_has_accelerated_tpr(void)
3989 static u64
svm_get_mt_mask(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool is_mmio
)
3994 static void svm_cpuid_update(struct kvm_vcpu
*vcpu
)
3998 static void svm_set_supported_cpuid(u32 func
, struct kvm_cpuid_entry2
*entry
)
4003 entry
->ecx
|= (1 << 2); /* Set SVM bit */
4006 entry
->eax
= 1; /* SVM revision 1 */
4007 entry
->ebx
= 8; /* Lets support 8 ASIDs in case we add proper
4008 ASID emulation to nested SVM */
4009 entry
->ecx
= 0; /* Reserved */
4010 entry
->edx
= 0; /* Per default do not support any
4011 additional features */
4013 /* Support next_rip if host supports it */
4014 if (boot_cpu_has(X86_FEATURE_NRIPS
))
4015 entry
->edx
|= SVM_FEATURE_NRIP
;
4017 /* Support NPT for the guest if enabled */
4019 entry
->edx
|= SVM_FEATURE_NPT
;
4025 static int svm_get_lpage_level(void)
4027 return PT_PDPE_LEVEL
;
4030 static bool svm_rdtscp_supported(void)
4035 static bool svm_has_wbinvd_exit(void)
4040 static void svm_fpu_deactivate(struct kvm_vcpu
*vcpu
)
4042 struct vcpu_svm
*svm
= to_svm(vcpu
);
4044 set_exception_intercept(svm
, NM_VECTOR
);
4045 update_cr0_intercept(svm
);
4048 #define PRE_EX(exit) { .exit_code = (exit), \
4049 .stage = X86_ICPT_PRE_EXCEPT, }
4050 #define POST_EX(exit) { .exit_code = (exit), \
4051 .stage = X86_ICPT_POST_EXCEPT, }
4052 #define POST_MEM(exit) { .exit_code = (exit), \
4053 .stage = X86_ICPT_POST_MEMACCESS, }
4055 static struct __x86_intercept
{
4057 enum x86_intercept_stage stage
;
4058 } x86_intercept_map
[] = {
4059 [x86_intercept_cr_read
] = POST_EX(SVM_EXIT_READ_CR0
),
4060 [x86_intercept_cr_write
] = POST_EX(SVM_EXIT_WRITE_CR0
),
4061 [x86_intercept_clts
] = POST_EX(SVM_EXIT_WRITE_CR0
),
4062 [x86_intercept_lmsw
] = POST_EX(SVM_EXIT_WRITE_CR0
),
4063 [x86_intercept_smsw
] = POST_EX(SVM_EXIT_READ_CR0
),
4064 [x86_intercept_dr_read
] = POST_EX(SVM_EXIT_READ_DR0
),
4065 [x86_intercept_dr_write
] = POST_EX(SVM_EXIT_WRITE_DR0
),
4066 [x86_intercept_sldt
] = POST_EX(SVM_EXIT_LDTR_READ
),
4067 [x86_intercept_str
] = POST_EX(SVM_EXIT_TR_READ
),
4068 [x86_intercept_lldt
] = POST_EX(SVM_EXIT_LDTR_WRITE
),
4069 [x86_intercept_ltr
] = POST_EX(SVM_EXIT_TR_WRITE
),
4070 [x86_intercept_sgdt
] = POST_EX(SVM_EXIT_GDTR_READ
),
4071 [x86_intercept_sidt
] = POST_EX(SVM_EXIT_IDTR_READ
),
4072 [x86_intercept_lgdt
] = POST_EX(SVM_EXIT_GDTR_WRITE
),
4073 [x86_intercept_lidt
] = POST_EX(SVM_EXIT_IDTR_WRITE
),
4074 [x86_intercept_vmrun
] = POST_EX(SVM_EXIT_VMRUN
),
4075 [x86_intercept_vmmcall
] = POST_EX(SVM_EXIT_VMMCALL
),
4076 [x86_intercept_vmload
] = POST_EX(SVM_EXIT_VMLOAD
),
4077 [x86_intercept_vmsave
] = POST_EX(SVM_EXIT_VMSAVE
),
4078 [x86_intercept_stgi
] = POST_EX(SVM_EXIT_STGI
),
4079 [x86_intercept_clgi
] = POST_EX(SVM_EXIT_CLGI
),
4080 [x86_intercept_skinit
] = POST_EX(SVM_EXIT_SKINIT
),
4081 [x86_intercept_invlpga
] = POST_EX(SVM_EXIT_INVLPGA
),
4082 [x86_intercept_rdtscp
] = POST_EX(SVM_EXIT_RDTSCP
),
4083 [x86_intercept_monitor
] = POST_MEM(SVM_EXIT_MONITOR
),
4084 [x86_intercept_mwait
] = POST_EX(SVM_EXIT_MWAIT
),
4085 [x86_intercept_invlpg
] = POST_EX(SVM_EXIT_INVLPG
),
4086 [x86_intercept_invd
] = POST_EX(SVM_EXIT_INVD
),
4087 [x86_intercept_wbinvd
] = POST_EX(SVM_EXIT_WBINVD
),
4088 [x86_intercept_wrmsr
] = POST_EX(SVM_EXIT_MSR
),
4089 [x86_intercept_rdtsc
] = POST_EX(SVM_EXIT_RDTSC
),
4090 [x86_intercept_rdmsr
] = POST_EX(SVM_EXIT_MSR
),
4091 [x86_intercept_rdpmc
] = POST_EX(SVM_EXIT_RDPMC
),
4092 [x86_intercept_cpuid
] = PRE_EX(SVM_EXIT_CPUID
),
4093 [x86_intercept_rsm
] = PRE_EX(SVM_EXIT_RSM
),
4094 [x86_intercept_pause
] = PRE_EX(SVM_EXIT_PAUSE
),
4095 [x86_intercept_pushf
] = PRE_EX(SVM_EXIT_PUSHF
),
4096 [x86_intercept_popf
] = PRE_EX(SVM_EXIT_POPF
),
4097 [x86_intercept_intn
] = PRE_EX(SVM_EXIT_SWINT
),
4098 [x86_intercept_iret
] = PRE_EX(SVM_EXIT_IRET
),
4099 [x86_intercept_icebp
] = PRE_EX(SVM_EXIT_ICEBP
),
4100 [x86_intercept_hlt
] = POST_EX(SVM_EXIT_HLT
),
4101 [x86_intercept_in
] = POST_EX(SVM_EXIT_IOIO
),
4102 [x86_intercept_ins
] = POST_EX(SVM_EXIT_IOIO
),
4103 [x86_intercept_out
] = POST_EX(SVM_EXIT_IOIO
),
4104 [x86_intercept_outs
] = POST_EX(SVM_EXIT_IOIO
),
4111 static int svm_check_intercept(struct kvm_vcpu
*vcpu
,
4112 struct x86_instruction_info
*info
,
4113 enum x86_intercept_stage stage
)
4115 struct vcpu_svm
*svm
= to_svm(vcpu
);
4116 int vmexit
, ret
= X86EMUL_CONTINUE
;
4117 struct __x86_intercept icpt_info
;
4118 struct vmcb
*vmcb
= svm
->vmcb
;
4120 if (info
->intercept
>= ARRAY_SIZE(x86_intercept_map
))
4123 icpt_info
= x86_intercept_map
[info
->intercept
];
4125 if (stage
!= icpt_info
.stage
)
4128 switch (icpt_info
.exit_code
) {
4129 case SVM_EXIT_READ_CR0
:
4130 if (info
->intercept
== x86_intercept_cr_read
)
4131 icpt_info
.exit_code
+= info
->modrm_reg
;
4133 case SVM_EXIT_WRITE_CR0
: {
4134 unsigned long cr0
, val
;
4137 if (info
->intercept
== x86_intercept_cr_write
)
4138 icpt_info
.exit_code
+= info
->modrm_reg
;
4140 if (icpt_info
.exit_code
!= SVM_EXIT_WRITE_CR0
)
4143 intercept
= svm
->nested
.intercept
;
4145 if (!(intercept
& (1ULL << INTERCEPT_SELECTIVE_CR0
)))
4148 cr0
= vcpu
->arch
.cr0
& ~SVM_CR0_SELECTIVE_MASK
;
4149 val
= info
->src_val
& ~SVM_CR0_SELECTIVE_MASK
;
4151 if (info
->intercept
== x86_intercept_lmsw
) {
4154 /* lmsw can't clear PE - catch this here */
4155 if (cr0
& X86_CR0_PE
)
4160 icpt_info
.exit_code
= SVM_EXIT_CR0_SEL_WRITE
;
4164 case SVM_EXIT_READ_DR0
:
4165 case SVM_EXIT_WRITE_DR0
:
4166 icpt_info
.exit_code
+= info
->modrm_reg
;
4169 if (info
->intercept
== x86_intercept_wrmsr
)
4170 vmcb
->control
.exit_info_1
= 1;
4172 vmcb
->control
.exit_info_1
= 0;
4174 case SVM_EXIT_PAUSE
:
4176 * We get this for NOP only, but pause
4177 * is rep not, check this here
4179 if (info
->rep_prefix
!= REPE_PREFIX
)
4181 case SVM_EXIT_IOIO
: {
4185 exit_info
= (vcpu
->arch
.regs
[VCPU_REGS_RDX
] & 0xffff) << 16;
4187 if (info
->intercept
== x86_intercept_in
||
4188 info
->intercept
== x86_intercept_ins
) {
4189 exit_info
|= SVM_IOIO_TYPE_MASK
;
4190 bytes
= info
->src_bytes
;
4192 bytes
= info
->dst_bytes
;
4195 if (info
->intercept
== x86_intercept_outs
||
4196 info
->intercept
== x86_intercept_ins
)
4197 exit_info
|= SVM_IOIO_STR_MASK
;
4199 if (info
->rep_prefix
)
4200 exit_info
|= SVM_IOIO_REP_MASK
;
4202 bytes
= min(bytes
, 4u);
4204 exit_info
|= bytes
<< SVM_IOIO_SIZE_SHIFT
;
4206 exit_info
|= (u32
)info
->ad_bytes
<< (SVM_IOIO_ASIZE_SHIFT
- 1);
4208 vmcb
->control
.exit_info_1
= exit_info
;
4209 vmcb
->control
.exit_info_2
= info
->next_rip
;
4217 vmcb
->control
.next_rip
= info
->next_rip
;
4218 vmcb
->control
.exit_code
= icpt_info
.exit_code
;
4219 vmexit
= nested_svm_exit_handled(svm
);
4221 ret
= (vmexit
== NESTED_EXIT_DONE
) ? X86EMUL_INTERCEPTED
4228 static struct kvm_x86_ops svm_x86_ops
= {
4229 .cpu_has_kvm_support
= has_svm
,
4230 .disabled_by_bios
= is_disabled
,
4231 .hardware_setup
= svm_hardware_setup
,
4232 .hardware_unsetup
= svm_hardware_unsetup
,
4233 .check_processor_compatibility
= svm_check_processor_compat
,
4234 .hardware_enable
= svm_hardware_enable
,
4235 .hardware_disable
= svm_hardware_disable
,
4236 .cpu_has_accelerated_tpr
= svm_cpu_has_accelerated_tpr
,
4238 .vcpu_create
= svm_create_vcpu
,
4239 .vcpu_free
= svm_free_vcpu
,
4240 .vcpu_reset
= svm_vcpu_reset
,
4242 .prepare_guest_switch
= svm_prepare_guest_switch
,
4243 .vcpu_load
= svm_vcpu_load
,
4244 .vcpu_put
= svm_vcpu_put
,
4246 .set_guest_debug
= svm_guest_debug
,
4247 .get_msr
= svm_get_msr
,
4248 .set_msr
= svm_set_msr
,
4249 .get_segment_base
= svm_get_segment_base
,
4250 .get_segment
= svm_get_segment
,
4251 .set_segment
= svm_set_segment
,
4252 .get_cpl
= svm_get_cpl
,
4253 .get_cs_db_l_bits
= kvm_get_cs_db_l_bits
,
4254 .decache_cr0_guest_bits
= svm_decache_cr0_guest_bits
,
4255 .decache_cr3
= svm_decache_cr3
,
4256 .decache_cr4_guest_bits
= svm_decache_cr4_guest_bits
,
4257 .set_cr0
= svm_set_cr0
,
4258 .set_cr3
= svm_set_cr3
,
4259 .set_cr4
= svm_set_cr4
,
4260 .set_efer
= svm_set_efer
,
4261 .get_idt
= svm_get_idt
,
4262 .set_idt
= svm_set_idt
,
4263 .get_gdt
= svm_get_gdt
,
4264 .set_gdt
= svm_set_gdt
,
4265 .set_dr7
= svm_set_dr7
,
4266 .cache_reg
= svm_cache_reg
,
4267 .get_rflags
= svm_get_rflags
,
4268 .set_rflags
= svm_set_rflags
,
4269 .fpu_activate
= svm_fpu_activate
,
4270 .fpu_deactivate
= svm_fpu_deactivate
,
4272 .tlb_flush
= svm_flush_tlb
,
4274 .run
= svm_vcpu_run
,
4275 .handle_exit
= handle_exit
,
4276 .skip_emulated_instruction
= skip_emulated_instruction
,
4277 .set_interrupt_shadow
= svm_set_interrupt_shadow
,
4278 .get_interrupt_shadow
= svm_get_interrupt_shadow
,
4279 .patch_hypercall
= svm_patch_hypercall
,
4280 .set_irq
= svm_set_irq
,
4281 .set_nmi
= svm_inject_nmi
,
4282 .queue_exception
= svm_queue_exception
,
4283 .cancel_injection
= svm_cancel_injection
,
4284 .interrupt_allowed
= svm_interrupt_allowed
,
4285 .nmi_allowed
= svm_nmi_allowed
,
4286 .get_nmi_mask
= svm_get_nmi_mask
,
4287 .set_nmi_mask
= svm_set_nmi_mask
,
4288 .enable_nmi_window
= enable_nmi_window
,
4289 .enable_irq_window
= enable_irq_window
,
4290 .update_cr8_intercept
= update_cr8_intercept
,
4292 .set_tss_addr
= svm_set_tss_addr
,
4293 .get_tdp_level
= get_npt_level
,
4294 .get_mt_mask
= svm_get_mt_mask
,
4296 .get_exit_info
= svm_get_exit_info
,
4298 .get_lpage_level
= svm_get_lpage_level
,
4300 .cpuid_update
= svm_cpuid_update
,
4302 .rdtscp_supported
= svm_rdtscp_supported
,
4304 .set_supported_cpuid
= svm_set_supported_cpuid
,
4306 .has_wbinvd_exit
= svm_has_wbinvd_exit
,
4308 .set_tsc_khz
= svm_set_tsc_khz
,
4309 .write_tsc_offset
= svm_write_tsc_offset
,
4310 .adjust_tsc_offset
= svm_adjust_tsc_offset
,
4311 .compute_tsc_offset
= svm_compute_tsc_offset
,
4312 .read_l1_tsc
= svm_read_l1_tsc
,
4314 .set_tdp_cr3
= set_tdp_cr3
,
4316 .check_intercept
= svm_check_intercept
,
4319 static int __init
svm_init(void)
4321 return kvm_init(&svm_x86_ops
, sizeof(struct vcpu_svm
),
4322 __alignof__(struct vcpu_svm
), THIS_MODULE
);
4325 static void __exit
svm_exit(void)
4330 module_init(svm_init
)
4331 module_exit(svm_exit
)