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"
26 #include <linux/module.h>
27 #include <linux/mod_devicetable.h>
28 #include <linux/kernel.h>
29 #include <linux/vmalloc.h>
30 #include <linux/highmem.h>
31 #include <linux/sched.h>
32 #include <linux/trace_events.h>
33 #include <linux/slab.h>
35 #include <asm/perf_event.h>
36 #include <asm/tlbflush.h>
38 #include <asm/debugreg.h>
39 #include <asm/kvm_para.h>
41 #include <asm/virtext.h>
44 #define __ex(x) __kvm_handle_fault_on_reboot(x)
46 MODULE_AUTHOR("Qumranet");
47 MODULE_LICENSE("GPL");
49 static const struct x86_cpu_id svm_cpu_id
[] = {
50 X86_FEATURE_MATCH(X86_FEATURE_SVM
),
53 MODULE_DEVICE_TABLE(x86cpu
, svm_cpu_id
);
55 #define IOPM_ALLOC_ORDER 2
56 #define MSRPM_ALLOC_ORDER 1
58 #define SEG_TYPE_LDT 2
59 #define SEG_TYPE_BUSY_TSS16 3
61 #define SVM_FEATURE_NPT (1 << 0)
62 #define SVM_FEATURE_LBRV (1 << 1)
63 #define SVM_FEATURE_SVML (1 << 2)
64 #define SVM_FEATURE_NRIP (1 << 3)
65 #define SVM_FEATURE_TSC_RATE (1 << 4)
66 #define SVM_FEATURE_VMCB_CLEAN (1 << 5)
67 #define SVM_FEATURE_FLUSH_ASID (1 << 6)
68 #define SVM_FEATURE_DECODE_ASSIST (1 << 7)
69 #define SVM_FEATURE_PAUSE_FILTER (1 << 10)
71 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
72 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
73 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
75 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
77 #define TSC_RATIO_RSVD 0xffffff0000000000ULL
78 #define TSC_RATIO_MIN 0x0000000000000001ULL
79 #define TSC_RATIO_MAX 0x000000ffffffffffULL
81 static bool erratum_383_found __read_mostly
;
83 static const u32 host_save_user_msrs
[] = {
85 MSR_STAR
, MSR_LSTAR
, MSR_CSTAR
, MSR_SYSCALL_MASK
, MSR_KERNEL_GS_BASE
,
88 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
92 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
102 /* These are the merged vectors */
105 /* gpa pointers to the real vectors */
109 /* A VMEXIT is required but not yet emulated */
112 /* cache for intercepts of the guest */
115 u32 intercept_exceptions
;
118 /* Nested Paging related state */
122 #define MSRPM_OFFSETS 16
123 static u32 msrpm_offsets
[MSRPM_OFFSETS
] __read_mostly
;
126 * Set osvw_len to higher value when updated Revision Guides
127 * are published and we know what the new status bits are
129 static uint64_t osvw_len
= 4, osvw_status
;
132 struct kvm_vcpu vcpu
;
134 unsigned long vmcb_pa
;
135 struct svm_cpu_data
*svm_data
;
136 uint64_t asid_generation
;
137 uint64_t sysenter_esp
;
138 uint64_t sysenter_eip
;
143 u64 host_user_msrs
[NR_HOST_SAVE_USER_MSRS
];
155 struct nested_state nested
;
159 unsigned int3_injected
;
160 unsigned long int3_rip
;
163 /* cached guest cpuid flags for faster access */
164 bool nrips_enabled
: 1;
167 static DEFINE_PER_CPU(u64
, current_tsc_ratio
);
168 #define TSC_RATIO_DEFAULT 0x0100000000ULL
170 #define MSR_INVALID 0xffffffffU
172 static const struct svm_direct_access_msrs
{
173 u32 index
; /* Index of the MSR */
174 bool always
; /* True if intercept is always on */
175 } direct_access_msrs
[] = {
176 { .index
= MSR_STAR
, .always
= true },
177 { .index
= MSR_IA32_SYSENTER_CS
, .always
= true },
179 { .index
= MSR_GS_BASE
, .always
= true },
180 { .index
= MSR_FS_BASE
, .always
= true },
181 { .index
= MSR_KERNEL_GS_BASE
, .always
= true },
182 { .index
= MSR_LSTAR
, .always
= true },
183 { .index
= MSR_CSTAR
, .always
= true },
184 { .index
= MSR_SYSCALL_MASK
, .always
= true },
186 { .index
= MSR_IA32_LASTBRANCHFROMIP
, .always
= false },
187 { .index
= MSR_IA32_LASTBRANCHTOIP
, .always
= false },
188 { .index
= MSR_IA32_LASTINTFROMIP
, .always
= false },
189 { .index
= MSR_IA32_LASTINTTOIP
, .always
= false },
190 { .index
= MSR_INVALID
, .always
= false },
193 /* enable NPT for AMD64 and X86 with PAE */
194 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
195 static bool npt_enabled
= true;
197 static bool npt_enabled
;
200 /* allow nested paging (virtualized MMU) for all guests */
201 static int npt
= true;
202 module_param(npt
, int, S_IRUGO
);
204 /* allow nested virtualization in KVM/SVM */
205 static int nested
= true;
206 module_param(nested
, int, S_IRUGO
);
208 static void svm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
);
209 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
);
210 static void svm_complete_interrupts(struct vcpu_svm
*svm
);
212 static int nested_svm_exit_handled(struct vcpu_svm
*svm
);
213 static int nested_svm_intercept(struct vcpu_svm
*svm
);
214 static int nested_svm_vmexit(struct vcpu_svm
*svm
);
215 static int nested_svm_check_exception(struct vcpu_svm
*svm
, unsigned nr
,
216 bool has_error_code
, u32 error_code
);
219 VMCB_INTERCEPTS
, /* Intercept vectors, TSC offset,
220 pause filter count */
221 VMCB_PERM_MAP
, /* IOPM Base and MSRPM Base */
222 VMCB_ASID
, /* ASID */
223 VMCB_INTR
, /* int_ctl, int_vector */
224 VMCB_NPT
, /* npt_en, nCR3, gPAT */
225 VMCB_CR
, /* CR0, CR3, CR4, EFER */
226 VMCB_DR
, /* DR6, DR7 */
227 VMCB_DT
, /* GDT, IDT */
228 VMCB_SEG
, /* CS, DS, SS, ES, CPL */
229 VMCB_CR2
, /* CR2 only */
230 VMCB_LBR
, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
234 /* TPR and CR2 are always written before VMRUN */
235 #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2))
237 static inline void mark_all_dirty(struct vmcb
*vmcb
)
239 vmcb
->control
.clean
= 0;
242 static inline void mark_all_clean(struct vmcb
*vmcb
)
244 vmcb
->control
.clean
= ((1 << VMCB_DIRTY_MAX
) - 1)
245 & ~VMCB_ALWAYS_DIRTY_MASK
;
248 static inline void mark_dirty(struct vmcb
*vmcb
, int bit
)
250 vmcb
->control
.clean
&= ~(1 << bit
);
253 static inline struct vcpu_svm
*to_svm(struct kvm_vcpu
*vcpu
)
255 return container_of(vcpu
, struct vcpu_svm
, vcpu
);
258 static void recalc_intercepts(struct vcpu_svm
*svm
)
260 struct vmcb_control_area
*c
, *h
;
261 struct nested_state
*g
;
263 mark_dirty(svm
->vmcb
, VMCB_INTERCEPTS
);
265 if (!is_guest_mode(&svm
->vcpu
))
268 c
= &svm
->vmcb
->control
;
269 h
= &svm
->nested
.hsave
->control
;
272 c
->intercept_cr
= h
->intercept_cr
| g
->intercept_cr
;
273 c
->intercept_dr
= h
->intercept_dr
| g
->intercept_dr
;
274 c
->intercept_exceptions
= h
->intercept_exceptions
| g
->intercept_exceptions
;
275 c
->intercept
= h
->intercept
| g
->intercept
;
278 static inline struct vmcb
*get_host_vmcb(struct vcpu_svm
*svm
)
280 if (is_guest_mode(&svm
->vcpu
))
281 return svm
->nested
.hsave
;
286 static inline void set_cr_intercept(struct vcpu_svm
*svm
, int bit
)
288 struct vmcb
*vmcb
= get_host_vmcb(svm
);
290 vmcb
->control
.intercept_cr
|= (1U << bit
);
292 recalc_intercepts(svm
);
295 static inline void clr_cr_intercept(struct vcpu_svm
*svm
, int bit
)
297 struct vmcb
*vmcb
= get_host_vmcb(svm
);
299 vmcb
->control
.intercept_cr
&= ~(1U << bit
);
301 recalc_intercepts(svm
);
304 static inline bool is_cr_intercept(struct vcpu_svm
*svm
, int bit
)
306 struct vmcb
*vmcb
= get_host_vmcb(svm
);
308 return vmcb
->control
.intercept_cr
& (1U << bit
);
311 static inline void set_dr_intercepts(struct vcpu_svm
*svm
)
313 struct vmcb
*vmcb
= get_host_vmcb(svm
);
315 vmcb
->control
.intercept_dr
= (1 << INTERCEPT_DR0_READ
)
316 | (1 << INTERCEPT_DR1_READ
)
317 | (1 << INTERCEPT_DR2_READ
)
318 | (1 << INTERCEPT_DR3_READ
)
319 | (1 << INTERCEPT_DR4_READ
)
320 | (1 << INTERCEPT_DR5_READ
)
321 | (1 << INTERCEPT_DR6_READ
)
322 | (1 << INTERCEPT_DR7_READ
)
323 | (1 << INTERCEPT_DR0_WRITE
)
324 | (1 << INTERCEPT_DR1_WRITE
)
325 | (1 << INTERCEPT_DR2_WRITE
)
326 | (1 << INTERCEPT_DR3_WRITE
)
327 | (1 << INTERCEPT_DR4_WRITE
)
328 | (1 << INTERCEPT_DR5_WRITE
)
329 | (1 << INTERCEPT_DR6_WRITE
)
330 | (1 << INTERCEPT_DR7_WRITE
);
332 recalc_intercepts(svm
);
335 static inline void clr_dr_intercepts(struct vcpu_svm
*svm
)
337 struct vmcb
*vmcb
= get_host_vmcb(svm
);
339 vmcb
->control
.intercept_dr
= 0;
341 recalc_intercepts(svm
);
344 static inline void set_exception_intercept(struct vcpu_svm
*svm
, int bit
)
346 struct vmcb
*vmcb
= get_host_vmcb(svm
);
348 vmcb
->control
.intercept_exceptions
|= (1U << bit
);
350 recalc_intercepts(svm
);
353 static inline void clr_exception_intercept(struct vcpu_svm
*svm
, int bit
)
355 struct vmcb
*vmcb
= get_host_vmcb(svm
);
357 vmcb
->control
.intercept_exceptions
&= ~(1U << bit
);
359 recalc_intercepts(svm
);
362 static inline void set_intercept(struct vcpu_svm
*svm
, int bit
)
364 struct vmcb
*vmcb
= get_host_vmcb(svm
);
366 vmcb
->control
.intercept
|= (1ULL << bit
);
368 recalc_intercepts(svm
);
371 static inline void clr_intercept(struct vcpu_svm
*svm
, int bit
)
373 struct vmcb
*vmcb
= get_host_vmcb(svm
);
375 vmcb
->control
.intercept
&= ~(1ULL << bit
);
377 recalc_intercepts(svm
);
380 static inline void enable_gif(struct vcpu_svm
*svm
)
382 svm
->vcpu
.arch
.hflags
|= HF_GIF_MASK
;
385 static inline void disable_gif(struct vcpu_svm
*svm
)
387 svm
->vcpu
.arch
.hflags
&= ~HF_GIF_MASK
;
390 static inline bool gif_set(struct vcpu_svm
*svm
)
392 return !!(svm
->vcpu
.arch
.hflags
& HF_GIF_MASK
);
395 static unsigned long iopm_base
;
397 struct kvm_ldttss_desc
{
400 unsigned base1
:8, type
:5, dpl
:2, p
:1;
401 unsigned limit1
:4, zero0
:3, g
:1, base2
:8;
404 } __attribute__((packed
));
406 struct svm_cpu_data
{
412 struct kvm_ldttss_desc
*tss_desc
;
414 struct page
*save_area
;
417 static DEFINE_PER_CPU(struct svm_cpu_data
*, svm_data
);
419 struct svm_init_data
{
424 static const u32 msrpm_ranges
[] = {0, 0xc0000000, 0xc0010000};
426 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
427 #define MSRS_RANGE_SIZE 2048
428 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
430 static u32
svm_msrpm_offset(u32 msr
)
435 for (i
= 0; i
< NUM_MSR_MAPS
; i
++) {
436 if (msr
< msrpm_ranges
[i
] ||
437 msr
>= msrpm_ranges
[i
] + MSRS_IN_RANGE
)
440 offset
= (msr
- msrpm_ranges
[i
]) / 4; /* 4 msrs per u8 */
441 offset
+= (i
* MSRS_RANGE_SIZE
); /* add range offset */
443 /* Now we have the u8 offset - but need the u32 offset */
447 /* MSR not in any range */
451 #define MAX_INST_SIZE 15
453 static inline void clgi(void)
455 asm volatile (__ex(SVM_CLGI
));
458 static inline void stgi(void)
460 asm volatile (__ex(SVM_STGI
));
463 static inline void invlpga(unsigned long addr
, u32 asid
)
465 asm volatile (__ex(SVM_INVLPGA
) : : "a"(addr
), "c"(asid
));
468 static int get_npt_level(void)
471 return PT64_ROOT_LEVEL
;
473 return PT32E_ROOT_LEVEL
;
477 static void svm_set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
479 vcpu
->arch
.efer
= efer
;
480 if (!npt_enabled
&& !(efer
& EFER_LMA
))
483 to_svm(vcpu
)->vmcb
->save
.efer
= efer
| EFER_SVME
;
484 mark_dirty(to_svm(vcpu
)->vmcb
, VMCB_CR
);
487 static int is_external_interrupt(u32 info
)
489 info
&= SVM_EVTINJ_TYPE_MASK
| SVM_EVTINJ_VALID
;
490 return info
== (SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
);
493 static u32
svm_get_interrupt_shadow(struct kvm_vcpu
*vcpu
)
495 struct vcpu_svm
*svm
= to_svm(vcpu
);
498 if (svm
->vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
)
499 ret
= KVM_X86_SHADOW_INT_STI
| KVM_X86_SHADOW_INT_MOV_SS
;
503 static void svm_set_interrupt_shadow(struct kvm_vcpu
*vcpu
, int mask
)
505 struct vcpu_svm
*svm
= to_svm(vcpu
);
508 svm
->vmcb
->control
.int_state
&= ~SVM_INTERRUPT_SHADOW_MASK
;
510 svm
->vmcb
->control
.int_state
|= SVM_INTERRUPT_SHADOW_MASK
;
514 static void skip_emulated_instruction(struct kvm_vcpu
*vcpu
)
516 struct vcpu_svm
*svm
= to_svm(vcpu
);
518 if (svm
->vmcb
->control
.next_rip
!= 0) {
519 WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS
));
520 svm
->next_rip
= svm
->vmcb
->control
.next_rip
;
523 if (!svm
->next_rip
) {
524 if (emulate_instruction(vcpu
, EMULTYPE_SKIP
) !=
526 printk(KERN_DEBUG
"%s: NOP\n", __func__
);
529 if (svm
->next_rip
- kvm_rip_read(vcpu
) > MAX_INST_SIZE
)
530 printk(KERN_ERR
"%s: ip 0x%lx next 0x%llx\n",
531 __func__
, kvm_rip_read(vcpu
), svm
->next_rip
);
533 kvm_rip_write(vcpu
, svm
->next_rip
);
534 svm_set_interrupt_shadow(vcpu
, 0);
537 static void svm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
,
538 bool has_error_code
, u32 error_code
,
541 struct vcpu_svm
*svm
= to_svm(vcpu
);
544 * If we are within a nested VM we'd better #VMEXIT and let the guest
545 * handle the exception
548 nested_svm_check_exception(svm
, nr
, has_error_code
, error_code
))
551 if (nr
== BP_VECTOR
&& !static_cpu_has(X86_FEATURE_NRIPS
)) {
552 unsigned long rip
, old_rip
= kvm_rip_read(&svm
->vcpu
);
555 * For guest debugging where we have to reinject #BP if some
556 * INT3 is guest-owned:
557 * Emulate nRIP by moving RIP forward. Will fail if injection
558 * raises a fault that is not intercepted. Still better than
559 * failing in all cases.
561 skip_emulated_instruction(&svm
->vcpu
);
562 rip
= kvm_rip_read(&svm
->vcpu
);
563 svm
->int3_rip
= rip
+ svm
->vmcb
->save
.cs
.base
;
564 svm
->int3_injected
= rip
- old_rip
;
567 svm
->vmcb
->control
.event_inj
= nr
569 | (has_error_code
? SVM_EVTINJ_VALID_ERR
: 0)
570 | SVM_EVTINJ_TYPE_EXEPT
;
571 svm
->vmcb
->control
.event_inj_err
= error_code
;
574 static void svm_init_erratum_383(void)
580 if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH
))
583 /* Use _safe variants to not break nested virtualization */
584 val
= native_read_msr_safe(MSR_AMD64_DC_CFG
, &err
);
590 low
= lower_32_bits(val
);
591 high
= upper_32_bits(val
);
593 native_write_msr_safe(MSR_AMD64_DC_CFG
, low
, high
);
595 erratum_383_found
= true;
598 static void svm_init_osvw(struct kvm_vcpu
*vcpu
)
601 * Guests should see errata 400 and 415 as fixed (assuming that
602 * HLT and IO instructions are intercepted).
604 vcpu
->arch
.osvw
.length
= (osvw_len
>= 3) ? (osvw_len
) : 3;
605 vcpu
->arch
.osvw
.status
= osvw_status
& ~(6ULL);
608 * By increasing VCPU's osvw.length to 3 we are telling the guest that
609 * all osvw.status bits inside that length, including bit 0 (which is
610 * reserved for erratum 298), are valid. However, if host processor's
611 * osvw_len is 0 then osvw_status[0] carries no information. We need to
612 * be conservative here and therefore we tell the guest that erratum 298
613 * is present (because we really don't know).
615 if (osvw_len
== 0 && boot_cpu_data
.x86
== 0x10)
616 vcpu
->arch
.osvw
.status
|= 1;
619 static int has_svm(void)
623 if (!cpu_has_svm(&msg
)) {
624 printk(KERN_INFO
"has_svm: %s\n", msg
);
631 static void svm_hardware_disable(void)
633 /* Make sure we clean up behind us */
634 if (static_cpu_has(X86_FEATURE_TSCRATEMSR
))
635 wrmsrl(MSR_AMD64_TSC_RATIO
, TSC_RATIO_DEFAULT
);
639 amd_pmu_disable_virt();
642 static int svm_hardware_enable(void)
645 struct svm_cpu_data
*sd
;
647 struct desc_ptr gdt_descr
;
648 struct desc_struct
*gdt
;
649 int me
= raw_smp_processor_id();
651 rdmsrl(MSR_EFER
, efer
);
652 if (efer
& EFER_SVME
)
656 pr_err("%s: err EOPNOTSUPP on %d\n", __func__
, me
);
659 sd
= per_cpu(svm_data
, me
);
661 pr_err("%s: svm_data is NULL on %d\n", __func__
, me
);
665 sd
->asid_generation
= 1;
666 sd
->max_asid
= cpuid_ebx(SVM_CPUID_FUNC
) - 1;
667 sd
->next_asid
= sd
->max_asid
+ 1;
669 native_store_gdt(&gdt_descr
);
670 gdt
= (struct desc_struct
*)gdt_descr
.address
;
671 sd
->tss_desc
= (struct kvm_ldttss_desc
*)(gdt
+ GDT_ENTRY_TSS
);
673 wrmsrl(MSR_EFER
, efer
| EFER_SVME
);
675 wrmsrl(MSR_VM_HSAVE_PA
, page_to_pfn(sd
->save_area
) << PAGE_SHIFT
);
677 if (static_cpu_has(X86_FEATURE_TSCRATEMSR
)) {
678 wrmsrl(MSR_AMD64_TSC_RATIO
, TSC_RATIO_DEFAULT
);
679 __this_cpu_write(current_tsc_ratio
, TSC_RATIO_DEFAULT
);
686 * Note that it is possible to have a system with mixed processor
687 * revisions and therefore different OSVW bits. If bits are not the same
688 * on different processors then choose the worst case (i.e. if erratum
689 * is present on one processor and not on another then assume that the
690 * erratum is present everywhere).
692 if (cpu_has(&boot_cpu_data
, X86_FEATURE_OSVW
)) {
693 uint64_t len
, status
= 0;
696 len
= native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH
, &err
);
698 status
= native_read_msr_safe(MSR_AMD64_OSVW_STATUS
,
702 osvw_status
= osvw_len
= 0;
706 osvw_status
|= status
;
707 osvw_status
&= (1ULL << osvw_len
) - 1;
710 osvw_status
= osvw_len
= 0;
712 svm_init_erratum_383();
714 amd_pmu_enable_virt();
719 static void svm_cpu_uninit(int cpu
)
721 struct svm_cpu_data
*sd
= per_cpu(svm_data
, raw_smp_processor_id());
726 per_cpu(svm_data
, raw_smp_processor_id()) = NULL
;
727 __free_page(sd
->save_area
);
731 static int svm_cpu_init(int cpu
)
733 struct svm_cpu_data
*sd
;
736 sd
= kzalloc(sizeof(struct svm_cpu_data
), GFP_KERNEL
);
740 sd
->save_area
= alloc_page(GFP_KERNEL
);
745 per_cpu(svm_data
, cpu
) = sd
;
755 static bool valid_msr_intercept(u32 index
)
759 for (i
= 0; direct_access_msrs
[i
].index
!= MSR_INVALID
; i
++)
760 if (direct_access_msrs
[i
].index
== index
)
766 static void set_msr_interception(u32
*msrpm
, unsigned msr
,
769 u8 bit_read
, bit_write
;
774 * If this warning triggers extend the direct_access_msrs list at the
775 * beginning of the file
777 WARN_ON(!valid_msr_intercept(msr
));
779 offset
= svm_msrpm_offset(msr
);
780 bit_read
= 2 * (msr
& 0x0f);
781 bit_write
= 2 * (msr
& 0x0f) + 1;
784 BUG_ON(offset
== MSR_INVALID
);
786 read
? clear_bit(bit_read
, &tmp
) : set_bit(bit_read
, &tmp
);
787 write
? clear_bit(bit_write
, &tmp
) : set_bit(bit_write
, &tmp
);
792 static void svm_vcpu_init_msrpm(u32
*msrpm
)
796 memset(msrpm
, 0xff, PAGE_SIZE
* (1 << MSRPM_ALLOC_ORDER
));
798 for (i
= 0; direct_access_msrs
[i
].index
!= MSR_INVALID
; i
++) {
799 if (!direct_access_msrs
[i
].always
)
802 set_msr_interception(msrpm
, direct_access_msrs
[i
].index
, 1, 1);
806 static void add_msr_offset(u32 offset
)
810 for (i
= 0; i
< MSRPM_OFFSETS
; ++i
) {
812 /* Offset already in list? */
813 if (msrpm_offsets
[i
] == offset
)
816 /* Slot used by another offset? */
817 if (msrpm_offsets
[i
] != MSR_INVALID
)
820 /* Add offset to list */
821 msrpm_offsets
[i
] = offset
;
827 * If this BUG triggers the msrpm_offsets table has an overflow. Just
828 * increase MSRPM_OFFSETS in this case.
833 static void init_msrpm_offsets(void)
837 memset(msrpm_offsets
, 0xff, sizeof(msrpm_offsets
));
839 for (i
= 0; direct_access_msrs
[i
].index
!= MSR_INVALID
; i
++) {
842 offset
= svm_msrpm_offset(direct_access_msrs
[i
].index
);
843 BUG_ON(offset
== MSR_INVALID
);
845 add_msr_offset(offset
);
849 static void svm_enable_lbrv(struct vcpu_svm
*svm
)
851 u32
*msrpm
= svm
->msrpm
;
853 svm
->vmcb
->control
.lbr_ctl
= 1;
854 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHFROMIP
, 1, 1);
855 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHTOIP
, 1, 1);
856 set_msr_interception(msrpm
, MSR_IA32_LASTINTFROMIP
, 1, 1);
857 set_msr_interception(msrpm
, MSR_IA32_LASTINTTOIP
, 1, 1);
860 static void svm_disable_lbrv(struct vcpu_svm
*svm
)
862 u32
*msrpm
= svm
->msrpm
;
864 svm
->vmcb
->control
.lbr_ctl
= 0;
865 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHFROMIP
, 0, 0);
866 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHTOIP
, 0, 0);
867 set_msr_interception(msrpm
, MSR_IA32_LASTINTFROMIP
, 0, 0);
868 set_msr_interception(msrpm
, MSR_IA32_LASTINTTOIP
, 0, 0);
871 static __init
int svm_hardware_setup(void)
874 struct page
*iopm_pages
;
878 iopm_pages
= alloc_pages(GFP_KERNEL
, IOPM_ALLOC_ORDER
);
883 iopm_va
= page_address(iopm_pages
);
884 memset(iopm_va
, 0xff, PAGE_SIZE
* (1 << IOPM_ALLOC_ORDER
));
885 iopm_base
= page_to_pfn(iopm_pages
) << PAGE_SHIFT
;
887 init_msrpm_offsets();
889 if (boot_cpu_has(X86_FEATURE_NX
))
890 kvm_enable_efer_bits(EFER_NX
);
892 if (boot_cpu_has(X86_FEATURE_FXSR_OPT
))
893 kvm_enable_efer_bits(EFER_FFXSR
);
895 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR
)) {
896 kvm_has_tsc_control
= true;
897 kvm_max_tsc_scaling_ratio
= TSC_RATIO_MAX
;
898 kvm_tsc_scaling_ratio_frac_bits
= 32;
902 printk(KERN_INFO
"kvm: Nested Virtualization enabled\n");
903 kvm_enable_efer_bits(EFER_SVME
| EFER_LMSLE
);
906 for_each_possible_cpu(cpu
) {
907 r
= svm_cpu_init(cpu
);
912 if (!boot_cpu_has(X86_FEATURE_NPT
))
915 if (npt_enabled
&& !npt
) {
916 printk(KERN_INFO
"kvm: Nested Paging disabled\n");
921 printk(KERN_INFO
"kvm: Nested Paging enabled\n");
929 __free_pages(iopm_pages
, IOPM_ALLOC_ORDER
);
934 static __exit
void svm_hardware_unsetup(void)
938 for_each_possible_cpu(cpu
)
941 __free_pages(pfn_to_page(iopm_base
>> PAGE_SHIFT
), IOPM_ALLOC_ORDER
);
945 static void init_seg(struct vmcb_seg
*seg
)
948 seg
->attrib
= SVM_SELECTOR_P_MASK
| SVM_SELECTOR_S_MASK
|
949 SVM_SELECTOR_WRITE_MASK
; /* Read/Write Data Segment */
954 static void init_sys_seg(struct vmcb_seg
*seg
, uint32_t type
)
957 seg
->attrib
= SVM_SELECTOR_P_MASK
| type
;
962 static u64
svm_read_tsc_offset(struct kvm_vcpu
*vcpu
)
964 struct vcpu_svm
*svm
= to_svm(vcpu
);
966 return svm
->vmcb
->control
.tsc_offset
;
969 static void svm_write_tsc_offset(struct kvm_vcpu
*vcpu
, u64 offset
)
971 struct vcpu_svm
*svm
= to_svm(vcpu
);
972 u64 g_tsc_offset
= 0;
974 if (is_guest_mode(vcpu
)) {
975 g_tsc_offset
= svm
->vmcb
->control
.tsc_offset
-
976 svm
->nested
.hsave
->control
.tsc_offset
;
977 svm
->nested
.hsave
->control
.tsc_offset
= offset
;
979 trace_kvm_write_tsc_offset(vcpu
->vcpu_id
,
980 svm
->vmcb
->control
.tsc_offset
,
983 svm
->vmcb
->control
.tsc_offset
= offset
+ g_tsc_offset
;
985 mark_dirty(svm
->vmcb
, VMCB_INTERCEPTS
);
988 static void svm_adjust_tsc_offset_guest(struct kvm_vcpu
*vcpu
, s64 adjustment
)
990 struct vcpu_svm
*svm
= to_svm(vcpu
);
992 svm
->vmcb
->control
.tsc_offset
+= adjustment
;
993 if (is_guest_mode(vcpu
))
994 svm
->nested
.hsave
->control
.tsc_offset
+= adjustment
;
996 trace_kvm_write_tsc_offset(vcpu
->vcpu_id
,
997 svm
->vmcb
->control
.tsc_offset
- adjustment
,
998 svm
->vmcb
->control
.tsc_offset
);
1000 mark_dirty(svm
->vmcb
, VMCB_INTERCEPTS
);
1003 static void init_vmcb(struct vcpu_svm
*svm
)
1005 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
1006 struct vmcb_save_area
*save
= &svm
->vmcb
->save
;
1008 svm
->vcpu
.fpu_active
= 1;
1009 svm
->vcpu
.arch
.hflags
= 0;
1011 set_cr_intercept(svm
, INTERCEPT_CR0_READ
);
1012 set_cr_intercept(svm
, INTERCEPT_CR3_READ
);
1013 set_cr_intercept(svm
, INTERCEPT_CR4_READ
);
1014 set_cr_intercept(svm
, INTERCEPT_CR0_WRITE
);
1015 set_cr_intercept(svm
, INTERCEPT_CR3_WRITE
);
1016 set_cr_intercept(svm
, INTERCEPT_CR4_WRITE
);
1017 set_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
1019 set_dr_intercepts(svm
);
1021 set_exception_intercept(svm
, PF_VECTOR
);
1022 set_exception_intercept(svm
, UD_VECTOR
);
1023 set_exception_intercept(svm
, MC_VECTOR
);
1024 set_exception_intercept(svm
, AC_VECTOR
);
1025 set_exception_intercept(svm
, DB_VECTOR
);
1027 set_intercept(svm
, INTERCEPT_INTR
);
1028 set_intercept(svm
, INTERCEPT_NMI
);
1029 set_intercept(svm
, INTERCEPT_SMI
);
1030 set_intercept(svm
, INTERCEPT_SELECTIVE_CR0
);
1031 set_intercept(svm
, INTERCEPT_RDPMC
);
1032 set_intercept(svm
, INTERCEPT_CPUID
);
1033 set_intercept(svm
, INTERCEPT_INVD
);
1034 set_intercept(svm
, INTERCEPT_HLT
);
1035 set_intercept(svm
, INTERCEPT_INVLPG
);
1036 set_intercept(svm
, INTERCEPT_INVLPGA
);
1037 set_intercept(svm
, INTERCEPT_IOIO_PROT
);
1038 set_intercept(svm
, INTERCEPT_MSR_PROT
);
1039 set_intercept(svm
, INTERCEPT_TASK_SWITCH
);
1040 set_intercept(svm
, INTERCEPT_SHUTDOWN
);
1041 set_intercept(svm
, INTERCEPT_VMRUN
);
1042 set_intercept(svm
, INTERCEPT_VMMCALL
);
1043 set_intercept(svm
, INTERCEPT_VMLOAD
);
1044 set_intercept(svm
, INTERCEPT_VMSAVE
);
1045 set_intercept(svm
, INTERCEPT_STGI
);
1046 set_intercept(svm
, INTERCEPT_CLGI
);
1047 set_intercept(svm
, INTERCEPT_SKINIT
);
1048 set_intercept(svm
, INTERCEPT_WBINVD
);
1049 set_intercept(svm
, INTERCEPT_MONITOR
);
1050 set_intercept(svm
, INTERCEPT_MWAIT
);
1051 set_intercept(svm
, INTERCEPT_XSETBV
);
1053 control
->iopm_base_pa
= iopm_base
;
1054 control
->msrpm_base_pa
= __pa(svm
->msrpm
);
1055 control
->int_ctl
= V_INTR_MASKING_MASK
;
1057 init_seg(&save
->es
);
1058 init_seg(&save
->ss
);
1059 init_seg(&save
->ds
);
1060 init_seg(&save
->fs
);
1061 init_seg(&save
->gs
);
1063 save
->cs
.selector
= 0xf000;
1064 save
->cs
.base
= 0xffff0000;
1065 /* Executable/Readable Code Segment */
1066 save
->cs
.attrib
= SVM_SELECTOR_READ_MASK
| SVM_SELECTOR_P_MASK
|
1067 SVM_SELECTOR_S_MASK
| SVM_SELECTOR_CODE_MASK
;
1068 save
->cs
.limit
= 0xffff;
1070 save
->gdtr
.limit
= 0xffff;
1071 save
->idtr
.limit
= 0xffff;
1073 init_sys_seg(&save
->ldtr
, SEG_TYPE_LDT
);
1074 init_sys_seg(&save
->tr
, SEG_TYPE_BUSY_TSS16
);
1076 svm_set_efer(&svm
->vcpu
, 0);
1077 save
->dr6
= 0xffff0ff0;
1078 kvm_set_rflags(&svm
->vcpu
, 2);
1079 save
->rip
= 0x0000fff0;
1080 svm
->vcpu
.arch
.regs
[VCPU_REGS_RIP
] = save
->rip
;
1083 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1084 * It also updates the guest-visible cr0 value.
1086 svm_set_cr0(&svm
->vcpu
, X86_CR0_NW
| X86_CR0_CD
| X86_CR0_ET
);
1087 kvm_mmu_reset_context(&svm
->vcpu
);
1089 save
->cr4
= X86_CR4_PAE
;
1093 /* Setup VMCB for Nested Paging */
1094 control
->nested_ctl
= 1;
1095 clr_intercept(svm
, INTERCEPT_INVLPG
);
1096 clr_exception_intercept(svm
, PF_VECTOR
);
1097 clr_cr_intercept(svm
, INTERCEPT_CR3_READ
);
1098 clr_cr_intercept(svm
, INTERCEPT_CR3_WRITE
);
1099 save
->g_pat
= svm
->vcpu
.arch
.pat
;
1103 svm
->asid_generation
= 0;
1105 svm
->nested
.vmcb
= 0;
1106 svm
->vcpu
.arch
.hflags
= 0;
1108 if (boot_cpu_has(X86_FEATURE_PAUSEFILTER
)) {
1109 control
->pause_filter_count
= 3000;
1110 set_intercept(svm
, INTERCEPT_PAUSE
);
1113 mark_all_dirty(svm
->vmcb
);
1118 static void svm_vcpu_reset(struct kvm_vcpu
*vcpu
, bool init_event
)
1120 struct vcpu_svm
*svm
= to_svm(vcpu
);
1125 svm
->vcpu
.arch
.apic_base
= APIC_DEFAULT_PHYS_BASE
|
1126 MSR_IA32_APICBASE_ENABLE
;
1127 if (kvm_vcpu_is_reset_bsp(&svm
->vcpu
))
1128 svm
->vcpu
.arch
.apic_base
|= MSR_IA32_APICBASE_BSP
;
1132 kvm_cpuid(vcpu
, &eax
, &dummy
, &dummy
, &dummy
);
1133 kvm_register_write(vcpu
, VCPU_REGS_RDX
, eax
);
1136 static struct kvm_vcpu
*svm_create_vcpu(struct kvm
*kvm
, unsigned int id
)
1138 struct vcpu_svm
*svm
;
1140 struct page
*msrpm_pages
;
1141 struct page
*hsave_page
;
1142 struct page
*nested_msrpm_pages
;
1145 svm
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
1151 err
= kvm_vcpu_init(&svm
->vcpu
, kvm
, id
);
1156 page
= alloc_page(GFP_KERNEL
);
1160 msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
1164 nested_msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
1165 if (!nested_msrpm_pages
)
1168 hsave_page
= alloc_page(GFP_KERNEL
);
1172 svm
->nested
.hsave
= page_address(hsave_page
);
1174 svm
->msrpm
= page_address(msrpm_pages
);
1175 svm_vcpu_init_msrpm(svm
->msrpm
);
1177 svm
->nested
.msrpm
= page_address(nested_msrpm_pages
);
1178 svm_vcpu_init_msrpm(svm
->nested
.msrpm
);
1180 svm
->vmcb
= page_address(page
);
1181 clear_page(svm
->vmcb
);
1182 svm
->vmcb_pa
= page_to_pfn(page
) << PAGE_SHIFT
;
1183 svm
->asid_generation
= 0;
1186 svm_init_osvw(&svm
->vcpu
);
1191 __free_pages(nested_msrpm_pages
, MSRPM_ALLOC_ORDER
);
1193 __free_pages(msrpm_pages
, MSRPM_ALLOC_ORDER
);
1197 kvm_vcpu_uninit(&svm
->vcpu
);
1199 kmem_cache_free(kvm_vcpu_cache
, svm
);
1201 return ERR_PTR(err
);
1204 static void svm_free_vcpu(struct kvm_vcpu
*vcpu
)
1206 struct vcpu_svm
*svm
= to_svm(vcpu
);
1208 __free_page(pfn_to_page(svm
->vmcb_pa
>> PAGE_SHIFT
));
1209 __free_pages(virt_to_page(svm
->msrpm
), MSRPM_ALLOC_ORDER
);
1210 __free_page(virt_to_page(svm
->nested
.hsave
));
1211 __free_pages(virt_to_page(svm
->nested
.msrpm
), MSRPM_ALLOC_ORDER
);
1212 kvm_vcpu_uninit(vcpu
);
1213 kmem_cache_free(kvm_vcpu_cache
, svm
);
1216 static void svm_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
1218 struct vcpu_svm
*svm
= to_svm(vcpu
);
1221 if (unlikely(cpu
!= vcpu
->cpu
)) {
1222 svm
->asid_generation
= 0;
1223 mark_all_dirty(svm
->vmcb
);
1226 #ifdef CONFIG_X86_64
1227 rdmsrl(MSR_GS_BASE
, to_svm(vcpu
)->host
.gs_base
);
1229 savesegment(fs
, svm
->host
.fs
);
1230 savesegment(gs
, svm
->host
.gs
);
1231 svm
->host
.ldt
= kvm_read_ldt();
1233 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
1234 rdmsrl(host_save_user_msrs
[i
], svm
->host_user_msrs
[i
]);
1236 if (static_cpu_has(X86_FEATURE_TSCRATEMSR
)) {
1237 u64 tsc_ratio
= vcpu
->arch
.tsc_scaling_ratio
;
1238 if (tsc_ratio
!= __this_cpu_read(current_tsc_ratio
)) {
1239 __this_cpu_write(current_tsc_ratio
, tsc_ratio
);
1240 wrmsrl(MSR_AMD64_TSC_RATIO
, tsc_ratio
);
1243 /* This assumes that the kernel never uses MSR_TSC_AUX */
1244 if (static_cpu_has(X86_FEATURE_RDTSCP
))
1245 wrmsrl(MSR_TSC_AUX
, svm
->tsc_aux
);
1248 static void svm_vcpu_put(struct kvm_vcpu
*vcpu
)
1250 struct vcpu_svm
*svm
= to_svm(vcpu
);
1253 ++vcpu
->stat
.host_state_reload
;
1254 kvm_load_ldt(svm
->host
.ldt
);
1255 #ifdef CONFIG_X86_64
1256 loadsegment(fs
, svm
->host
.fs
);
1257 wrmsrl(MSR_KERNEL_GS_BASE
, current
->thread
.gs
);
1258 load_gs_index(svm
->host
.gs
);
1260 #ifdef CONFIG_X86_32_LAZY_GS
1261 loadsegment(gs
, svm
->host
.gs
);
1264 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
1265 wrmsrl(host_save_user_msrs
[i
], svm
->host_user_msrs
[i
]);
1268 static unsigned long svm_get_rflags(struct kvm_vcpu
*vcpu
)
1270 return to_svm(vcpu
)->vmcb
->save
.rflags
;
1273 static void svm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
1276 * Any change of EFLAGS.VM is accompained by a reload of SS
1277 * (caused by either a task switch or an inter-privilege IRET),
1278 * so we do not need to update the CPL here.
1280 to_svm(vcpu
)->vmcb
->save
.rflags
= rflags
;
1283 static void svm_cache_reg(struct kvm_vcpu
*vcpu
, enum kvm_reg reg
)
1286 case VCPU_EXREG_PDPTR
:
1287 BUG_ON(!npt_enabled
);
1288 load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, kvm_read_cr3(vcpu
));
1295 static void svm_set_vintr(struct vcpu_svm
*svm
)
1297 set_intercept(svm
, INTERCEPT_VINTR
);
1300 static void svm_clear_vintr(struct vcpu_svm
*svm
)
1302 clr_intercept(svm
, INTERCEPT_VINTR
);
1305 static struct vmcb_seg
*svm_seg(struct kvm_vcpu
*vcpu
, int seg
)
1307 struct vmcb_save_area
*save
= &to_svm(vcpu
)->vmcb
->save
;
1310 case VCPU_SREG_CS
: return &save
->cs
;
1311 case VCPU_SREG_DS
: return &save
->ds
;
1312 case VCPU_SREG_ES
: return &save
->es
;
1313 case VCPU_SREG_FS
: return &save
->fs
;
1314 case VCPU_SREG_GS
: return &save
->gs
;
1315 case VCPU_SREG_SS
: return &save
->ss
;
1316 case VCPU_SREG_TR
: return &save
->tr
;
1317 case VCPU_SREG_LDTR
: return &save
->ldtr
;
1323 static u64
svm_get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
1325 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
1330 static void svm_get_segment(struct kvm_vcpu
*vcpu
,
1331 struct kvm_segment
*var
, int seg
)
1333 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
1335 var
->base
= s
->base
;
1336 var
->limit
= s
->limit
;
1337 var
->selector
= s
->selector
;
1338 var
->type
= s
->attrib
& SVM_SELECTOR_TYPE_MASK
;
1339 var
->s
= (s
->attrib
>> SVM_SELECTOR_S_SHIFT
) & 1;
1340 var
->dpl
= (s
->attrib
>> SVM_SELECTOR_DPL_SHIFT
) & 3;
1341 var
->present
= (s
->attrib
>> SVM_SELECTOR_P_SHIFT
) & 1;
1342 var
->avl
= (s
->attrib
>> SVM_SELECTOR_AVL_SHIFT
) & 1;
1343 var
->l
= (s
->attrib
>> SVM_SELECTOR_L_SHIFT
) & 1;
1344 var
->db
= (s
->attrib
>> SVM_SELECTOR_DB_SHIFT
) & 1;
1347 * AMD CPUs circa 2014 track the G bit for all segments except CS.
1348 * However, the SVM spec states that the G bit is not observed by the
1349 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
1350 * So let's synthesize a legal G bit for all segments, this helps
1351 * running KVM nested. It also helps cross-vendor migration, because
1352 * Intel's vmentry has a check on the 'G' bit.
1354 var
->g
= s
->limit
> 0xfffff;
1357 * AMD's VMCB does not have an explicit unusable field, so emulate it
1358 * for cross vendor migration purposes by "not present"
1360 var
->unusable
= !var
->present
|| (var
->type
== 0);
1365 * Work around a bug where the busy flag in the tr selector
1375 * The accessed bit must always be set in the segment
1376 * descriptor cache, although it can be cleared in the
1377 * descriptor, the cached bit always remains at 1. Since
1378 * Intel has a check on this, set it here to support
1379 * cross-vendor migration.
1386 * On AMD CPUs sometimes the DB bit in the segment
1387 * descriptor is left as 1, although the whole segment has
1388 * been made unusable. Clear it here to pass an Intel VMX
1389 * entry check when cross vendor migrating.
1393 var
->dpl
= to_svm(vcpu
)->vmcb
->save
.cpl
;
1398 static int svm_get_cpl(struct kvm_vcpu
*vcpu
)
1400 struct vmcb_save_area
*save
= &to_svm(vcpu
)->vmcb
->save
;
1405 static void svm_get_idt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1407 struct vcpu_svm
*svm
= to_svm(vcpu
);
1409 dt
->size
= svm
->vmcb
->save
.idtr
.limit
;
1410 dt
->address
= svm
->vmcb
->save
.idtr
.base
;
1413 static void svm_set_idt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1415 struct vcpu_svm
*svm
= to_svm(vcpu
);
1417 svm
->vmcb
->save
.idtr
.limit
= dt
->size
;
1418 svm
->vmcb
->save
.idtr
.base
= dt
->address
;
1419 mark_dirty(svm
->vmcb
, VMCB_DT
);
1422 static void svm_get_gdt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1424 struct vcpu_svm
*svm
= to_svm(vcpu
);
1426 dt
->size
= svm
->vmcb
->save
.gdtr
.limit
;
1427 dt
->address
= svm
->vmcb
->save
.gdtr
.base
;
1430 static void svm_set_gdt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1432 struct vcpu_svm
*svm
= to_svm(vcpu
);
1434 svm
->vmcb
->save
.gdtr
.limit
= dt
->size
;
1435 svm
->vmcb
->save
.gdtr
.base
= dt
->address
;
1436 mark_dirty(svm
->vmcb
, VMCB_DT
);
1439 static void svm_decache_cr0_guest_bits(struct kvm_vcpu
*vcpu
)
1443 static void svm_decache_cr3(struct kvm_vcpu
*vcpu
)
1447 static void svm_decache_cr4_guest_bits(struct kvm_vcpu
*vcpu
)
1451 static void update_cr0_intercept(struct vcpu_svm
*svm
)
1453 ulong gcr0
= svm
->vcpu
.arch
.cr0
;
1454 u64
*hcr0
= &svm
->vmcb
->save
.cr0
;
1456 if (!svm
->vcpu
.fpu_active
)
1457 *hcr0
|= SVM_CR0_SELECTIVE_MASK
;
1459 *hcr0
= (*hcr0
& ~SVM_CR0_SELECTIVE_MASK
)
1460 | (gcr0
& SVM_CR0_SELECTIVE_MASK
);
1462 mark_dirty(svm
->vmcb
, VMCB_CR
);
1464 if (gcr0
== *hcr0
&& svm
->vcpu
.fpu_active
) {
1465 clr_cr_intercept(svm
, INTERCEPT_CR0_READ
);
1466 clr_cr_intercept(svm
, INTERCEPT_CR0_WRITE
);
1468 set_cr_intercept(svm
, INTERCEPT_CR0_READ
);
1469 set_cr_intercept(svm
, INTERCEPT_CR0_WRITE
);
1473 static void svm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
1475 struct vcpu_svm
*svm
= to_svm(vcpu
);
1477 #ifdef CONFIG_X86_64
1478 if (vcpu
->arch
.efer
& EFER_LME
) {
1479 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
1480 vcpu
->arch
.efer
|= EFER_LMA
;
1481 svm
->vmcb
->save
.efer
|= EFER_LMA
| EFER_LME
;
1484 if (is_paging(vcpu
) && !(cr0
& X86_CR0_PG
)) {
1485 vcpu
->arch
.efer
&= ~EFER_LMA
;
1486 svm
->vmcb
->save
.efer
&= ~(EFER_LMA
| EFER_LME
);
1490 vcpu
->arch
.cr0
= cr0
;
1493 cr0
|= X86_CR0_PG
| X86_CR0_WP
;
1495 if (!vcpu
->fpu_active
)
1498 * re-enable caching here because the QEMU bios
1499 * does not do it - this results in some delay at
1502 if (kvm_check_has_quirk(vcpu
->kvm
, KVM_X86_QUIRK_CD_NW_CLEARED
))
1503 cr0
&= ~(X86_CR0_CD
| X86_CR0_NW
);
1504 svm
->vmcb
->save
.cr0
= cr0
;
1505 mark_dirty(svm
->vmcb
, VMCB_CR
);
1506 update_cr0_intercept(svm
);
1509 static int svm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
1511 unsigned long host_cr4_mce
= cr4_read_shadow() & X86_CR4_MCE
;
1512 unsigned long old_cr4
= to_svm(vcpu
)->vmcb
->save
.cr4
;
1514 if (cr4
& X86_CR4_VMXE
)
1517 if (npt_enabled
&& ((old_cr4
^ cr4
) & X86_CR4_PGE
))
1518 svm_flush_tlb(vcpu
);
1520 vcpu
->arch
.cr4
= cr4
;
1523 cr4
|= host_cr4_mce
;
1524 to_svm(vcpu
)->vmcb
->save
.cr4
= cr4
;
1525 mark_dirty(to_svm(vcpu
)->vmcb
, VMCB_CR
);
1529 static void svm_set_segment(struct kvm_vcpu
*vcpu
,
1530 struct kvm_segment
*var
, int seg
)
1532 struct vcpu_svm
*svm
= to_svm(vcpu
);
1533 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
1535 s
->base
= var
->base
;
1536 s
->limit
= var
->limit
;
1537 s
->selector
= var
->selector
;
1541 s
->attrib
= (var
->type
& SVM_SELECTOR_TYPE_MASK
);
1542 s
->attrib
|= (var
->s
& 1) << SVM_SELECTOR_S_SHIFT
;
1543 s
->attrib
|= (var
->dpl
& 3) << SVM_SELECTOR_DPL_SHIFT
;
1544 s
->attrib
|= (var
->present
& 1) << SVM_SELECTOR_P_SHIFT
;
1545 s
->attrib
|= (var
->avl
& 1) << SVM_SELECTOR_AVL_SHIFT
;
1546 s
->attrib
|= (var
->l
& 1) << SVM_SELECTOR_L_SHIFT
;
1547 s
->attrib
|= (var
->db
& 1) << SVM_SELECTOR_DB_SHIFT
;
1548 s
->attrib
|= (var
->g
& 1) << SVM_SELECTOR_G_SHIFT
;
1552 * This is always accurate, except if SYSRET returned to a segment
1553 * with SS.DPL != 3. Intel does not have this quirk, and always
1554 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
1555 * would entail passing the CPL to userspace and back.
1557 if (seg
== VCPU_SREG_SS
)
1558 svm
->vmcb
->save
.cpl
= (s
->attrib
>> SVM_SELECTOR_DPL_SHIFT
) & 3;
1560 mark_dirty(svm
->vmcb
, VMCB_SEG
);
1563 static void update_bp_intercept(struct kvm_vcpu
*vcpu
)
1565 struct vcpu_svm
*svm
= to_svm(vcpu
);
1567 clr_exception_intercept(svm
, BP_VECTOR
);
1569 if (vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
) {
1570 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_SW_BP
)
1571 set_exception_intercept(svm
, BP_VECTOR
);
1573 vcpu
->guest_debug
= 0;
1576 static void new_asid(struct vcpu_svm
*svm
, struct svm_cpu_data
*sd
)
1578 if (sd
->next_asid
> sd
->max_asid
) {
1579 ++sd
->asid_generation
;
1581 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ALL_ASID
;
1584 svm
->asid_generation
= sd
->asid_generation
;
1585 svm
->vmcb
->control
.asid
= sd
->next_asid
++;
1587 mark_dirty(svm
->vmcb
, VMCB_ASID
);
1590 static u64
svm_get_dr6(struct kvm_vcpu
*vcpu
)
1592 return to_svm(vcpu
)->vmcb
->save
.dr6
;
1595 static void svm_set_dr6(struct kvm_vcpu
*vcpu
, unsigned long value
)
1597 struct vcpu_svm
*svm
= to_svm(vcpu
);
1599 svm
->vmcb
->save
.dr6
= value
;
1600 mark_dirty(svm
->vmcb
, VMCB_DR
);
1603 static void svm_sync_dirty_debug_regs(struct kvm_vcpu
*vcpu
)
1605 struct vcpu_svm
*svm
= to_svm(vcpu
);
1607 get_debugreg(vcpu
->arch
.db
[0], 0);
1608 get_debugreg(vcpu
->arch
.db
[1], 1);
1609 get_debugreg(vcpu
->arch
.db
[2], 2);
1610 get_debugreg(vcpu
->arch
.db
[3], 3);
1611 vcpu
->arch
.dr6
= svm_get_dr6(vcpu
);
1612 vcpu
->arch
.dr7
= svm
->vmcb
->save
.dr7
;
1614 vcpu
->arch
.switch_db_regs
&= ~KVM_DEBUGREG_WONT_EXIT
;
1615 set_dr_intercepts(svm
);
1618 static void svm_set_dr7(struct kvm_vcpu
*vcpu
, unsigned long value
)
1620 struct vcpu_svm
*svm
= to_svm(vcpu
);
1622 svm
->vmcb
->save
.dr7
= value
;
1623 mark_dirty(svm
->vmcb
, VMCB_DR
);
1626 static int pf_interception(struct vcpu_svm
*svm
)
1628 u64 fault_address
= svm
->vmcb
->control
.exit_info_2
;
1632 switch (svm
->apf_reason
) {
1634 error_code
= svm
->vmcb
->control
.exit_info_1
;
1636 trace_kvm_page_fault(fault_address
, error_code
);
1637 if (!npt_enabled
&& kvm_event_needs_reinjection(&svm
->vcpu
))
1638 kvm_mmu_unprotect_page_virt(&svm
->vcpu
, fault_address
);
1639 r
= kvm_mmu_page_fault(&svm
->vcpu
, fault_address
, error_code
,
1640 svm
->vmcb
->control
.insn_bytes
,
1641 svm
->vmcb
->control
.insn_len
);
1643 case KVM_PV_REASON_PAGE_NOT_PRESENT
:
1644 svm
->apf_reason
= 0;
1645 local_irq_disable();
1646 kvm_async_pf_task_wait(fault_address
);
1649 case KVM_PV_REASON_PAGE_READY
:
1650 svm
->apf_reason
= 0;
1651 local_irq_disable();
1652 kvm_async_pf_task_wake(fault_address
);
1659 static int db_interception(struct vcpu_svm
*svm
)
1661 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
1663 if (!(svm
->vcpu
.guest_debug
&
1664 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
)) &&
1665 !svm
->nmi_singlestep
) {
1666 kvm_queue_exception(&svm
->vcpu
, DB_VECTOR
);
1670 if (svm
->nmi_singlestep
) {
1671 svm
->nmi_singlestep
= false;
1672 if (!(svm
->vcpu
.guest_debug
& KVM_GUESTDBG_SINGLESTEP
))
1673 svm
->vmcb
->save
.rflags
&=
1674 ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
1677 if (svm
->vcpu
.guest_debug
&
1678 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
)) {
1679 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
1680 kvm_run
->debug
.arch
.pc
=
1681 svm
->vmcb
->save
.cs
.base
+ svm
->vmcb
->save
.rip
;
1682 kvm_run
->debug
.arch
.exception
= DB_VECTOR
;
1689 static int bp_interception(struct vcpu_svm
*svm
)
1691 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
1693 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
1694 kvm_run
->debug
.arch
.pc
= svm
->vmcb
->save
.cs
.base
+ svm
->vmcb
->save
.rip
;
1695 kvm_run
->debug
.arch
.exception
= BP_VECTOR
;
1699 static int ud_interception(struct vcpu_svm
*svm
)
1703 er
= emulate_instruction(&svm
->vcpu
, EMULTYPE_TRAP_UD
);
1704 if (er
!= EMULATE_DONE
)
1705 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
1709 static int ac_interception(struct vcpu_svm
*svm
)
1711 kvm_queue_exception_e(&svm
->vcpu
, AC_VECTOR
, 0);
1715 static void svm_fpu_activate(struct kvm_vcpu
*vcpu
)
1717 struct vcpu_svm
*svm
= to_svm(vcpu
);
1719 clr_exception_intercept(svm
, NM_VECTOR
);
1721 svm
->vcpu
.fpu_active
= 1;
1722 update_cr0_intercept(svm
);
1725 static int nm_interception(struct vcpu_svm
*svm
)
1727 svm_fpu_activate(&svm
->vcpu
);
1731 static bool is_erratum_383(void)
1736 if (!erratum_383_found
)
1739 value
= native_read_msr_safe(MSR_IA32_MC0_STATUS
, &err
);
1743 /* Bit 62 may or may not be set for this mce */
1744 value
&= ~(1ULL << 62);
1746 if (value
!= 0xb600000000010015ULL
)
1749 /* Clear MCi_STATUS registers */
1750 for (i
= 0; i
< 6; ++i
)
1751 native_write_msr_safe(MSR_IA32_MCx_STATUS(i
), 0, 0);
1753 value
= native_read_msr_safe(MSR_IA32_MCG_STATUS
, &err
);
1757 value
&= ~(1ULL << 2);
1758 low
= lower_32_bits(value
);
1759 high
= upper_32_bits(value
);
1761 native_write_msr_safe(MSR_IA32_MCG_STATUS
, low
, high
);
1764 /* Flush tlb to evict multi-match entries */
1770 static void svm_handle_mce(struct vcpu_svm
*svm
)
1772 if (is_erratum_383()) {
1774 * Erratum 383 triggered. Guest state is corrupt so kill the
1777 pr_err("KVM: Guest triggered AMD Erratum 383\n");
1779 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, &svm
->vcpu
);
1785 * On an #MC intercept the MCE handler is not called automatically in
1786 * the host. So do it by hand here.
1790 /* not sure if we ever come back to this point */
1795 static int mc_interception(struct vcpu_svm
*svm
)
1800 static int shutdown_interception(struct vcpu_svm
*svm
)
1802 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
1805 * VMCB is undefined after a SHUTDOWN intercept
1806 * so reinitialize it.
1808 clear_page(svm
->vmcb
);
1811 kvm_run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
1815 static int io_interception(struct vcpu_svm
*svm
)
1817 struct kvm_vcpu
*vcpu
= &svm
->vcpu
;
1818 u32 io_info
= svm
->vmcb
->control
.exit_info_1
; /* address size bug? */
1819 int size
, in
, string
;
1822 ++svm
->vcpu
.stat
.io_exits
;
1823 string
= (io_info
& SVM_IOIO_STR_MASK
) != 0;
1824 in
= (io_info
& SVM_IOIO_TYPE_MASK
) != 0;
1826 return emulate_instruction(vcpu
, 0) == EMULATE_DONE
;
1828 port
= io_info
>> 16;
1829 size
= (io_info
& SVM_IOIO_SIZE_MASK
) >> SVM_IOIO_SIZE_SHIFT
;
1830 svm
->next_rip
= svm
->vmcb
->control
.exit_info_2
;
1831 skip_emulated_instruction(&svm
->vcpu
);
1833 return kvm_fast_pio_out(vcpu
, size
, port
);
1836 static int nmi_interception(struct vcpu_svm
*svm
)
1841 static int intr_interception(struct vcpu_svm
*svm
)
1843 ++svm
->vcpu
.stat
.irq_exits
;
1847 static int nop_on_interception(struct vcpu_svm
*svm
)
1852 static int halt_interception(struct vcpu_svm
*svm
)
1854 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 1;
1855 return kvm_emulate_halt(&svm
->vcpu
);
1858 static int vmmcall_interception(struct vcpu_svm
*svm
)
1860 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
1861 kvm_emulate_hypercall(&svm
->vcpu
);
1865 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu
*vcpu
)
1867 struct vcpu_svm
*svm
= to_svm(vcpu
);
1869 return svm
->nested
.nested_cr3
;
1872 static u64
nested_svm_get_tdp_pdptr(struct kvm_vcpu
*vcpu
, int index
)
1874 struct vcpu_svm
*svm
= to_svm(vcpu
);
1875 u64 cr3
= svm
->nested
.nested_cr3
;
1879 ret
= kvm_vcpu_read_guest_page(vcpu
, gpa_to_gfn(cr3
), &pdpte
,
1880 offset_in_page(cr3
) + index
* 8, 8);
1886 static void nested_svm_set_tdp_cr3(struct kvm_vcpu
*vcpu
,
1889 struct vcpu_svm
*svm
= to_svm(vcpu
);
1891 svm
->vmcb
->control
.nested_cr3
= root
;
1892 mark_dirty(svm
->vmcb
, VMCB_NPT
);
1893 svm_flush_tlb(vcpu
);
1896 static void nested_svm_inject_npf_exit(struct kvm_vcpu
*vcpu
,
1897 struct x86_exception
*fault
)
1899 struct vcpu_svm
*svm
= to_svm(vcpu
);
1901 if (svm
->vmcb
->control
.exit_code
!= SVM_EXIT_NPF
) {
1903 * TODO: track the cause of the nested page fault, and
1904 * correctly fill in the high bits of exit_info_1.
1906 svm
->vmcb
->control
.exit_code
= SVM_EXIT_NPF
;
1907 svm
->vmcb
->control
.exit_code_hi
= 0;
1908 svm
->vmcb
->control
.exit_info_1
= (1ULL << 32);
1909 svm
->vmcb
->control
.exit_info_2
= fault
->address
;
1912 svm
->vmcb
->control
.exit_info_1
&= ~0xffffffffULL
;
1913 svm
->vmcb
->control
.exit_info_1
|= fault
->error_code
;
1916 * The present bit is always zero for page structure faults on real
1919 if (svm
->vmcb
->control
.exit_info_1
& (2ULL << 32))
1920 svm
->vmcb
->control
.exit_info_1
&= ~1;
1922 nested_svm_vmexit(svm
);
1925 static void nested_svm_init_mmu_context(struct kvm_vcpu
*vcpu
)
1927 WARN_ON(mmu_is_nested(vcpu
));
1928 kvm_init_shadow_mmu(vcpu
);
1929 vcpu
->arch
.mmu
.set_cr3
= nested_svm_set_tdp_cr3
;
1930 vcpu
->arch
.mmu
.get_cr3
= nested_svm_get_tdp_cr3
;
1931 vcpu
->arch
.mmu
.get_pdptr
= nested_svm_get_tdp_pdptr
;
1932 vcpu
->arch
.mmu
.inject_page_fault
= nested_svm_inject_npf_exit
;
1933 vcpu
->arch
.mmu
.shadow_root_level
= get_npt_level();
1934 reset_shadow_zero_bits_mask(vcpu
, &vcpu
->arch
.mmu
);
1935 vcpu
->arch
.walk_mmu
= &vcpu
->arch
.nested_mmu
;
1938 static void nested_svm_uninit_mmu_context(struct kvm_vcpu
*vcpu
)
1940 vcpu
->arch
.walk_mmu
= &vcpu
->arch
.mmu
;
1943 static int nested_svm_check_permissions(struct vcpu_svm
*svm
)
1945 if (!(svm
->vcpu
.arch
.efer
& EFER_SVME
)
1946 || !is_paging(&svm
->vcpu
)) {
1947 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
1951 if (svm
->vmcb
->save
.cpl
) {
1952 kvm_inject_gp(&svm
->vcpu
, 0);
1959 static int nested_svm_check_exception(struct vcpu_svm
*svm
, unsigned nr
,
1960 bool has_error_code
, u32 error_code
)
1964 if (!is_guest_mode(&svm
->vcpu
))
1967 svm
->vmcb
->control
.exit_code
= SVM_EXIT_EXCP_BASE
+ nr
;
1968 svm
->vmcb
->control
.exit_code_hi
= 0;
1969 svm
->vmcb
->control
.exit_info_1
= error_code
;
1970 svm
->vmcb
->control
.exit_info_2
= svm
->vcpu
.arch
.cr2
;
1972 vmexit
= nested_svm_intercept(svm
);
1973 if (vmexit
== NESTED_EXIT_DONE
)
1974 svm
->nested
.exit_required
= true;
1979 /* This function returns true if it is save to enable the irq window */
1980 static inline bool nested_svm_intr(struct vcpu_svm
*svm
)
1982 if (!is_guest_mode(&svm
->vcpu
))
1985 if (!(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
))
1988 if (!(svm
->vcpu
.arch
.hflags
& HF_HIF_MASK
))
1992 * if vmexit was already requested (by intercepted exception
1993 * for instance) do not overwrite it with "external interrupt"
1996 if (svm
->nested
.exit_required
)
1999 svm
->vmcb
->control
.exit_code
= SVM_EXIT_INTR
;
2000 svm
->vmcb
->control
.exit_info_1
= 0;
2001 svm
->vmcb
->control
.exit_info_2
= 0;
2003 if (svm
->nested
.intercept
& 1ULL) {
2005 * The #vmexit can't be emulated here directly because this
2006 * code path runs with irqs and preemption disabled. A
2007 * #vmexit emulation might sleep. Only signal request for
2010 svm
->nested
.exit_required
= true;
2011 trace_kvm_nested_intr_vmexit(svm
->vmcb
->save
.rip
);
2018 /* This function returns true if it is save to enable the nmi window */
2019 static inline bool nested_svm_nmi(struct vcpu_svm
*svm
)
2021 if (!is_guest_mode(&svm
->vcpu
))
2024 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_NMI
)))
2027 svm
->vmcb
->control
.exit_code
= SVM_EXIT_NMI
;
2028 svm
->nested
.exit_required
= true;
2033 static void *nested_svm_map(struct vcpu_svm
*svm
, u64 gpa
, struct page
**_page
)
2039 page
= kvm_vcpu_gfn_to_page(&svm
->vcpu
, gpa
>> PAGE_SHIFT
);
2040 if (is_error_page(page
))
2048 kvm_inject_gp(&svm
->vcpu
, 0);
2053 static void nested_svm_unmap(struct page
*page
)
2056 kvm_release_page_dirty(page
);
2059 static int nested_svm_intercept_ioio(struct vcpu_svm
*svm
)
2061 unsigned port
, size
, iopm_len
;
2066 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_IOIO_PROT
)))
2067 return NESTED_EXIT_HOST
;
2069 port
= svm
->vmcb
->control
.exit_info_1
>> 16;
2070 size
= (svm
->vmcb
->control
.exit_info_1
& SVM_IOIO_SIZE_MASK
) >>
2071 SVM_IOIO_SIZE_SHIFT
;
2072 gpa
= svm
->nested
.vmcb_iopm
+ (port
/ 8);
2073 start_bit
= port
% 8;
2074 iopm_len
= (start_bit
+ size
> 8) ? 2 : 1;
2075 mask
= (0xf >> (4 - size
)) << start_bit
;
2078 if (kvm_vcpu_read_guest(&svm
->vcpu
, gpa
, &val
, iopm_len
))
2079 return NESTED_EXIT_DONE
;
2081 return (val
& mask
) ? NESTED_EXIT_DONE
: NESTED_EXIT_HOST
;
2084 static int nested_svm_exit_handled_msr(struct vcpu_svm
*svm
)
2086 u32 offset
, msr
, value
;
2089 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_MSR_PROT
)))
2090 return NESTED_EXIT_HOST
;
2092 msr
= svm
->vcpu
.arch
.regs
[VCPU_REGS_RCX
];
2093 offset
= svm_msrpm_offset(msr
);
2094 write
= svm
->vmcb
->control
.exit_info_1
& 1;
2095 mask
= 1 << ((2 * (msr
& 0xf)) + write
);
2097 if (offset
== MSR_INVALID
)
2098 return NESTED_EXIT_DONE
;
2100 /* Offset is in 32 bit units but need in 8 bit units */
2103 if (kvm_vcpu_read_guest(&svm
->vcpu
, svm
->nested
.vmcb_msrpm
+ offset
, &value
, 4))
2104 return NESTED_EXIT_DONE
;
2106 return (value
& mask
) ? NESTED_EXIT_DONE
: NESTED_EXIT_HOST
;
2109 static int nested_svm_exit_special(struct vcpu_svm
*svm
)
2111 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
2113 switch (exit_code
) {
2116 case SVM_EXIT_EXCP_BASE
+ MC_VECTOR
:
2117 return NESTED_EXIT_HOST
;
2119 /* For now we are always handling NPFs when using them */
2121 return NESTED_EXIT_HOST
;
2123 case SVM_EXIT_EXCP_BASE
+ PF_VECTOR
:
2124 /* When we're shadowing, trap PFs, but not async PF */
2125 if (!npt_enabled
&& svm
->apf_reason
== 0)
2126 return NESTED_EXIT_HOST
;
2128 case SVM_EXIT_EXCP_BASE
+ NM_VECTOR
:
2129 nm_interception(svm
);
2135 return NESTED_EXIT_CONTINUE
;
2139 * If this function returns true, this #vmexit was already handled
2141 static int nested_svm_intercept(struct vcpu_svm
*svm
)
2143 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
2144 int vmexit
= NESTED_EXIT_HOST
;
2146 switch (exit_code
) {
2148 vmexit
= nested_svm_exit_handled_msr(svm
);
2151 vmexit
= nested_svm_intercept_ioio(svm
);
2153 case SVM_EXIT_READ_CR0
... SVM_EXIT_WRITE_CR8
: {
2154 u32 bit
= 1U << (exit_code
- SVM_EXIT_READ_CR0
);
2155 if (svm
->nested
.intercept_cr
& bit
)
2156 vmexit
= NESTED_EXIT_DONE
;
2159 case SVM_EXIT_READ_DR0
... SVM_EXIT_WRITE_DR7
: {
2160 u32 bit
= 1U << (exit_code
- SVM_EXIT_READ_DR0
);
2161 if (svm
->nested
.intercept_dr
& bit
)
2162 vmexit
= NESTED_EXIT_DONE
;
2165 case SVM_EXIT_EXCP_BASE
... SVM_EXIT_EXCP_BASE
+ 0x1f: {
2166 u32 excp_bits
= 1 << (exit_code
- SVM_EXIT_EXCP_BASE
);
2167 if (svm
->nested
.intercept_exceptions
& excp_bits
)
2168 vmexit
= NESTED_EXIT_DONE
;
2169 /* async page fault always cause vmexit */
2170 else if ((exit_code
== SVM_EXIT_EXCP_BASE
+ PF_VECTOR
) &&
2171 svm
->apf_reason
!= 0)
2172 vmexit
= NESTED_EXIT_DONE
;
2175 case SVM_EXIT_ERR
: {
2176 vmexit
= NESTED_EXIT_DONE
;
2180 u64 exit_bits
= 1ULL << (exit_code
- SVM_EXIT_INTR
);
2181 if (svm
->nested
.intercept
& exit_bits
)
2182 vmexit
= NESTED_EXIT_DONE
;
2189 static int nested_svm_exit_handled(struct vcpu_svm
*svm
)
2193 vmexit
= nested_svm_intercept(svm
);
2195 if (vmexit
== NESTED_EXIT_DONE
)
2196 nested_svm_vmexit(svm
);
2201 static inline void copy_vmcb_control_area(struct vmcb
*dst_vmcb
, struct vmcb
*from_vmcb
)
2203 struct vmcb_control_area
*dst
= &dst_vmcb
->control
;
2204 struct vmcb_control_area
*from
= &from_vmcb
->control
;
2206 dst
->intercept_cr
= from
->intercept_cr
;
2207 dst
->intercept_dr
= from
->intercept_dr
;
2208 dst
->intercept_exceptions
= from
->intercept_exceptions
;
2209 dst
->intercept
= from
->intercept
;
2210 dst
->iopm_base_pa
= from
->iopm_base_pa
;
2211 dst
->msrpm_base_pa
= from
->msrpm_base_pa
;
2212 dst
->tsc_offset
= from
->tsc_offset
;
2213 dst
->asid
= from
->asid
;
2214 dst
->tlb_ctl
= from
->tlb_ctl
;
2215 dst
->int_ctl
= from
->int_ctl
;
2216 dst
->int_vector
= from
->int_vector
;
2217 dst
->int_state
= from
->int_state
;
2218 dst
->exit_code
= from
->exit_code
;
2219 dst
->exit_code_hi
= from
->exit_code_hi
;
2220 dst
->exit_info_1
= from
->exit_info_1
;
2221 dst
->exit_info_2
= from
->exit_info_2
;
2222 dst
->exit_int_info
= from
->exit_int_info
;
2223 dst
->exit_int_info_err
= from
->exit_int_info_err
;
2224 dst
->nested_ctl
= from
->nested_ctl
;
2225 dst
->event_inj
= from
->event_inj
;
2226 dst
->event_inj_err
= from
->event_inj_err
;
2227 dst
->nested_cr3
= from
->nested_cr3
;
2228 dst
->lbr_ctl
= from
->lbr_ctl
;
2231 static int nested_svm_vmexit(struct vcpu_svm
*svm
)
2233 struct vmcb
*nested_vmcb
;
2234 struct vmcb
*hsave
= svm
->nested
.hsave
;
2235 struct vmcb
*vmcb
= svm
->vmcb
;
2238 trace_kvm_nested_vmexit_inject(vmcb
->control
.exit_code
,
2239 vmcb
->control
.exit_info_1
,
2240 vmcb
->control
.exit_info_2
,
2241 vmcb
->control
.exit_int_info
,
2242 vmcb
->control
.exit_int_info_err
,
2245 nested_vmcb
= nested_svm_map(svm
, svm
->nested
.vmcb
, &page
);
2249 /* Exit Guest-Mode */
2250 leave_guest_mode(&svm
->vcpu
);
2251 svm
->nested
.vmcb
= 0;
2253 /* Give the current vmcb to the guest */
2256 nested_vmcb
->save
.es
= vmcb
->save
.es
;
2257 nested_vmcb
->save
.cs
= vmcb
->save
.cs
;
2258 nested_vmcb
->save
.ss
= vmcb
->save
.ss
;
2259 nested_vmcb
->save
.ds
= vmcb
->save
.ds
;
2260 nested_vmcb
->save
.gdtr
= vmcb
->save
.gdtr
;
2261 nested_vmcb
->save
.idtr
= vmcb
->save
.idtr
;
2262 nested_vmcb
->save
.efer
= svm
->vcpu
.arch
.efer
;
2263 nested_vmcb
->save
.cr0
= kvm_read_cr0(&svm
->vcpu
);
2264 nested_vmcb
->save
.cr3
= kvm_read_cr3(&svm
->vcpu
);
2265 nested_vmcb
->save
.cr2
= vmcb
->save
.cr2
;
2266 nested_vmcb
->save
.cr4
= svm
->vcpu
.arch
.cr4
;
2267 nested_vmcb
->save
.rflags
= kvm_get_rflags(&svm
->vcpu
);
2268 nested_vmcb
->save
.rip
= vmcb
->save
.rip
;
2269 nested_vmcb
->save
.rsp
= vmcb
->save
.rsp
;
2270 nested_vmcb
->save
.rax
= vmcb
->save
.rax
;
2271 nested_vmcb
->save
.dr7
= vmcb
->save
.dr7
;
2272 nested_vmcb
->save
.dr6
= vmcb
->save
.dr6
;
2273 nested_vmcb
->save
.cpl
= vmcb
->save
.cpl
;
2275 nested_vmcb
->control
.int_ctl
= vmcb
->control
.int_ctl
;
2276 nested_vmcb
->control
.int_vector
= vmcb
->control
.int_vector
;
2277 nested_vmcb
->control
.int_state
= vmcb
->control
.int_state
;
2278 nested_vmcb
->control
.exit_code
= vmcb
->control
.exit_code
;
2279 nested_vmcb
->control
.exit_code_hi
= vmcb
->control
.exit_code_hi
;
2280 nested_vmcb
->control
.exit_info_1
= vmcb
->control
.exit_info_1
;
2281 nested_vmcb
->control
.exit_info_2
= vmcb
->control
.exit_info_2
;
2282 nested_vmcb
->control
.exit_int_info
= vmcb
->control
.exit_int_info
;
2283 nested_vmcb
->control
.exit_int_info_err
= vmcb
->control
.exit_int_info_err
;
2285 if (svm
->nrips_enabled
)
2286 nested_vmcb
->control
.next_rip
= vmcb
->control
.next_rip
;
2289 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2290 * to make sure that we do not lose injected events. So check event_inj
2291 * here and copy it to exit_int_info if it is valid.
2292 * Exit_int_info and event_inj can't be both valid because the case
2293 * below only happens on a VMRUN instruction intercept which has
2294 * no valid exit_int_info set.
2296 if (vmcb
->control
.event_inj
& SVM_EVTINJ_VALID
) {
2297 struct vmcb_control_area
*nc
= &nested_vmcb
->control
;
2299 nc
->exit_int_info
= vmcb
->control
.event_inj
;
2300 nc
->exit_int_info_err
= vmcb
->control
.event_inj_err
;
2303 nested_vmcb
->control
.tlb_ctl
= 0;
2304 nested_vmcb
->control
.event_inj
= 0;
2305 nested_vmcb
->control
.event_inj_err
= 0;
2307 /* We always set V_INTR_MASKING and remember the old value in hflags */
2308 if (!(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
))
2309 nested_vmcb
->control
.int_ctl
&= ~V_INTR_MASKING_MASK
;
2311 /* Restore the original control entries */
2312 copy_vmcb_control_area(vmcb
, hsave
);
2314 kvm_clear_exception_queue(&svm
->vcpu
);
2315 kvm_clear_interrupt_queue(&svm
->vcpu
);
2317 svm
->nested
.nested_cr3
= 0;
2319 /* Restore selected save entries */
2320 svm
->vmcb
->save
.es
= hsave
->save
.es
;
2321 svm
->vmcb
->save
.cs
= hsave
->save
.cs
;
2322 svm
->vmcb
->save
.ss
= hsave
->save
.ss
;
2323 svm
->vmcb
->save
.ds
= hsave
->save
.ds
;
2324 svm
->vmcb
->save
.gdtr
= hsave
->save
.gdtr
;
2325 svm
->vmcb
->save
.idtr
= hsave
->save
.idtr
;
2326 kvm_set_rflags(&svm
->vcpu
, hsave
->save
.rflags
);
2327 svm_set_efer(&svm
->vcpu
, hsave
->save
.efer
);
2328 svm_set_cr0(&svm
->vcpu
, hsave
->save
.cr0
| X86_CR0_PE
);
2329 svm_set_cr4(&svm
->vcpu
, hsave
->save
.cr4
);
2331 svm
->vmcb
->save
.cr3
= hsave
->save
.cr3
;
2332 svm
->vcpu
.arch
.cr3
= hsave
->save
.cr3
;
2334 (void)kvm_set_cr3(&svm
->vcpu
, hsave
->save
.cr3
);
2336 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
, hsave
->save
.rax
);
2337 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RSP
, hsave
->save
.rsp
);
2338 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RIP
, hsave
->save
.rip
);
2339 svm
->vmcb
->save
.dr7
= 0;
2340 svm
->vmcb
->save
.cpl
= 0;
2341 svm
->vmcb
->control
.exit_int_info
= 0;
2343 mark_all_dirty(svm
->vmcb
);
2345 nested_svm_unmap(page
);
2347 nested_svm_uninit_mmu_context(&svm
->vcpu
);
2348 kvm_mmu_reset_context(&svm
->vcpu
);
2349 kvm_mmu_load(&svm
->vcpu
);
2354 static bool nested_svm_vmrun_msrpm(struct vcpu_svm
*svm
)
2357 * This function merges the msr permission bitmaps of kvm and the
2358 * nested vmcb. It is optimized in that it only merges the parts where
2359 * the kvm msr permission bitmap may contain zero bits
2363 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_MSR_PROT
)))
2366 for (i
= 0; i
< MSRPM_OFFSETS
; i
++) {
2370 if (msrpm_offsets
[i
] == 0xffffffff)
2373 p
= msrpm_offsets
[i
];
2374 offset
= svm
->nested
.vmcb_msrpm
+ (p
* 4);
2376 if (kvm_vcpu_read_guest(&svm
->vcpu
, offset
, &value
, 4))
2379 svm
->nested
.msrpm
[p
] = svm
->msrpm
[p
] | value
;
2382 svm
->vmcb
->control
.msrpm_base_pa
= __pa(svm
->nested
.msrpm
);
2387 static bool nested_vmcb_checks(struct vmcb
*vmcb
)
2389 if ((vmcb
->control
.intercept
& (1ULL << INTERCEPT_VMRUN
)) == 0)
2392 if (vmcb
->control
.asid
== 0)
2395 if (vmcb
->control
.nested_ctl
&& !npt_enabled
)
2401 static bool nested_svm_vmrun(struct vcpu_svm
*svm
)
2403 struct vmcb
*nested_vmcb
;
2404 struct vmcb
*hsave
= svm
->nested
.hsave
;
2405 struct vmcb
*vmcb
= svm
->vmcb
;
2409 vmcb_gpa
= svm
->vmcb
->save
.rax
;
2411 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, &page
);
2415 if (!nested_vmcb_checks(nested_vmcb
)) {
2416 nested_vmcb
->control
.exit_code
= SVM_EXIT_ERR
;
2417 nested_vmcb
->control
.exit_code_hi
= 0;
2418 nested_vmcb
->control
.exit_info_1
= 0;
2419 nested_vmcb
->control
.exit_info_2
= 0;
2421 nested_svm_unmap(page
);
2426 trace_kvm_nested_vmrun(svm
->vmcb
->save
.rip
, vmcb_gpa
,
2427 nested_vmcb
->save
.rip
,
2428 nested_vmcb
->control
.int_ctl
,
2429 nested_vmcb
->control
.event_inj
,
2430 nested_vmcb
->control
.nested_ctl
);
2432 trace_kvm_nested_intercepts(nested_vmcb
->control
.intercept_cr
& 0xffff,
2433 nested_vmcb
->control
.intercept_cr
>> 16,
2434 nested_vmcb
->control
.intercept_exceptions
,
2435 nested_vmcb
->control
.intercept
);
2437 /* Clear internal status */
2438 kvm_clear_exception_queue(&svm
->vcpu
);
2439 kvm_clear_interrupt_queue(&svm
->vcpu
);
2442 * Save the old vmcb, so we don't need to pick what we save, but can
2443 * restore everything when a VMEXIT occurs
2445 hsave
->save
.es
= vmcb
->save
.es
;
2446 hsave
->save
.cs
= vmcb
->save
.cs
;
2447 hsave
->save
.ss
= vmcb
->save
.ss
;
2448 hsave
->save
.ds
= vmcb
->save
.ds
;
2449 hsave
->save
.gdtr
= vmcb
->save
.gdtr
;
2450 hsave
->save
.idtr
= vmcb
->save
.idtr
;
2451 hsave
->save
.efer
= svm
->vcpu
.arch
.efer
;
2452 hsave
->save
.cr0
= kvm_read_cr0(&svm
->vcpu
);
2453 hsave
->save
.cr4
= svm
->vcpu
.arch
.cr4
;
2454 hsave
->save
.rflags
= kvm_get_rflags(&svm
->vcpu
);
2455 hsave
->save
.rip
= kvm_rip_read(&svm
->vcpu
);
2456 hsave
->save
.rsp
= vmcb
->save
.rsp
;
2457 hsave
->save
.rax
= vmcb
->save
.rax
;
2459 hsave
->save
.cr3
= vmcb
->save
.cr3
;
2461 hsave
->save
.cr3
= kvm_read_cr3(&svm
->vcpu
);
2463 copy_vmcb_control_area(hsave
, vmcb
);
2465 if (kvm_get_rflags(&svm
->vcpu
) & X86_EFLAGS_IF
)
2466 svm
->vcpu
.arch
.hflags
|= HF_HIF_MASK
;
2468 svm
->vcpu
.arch
.hflags
&= ~HF_HIF_MASK
;
2470 if (nested_vmcb
->control
.nested_ctl
) {
2471 kvm_mmu_unload(&svm
->vcpu
);
2472 svm
->nested
.nested_cr3
= nested_vmcb
->control
.nested_cr3
;
2473 nested_svm_init_mmu_context(&svm
->vcpu
);
2476 /* Load the nested guest state */
2477 svm
->vmcb
->save
.es
= nested_vmcb
->save
.es
;
2478 svm
->vmcb
->save
.cs
= nested_vmcb
->save
.cs
;
2479 svm
->vmcb
->save
.ss
= nested_vmcb
->save
.ss
;
2480 svm
->vmcb
->save
.ds
= nested_vmcb
->save
.ds
;
2481 svm
->vmcb
->save
.gdtr
= nested_vmcb
->save
.gdtr
;
2482 svm
->vmcb
->save
.idtr
= nested_vmcb
->save
.idtr
;
2483 kvm_set_rflags(&svm
->vcpu
, nested_vmcb
->save
.rflags
);
2484 svm_set_efer(&svm
->vcpu
, nested_vmcb
->save
.efer
);
2485 svm_set_cr0(&svm
->vcpu
, nested_vmcb
->save
.cr0
);
2486 svm_set_cr4(&svm
->vcpu
, nested_vmcb
->save
.cr4
);
2488 svm
->vmcb
->save
.cr3
= nested_vmcb
->save
.cr3
;
2489 svm
->vcpu
.arch
.cr3
= nested_vmcb
->save
.cr3
;
2491 (void)kvm_set_cr3(&svm
->vcpu
, nested_vmcb
->save
.cr3
);
2493 /* Guest paging mode is active - reset mmu */
2494 kvm_mmu_reset_context(&svm
->vcpu
);
2496 svm
->vmcb
->save
.cr2
= svm
->vcpu
.arch
.cr2
= nested_vmcb
->save
.cr2
;
2497 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
, nested_vmcb
->save
.rax
);
2498 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RSP
, nested_vmcb
->save
.rsp
);
2499 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RIP
, nested_vmcb
->save
.rip
);
2501 /* In case we don't even reach vcpu_run, the fields are not updated */
2502 svm
->vmcb
->save
.rax
= nested_vmcb
->save
.rax
;
2503 svm
->vmcb
->save
.rsp
= nested_vmcb
->save
.rsp
;
2504 svm
->vmcb
->save
.rip
= nested_vmcb
->save
.rip
;
2505 svm
->vmcb
->save
.dr7
= nested_vmcb
->save
.dr7
;
2506 svm
->vmcb
->save
.dr6
= nested_vmcb
->save
.dr6
;
2507 svm
->vmcb
->save
.cpl
= nested_vmcb
->save
.cpl
;
2509 svm
->nested
.vmcb_msrpm
= nested_vmcb
->control
.msrpm_base_pa
& ~0x0fffULL
;
2510 svm
->nested
.vmcb_iopm
= nested_vmcb
->control
.iopm_base_pa
& ~0x0fffULL
;
2512 /* cache intercepts */
2513 svm
->nested
.intercept_cr
= nested_vmcb
->control
.intercept_cr
;
2514 svm
->nested
.intercept_dr
= nested_vmcb
->control
.intercept_dr
;
2515 svm
->nested
.intercept_exceptions
= nested_vmcb
->control
.intercept_exceptions
;
2516 svm
->nested
.intercept
= nested_vmcb
->control
.intercept
;
2518 svm_flush_tlb(&svm
->vcpu
);
2519 svm
->vmcb
->control
.int_ctl
= nested_vmcb
->control
.int_ctl
| V_INTR_MASKING_MASK
;
2520 if (nested_vmcb
->control
.int_ctl
& V_INTR_MASKING_MASK
)
2521 svm
->vcpu
.arch
.hflags
|= HF_VINTR_MASK
;
2523 svm
->vcpu
.arch
.hflags
&= ~HF_VINTR_MASK
;
2525 if (svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
) {
2526 /* We only want the cr8 intercept bits of the guest */
2527 clr_cr_intercept(svm
, INTERCEPT_CR8_READ
);
2528 clr_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
2531 /* We don't want to see VMMCALLs from a nested guest */
2532 clr_intercept(svm
, INTERCEPT_VMMCALL
);
2534 svm
->vmcb
->control
.lbr_ctl
= nested_vmcb
->control
.lbr_ctl
;
2535 svm
->vmcb
->control
.int_vector
= nested_vmcb
->control
.int_vector
;
2536 svm
->vmcb
->control
.int_state
= nested_vmcb
->control
.int_state
;
2537 svm
->vmcb
->control
.tsc_offset
+= nested_vmcb
->control
.tsc_offset
;
2538 svm
->vmcb
->control
.event_inj
= nested_vmcb
->control
.event_inj
;
2539 svm
->vmcb
->control
.event_inj_err
= nested_vmcb
->control
.event_inj_err
;
2541 nested_svm_unmap(page
);
2543 /* Enter Guest-Mode */
2544 enter_guest_mode(&svm
->vcpu
);
2547 * Merge guest and host intercepts - must be called with vcpu in
2548 * guest-mode to take affect here
2550 recalc_intercepts(svm
);
2552 svm
->nested
.vmcb
= vmcb_gpa
;
2556 mark_all_dirty(svm
->vmcb
);
2561 static void nested_svm_vmloadsave(struct vmcb
*from_vmcb
, struct vmcb
*to_vmcb
)
2563 to_vmcb
->save
.fs
= from_vmcb
->save
.fs
;
2564 to_vmcb
->save
.gs
= from_vmcb
->save
.gs
;
2565 to_vmcb
->save
.tr
= from_vmcb
->save
.tr
;
2566 to_vmcb
->save
.ldtr
= from_vmcb
->save
.ldtr
;
2567 to_vmcb
->save
.kernel_gs_base
= from_vmcb
->save
.kernel_gs_base
;
2568 to_vmcb
->save
.star
= from_vmcb
->save
.star
;
2569 to_vmcb
->save
.lstar
= from_vmcb
->save
.lstar
;
2570 to_vmcb
->save
.cstar
= from_vmcb
->save
.cstar
;
2571 to_vmcb
->save
.sfmask
= from_vmcb
->save
.sfmask
;
2572 to_vmcb
->save
.sysenter_cs
= from_vmcb
->save
.sysenter_cs
;
2573 to_vmcb
->save
.sysenter_esp
= from_vmcb
->save
.sysenter_esp
;
2574 to_vmcb
->save
.sysenter_eip
= from_vmcb
->save
.sysenter_eip
;
2577 static int vmload_interception(struct vcpu_svm
*svm
)
2579 struct vmcb
*nested_vmcb
;
2582 if (nested_svm_check_permissions(svm
))
2585 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, &page
);
2589 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2590 skip_emulated_instruction(&svm
->vcpu
);
2592 nested_svm_vmloadsave(nested_vmcb
, svm
->vmcb
);
2593 nested_svm_unmap(page
);
2598 static int vmsave_interception(struct vcpu_svm
*svm
)
2600 struct vmcb
*nested_vmcb
;
2603 if (nested_svm_check_permissions(svm
))
2606 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, &page
);
2610 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2611 skip_emulated_instruction(&svm
->vcpu
);
2613 nested_svm_vmloadsave(svm
->vmcb
, nested_vmcb
);
2614 nested_svm_unmap(page
);
2619 static int vmrun_interception(struct vcpu_svm
*svm
)
2621 if (nested_svm_check_permissions(svm
))
2624 /* Save rip after vmrun instruction */
2625 kvm_rip_write(&svm
->vcpu
, kvm_rip_read(&svm
->vcpu
) + 3);
2627 if (!nested_svm_vmrun(svm
))
2630 if (!nested_svm_vmrun_msrpm(svm
))
2637 svm
->vmcb
->control
.exit_code
= SVM_EXIT_ERR
;
2638 svm
->vmcb
->control
.exit_code_hi
= 0;
2639 svm
->vmcb
->control
.exit_info_1
= 0;
2640 svm
->vmcb
->control
.exit_info_2
= 0;
2642 nested_svm_vmexit(svm
);
2647 static int stgi_interception(struct vcpu_svm
*svm
)
2649 if (nested_svm_check_permissions(svm
))
2652 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2653 skip_emulated_instruction(&svm
->vcpu
);
2654 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
2661 static int clgi_interception(struct vcpu_svm
*svm
)
2663 if (nested_svm_check_permissions(svm
))
2666 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2667 skip_emulated_instruction(&svm
->vcpu
);
2671 /* After a CLGI no interrupts should come */
2672 svm_clear_vintr(svm
);
2673 svm
->vmcb
->control
.int_ctl
&= ~V_IRQ_MASK
;
2675 mark_dirty(svm
->vmcb
, VMCB_INTR
);
2680 static int invlpga_interception(struct vcpu_svm
*svm
)
2682 struct kvm_vcpu
*vcpu
= &svm
->vcpu
;
2684 trace_kvm_invlpga(svm
->vmcb
->save
.rip
, kvm_register_read(&svm
->vcpu
, VCPU_REGS_RCX
),
2685 kvm_register_read(&svm
->vcpu
, VCPU_REGS_RAX
));
2687 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2688 kvm_mmu_invlpg(vcpu
, kvm_register_read(&svm
->vcpu
, VCPU_REGS_RAX
));
2690 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2691 skip_emulated_instruction(&svm
->vcpu
);
2695 static int skinit_interception(struct vcpu_svm
*svm
)
2697 trace_kvm_skinit(svm
->vmcb
->save
.rip
, kvm_register_read(&svm
->vcpu
, VCPU_REGS_RAX
));
2699 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
2703 static int wbinvd_interception(struct vcpu_svm
*svm
)
2705 kvm_emulate_wbinvd(&svm
->vcpu
);
2709 static int xsetbv_interception(struct vcpu_svm
*svm
)
2711 u64 new_bv
= kvm_read_edx_eax(&svm
->vcpu
);
2712 u32 index
= kvm_register_read(&svm
->vcpu
, VCPU_REGS_RCX
);
2714 if (kvm_set_xcr(&svm
->vcpu
, index
, new_bv
) == 0) {
2715 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2716 skip_emulated_instruction(&svm
->vcpu
);
2722 static int task_switch_interception(struct vcpu_svm
*svm
)
2726 int int_type
= svm
->vmcb
->control
.exit_int_info
&
2727 SVM_EXITINTINFO_TYPE_MASK
;
2728 int int_vec
= svm
->vmcb
->control
.exit_int_info
& SVM_EVTINJ_VEC_MASK
;
2730 svm
->vmcb
->control
.exit_int_info
& SVM_EXITINTINFO_TYPE_MASK
;
2732 svm
->vmcb
->control
.exit_int_info
& SVM_EXITINTINFO_VALID
;
2733 bool has_error_code
= false;
2736 tss_selector
= (u16
)svm
->vmcb
->control
.exit_info_1
;
2738 if (svm
->vmcb
->control
.exit_info_2
&
2739 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET
))
2740 reason
= TASK_SWITCH_IRET
;
2741 else if (svm
->vmcb
->control
.exit_info_2
&
2742 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP
))
2743 reason
= TASK_SWITCH_JMP
;
2745 reason
= TASK_SWITCH_GATE
;
2747 reason
= TASK_SWITCH_CALL
;
2749 if (reason
== TASK_SWITCH_GATE
) {
2751 case SVM_EXITINTINFO_TYPE_NMI
:
2752 svm
->vcpu
.arch
.nmi_injected
= false;
2754 case SVM_EXITINTINFO_TYPE_EXEPT
:
2755 if (svm
->vmcb
->control
.exit_info_2
&
2756 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE
)) {
2757 has_error_code
= true;
2759 (u32
)svm
->vmcb
->control
.exit_info_2
;
2761 kvm_clear_exception_queue(&svm
->vcpu
);
2763 case SVM_EXITINTINFO_TYPE_INTR
:
2764 kvm_clear_interrupt_queue(&svm
->vcpu
);
2771 if (reason
!= TASK_SWITCH_GATE
||
2772 int_type
== SVM_EXITINTINFO_TYPE_SOFT
||
2773 (int_type
== SVM_EXITINTINFO_TYPE_EXEPT
&&
2774 (int_vec
== OF_VECTOR
|| int_vec
== BP_VECTOR
)))
2775 skip_emulated_instruction(&svm
->vcpu
);
2777 if (int_type
!= SVM_EXITINTINFO_TYPE_SOFT
)
2780 if (kvm_task_switch(&svm
->vcpu
, tss_selector
, int_vec
, reason
,
2781 has_error_code
, error_code
) == EMULATE_FAIL
) {
2782 svm
->vcpu
.run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
2783 svm
->vcpu
.run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
2784 svm
->vcpu
.run
->internal
.ndata
= 0;
2790 static int cpuid_interception(struct vcpu_svm
*svm
)
2792 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
2793 kvm_emulate_cpuid(&svm
->vcpu
);
2797 static int iret_interception(struct vcpu_svm
*svm
)
2799 ++svm
->vcpu
.stat
.nmi_window_exits
;
2800 clr_intercept(svm
, INTERCEPT_IRET
);
2801 svm
->vcpu
.arch
.hflags
|= HF_IRET_MASK
;
2802 svm
->nmi_iret_rip
= kvm_rip_read(&svm
->vcpu
);
2803 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
2807 static int invlpg_interception(struct vcpu_svm
*svm
)
2809 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS
))
2810 return emulate_instruction(&svm
->vcpu
, 0) == EMULATE_DONE
;
2812 kvm_mmu_invlpg(&svm
->vcpu
, svm
->vmcb
->control
.exit_info_1
);
2813 skip_emulated_instruction(&svm
->vcpu
);
2817 static int emulate_on_interception(struct vcpu_svm
*svm
)
2819 return emulate_instruction(&svm
->vcpu
, 0) == EMULATE_DONE
;
2822 static int rdpmc_interception(struct vcpu_svm
*svm
)
2826 if (!static_cpu_has(X86_FEATURE_NRIPS
))
2827 return emulate_on_interception(svm
);
2829 err
= kvm_rdpmc(&svm
->vcpu
);
2830 kvm_complete_insn_gp(&svm
->vcpu
, err
);
2835 static bool check_selective_cr0_intercepted(struct vcpu_svm
*svm
,
2838 unsigned long cr0
= svm
->vcpu
.arch
.cr0
;
2842 intercept
= svm
->nested
.intercept
;
2844 if (!is_guest_mode(&svm
->vcpu
) ||
2845 (!(intercept
& (1ULL << INTERCEPT_SELECTIVE_CR0
))))
2848 cr0
&= ~SVM_CR0_SELECTIVE_MASK
;
2849 val
&= ~SVM_CR0_SELECTIVE_MASK
;
2852 svm
->vmcb
->control
.exit_code
= SVM_EXIT_CR0_SEL_WRITE
;
2853 ret
= (nested_svm_exit_handled(svm
) == NESTED_EXIT_DONE
);
2859 #define CR_VALID (1ULL << 63)
2861 static int cr_interception(struct vcpu_svm
*svm
)
2867 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS
))
2868 return emulate_on_interception(svm
);
2870 if (unlikely((svm
->vmcb
->control
.exit_info_1
& CR_VALID
) == 0))
2871 return emulate_on_interception(svm
);
2873 reg
= svm
->vmcb
->control
.exit_info_1
& SVM_EXITINFO_REG_MASK
;
2874 if (svm
->vmcb
->control
.exit_code
== SVM_EXIT_CR0_SEL_WRITE
)
2875 cr
= SVM_EXIT_WRITE_CR0
- SVM_EXIT_READ_CR0
;
2877 cr
= svm
->vmcb
->control
.exit_code
- SVM_EXIT_READ_CR0
;
2880 if (cr
>= 16) { /* mov to cr */
2882 val
= kvm_register_read(&svm
->vcpu
, reg
);
2885 if (!check_selective_cr0_intercepted(svm
, val
))
2886 err
= kvm_set_cr0(&svm
->vcpu
, val
);
2892 err
= kvm_set_cr3(&svm
->vcpu
, val
);
2895 err
= kvm_set_cr4(&svm
->vcpu
, val
);
2898 err
= kvm_set_cr8(&svm
->vcpu
, val
);
2901 WARN(1, "unhandled write to CR%d", cr
);
2902 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
2905 } else { /* mov from cr */
2908 val
= kvm_read_cr0(&svm
->vcpu
);
2911 val
= svm
->vcpu
.arch
.cr2
;
2914 val
= kvm_read_cr3(&svm
->vcpu
);
2917 val
= kvm_read_cr4(&svm
->vcpu
);
2920 val
= kvm_get_cr8(&svm
->vcpu
);
2923 WARN(1, "unhandled read from CR%d", cr
);
2924 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
2927 kvm_register_write(&svm
->vcpu
, reg
, val
);
2929 kvm_complete_insn_gp(&svm
->vcpu
, err
);
2934 static int dr_interception(struct vcpu_svm
*svm
)
2939 if (svm
->vcpu
.guest_debug
== 0) {
2941 * No more DR vmexits; force a reload of the debug registers
2942 * and reenter on this instruction. The next vmexit will
2943 * retrieve the full state of the debug registers.
2945 clr_dr_intercepts(svm
);
2946 svm
->vcpu
.arch
.switch_db_regs
|= KVM_DEBUGREG_WONT_EXIT
;
2950 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS
))
2951 return emulate_on_interception(svm
);
2953 reg
= svm
->vmcb
->control
.exit_info_1
& SVM_EXITINFO_REG_MASK
;
2954 dr
= svm
->vmcb
->control
.exit_code
- SVM_EXIT_READ_DR0
;
2956 if (dr
>= 16) { /* mov to DRn */
2957 if (!kvm_require_dr(&svm
->vcpu
, dr
- 16))
2959 val
= kvm_register_read(&svm
->vcpu
, reg
);
2960 kvm_set_dr(&svm
->vcpu
, dr
- 16, val
);
2962 if (!kvm_require_dr(&svm
->vcpu
, dr
))
2964 kvm_get_dr(&svm
->vcpu
, dr
, &val
);
2965 kvm_register_write(&svm
->vcpu
, reg
, val
);
2968 skip_emulated_instruction(&svm
->vcpu
);
2973 static int cr8_write_interception(struct vcpu_svm
*svm
)
2975 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
2978 u8 cr8_prev
= kvm_get_cr8(&svm
->vcpu
);
2979 /* instruction emulation calls kvm_set_cr8() */
2980 r
= cr_interception(svm
);
2981 if (lapic_in_kernel(&svm
->vcpu
))
2983 if (cr8_prev
<= kvm_get_cr8(&svm
->vcpu
))
2985 kvm_run
->exit_reason
= KVM_EXIT_SET_TPR
;
2989 static u64
svm_read_l1_tsc(struct kvm_vcpu
*vcpu
, u64 host_tsc
)
2991 struct vmcb
*vmcb
= get_host_vmcb(to_svm(vcpu
));
2992 return vmcb
->control
.tsc_offset
+ host_tsc
;
2995 static int svm_get_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
2997 struct vcpu_svm
*svm
= to_svm(vcpu
);
2999 switch (msr_info
->index
) {
3000 case MSR_IA32_TSC
: {
3001 msr_info
->data
= svm
->vmcb
->control
.tsc_offset
+
3002 kvm_scale_tsc(vcpu
, rdtsc());
3007 msr_info
->data
= svm
->vmcb
->save
.star
;
3009 #ifdef CONFIG_X86_64
3011 msr_info
->data
= svm
->vmcb
->save
.lstar
;
3014 msr_info
->data
= svm
->vmcb
->save
.cstar
;
3016 case MSR_KERNEL_GS_BASE
:
3017 msr_info
->data
= svm
->vmcb
->save
.kernel_gs_base
;
3019 case MSR_SYSCALL_MASK
:
3020 msr_info
->data
= svm
->vmcb
->save
.sfmask
;
3023 case MSR_IA32_SYSENTER_CS
:
3024 msr_info
->data
= svm
->vmcb
->save
.sysenter_cs
;
3026 case MSR_IA32_SYSENTER_EIP
:
3027 msr_info
->data
= svm
->sysenter_eip
;
3029 case MSR_IA32_SYSENTER_ESP
:
3030 msr_info
->data
= svm
->sysenter_esp
;
3033 if (!boot_cpu_has(X86_FEATURE_RDTSCP
))
3035 msr_info
->data
= svm
->tsc_aux
;
3038 * Nobody will change the following 5 values in the VMCB so we can
3039 * safely return them on rdmsr. They will always be 0 until LBRV is
3042 case MSR_IA32_DEBUGCTLMSR
:
3043 msr_info
->data
= svm
->vmcb
->save
.dbgctl
;
3045 case MSR_IA32_LASTBRANCHFROMIP
:
3046 msr_info
->data
= svm
->vmcb
->save
.br_from
;
3048 case MSR_IA32_LASTBRANCHTOIP
:
3049 msr_info
->data
= svm
->vmcb
->save
.br_to
;
3051 case MSR_IA32_LASTINTFROMIP
:
3052 msr_info
->data
= svm
->vmcb
->save
.last_excp_from
;
3054 case MSR_IA32_LASTINTTOIP
:
3055 msr_info
->data
= svm
->vmcb
->save
.last_excp_to
;
3057 case MSR_VM_HSAVE_PA
:
3058 msr_info
->data
= svm
->nested
.hsave_msr
;
3061 msr_info
->data
= svm
->nested
.vm_cr_msr
;
3063 case MSR_IA32_UCODE_REV
:
3064 msr_info
->data
= 0x01000065;
3067 return kvm_get_msr_common(vcpu
, msr_info
);
3072 static int rdmsr_interception(struct vcpu_svm
*svm
)
3074 u32 ecx
= kvm_register_read(&svm
->vcpu
, VCPU_REGS_RCX
);
3075 struct msr_data msr_info
;
3077 msr_info
.index
= ecx
;
3078 msr_info
.host_initiated
= false;
3079 if (svm_get_msr(&svm
->vcpu
, &msr_info
)) {
3080 trace_kvm_msr_read_ex(ecx
);
3081 kvm_inject_gp(&svm
->vcpu
, 0);
3083 trace_kvm_msr_read(ecx
, msr_info
.data
);
3085 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
,
3086 msr_info
.data
& 0xffffffff);
3087 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RDX
,
3088 msr_info
.data
>> 32);
3089 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
3090 skip_emulated_instruction(&svm
->vcpu
);
3095 static int svm_set_vm_cr(struct kvm_vcpu
*vcpu
, u64 data
)
3097 struct vcpu_svm
*svm
= to_svm(vcpu
);
3098 int svm_dis
, chg_mask
;
3100 if (data
& ~SVM_VM_CR_VALID_MASK
)
3103 chg_mask
= SVM_VM_CR_VALID_MASK
;
3105 if (svm
->nested
.vm_cr_msr
& SVM_VM_CR_SVM_DIS_MASK
)
3106 chg_mask
&= ~(SVM_VM_CR_SVM_LOCK_MASK
| SVM_VM_CR_SVM_DIS_MASK
);
3108 svm
->nested
.vm_cr_msr
&= ~chg_mask
;
3109 svm
->nested
.vm_cr_msr
|= (data
& chg_mask
);
3111 svm_dis
= svm
->nested
.vm_cr_msr
& SVM_VM_CR_SVM_DIS_MASK
;
3113 /* check for svm_disable while efer.svme is set */
3114 if (svm_dis
&& (vcpu
->arch
.efer
& EFER_SVME
))
3120 static int svm_set_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
3122 struct vcpu_svm
*svm
= to_svm(vcpu
);
3124 u32 ecx
= msr
->index
;
3125 u64 data
= msr
->data
;
3128 kvm_write_tsc(vcpu
, msr
);
3131 svm
->vmcb
->save
.star
= data
;
3133 #ifdef CONFIG_X86_64
3135 svm
->vmcb
->save
.lstar
= data
;
3138 svm
->vmcb
->save
.cstar
= data
;
3140 case MSR_KERNEL_GS_BASE
:
3141 svm
->vmcb
->save
.kernel_gs_base
= data
;
3143 case MSR_SYSCALL_MASK
:
3144 svm
->vmcb
->save
.sfmask
= data
;
3147 case MSR_IA32_SYSENTER_CS
:
3148 svm
->vmcb
->save
.sysenter_cs
= data
;
3150 case MSR_IA32_SYSENTER_EIP
:
3151 svm
->sysenter_eip
= data
;
3152 svm
->vmcb
->save
.sysenter_eip
= data
;
3154 case MSR_IA32_SYSENTER_ESP
:
3155 svm
->sysenter_esp
= data
;
3156 svm
->vmcb
->save
.sysenter_esp
= data
;
3159 if (!boot_cpu_has(X86_FEATURE_RDTSCP
))
3163 * This is rare, so we update the MSR here instead of using
3164 * direct_access_msrs. Doing that would require a rdmsr in
3167 svm
->tsc_aux
= data
;
3168 wrmsrl(MSR_TSC_AUX
, svm
->tsc_aux
);
3170 case MSR_IA32_DEBUGCTLMSR
:
3171 if (!boot_cpu_has(X86_FEATURE_LBRV
)) {
3172 vcpu_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3176 if (data
& DEBUGCTL_RESERVED_BITS
)
3179 svm
->vmcb
->save
.dbgctl
= data
;
3180 mark_dirty(svm
->vmcb
, VMCB_LBR
);
3181 if (data
& (1ULL<<0))
3182 svm_enable_lbrv(svm
);
3184 svm_disable_lbrv(svm
);
3186 case MSR_VM_HSAVE_PA
:
3187 svm
->nested
.hsave_msr
= data
;
3190 return svm_set_vm_cr(vcpu
, data
);
3192 vcpu_unimpl(vcpu
, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx
, data
);
3195 return kvm_set_msr_common(vcpu
, msr
);
3200 static int wrmsr_interception(struct vcpu_svm
*svm
)
3202 struct msr_data msr
;
3203 u32 ecx
= kvm_register_read(&svm
->vcpu
, VCPU_REGS_RCX
);
3204 u64 data
= kvm_read_edx_eax(&svm
->vcpu
);
3208 msr
.host_initiated
= false;
3210 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
3211 if (kvm_set_msr(&svm
->vcpu
, &msr
)) {
3212 trace_kvm_msr_write_ex(ecx
, data
);
3213 kvm_inject_gp(&svm
->vcpu
, 0);
3215 trace_kvm_msr_write(ecx
, data
);
3216 skip_emulated_instruction(&svm
->vcpu
);
3221 static int msr_interception(struct vcpu_svm
*svm
)
3223 if (svm
->vmcb
->control
.exit_info_1
)
3224 return wrmsr_interception(svm
);
3226 return rdmsr_interception(svm
);
3229 static int interrupt_window_interception(struct vcpu_svm
*svm
)
3231 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
3232 svm_clear_vintr(svm
);
3233 svm
->vmcb
->control
.int_ctl
&= ~V_IRQ_MASK
;
3234 mark_dirty(svm
->vmcb
, VMCB_INTR
);
3235 ++svm
->vcpu
.stat
.irq_window_exits
;
3239 static int pause_interception(struct vcpu_svm
*svm
)
3241 kvm_vcpu_on_spin(&(svm
->vcpu
));
3245 static int nop_interception(struct vcpu_svm
*svm
)
3247 skip_emulated_instruction(&(svm
->vcpu
));
3251 static int monitor_interception(struct vcpu_svm
*svm
)
3253 printk_once(KERN_WARNING
"kvm: MONITOR instruction emulated as NOP!\n");
3254 return nop_interception(svm
);
3257 static int mwait_interception(struct vcpu_svm
*svm
)
3259 printk_once(KERN_WARNING
"kvm: MWAIT instruction emulated as NOP!\n");
3260 return nop_interception(svm
);
3263 static int (*const svm_exit_handlers
[])(struct vcpu_svm
*svm
) = {
3264 [SVM_EXIT_READ_CR0
] = cr_interception
,
3265 [SVM_EXIT_READ_CR3
] = cr_interception
,
3266 [SVM_EXIT_READ_CR4
] = cr_interception
,
3267 [SVM_EXIT_READ_CR8
] = cr_interception
,
3268 [SVM_EXIT_CR0_SEL_WRITE
] = cr_interception
,
3269 [SVM_EXIT_WRITE_CR0
] = cr_interception
,
3270 [SVM_EXIT_WRITE_CR3
] = cr_interception
,
3271 [SVM_EXIT_WRITE_CR4
] = cr_interception
,
3272 [SVM_EXIT_WRITE_CR8
] = cr8_write_interception
,
3273 [SVM_EXIT_READ_DR0
] = dr_interception
,
3274 [SVM_EXIT_READ_DR1
] = dr_interception
,
3275 [SVM_EXIT_READ_DR2
] = dr_interception
,
3276 [SVM_EXIT_READ_DR3
] = dr_interception
,
3277 [SVM_EXIT_READ_DR4
] = dr_interception
,
3278 [SVM_EXIT_READ_DR5
] = dr_interception
,
3279 [SVM_EXIT_READ_DR6
] = dr_interception
,
3280 [SVM_EXIT_READ_DR7
] = dr_interception
,
3281 [SVM_EXIT_WRITE_DR0
] = dr_interception
,
3282 [SVM_EXIT_WRITE_DR1
] = dr_interception
,
3283 [SVM_EXIT_WRITE_DR2
] = dr_interception
,
3284 [SVM_EXIT_WRITE_DR3
] = dr_interception
,
3285 [SVM_EXIT_WRITE_DR4
] = dr_interception
,
3286 [SVM_EXIT_WRITE_DR5
] = dr_interception
,
3287 [SVM_EXIT_WRITE_DR6
] = dr_interception
,
3288 [SVM_EXIT_WRITE_DR7
] = dr_interception
,
3289 [SVM_EXIT_EXCP_BASE
+ DB_VECTOR
] = db_interception
,
3290 [SVM_EXIT_EXCP_BASE
+ BP_VECTOR
] = bp_interception
,
3291 [SVM_EXIT_EXCP_BASE
+ UD_VECTOR
] = ud_interception
,
3292 [SVM_EXIT_EXCP_BASE
+ PF_VECTOR
] = pf_interception
,
3293 [SVM_EXIT_EXCP_BASE
+ NM_VECTOR
] = nm_interception
,
3294 [SVM_EXIT_EXCP_BASE
+ MC_VECTOR
] = mc_interception
,
3295 [SVM_EXIT_EXCP_BASE
+ AC_VECTOR
] = ac_interception
,
3296 [SVM_EXIT_INTR
] = intr_interception
,
3297 [SVM_EXIT_NMI
] = nmi_interception
,
3298 [SVM_EXIT_SMI
] = nop_on_interception
,
3299 [SVM_EXIT_INIT
] = nop_on_interception
,
3300 [SVM_EXIT_VINTR
] = interrupt_window_interception
,
3301 [SVM_EXIT_RDPMC
] = rdpmc_interception
,
3302 [SVM_EXIT_CPUID
] = cpuid_interception
,
3303 [SVM_EXIT_IRET
] = iret_interception
,
3304 [SVM_EXIT_INVD
] = emulate_on_interception
,
3305 [SVM_EXIT_PAUSE
] = pause_interception
,
3306 [SVM_EXIT_HLT
] = halt_interception
,
3307 [SVM_EXIT_INVLPG
] = invlpg_interception
,
3308 [SVM_EXIT_INVLPGA
] = invlpga_interception
,
3309 [SVM_EXIT_IOIO
] = io_interception
,
3310 [SVM_EXIT_MSR
] = msr_interception
,
3311 [SVM_EXIT_TASK_SWITCH
] = task_switch_interception
,
3312 [SVM_EXIT_SHUTDOWN
] = shutdown_interception
,
3313 [SVM_EXIT_VMRUN
] = vmrun_interception
,
3314 [SVM_EXIT_VMMCALL
] = vmmcall_interception
,
3315 [SVM_EXIT_VMLOAD
] = vmload_interception
,
3316 [SVM_EXIT_VMSAVE
] = vmsave_interception
,
3317 [SVM_EXIT_STGI
] = stgi_interception
,
3318 [SVM_EXIT_CLGI
] = clgi_interception
,
3319 [SVM_EXIT_SKINIT
] = skinit_interception
,
3320 [SVM_EXIT_WBINVD
] = wbinvd_interception
,
3321 [SVM_EXIT_MONITOR
] = monitor_interception
,
3322 [SVM_EXIT_MWAIT
] = mwait_interception
,
3323 [SVM_EXIT_XSETBV
] = xsetbv_interception
,
3324 [SVM_EXIT_NPF
] = pf_interception
,
3325 [SVM_EXIT_RSM
] = emulate_on_interception
,
3328 static void dump_vmcb(struct kvm_vcpu
*vcpu
)
3330 struct vcpu_svm
*svm
= to_svm(vcpu
);
3331 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
3332 struct vmcb_save_area
*save
= &svm
->vmcb
->save
;
3334 pr_err("VMCB Control Area:\n");
3335 pr_err("%-20s%04x\n", "cr_read:", control
->intercept_cr
& 0xffff);
3336 pr_err("%-20s%04x\n", "cr_write:", control
->intercept_cr
>> 16);
3337 pr_err("%-20s%04x\n", "dr_read:", control
->intercept_dr
& 0xffff);
3338 pr_err("%-20s%04x\n", "dr_write:", control
->intercept_dr
>> 16);
3339 pr_err("%-20s%08x\n", "exceptions:", control
->intercept_exceptions
);
3340 pr_err("%-20s%016llx\n", "intercepts:", control
->intercept
);
3341 pr_err("%-20s%d\n", "pause filter count:", control
->pause_filter_count
);
3342 pr_err("%-20s%016llx\n", "iopm_base_pa:", control
->iopm_base_pa
);
3343 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control
->msrpm_base_pa
);
3344 pr_err("%-20s%016llx\n", "tsc_offset:", control
->tsc_offset
);
3345 pr_err("%-20s%d\n", "asid:", control
->asid
);
3346 pr_err("%-20s%d\n", "tlb_ctl:", control
->tlb_ctl
);
3347 pr_err("%-20s%08x\n", "int_ctl:", control
->int_ctl
);
3348 pr_err("%-20s%08x\n", "int_vector:", control
->int_vector
);
3349 pr_err("%-20s%08x\n", "int_state:", control
->int_state
);
3350 pr_err("%-20s%08x\n", "exit_code:", control
->exit_code
);
3351 pr_err("%-20s%016llx\n", "exit_info1:", control
->exit_info_1
);
3352 pr_err("%-20s%016llx\n", "exit_info2:", control
->exit_info_2
);
3353 pr_err("%-20s%08x\n", "exit_int_info:", control
->exit_int_info
);
3354 pr_err("%-20s%08x\n", "exit_int_info_err:", control
->exit_int_info_err
);
3355 pr_err("%-20s%lld\n", "nested_ctl:", control
->nested_ctl
);
3356 pr_err("%-20s%016llx\n", "nested_cr3:", control
->nested_cr3
);
3357 pr_err("%-20s%08x\n", "event_inj:", control
->event_inj
);
3358 pr_err("%-20s%08x\n", "event_inj_err:", control
->event_inj_err
);
3359 pr_err("%-20s%lld\n", "lbr_ctl:", control
->lbr_ctl
);
3360 pr_err("%-20s%016llx\n", "next_rip:", control
->next_rip
);
3361 pr_err("VMCB State Save Area:\n");
3362 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3364 save
->es
.selector
, save
->es
.attrib
,
3365 save
->es
.limit
, save
->es
.base
);
3366 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3368 save
->cs
.selector
, save
->cs
.attrib
,
3369 save
->cs
.limit
, save
->cs
.base
);
3370 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3372 save
->ss
.selector
, save
->ss
.attrib
,
3373 save
->ss
.limit
, save
->ss
.base
);
3374 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3376 save
->ds
.selector
, save
->ds
.attrib
,
3377 save
->ds
.limit
, save
->ds
.base
);
3378 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3380 save
->fs
.selector
, save
->fs
.attrib
,
3381 save
->fs
.limit
, save
->fs
.base
);
3382 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3384 save
->gs
.selector
, save
->gs
.attrib
,
3385 save
->gs
.limit
, save
->gs
.base
);
3386 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3388 save
->gdtr
.selector
, save
->gdtr
.attrib
,
3389 save
->gdtr
.limit
, save
->gdtr
.base
);
3390 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3392 save
->ldtr
.selector
, save
->ldtr
.attrib
,
3393 save
->ldtr
.limit
, save
->ldtr
.base
);
3394 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3396 save
->idtr
.selector
, save
->idtr
.attrib
,
3397 save
->idtr
.limit
, save
->idtr
.base
);
3398 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3400 save
->tr
.selector
, save
->tr
.attrib
,
3401 save
->tr
.limit
, save
->tr
.base
);
3402 pr_err("cpl: %d efer: %016llx\n",
3403 save
->cpl
, save
->efer
);
3404 pr_err("%-15s %016llx %-13s %016llx\n",
3405 "cr0:", save
->cr0
, "cr2:", save
->cr2
);
3406 pr_err("%-15s %016llx %-13s %016llx\n",
3407 "cr3:", save
->cr3
, "cr4:", save
->cr4
);
3408 pr_err("%-15s %016llx %-13s %016llx\n",
3409 "dr6:", save
->dr6
, "dr7:", save
->dr7
);
3410 pr_err("%-15s %016llx %-13s %016llx\n",
3411 "rip:", save
->rip
, "rflags:", save
->rflags
);
3412 pr_err("%-15s %016llx %-13s %016llx\n",
3413 "rsp:", save
->rsp
, "rax:", save
->rax
);
3414 pr_err("%-15s %016llx %-13s %016llx\n",
3415 "star:", save
->star
, "lstar:", save
->lstar
);
3416 pr_err("%-15s %016llx %-13s %016llx\n",
3417 "cstar:", save
->cstar
, "sfmask:", save
->sfmask
);
3418 pr_err("%-15s %016llx %-13s %016llx\n",
3419 "kernel_gs_base:", save
->kernel_gs_base
,
3420 "sysenter_cs:", save
->sysenter_cs
);
3421 pr_err("%-15s %016llx %-13s %016llx\n",
3422 "sysenter_esp:", save
->sysenter_esp
,
3423 "sysenter_eip:", save
->sysenter_eip
);
3424 pr_err("%-15s %016llx %-13s %016llx\n",
3425 "gpat:", save
->g_pat
, "dbgctl:", save
->dbgctl
);
3426 pr_err("%-15s %016llx %-13s %016llx\n",
3427 "br_from:", save
->br_from
, "br_to:", save
->br_to
);
3428 pr_err("%-15s %016llx %-13s %016llx\n",
3429 "excp_from:", save
->last_excp_from
,
3430 "excp_to:", save
->last_excp_to
);
3433 static void svm_get_exit_info(struct kvm_vcpu
*vcpu
, u64
*info1
, u64
*info2
)
3435 struct vmcb_control_area
*control
= &to_svm(vcpu
)->vmcb
->control
;
3437 *info1
= control
->exit_info_1
;
3438 *info2
= control
->exit_info_2
;
3441 static int handle_exit(struct kvm_vcpu
*vcpu
)
3443 struct vcpu_svm
*svm
= to_svm(vcpu
);
3444 struct kvm_run
*kvm_run
= vcpu
->run
;
3445 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
3447 if (!is_cr_intercept(svm
, INTERCEPT_CR0_WRITE
))
3448 vcpu
->arch
.cr0
= svm
->vmcb
->save
.cr0
;
3450 vcpu
->arch
.cr3
= svm
->vmcb
->save
.cr3
;
3452 if (unlikely(svm
->nested
.exit_required
)) {
3453 nested_svm_vmexit(svm
);
3454 svm
->nested
.exit_required
= false;
3459 if (is_guest_mode(vcpu
)) {
3462 trace_kvm_nested_vmexit(svm
->vmcb
->save
.rip
, exit_code
,
3463 svm
->vmcb
->control
.exit_info_1
,
3464 svm
->vmcb
->control
.exit_info_2
,
3465 svm
->vmcb
->control
.exit_int_info
,
3466 svm
->vmcb
->control
.exit_int_info_err
,
3469 vmexit
= nested_svm_exit_special(svm
);
3471 if (vmexit
== NESTED_EXIT_CONTINUE
)
3472 vmexit
= nested_svm_exit_handled(svm
);
3474 if (vmexit
== NESTED_EXIT_DONE
)
3478 svm_complete_interrupts(svm
);
3480 if (svm
->vmcb
->control
.exit_code
== SVM_EXIT_ERR
) {
3481 kvm_run
->exit_reason
= KVM_EXIT_FAIL_ENTRY
;
3482 kvm_run
->fail_entry
.hardware_entry_failure_reason
3483 = svm
->vmcb
->control
.exit_code
;
3484 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
3489 if (is_external_interrupt(svm
->vmcb
->control
.exit_int_info
) &&
3490 exit_code
!= SVM_EXIT_EXCP_BASE
+ PF_VECTOR
&&
3491 exit_code
!= SVM_EXIT_NPF
&& exit_code
!= SVM_EXIT_TASK_SWITCH
&&
3492 exit_code
!= SVM_EXIT_INTR
&& exit_code
!= SVM_EXIT_NMI
)
3493 printk(KERN_ERR
"%s: unexpected exit_int_info 0x%x "
3495 __func__
, svm
->vmcb
->control
.exit_int_info
,
3498 if (exit_code
>= ARRAY_SIZE(svm_exit_handlers
)
3499 || !svm_exit_handlers
[exit_code
]) {
3500 WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code
);
3501 kvm_queue_exception(vcpu
, UD_VECTOR
);
3505 return svm_exit_handlers
[exit_code
](svm
);
3508 static void reload_tss(struct kvm_vcpu
*vcpu
)
3510 int cpu
= raw_smp_processor_id();
3512 struct svm_cpu_data
*sd
= per_cpu(svm_data
, cpu
);
3513 sd
->tss_desc
->type
= 9; /* available 32/64-bit TSS */
3517 static void pre_svm_run(struct vcpu_svm
*svm
)
3519 int cpu
= raw_smp_processor_id();
3521 struct svm_cpu_data
*sd
= per_cpu(svm_data
, cpu
);
3523 /* FIXME: handle wraparound of asid_generation */
3524 if (svm
->asid_generation
!= sd
->asid_generation
)
3528 static void svm_inject_nmi(struct kvm_vcpu
*vcpu
)
3530 struct vcpu_svm
*svm
= to_svm(vcpu
);
3532 svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_NMI
;
3533 vcpu
->arch
.hflags
|= HF_NMI_MASK
;
3534 set_intercept(svm
, INTERCEPT_IRET
);
3535 ++vcpu
->stat
.nmi_injections
;
3538 static inline void svm_inject_irq(struct vcpu_svm
*svm
, int irq
)
3540 struct vmcb_control_area
*control
;
3542 control
= &svm
->vmcb
->control
;
3543 control
->int_vector
= irq
;
3544 control
->int_ctl
&= ~V_INTR_PRIO_MASK
;
3545 control
->int_ctl
|= V_IRQ_MASK
|
3546 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT
);
3547 mark_dirty(svm
->vmcb
, VMCB_INTR
);
3550 static void svm_set_irq(struct kvm_vcpu
*vcpu
)
3552 struct vcpu_svm
*svm
= to_svm(vcpu
);
3554 BUG_ON(!(gif_set(svm
)));
3556 trace_kvm_inj_virq(vcpu
->arch
.interrupt
.nr
);
3557 ++vcpu
->stat
.irq_injections
;
3559 svm
->vmcb
->control
.event_inj
= vcpu
->arch
.interrupt
.nr
|
3560 SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
;
3563 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
, int tpr
, int irr
)
3565 struct vcpu_svm
*svm
= to_svm(vcpu
);
3567 if (is_guest_mode(vcpu
) && (vcpu
->arch
.hflags
& HF_VINTR_MASK
))
3570 clr_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
3576 set_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
3579 static void svm_set_virtual_x2apic_mode(struct kvm_vcpu
*vcpu
, bool set
)
3584 static bool svm_get_enable_apicv(void)
3589 static void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu
*vcpu
)
3593 static void svm_load_eoi_exitmap(struct kvm_vcpu
*vcpu
, u64
*eoi_exit_bitmap
)
3598 static void svm_sync_pir_to_irr(struct kvm_vcpu
*vcpu
)
3603 static int svm_nmi_allowed(struct kvm_vcpu
*vcpu
)
3605 struct vcpu_svm
*svm
= to_svm(vcpu
);
3606 struct vmcb
*vmcb
= svm
->vmcb
;
3608 ret
= !(vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
) &&
3609 !(svm
->vcpu
.arch
.hflags
& HF_NMI_MASK
);
3610 ret
= ret
&& gif_set(svm
) && nested_svm_nmi(svm
);
3615 static bool svm_get_nmi_mask(struct kvm_vcpu
*vcpu
)
3617 struct vcpu_svm
*svm
= to_svm(vcpu
);
3619 return !!(svm
->vcpu
.arch
.hflags
& HF_NMI_MASK
);
3622 static void svm_set_nmi_mask(struct kvm_vcpu
*vcpu
, bool masked
)
3624 struct vcpu_svm
*svm
= to_svm(vcpu
);
3627 svm
->vcpu
.arch
.hflags
|= HF_NMI_MASK
;
3628 set_intercept(svm
, INTERCEPT_IRET
);
3630 svm
->vcpu
.arch
.hflags
&= ~HF_NMI_MASK
;
3631 clr_intercept(svm
, INTERCEPT_IRET
);
3635 static int svm_interrupt_allowed(struct kvm_vcpu
*vcpu
)
3637 struct vcpu_svm
*svm
= to_svm(vcpu
);
3638 struct vmcb
*vmcb
= svm
->vmcb
;
3641 if (!gif_set(svm
) ||
3642 (vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
))
3645 ret
= !!(kvm_get_rflags(vcpu
) & X86_EFLAGS_IF
);
3647 if (is_guest_mode(vcpu
))
3648 return ret
&& !(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
);
3653 static void enable_irq_window(struct kvm_vcpu
*vcpu
)
3655 struct vcpu_svm
*svm
= to_svm(vcpu
);
3658 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3659 * 1, because that's a separate STGI/VMRUN intercept. The next time we
3660 * get that intercept, this function will be called again though and
3661 * we'll get the vintr intercept.
3663 if (gif_set(svm
) && nested_svm_intr(svm
)) {
3665 svm_inject_irq(svm
, 0x0);
3669 static void enable_nmi_window(struct kvm_vcpu
*vcpu
)
3671 struct vcpu_svm
*svm
= to_svm(vcpu
);
3673 if ((svm
->vcpu
.arch
.hflags
& (HF_NMI_MASK
| HF_IRET_MASK
))
3675 return; /* IRET will cause a vm exit */
3678 * Something prevents NMI from been injected. Single step over possible
3679 * problem (IRET or exception injection or interrupt shadow)
3681 svm
->nmi_singlestep
= true;
3682 svm
->vmcb
->save
.rflags
|= (X86_EFLAGS_TF
| X86_EFLAGS_RF
);
3685 static int svm_set_tss_addr(struct kvm
*kvm
, unsigned int addr
)
3690 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
)
3692 struct vcpu_svm
*svm
= to_svm(vcpu
);
3694 if (static_cpu_has(X86_FEATURE_FLUSHBYASID
))
3695 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ASID
;
3697 svm
->asid_generation
--;
3700 static void svm_prepare_guest_switch(struct kvm_vcpu
*vcpu
)
3704 static inline void sync_cr8_to_lapic(struct kvm_vcpu
*vcpu
)
3706 struct vcpu_svm
*svm
= to_svm(vcpu
);
3708 if (is_guest_mode(vcpu
) && (vcpu
->arch
.hflags
& HF_VINTR_MASK
))
3711 if (!is_cr_intercept(svm
, INTERCEPT_CR8_WRITE
)) {
3712 int cr8
= svm
->vmcb
->control
.int_ctl
& V_TPR_MASK
;
3713 kvm_set_cr8(vcpu
, cr8
);
3717 static inline void sync_lapic_to_cr8(struct kvm_vcpu
*vcpu
)
3719 struct vcpu_svm
*svm
= to_svm(vcpu
);
3722 if (is_guest_mode(vcpu
) && (vcpu
->arch
.hflags
& HF_VINTR_MASK
))
3725 cr8
= kvm_get_cr8(vcpu
);
3726 svm
->vmcb
->control
.int_ctl
&= ~V_TPR_MASK
;
3727 svm
->vmcb
->control
.int_ctl
|= cr8
& V_TPR_MASK
;
3730 static void svm_complete_interrupts(struct vcpu_svm
*svm
)
3734 u32 exitintinfo
= svm
->vmcb
->control
.exit_int_info
;
3735 unsigned int3_injected
= svm
->int3_injected
;
3737 svm
->int3_injected
= 0;
3740 * If we've made progress since setting HF_IRET_MASK, we've
3741 * executed an IRET and can allow NMI injection.
3743 if ((svm
->vcpu
.arch
.hflags
& HF_IRET_MASK
)
3744 && kvm_rip_read(&svm
->vcpu
) != svm
->nmi_iret_rip
) {
3745 svm
->vcpu
.arch
.hflags
&= ~(HF_NMI_MASK
| HF_IRET_MASK
);
3746 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
3749 svm
->vcpu
.arch
.nmi_injected
= false;
3750 kvm_clear_exception_queue(&svm
->vcpu
);
3751 kvm_clear_interrupt_queue(&svm
->vcpu
);
3753 if (!(exitintinfo
& SVM_EXITINTINFO_VALID
))
3756 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
3758 vector
= exitintinfo
& SVM_EXITINTINFO_VEC_MASK
;
3759 type
= exitintinfo
& SVM_EXITINTINFO_TYPE_MASK
;
3762 case SVM_EXITINTINFO_TYPE_NMI
:
3763 svm
->vcpu
.arch
.nmi_injected
= true;
3765 case SVM_EXITINTINFO_TYPE_EXEPT
:
3767 * In case of software exceptions, do not reinject the vector,
3768 * but re-execute the instruction instead. Rewind RIP first
3769 * if we emulated INT3 before.
3771 if (kvm_exception_is_soft(vector
)) {
3772 if (vector
== BP_VECTOR
&& int3_injected
&&
3773 kvm_is_linear_rip(&svm
->vcpu
, svm
->int3_rip
))
3774 kvm_rip_write(&svm
->vcpu
,
3775 kvm_rip_read(&svm
->vcpu
) -
3779 if (exitintinfo
& SVM_EXITINTINFO_VALID_ERR
) {
3780 u32 err
= svm
->vmcb
->control
.exit_int_info_err
;
3781 kvm_requeue_exception_e(&svm
->vcpu
, vector
, err
);
3784 kvm_requeue_exception(&svm
->vcpu
, vector
);
3786 case SVM_EXITINTINFO_TYPE_INTR
:
3787 kvm_queue_interrupt(&svm
->vcpu
, vector
, false);
3794 static void svm_cancel_injection(struct kvm_vcpu
*vcpu
)
3796 struct vcpu_svm
*svm
= to_svm(vcpu
);
3797 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
3799 control
->exit_int_info
= control
->event_inj
;
3800 control
->exit_int_info_err
= control
->event_inj_err
;
3801 control
->event_inj
= 0;
3802 svm_complete_interrupts(svm
);
3805 static void svm_vcpu_run(struct kvm_vcpu
*vcpu
)
3807 struct vcpu_svm
*svm
= to_svm(vcpu
);
3809 svm
->vmcb
->save
.rax
= vcpu
->arch
.regs
[VCPU_REGS_RAX
];
3810 svm
->vmcb
->save
.rsp
= vcpu
->arch
.regs
[VCPU_REGS_RSP
];
3811 svm
->vmcb
->save
.rip
= vcpu
->arch
.regs
[VCPU_REGS_RIP
];
3814 * A vmexit emulation is required before the vcpu can be executed
3817 if (unlikely(svm
->nested
.exit_required
))
3822 sync_lapic_to_cr8(vcpu
);
3824 svm
->vmcb
->save
.cr2
= vcpu
->arch
.cr2
;
3831 "push %%" _ASM_BP
"; \n\t"
3832 "mov %c[rbx](%[svm]), %%" _ASM_BX
" \n\t"
3833 "mov %c[rcx](%[svm]), %%" _ASM_CX
" \n\t"
3834 "mov %c[rdx](%[svm]), %%" _ASM_DX
" \n\t"
3835 "mov %c[rsi](%[svm]), %%" _ASM_SI
" \n\t"
3836 "mov %c[rdi](%[svm]), %%" _ASM_DI
" \n\t"
3837 "mov %c[rbp](%[svm]), %%" _ASM_BP
" \n\t"
3838 #ifdef CONFIG_X86_64
3839 "mov %c[r8](%[svm]), %%r8 \n\t"
3840 "mov %c[r9](%[svm]), %%r9 \n\t"
3841 "mov %c[r10](%[svm]), %%r10 \n\t"
3842 "mov %c[r11](%[svm]), %%r11 \n\t"
3843 "mov %c[r12](%[svm]), %%r12 \n\t"
3844 "mov %c[r13](%[svm]), %%r13 \n\t"
3845 "mov %c[r14](%[svm]), %%r14 \n\t"
3846 "mov %c[r15](%[svm]), %%r15 \n\t"
3849 /* Enter guest mode */
3850 "push %%" _ASM_AX
" \n\t"
3851 "mov %c[vmcb](%[svm]), %%" _ASM_AX
" \n\t"
3852 __ex(SVM_VMLOAD
) "\n\t"
3853 __ex(SVM_VMRUN
) "\n\t"
3854 __ex(SVM_VMSAVE
) "\n\t"
3855 "pop %%" _ASM_AX
" \n\t"
3857 /* Save guest registers, load host registers */
3858 "mov %%" _ASM_BX
", %c[rbx](%[svm]) \n\t"
3859 "mov %%" _ASM_CX
", %c[rcx](%[svm]) \n\t"
3860 "mov %%" _ASM_DX
", %c[rdx](%[svm]) \n\t"
3861 "mov %%" _ASM_SI
", %c[rsi](%[svm]) \n\t"
3862 "mov %%" _ASM_DI
", %c[rdi](%[svm]) \n\t"
3863 "mov %%" _ASM_BP
", %c[rbp](%[svm]) \n\t"
3864 #ifdef CONFIG_X86_64
3865 "mov %%r8, %c[r8](%[svm]) \n\t"
3866 "mov %%r9, %c[r9](%[svm]) \n\t"
3867 "mov %%r10, %c[r10](%[svm]) \n\t"
3868 "mov %%r11, %c[r11](%[svm]) \n\t"
3869 "mov %%r12, %c[r12](%[svm]) \n\t"
3870 "mov %%r13, %c[r13](%[svm]) \n\t"
3871 "mov %%r14, %c[r14](%[svm]) \n\t"
3872 "mov %%r15, %c[r15](%[svm]) \n\t"
3877 [vmcb
]"i"(offsetof(struct vcpu_svm
, vmcb_pa
)),
3878 [rbx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RBX
])),
3879 [rcx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RCX
])),
3880 [rdx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RDX
])),
3881 [rsi
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RSI
])),
3882 [rdi
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RDI
])),
3883 [rbp
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RBP
]))
3884 #ifdef CONFIG_X86_64
3885 , [r8
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R8
])),
3886 [r9
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R9
])),
3887 [r10
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R10
])),
3888 [r11
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R11
])),
3889 [r12
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R12
])),
3890 [r13
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R13
])),
3891 [r14
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R14
])),
3892 [r15
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R15
]))
3895 #ifdef CONFIG_X86_64
3896 , "rbx", "rcx", "rdx", "rsi", "rdi"
3897 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
3899 , "ebx", "ecx", "edx", "esi", "edi"
3903 #ifdef CONFIG_X86_64
3904 wrmsrl(MSR_GS_BASE
, svm
->host
.gs_base
);
3906 loadsegment(fs
, svm
->host
.fs
);
3907 #ifndef CONFIG_X86_32_LAZY_GS
3908 loadsegment(gs
, svm
->host
.gs
);
3914 local_irq_disable();
3916 vcpu
->arch
.cr2
= svm
->vmcb
->save
.cr2
;
3917 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = svm
->vmcb
->save
.rax
;
3918 vcpu
->arch
.regs
[VCPU_REGS_RSP
] = svm
->vmcb
->save
.rsp
;
3919 vcpu
->arch
.regs
[VCPU_REGS_RIP
] = svm
->vmcb
->save
.rip
;
3921 trace_kvm_exit(svm
->vmcb
->control
.exit_code
, vcpu
, KVM_ISA_SVM
);
3923 if (unlikely(svm
->vmcb
->control
.exit_code
== SVM_EXIT_NMI
))
3924 kvm_before_handle_nmi(&svm
->vcpu
);
3928 /* Any pending NMI will happen here */
3930 if (unlikely(svm
->vmcb
->control
.exit_code
== SVM_EXIT_NMI
))
3931 kvm_after_handle_nmi(&svm
->vcpu
);
3933 sync_cr8_to_lapic(vcpu
);
3937 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_DO_NOTHING
;
3939 /* if exit due to PF check for async PF */
3940 if (svm
->vmcb
->control
.exit_code
== SVM_EXIT_EXCP_BASE
+ PF_VECTOR
)
3941 svm
->apf_reason
= kvm_read_and_reset_pf_reason();
3944 vcpu
->arch
.regs_avail
&= ~(1 << VCPU_EXREG_PDPTR
);
3945 vcpu
->arch
.regs_dirty
&= ~(1 << VCPU_EXREG_PDPTR
);
3949 * We need to handle MC intercepts here before the vcpu has a chance to
3950 * change the physical cpu
3952 if (unlikely(svm
->vmcb
->control
.exit_code
==
3953 SVM_EXIT_EXCP_BASE
+ MC_VECTOR
))
3954 svm_handle_mce(svm
);
3956 mark_all_clean(svm
->vmcb
);
3959 static void svm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
3961 struct vcpu_svm
*svm
= to_svm(vcpu
);
3963 svm
->vmcb
->save
.cr3
= root
;
3964 mark_dirty(svm
->vmcb
, VMCB_CR
);
3965 svm_flush_tlb(vcpu
);
3968 static void set_tdp_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
3970 struct vcpu_svm
*svm
= to_svm(vcpu
);
3972 svm
->vmcb
->control
.nested_cr3
= root
;
3973 mark_dirty(svm
->vmcb
, VMCB_NPT
);
3975 /* Also sync guest cr3 here in case we live migrate */
3976 svm
->vmcb
->save
.cr3
= kvm_read_cr3(vcpu
);
3977 mark_dirty(svm
->vmcb
, VMCB_CR
);
3979 svm_flush_tlb(vcpu
);
3982 static int is_disabled(void)
3986 rdmsrl(MSR_VM_CR
, vm_cr
);
3987 if (vm_cr
& (1 << SVM_VM_CR_SVM_DISABLE
))
3994 svm_patch_hypercall(struct kvm_vcpu
*vcpu
, unsigned char *hypercall
)
3997 * Patch in the VMMCALL instruction:
3999 hypercall
[0] = 0x0f;
4000 hypercall
[1] = 0x01;
4001 hypercall
[2] = 0xd9;
4004 static void svm_check_processor_compat(void *rtn
)
4009 static bool svm_cpu_has_accelerated_tpr(void)
4014 static bool svm_has_high_real_mode_segbase(void)
4019 static u64
svm_get_mt_mask(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool is_mmio
)
4024 static void svm_cpuid_update(struct kvm_vcpu
*vcpu
)
4026 struct vcpu_svm
*svm
= to_svm(vcpu
);
4028 /* Update nrips enabled cache */
4029 svm
->nrips_enabled
= !!guest_cpuid_has_nrips(&svm
->vcpu
);
4032 static void svm_set_supported_cpuid(u32 func
, struct kvm_cpuid_entry2
*entry
)
4037 entry
->ecx
|= (1 << 2); /* Set SVM bit */
4040 entry
->eax
= 1; /* SVM revision 1 */
4041 entry
->ebx
= 8; /* Lets support 8 ASIDs in case we add proper
4042 ASID emulation to nested SVM */
4043 entry
->ecx
= 0; /* Reserved */
4044 entry
->edx
= 0; /* Per default do not support any
4045 additional features */
4047 /* Support next_rip if host supports it */
4048 if (boot_cpu_has(X86_FEATURE_NRIPS
))
4049 entry
->edx
|= SVM_FEATURE_NRIP
;
4051 /* Support NPT for the guest if enabled */
4053 entry
->edx
|= SVM_FEATURE_NPT
;
4059 static int svm_get_lpage_level(void)
4061 return PT_PDPE_LEVEL
;
4064 static bool svm_rdtscp_supported(void)
4066 return boot_cpu_has(X86_FEATURE_RDTSCP
);
4069 static bool svm_invpcid_supported(void)
4074 static bool svm_mpx_supported(void)
4079 static bool svm_xsaves_supported(void)
4084 static bool svm_has_wbinvd_exit(void)
4089 static void svm_fpu_deactivate(struct kvm_vcpu
*vcpu
)
4091 struct vcpu_svm
*svm
= to_svm(vcpu
);
4093 set_exception_intercept(svm
, NM_VECTOR
);
4094 update_cr0_intercept(svm
);
4097 #define PRE_EX(exit) { .exit_code = (exit), \
4098 .stage = X86_ICPT_PRE_EXCEPT, }
4099 #define POST_EX(exit) { .exit_code = (exit), \
4100 .stage = X86_ICPT_POST_EXCEPT, }
4101 #define POST_MEM(exit) { .exit_code = (exit), \
4102 .stage = X86_ICPT_POST_MEMACCESS, }
4104 static const struct __x86_intercept
{
4106 enum x86_intercept_stage stage
;
4107 } x86_intercept_map
[] = {
4108 [x86_intercept_cr_read
] = POST_EX(SVM_EXIT_READ_CR0
),
4109 [x86_intercept_cr_write
] = POST_EX(SVM_EXIT_WRITE_CR0
),
4110 [x86_intercept_clts
] = POST_EX(SVM_EXIT_WRITE_CR0
),
4111 [x86_intercept_lmsw
] = POST_EX(SVM_EXIT_WRITE_CR0
),
4112 [x86_intercept_smsw
] = POST_EX(SVM_EXIT_READ_CR0
),
4113 [x86_intercept_dr_read
] = POST_EX(SVM_EXIT_READ_DR0
),
4114 [x86_intercept_dr_write
] = POST_EX(SVM_EXIT_WRITE_DR0
),
4115 [x86_intercept_sldt
] = POST_EX(SVM_EXIT_LDTR_READ
),
4116 [x86_intercept_str
] = POST_EX(SVM_EXIT_TR_READ
),
4117 [x86_intercept_lldt
] = POST_EX(SVM_EXIT_LDTR_WRITE
),
4118 [x86_intercept_ltr
] = POST_EX(SVM_EXIT_TR_WRITE
),
4119 [x86_intercept_sgdt
] = POST_EX(SVM_EXIT_GDTR_READ
),
4120 [x86_intercept_sidt
] = POST_EX(SVM_EXIT_IDTR_READ
),
4121 [x86_intercept_lgdt
] = POST_EX(SVM_EXIT_GDTR_WRITE
),
4122 [x86_intercept_lidt
] = POST_EX(SVM_EXIT_IDTR_WRITE
),
4123 [x86_intercept_vmrun
] = POST_EX(SVM_EXIT_VMRUN
),
4124 [x86_intercept_vmmcall
] = POST_EX(SVM_EXIT_VMMCALL
),
4125 [x86_intercept_vmload
] = POST_EX(SVM_EXIT_VMLOAD
),
4126 [x86_intercept_vmsave
] = POST_EX(SVM_EXIT_VMSAVE
),
4127 [x86_intercept_stgi
] = POST_EX(SVM_EXIT_STGI
),
4128 [x86_intercept_clgi
] = POST_EX(SVM_EXIT_CLGI
),
4129 [x86_intercept_skinit
] = POST_EX(SVM_EXIT_SKINIT
),
4130 [x86_intercept_invlpga
] = POST_EX(SVM_EXIT_INVLPGA
),
4131 [x86_intercept_rdtscp
] = POST_EX(SVM_EXIT_RDTSCP
),
4132 [x86_intercept_monitor
] = POST_MEM(SVM_EXIT_MONITOR
),
4133 [x86_intercept_mwait
] = POST_EX(SVM_EXIT_MWAIT
),
4134 [x86_intercept_invlpg
] = POST_EX(SVM_EXIT_INVLPG
),
4135 [x86_intercept_invd
] = POST_EX(SVM_EXIT_INVD
),
4136 [x86_intercept_wbinvd
] = POST_EX(SVM_EXIT_WBINVD
),
4137 [x86_intercept_wrmsr
] = POST_EX(SVM_EXIT_MSR
),
4138 [x86_intercept_rdtsc
] = POST_EX(SVM_EXIT_RDTSC
),
4139 [x86_intercept_rdmsr
] = POST_EX(SVM_EXIT_MSR
),
4140 [x86_intercept_rdpmc
] = POST_EX(SVM_EXIT_RDPMC
),
4141 [x86_intercept_cpuid
] = PRE_EX(SVM_EXIT_CPUID
),
4142 [x86_intercept_rsm
] = PRE_EX(SVM_EXIT_RSM
),
4143 [x86_intercept_pause
] = PRE_EX(SVM_EXIT_PAUSE
),
4144 [x86_intercept_pushf
] = PRE_EX(SVM_EXIT_PUSHF
),
4145 [x86_intercept_popf
] = PRE_EX(SVM_EXIT_POPF
),
4146 [x86_intercept_intn
] = PRE_EX(SVM_EXIT_SWINT
),
4147 [x86_intercept_iret
] = PRE_EX(SVM_EXIT_IRET
),
4148 [x86_intercept_icebp
] = PRE_EX(SVM_EXIT_ICEBP
),
4149 [x86_intercept_hlt
] = POST_EX(SVM_EXIT_HLT
),
4150 [x86_intercept_in
] = POST_EX(SVM_EXIT_IOIO
),
4151 [x86_intercept_ins
] = POST_EX(SVM_EXIT_IOIO
),
4152 [x86_intercept_out
] = POST_EX(SVM_EXIT_IOIO
),
4153 [x86_intercept_outs
] = POST_EX(SVM_EXIT_IOIO
),
4160 static int svm_check_intercept(struct kvm_vcpu
*vcpu
,
4161 struct x86_instruction_info
*info
,
4162 enum x86_intercept_stage stage
)
4164 struct vcpu_svm
*svm
= to_svm(vcpu
);
4165 int vmexit
, ret
= X86EMUL_CONTINUE
;
4166 struct __x86_intercept icpt_info
;
4167 struct vmcb
*vmcb
= svm
->vmcb
;
4169 if (info
->intercept
>= ARRAY_SIZE(x86_intercept_map
))
4172 icpt_info
= x86_intercept_map
[info
->intercept
];
4174 if (stage
!= icpt_info
.stage
)
4177 switch (icpt_info
.exit_code
) {
4178 case SVM_EXIT_READ_CR0
:
4179 if (info
->intercept
== x86_intercept_cr_read
)
4180 icpt_info
.exit_code
+= info
->modrm_reg
;
4182 case SVM_EXIT_WRITE_CR0
: {
4183 unsigned long cr0
, val
;
4186 if (info
->intercept
== x86_intercept_cr_write
)
4187 icpt_info
.exit_code
+= info
->modrm_reg
;
4189 if (icpt_info
.exit_code
!= SVM_EXIT_WRITE_CR0
||
4190 info
->intercept
== x86_intercept_clts
)
4193 intercept
= svm
->nested
.intercept
;
4195 if (!(intercept
& (1ULL << INTERCEPT_SELECTIVE_CR0
)))
4198 cr0
= vcpu
->arch
.cr0
& ~SVM_CR0_SELECTIVE_MASK
;
4199 val
= info
->src_val
& ~SVM_CR0_SELECTIVE_MASK
;
4201 if (info
->intercept
== x86_intercept_lmsw
) {
4204 /* lmsw can't clear PE - catch this here */
4205 if (cr0
& X86_CR0_PE
)
4210 icpt_info
.exit_code
= SVM_EXIT_CR0_SEL_WRITE
;
4214 case SVM_EXIT_READ_DR0
:
4215 case SVM_EXIT_WRITE_DR0
:
4216 icpt_info
.exit_code
+= info
->modrm_reg
;
4219 if (info
->intercept
== x86_intercept_wrmsr
)
4220 vmcb
->control
.exit_info_1
= 1;
4222 vmcb
->control
.exit_info_1
= 0;
4224 case SVM_EXIT_PAUSE
:
4226 * We get this for NOP only, but pause
4227 * is rep not, check this here
4229 if (info
->rep_prefix
!= REPE_PREFIX
)
4231 case SVM_EXIT_IOIO
: {
4235 if (info
->intercept
== x86_intercept_in
||
4236 info
->intercept
== x86_intercept_ins
) {
4237 exit_info
= ((info
->src_val
& 0xffff) << 16) |
4239 bytes
= info
->dst_bytes
;
4241 exit_info
= (info
->dst_val
& 0xffff) << 16;
4242 bytes
= info
->src_bytes
;
4245 if (info
->intercept
== x86_intercept_outs
||
4246 info
->intercept
== x86_intercept_ins
)
4247 exit_info
|= SVM_IOIO_STR_MASK
;
4249 if (info
->rep_prefix
)
4250 exit_info
|= SVM_IOIO_REP_MASK
;
4252 bytes
= min(bytes
, 4u);
4254 exit_info
|= bytes
<< SVM_IOIO_SIZE_SHIFT
;
4256 exit_info
|= (u32
)info
->ad_bytes
<< (SVM_IOIO_ASIZE_SHIFT
- 1);
4258 vmcb
->control
.exit_info_1
= exit_info
;
4259 vmcb
->control
.exit_info_2
= info
->next_rip
;
4267 /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
4268 if (static_cpu_has(X86_FEATURE_NRIPS
))
4269 vmcb
->control
.next_rip
= info
->next_rip
;
4270 vmcb
->control
.exit_code
= icpt_info
.exit_code
;
4271 vmexit
= nested_svm_exit_handled(svm
);
4273 ret
= (vmexit
== NESTED_EXIT_DONE
) ? X86EMUL_INTERCEPTED
4280 static void svm_handle_external_intr(struct kvm_vcpu
*vcpu
)
4285 static void svm_sched_in(struct kvm_vcpu
*vcpu
, int cpu
)
4289 static struct kvm_x86_ops svm_x86_ops
= {
4290 .cpu_has_kvm_support
= has_svm
,
4291 .disabled_by_bios
= is_disabled
,
4292 .hardware_setup
= svm_hardware_setup
,
4293 .hardware_unsetup
= svm_hardware_unsetup
,
4294 .check_processor_compatibility
= svm_check_processor_compat
,
4295 .hardware_enable
= svm_hardware_enable
,
4296 .hardware_disable
= svm_hardware_disable
,
4297 .cpu_has_accelerated_tpr
= svm_cpu_has_accelerated_tpr
,
4298 .cpu_has_high_real_mode_segbase
= svm_has_high_real_mode_segbase
,
4300 .vcpu_create
= svm_create_vcpu
,
4301 .vcpu_free
= svm_free_vcpu
,
4302 .vcpu_reset
= svm_vcpu_reset
,
4304 .prepare_guest_switch
= svm_prepare_guest_switch
,
4305 .vcpu_load
= svm_vcpu_load
,
4306 .vcpu_put
= svm_vcpu_put
,
4308 .update_bp_intercept
= update_bp_intercept
,
4309 .get_msr
= svm_get_msr
,
4310 .set_msr
= svm_set_msr
,
4311 .get_segment_base
= svm_get_segment_base
,
4312 .get_segment
= svm_get_segment
,
4313 .set_segment
= svm_set_segment
,
4314 .get_cpl
= svm_get_cpl
,
4315 .get_cs_db_l_bits
= kvm_get_cs_db_l_bits
,
4316 .decache_cr0_guest_bits
= svm_decache_cr0_guest_bits
,
4317 .decache_cr3
= svm_decache_cr3
,
4318 .decache_cr4_guest_bits
= svm_decache_cr4_guest_bits
,
4319 .set_cr0
= svm_set_cr0
,
4320 .set_cr3
= svm_set_cr3
,
4321 .set_cr4
= svm_set_cr4
,
4322 .set_efer
= svm_set_efer
,
4323 .get_idt
= svm_get_idt
,
4324 .set_idt
= svm_set_idt
,
4325 .get_gdt
= svm_get_gdt
,
4326 .set_gdt
= svm_set_gdt
,
4327 .get_dr6
= svm_get_dr6
,
4328 .set_dr6
= svm_set_dr6
,
4329 .set_dr7
= svm_set_dr7
,
4330 .sync_dirty_debug_regs
= svm_sync_dirty_debug_regs
,
4331 .cache_reg
= svm_cache_reg
,
4332 .get_rflags
= svm_get_rflags
,
4333 .set_rflags
= svm_set_rflags
,
4334 .fpu_activate
= svm_fpu_activate
,
4335 .fpu_deactivate
= svm_fpu_deactivate
,
4337 .tlb_flush
= svm_flush_tlb
,
4339 .run
= svm_vcpu_run
,
4340 .handle_exit
= handle_exit
,
4341 .skip_emulated_instruction
= skip_emulated_instruction
,
4342 .set_interrupt_shadow
= svm_set_interrupt_shadow
,
4343 .get_interrupt_shadow
= svm_get_interrupt_shadow
,
4344 .patch_hypercall
= svm_patch_hypercall
,
4345 .set_irq
= svm_set_irq
,
4346 .set_nmi
= svm_inject_nmi
,
4347 .queue_exception
= svm_queue_exception
,
4348 .cancel_injection
= svm_cancel_injection
,
4349 .interrupt_allowed
= svm_interrupt_allowed
,
4350 .nmi_allowed
= svm_nmi_allowed
,
4351 .get_nmi_mask
= svm_get_nmi_mask
,
4352 .set_nmi_mask
= svm_set_nmi_mask
,
4353 .enable_nmi_window
= enable_nmi_window
,
4354 .enable_irq_window
= enable_irq_window
,
4355 .update_cr8_intercept
= update_cr8_intercept
,
4356 .set_virtual_x2apic_mode
= svm_set_virtual_x2apic_mode
,
4357 .get_enable_apicv
= svm_get_enable_apicv
,
4358 .refresh_apicv_exec_ctrl
= svm_refresh_apicv_exec_ctrl
,
4359 .load_eoi_exitmap
= svm_load_eoi_exitmap
,
4360 .sync_pir_to_irr
= svm_sync_pir_to_irr
,
4362 .set_tss_addr
= svm_set_tss_addr
,
4363 .get_tdp_level
= get_npt_level
,
4364 .get_mt_mask
= svm_get_mt_mask
,
4366 .get_exit_info
= svm_get_exit_info
,
4368 .get_lpage_level
= svm_get_lpage_level
,
4370 .cpuid_update
= svm_cpuid_update
,
4372 .rdtscp_supported
= svm_rdtscp_supported
,
4373 .invpcid_supported
= svm_invpcid_supported
,
4374 .mpx_supported
= svm_mpx_supported
,
4375 .xsaves_supported
= svm_xsaves_supported
,
4377 .set_supported_cpuid
= svm_set_supported_cpuid
,
4379 .has_wbinvd_exit
= svm_has_wbinvd_exit
,
4381 .read_tsc_offset
= svm_read_tsc_offset
,
4382 .write_tsc_offset
= svm_write_tsc_offset
,
4383 .adjust_tsc_offset_guest
= svm_adjust_tsc_offset_guest
,
4384 .read_l1_tsc
= svm_read_l1_tsc
,
4386 .set_tdp_cr3
= set_tdp_cr3
,
4388 .check_intercept
= svm_check_intercept
,
4389 .handle_external_intr
= svm_handle_external_intr
,
4391 .sched_in
= svm_sched_in
,
4393 .pmu_ops
= &amd_pmu_ops
,
4396 static int __init
svm_init(void)
4398 return kvm_init(&svm_x86_ops
, sizeof(struct vcpu_svm
),
4399 __alignof__(struct vcpu_svm
), THIS_MODULE
);
4402 static void __exit
svm_exit(void)
4407 module_init(svm_init
)
4408 module_exit(svm_exit
)