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.
18 #define pr_fmt(fmt) "SVM: " fmt
20 #include <linux/kvm_host.h>
24 #include "kvm_cache_regs.h"
29 #include <linux/module.h>
30 #include <linux/mod_devicetable.h>
31 #include <linux/kernel.h>
32 #include <linux/vmalloc.h>
33 #include <linux/highmem.h>
34 #include <linux/sched.h>
35 #include <linux/trace_events.h>
36 #include <linux/slab.h>
37 #include <linux/amd-iommu.h>
38 #include <linux/hashtable.h>
39 #include <linux/frame.h>
42 #include <asm/perf_event.h>
43 #include <asm/tlbflush.h>
45 #include <asm/debugreg.h>
46 #include <asm/kvm_para.h>
47 #include <asm/irq_remapping.h>
49 #include <asm/virtext.h>
52 #define __ex(x) __kvm_handle_fault_on_reboot(x)
54 MODULE_AUTHOR("Qumranet");
55 MODULE_LICENSE("GPL");
57 static const struct x86_cpu_id svm_cpu_id
[] = {
58 X86_FEATURE_MATCH(X86_FEATURE_SVM
),
61 MODULE_DEVICE_TABLE(x86cpu
, svm_cpu_id
);
63 #define IOPM_ALLOC_ORDER 2
64 #define MSRPM_ALLOC_ORDER 1
66 #define SEG_TYPE_LDT 2
67 #define SEG_TYPE_BUSY_TSS16 3
69 #define SVM_FEATURE_NPT (1 << 0)
70 #define SVM_FEATURE_LBRV (1 << 1)
71 #define SVM_FEATURE_SVML (1 << 2)
72 #define SVM_FEATURE_NRIP (1 << 3)
73 #define SVM_FEATURE_TSC_RATE (1 << 4)
74 #define SVM_FEATURE_VMCB_CLEAN (1 << 5)
75 #define SVM_FEATURE_FLUSH_ASID (1 << 6)
76 #define SVM_FEATURE_DECODE_ASSIST (1 << 7)
77 #define SVM_FEATURE_PAUSE_FILTER (1 << 10)
79 #define SVM_AVIC_DOORBELL 0xc001011b
81 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
82 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
83 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
85 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
87 #define TSC_RATIO_RSVD 0xffffff0000000000ULL
88 #define TSC_RATIO_MIN 0x0000000000000001ULL
89 #define TSC_RATIO_MAX 0x000000ffffffffffULL
91 #define AVIC_HPA_MASK ~((0xFFFULL << 52) | 0xFFF)
94 * 0xff is broadcast, so the max index allowed for physical APIC ID
95 * table is 0xfe. APIC IDs above 0xff are reserved.
97 #define AVIC_MAX_PHYSICAL_ID_COUNT 255
99 #define AVIC_UNACCEL_ACCESS_WRITE_MASK 1
100 #define AVIC_UNACCEL_ACCESS_OFFSET_MASK 0xFF0
101 #define AVIC_UNACCEL_ACCESS_VECTOR_MASK 0xFFFFFFFF
103 /* AVIC GATAG is encoded using VM and VCPU IDs */
104 #define AVIC_VCPU_ID_BITS 8
105 #define AVIC_VCPU_ID_MASK ((1 << AVIC_VCPU_ID_BITS) - 1)
107 #define AVIC_VM_ID_BITS 24
108 #define AVIC_VM_ID_NR (1 << AVIC_VM_ID_BITS)
109 #define AVIC_VM_ID_MASK ((1 << AVIC_VM_ID_BITS) - 1)
111 #define AVIC_GATAG(x, y) (((x & AVIC_VM_ID_MASK) << AVIC_VCPU_ID_BITS) | \
112 (y & AVIC_VCPU_ID_MASK))
113 #define AVIC_GATAG_TO_VMID(x) ((x >> AVIC_VCPU_ID_BITS) & AVIC_VM_ID_MASK)
114 #define AVIC_GATAG_TO_VCPUID(x) (x & AVIC_VCPU_ID_MASK)
116 static bool erratum_383_found __read_mostly
;
118 static const u32 host_save_user_msrs
[] = {
120 MSR_STAR
, MSR_LSTAR
, MSR_CSTAR
, MSR_SYSCALL_MASK
, MSR_KERNEL_GS_BASE
,
123 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
127 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
131 struct nested_state
{
137 /* These are the merged vectors */
140 /* gpa pointers to the real vectors */
144 /* A VMEXIT is required but not yet emulated */
147 /* cache for intercepts of the guest */
150 u32 intercept_exceptions
;
153 /* Nested Paging related state */
157 #define MSRPM_OFFSETS 16
158 static u32 msrpm_offsets
[MSRPM_OFFSETS
] __read_mostly
;
161 * Set osvw_len to higher value when updated Revision Guides
162 * are published and we know what the new status bits are
164 static uint64_t osvw_len
= 4, osvw_status
;
167 struct kvm_vcpu vcpu
;
169 unsigned long vmcb_pa
;
170 struct svm_cpu_data
*svm_data
;
171 uint64_t asid_generation
;
172 uint64_t sysenter_esp
;
173 uint64_t sysenter_eip
;
180 u64 host_user_msrs
[NR_HOST_SAVE_USER_MSRS
];
192 struct nested_state nested
;
195 u64 nmi_singlestep_guest_rflags
;
197 unsigned int3_injected
;
198 unsigned long int3_rip
;
200 /* cached guest cpuid flags for faster access */
201 bool nrips_enabled
: 1;
204 struct page
*avic_backing_page
;
205 u64
*avic_physical_id_cache
;
206 bool avic_is_running
;
209 * Per-vcpu list of struct amd_svm_iommu_ir:
210 * This is used mainly to store interrupt remapping information used
211 * when update the vcpu affinity. This avoids the need to scan for
212 * IRTE and try to match ga_tag in the IOMMU driver.
214 struct list_head ir_list
;
215 spinlock_t ir_list_lock
;
219 * This is a wrapper of struct amd_iommu_ir_data.
221 struct amd_svm_iommu_ir
{
222 struct list_head node
; /* Used by SVM for per-vcpu ir_list */
223 void *data
; /* Storing pointer to struct amd_ir_data */
226 #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK (0xFF)
227 #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK (1 << 31)
229 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK (0xFFULL)
230 #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12)
231 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62)
232 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63)
234 static DEFINE_PER_CPU(u64
, current_tsc_ratio
);
235 #define TSC_RATIO_DEFAULT 0x0100000000ULL
237 #define MSR_INVALID 0xffffffffU
239 static const struct svm_direct_access_msrs
{
240 u32 index
; /* Index of the MSR */
241 bool always
; /* True if intercept is always on */
242 } direct_access_msrs
[] = {
243 { .index
= MSR_STAR
, .always
= true },
244 { .index
= MSR_IA32_SYSENTER_CS
, .always
= true },
246 { .index
= MSR_GS_BASE
, .always
= true },
247 { .index
= MSR_FS_BASE
, .always
= true },
248 { .index
= MSR_KERNEL_GS_BASE
, .always
= true },
249 { .index
= MSR_LSTAR
, .always
= true },
250 { .index
= MSR_CSTAR
, .always
= true },
251 { .index
= MSR_SYSCALL_MASK
, .always
= true },
253 { .index
= MSR_IA32_SPEC_CTRL
, .always
= true },
254 { .index
= MSR_IA32_PRED_CMD
, .always
= true },
255 { .index
= MSR_IA32_LASTBRANCHFROMIP
, .always
= false },
256 { .index
= MSR_IA32_LASTBRANCHTOIP
, .always
= false },
257 { .index
= MSR_IA32_LASTINTFROMIP
, .always
= false },
258 { .index
= MSR_IA32_LASTINTTOIP
, .always
= false },
259 { .index
= MSR_INVALID
, .always
= false },
262 /* enable NPT for AMD64 and X86 with PAE */
263 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
264 static bool npt_enabled
= true;
266 static bool npt_enabled
;
269 /* allow nested paging (virtualized MMU) for all guests */
270 static int npt
= true;
271 module_param(npt
, int, S_IRUGO
);
273 /* allow nested virtualization in KVM/SVM */
274 static int nested
= true;
275 module_param(nested
, int, S_IRUGO
);
277 /* enable / disable AVIC */
279 #ifdef CONFIG_X86_LOCAL_APIC
280 module_param(avic
, int, S_IRUGO
);
283 /* enable/disable Virtual VMLOAD VMSAVE */
284 static int vls
= true;
285 module_param(vls
, int, 0444);
287 /* AVIC VM ID bit masks and lock */
288 static DECLARE_BITMAP(avic_vm_id_bitmap
, AVIC_VM_ID_NR
);
289 static DEFINE_SPINLOCK(avic_vm_id_lock
);
291 static void svm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
);
292 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
);
293 static void svm_complete_interrupts(struct vcpu_svm
*svm
);
295 static int nested_svm_exit_handled(struct vcpu_svm
*svm
);
296 static int nested_svm_intercept(struct vcpu_svm
*svm
);
297 static int nested_svm_vmexit(struct vcpu_svm
*svm
);
298 static int nested_svm_check_exception(struct vcpu_svm
*svm
, unsigned nr
,
299 bool has_error_code
, u32 error_code
);
302 VMCB_INTERCEPTS
, /* Intercept vectors, TSC offset,
303 pause filter count */
304 VMCB_PERM_MAP
, /* IOPM Base and MSRPM Base */
305 VMCB_ASID
, /* ASID */
306 VMCB_INTR
, /* int_ctl, int_vector */
307 VMCB_NPT
, /* npt_en, nCR3, gPAT */
308 VMCB_CR
, /* CR0, CR3, CR4, EFER */
309 VMCB_DR
, /* DR6, DR7 */
310 VMCB_DT
, /* GDT, IDT */
311 VMCB_SEG
, /* CS, DS, SS, ES, CPL */
312 VMCB_CR2
, /* CR2 only */
313 VMCB_LBR
, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
314 VMCB_AVIC
, /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE,
315 * AVIC PHYSICAL_TABLE pointer,
316 * AVIC LOGICAL_TABLE pointer
321 /* TPR and CR2 are always written before VMRUN */
322 #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2))
324 #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL
326 static inline void mark_all_dirty(struct vmcb
*vmcb
)
328 vmcb
->control
.clean
= 0;
331 static inline void mark_all_clean(struct vmcb
*vmcb
)
333 vmcb
->control
.clean
= ((1 << VMCB_DIRTY_MAX
) - 1)
334 & ~VMCB_ALWAYS_DIRTY_MASK
;
337 static inline void mark_dirty(struct vmcb
*vmcb
, int bit
)
339 vmcb
->control
.clean
&= ~(1 << bit
);
342 static inline struct vcpu_svm
*to_svm(struct kvm_vcpu
*vcpu
)
344 return container_of(vcpu
, struct vcpu_svm
, vcpu
);
347 static inline void avic_update_vapic_bar(struct vcpu_svm
*svm
, u64 data
)
349 svm
->vmcb
->control
.avic_vapic_bar
= data
& VMCB_AVIC_APIC_BAR_MASK
;
350 mark_dirty(svm
->vmcb
, VMCB_AVIC
);
353 static inline bool avic_vcpu_is_running(struct kvm_vcpu
*vcpu
)
355 struct vcpu_svm
*svm
= to_svm(vcpu
);
356 u64
*entry
= svm
->avic_physical_id_cache
;
361 return (READ_ONCE(*entry
) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK
);
364 static void recalc_intercepts(struct vcpu_svm
*svm
)
366 struct vmcb_control_area
*c
, *h
;
367 struct nested_state
*g
;
369 mark_dirty(svm
->vmcb
, VMCB_INTERCEPTS
);
371 if (!is_guest_mode(&svm
->vcpu
))
374 c
= &svm
->vmcb
->control
;
375 h
= &svm
->nested
.hsave
->control
;
378 c
->intercept_cr
= h
->intercept_cr
| g
->intercept_cr
;
379 c
->intercept_dr
= h
->intercept_dr
| g
->intercept_dr
;
380 c
->intercept_exceptions
= h
->intercept_exceptions
| g
->intercept_exceptions
;
381 c
->intercept
= h
->intercept
| g
->intercept
;
384 static inline struct vmcb
*get_host_vmcb(struct vcpu_svm
*svm
)
386 if (is_guest_mode(&svm
->vcpu
))
387 return svm
->nested
.hsave
;
392 static inline void set_cr_intercept(struct vcpu_svm
*svm
, int bit
)
394 struct vmcb
*vmcb
= get_host_vmcb(svm
);
396 vmcb
->control
.intercept_cr
|= (1U << bit
);
398 recalc_intercepts(svm
);
401 static inline void clr_cr_intercept(struct vcpu_svm
*svm
, int bit
)
403 struct vmcb
*vmcb
= get_host_vmcb(svm
);
405 vmcb
->control
.intercept_cr
&= ~(1U << bit
);
407 recalc_intercepts(svm
);
410 static inline bool is_cr_intercept(struct vcpu_svm
*svm
, int bit
)
412 struct vmcb
*vmcb
= get_host_vmcb(svm
);
414 return vmcb
->control
.intercept_cr
& (1U << bit
);
417 static inline void set_dr_intercepts(struct vcpu_svm
*svm
)
419 struct vmcb
*vmcb
= get_host_vmcb(svm
);
421 vmcb
->control
.intercept_dr
= (1 << INTERCEPT_DR0_READ
)
422 | (1 << INTERCEPT_DR1_READ
)
423 | (1 << INTERCEPT_DR2_READ
)
424 | (1 << INTERCEPT_DR3_READ
)
425 | (1 << INTERCEPT_DR4_READ
)
426 | (1 << INTERCEPT_DR5_READ
)
427 | (1 << INTERCEPT_DR6_READ
)
428 | (1 << INTERCEPT_DR7_READ
)
429 | (1 << INTERCEPT_DR0_WRITE
)
430 | (1 << INTERCEPT_DR1_WRITE
)
431 | (1 << INTERCEPT_DR2_WRITE
)
432 | (1 << INTERCEPT_DR3_WRITE
)
433 | (1 << INTERCEPT_DR4_WRITE
)
434 | (1 << INTERCEPT_DR5_WRITE
)
435 | (1 << INTERCEPT_DR6_WRITE
)
436 | (1 << INTERCEPT_DR7_WRITE
);
438 recalc_intercepts(svm
);
441 static inline void clr_dr_intercepts(struct vcpu_svm
*svm
)
443 struct vmcb
*vmcb
= get_host_vmcb(svm
);
445 vmcb
->control
.intercept_dr
= 0;
447 recalc_intercepts(svm
);
450 static inline void set_exception_intercept(struct vcpu_svm
*svm
, int bit
)
452 struct vmcb
*vmcb
= get_host_vmcb(svm
);
454 vmcb
->control
.intercept_exceptions
|= (1U << bit
);
456 recalc_intercepts(svm
);
459 static inline void clr_exception_intercept(struct vcpu_svm
*svm
, int bit
)
461 struct vmcb
*vmcb
= get_host_vmcb(svm
);
463 vmcb
->control
.intercept_exceptions
&= ~(1U << bit
);
465 recalc_intercepts(svm
);
468 static inline void set_intercept(struct vcpu_svm
*svm
, int bit
)
470 struct vmcb
*vmcb
= get_host_vmcb(svm
);
472 vmcb
->control
.intercept
|= (1ULL << bit
);
474 recalc_intercepts(svm
);
477 static inline void clr_intercept(struct vcpu_svm
*svm
, int bit
)
479 struct vmcb
*vmcb
= get_host_vmcb(svm
);
481 vmcb
->control
.intercept
&= ~(1ULL << bit
);
483 recalc_intercepts(svm
);
486 static inline void enable_gif(struct vcpu_svm
*svm
)
488 svm
->vcpu
.arch
.hflags
|= HF_GIF_MASK
;
491 static inline void disable_gif(struct vcpu_svm
*svm
)
493 svm
->vcpu
.arch
.hflags
&= ~HF_GIF_MASK
;
496 static inline bool gif_set(struct vcpu_svm
*svm
)
498 return !!(svm
->vcpu
.arch
.hflags
& HF_GIF_MASK
);
501 static unsigned long iopm_base
;
503 struct kvm_ldttss_desc
{
506 unsigned base1
:8, type
:5, dpl
:2, p
:1;
507 unsigned limit1
:4, zero0
:3, g
:1, base2
:8;
510 } __attribute__((packed
));
512 struct svm_cpu_data
{
518 struct kvm_ldttss_desc
*tss_desc
;
520 struct page
*save_area
;
523 static DEFINE_PER_CPU(struct svm_cpu_data
*, svm_data
);
525 struct svm_init_data
{
530 static const u32 msrpm_ranges
[] = {0, 0xc0000000, 0xc0010000};
532 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
533 #define MSRS_RANGE_SIZE 2048
534 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
536 static u32
svm_msrpm_offset(u32 msr
)
541 for (i
= 0; i
< NUM_MSR_MAPS
; i
++) {
542 if (msr
< msrpm_ranges
[i
] ||
543 msr
>= msrpm_ranges
[i
] + MSRS_IN_RANGE
)
546 offset
= (msr
- msrpm_ranges
[i
]) / 4; /* 4 msrs per u8 */
547 offset
+= (i
* MSRS_RANGE_SIZE
); /* add range offset */
549 /* Now we have the u8 offset - but need the u32 offset */
553 /* MSR not in any range */
557 #define MAX_INST_SIZE 15
559 static inline void clgi(void)
561 asm volatile (__ex(SVM_CLGI
));
564 static inline void stgi(void)
566 asm volatile (__ex(SVM_STGI
));
569 static inline void invlpga(unsigned long addr
, u32 asid
)
571 asm volatile (__ex(SVM_INVLPGA
) : : "a"(addr
), "c"(asid
));
574 static int get_npt_level(void)
577 return PT64_ROOT_LEVEL
;
579 return PT32E_ROOT_LEVEL
;
583 static void svm_set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
585 vcpu
->arch
.efer
= efer
;
586 if (!npt_enabled
&& !(efer
& EFER_LMA
))
589 to_svm(vcpu
)->vmcb
->save
.efer
= efer
| EFER_SVME
;
590 mark_dirty(to_svm(vcpu
)->vmcb
, VMCB_CR
);
593 static int is_external_interrupt(u32 info
)
595 info
&= SVM_EVTINJ_TYPE_MASK
| SVM_EVTINJ_VALID
;
596 return info
== (SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
);
599 static u32
svm_get_interrupt_shadow(struct kvm_vcpu
*vcpu
)
601 struct vcpu_svm
*svm
= to_svm(vcpu
);
604 if (svm
->vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
)
605 ret
= KVM_X86_SHADOW_INT_STI
| KVM_X86_SHADOW_INT_MOV_SS
;
609 static void svm_set_interrupt_shadow(struct kvm_vcpu
*vcpu
, int mask
)
611 struct vcpu_svm
*svm
= to_svm(vcpu
);
614 svm
->vmcb
->control
.int_state
&= ~SVM_INTERRUPT_SHADOW_MASK
;
616 svm
->vmcb
->control
.int_state
|= SVM_INTERRUPT_SHADOW_MASK
;
620 static void skip_emulated_instruction(struct kvm_vcpu
*vcpu
)
622 struct vcpu_svm
*svm
= to_svm(vcpu
);
624 if (svm
->vmcb
->control
.next_rip
!= 0) {
625 WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS
));
626 svm
->next_rip
= svm
->vmcb
->control
.next_rip
;
629 if (!svm
->next_rip
) {
630 if (emulate_instruction(vcpu
, EMULTYPE_SKIP
) !=
632 printk(KERN_DEBUG
"%s: NOP\n", __func__
);
635 if (svm
->next_rip
- kvm_rip_read(vcpu
) > MAX_INST_SIZE
)
636 printk(KERN_ERR
"%s: ip 0x%lx next 0x%llx\n",
637 __func__
, kvm_rip_read(vcpu
), svm
->next_rip
);
639 kvm_rip_write(vcpu
, svm
->next_rip
);
640 svm_set_interrupt_shadow(vcpu
, 0);
643 static void svm_queue_exception(struct kvm_vcpu
*vcpu
)
645 struct vcpu_svm
*svm
= to_svm(vcpu
);
646 unsigned nr
= vcpu
->arch
.exception
.nr
;
647 bool has_error_code
= vcpu
->arch
.exception
.has_error_code
;
648 bool reinject
= vcpu
->arch
.exception
.reinject
;
649 u32 error_code
= vcpu
->arch
.exception
.error_code
;
652 * If we are within a nested VM we'd better #VMEXIT and let the guest
653 * handle the exception
656 nested_svm_check_exception(svm
, nr
, has_error_code
, error_code
))
659 if (nr
== BP_VECTOR
&& !static_cpu_has(X86_FEATURE_NRIPS
)) {
660 unsigned long rip
, old_rip
= kvm_rip_read(&svm
->vcpu
);
663 * For guest debugging where we have to reinject #BP if some
664 * INT3 is guest-owned:
665 * Emulate nRIP by moving RIP forward. Will fail if injection
666 * raises a fault that is not intercepted. Still better than
667 * failing in all cases.
669 skip_emulated_instruction(&svm
->vcpu
);
670 rip
= kvm_rip_read(&svm
->vcpu
);
671 svm
->int3_rip
= rip
+ svm
->vmcb
->save
.cs
.base
;
672 svm
->int3_injected
= rip
- old_rip
;
675 svm
->vmcb
->control
.event_inj
= nr
677 | (has_error_code
? SVM_EVTINJ_VALID_ERR
: 0)
678 | SVM_EVTINJ_TYPE_EXEPT
;
679 svm
->vmcb
->control
.event_inj_err
= error_code
;
682 static void svm_init_erratum_383(void)
688 if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH
))
691 /* Use _safe variants to not break nested virtualization */
692 val
= native_read_msr_safe(MSR_AMD64_DC_CFG
, &err
);
698 low
= lower_32_bits(val
);
699 high
= upper_32_bits(val
);
701 native_write_msr_safe(MSR_AMD64_DC_CFG
, low
, high
);
703 erratum_383_found
= true;
706 static void svm_init_osvw(struct kvm_vcpu
*vcpu
)
709 * Guests should see errata 400 and 415 as fixed (assuming that
710 * HLT and IO instructions are intercepted).
712 vcpu
->arch
.osvw
.length
= (osvw_len
>= 3) ? (osvw_len
) : 3;
713 vcpu
->arch
.osvw
.status
= osvw_status
& ~(6ULL);
716 * By increasing VCPU's osvw.length to 3 we are telling the guest that
717 * all osvw.status bits inside that length, including bit 0 (which is
718 * reserved for erratum 298), are valid. However, if host processor's
719 * osvw_len is 0 then osvw_status[0] carries no information. We need to
720 * be conservative here and therefore we tell the guest that erratum 298
721 * is present (because we really don't know).
723 if (osvw_len
== 0 && boot_cpu_data
.x86
== 0x10)
724 vcpu
->arch
.osvw
.status
|= 1;
727 static int has_svm(void)
731 if (!cpu_has_svm(&msg
)) {
732 printk(KERN_INFO
"has_svm: %s\n", msg
);
739 static void svm_hardware_disable(void)
741 /* Make sure we clean up behind us */
742 if (static_cpu_has(X86_FEATURE_TSCRATEMSR
))
743 wrmsrl(MSR_AMD64_TSC_RATIO
, TSC_RATIO_DEFAULT
);
747 amd_pmu_disable_virt();
750 static int svm_hardware_enable(void)
753 struct svm_cpu_data
*sd
;
755 struct desc_struct
*gdt
;
756 int me
= raw_smp_processor_id();
758 rdmsrl(MSR_EFER
, efer
);
759 if (efer
& EFER_SVME
)
763 pr_err("%s: err EOPNOTSUPP on %d\n", __func__
, me
);
766 sd
= per_cpu(svm_data
, me
);
768 pr_err("%s: svm_data is NULL on %d\n", __func__
, me
);
772 sd
->asid_generation
= 1;
773 sd
->max_asid
= cpuid_ebx(SVM_CPUID_FUNC
) - 1;
774 sd
->next_asid
= sd
->max_asid
+ 1;
776 gdt
= get_current_gdt_rw();
777 sd
->tss_desc
= (struct kvm_ldttss_desc
*)(gdt
+ GDT_ENTRY_TSS
);
779 wrmsrl(MSR_EFER
, efer
| EFER_SVME
);
781 wrmsrl(MSR_VM_HSAVE_PA
, page_to_pfn(sd
->save_area
) << PAGE_SHIFT
);
783 if (static_cpu_has(X86_FEATURE_TSCRATEMSR
)) {
784 wrmsrl(MSR_AMD64_TSC_RATIO
, TSC_RATIO_DEFAULT
);
785 __this_cpu_write(current_tsc_ratio
, TSC_RATIO_DEFAULT
);
792 * Note that it is possible to have a system with mixed processor
793 * revisions and therefore different OSVW bits. If bits are not the same
794 * on different processors then choose the worst case (i.e. if erratum
795 * is present on one processor and not on another then assume that the
796 * erratum is present everywhere).
798 if (cpu_has(&boot_cpu_data
, X86_FEATURE_OSVW
)) {
799 uint64_t len
, status
= 0;
802 len
= native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH
, &err
);
804 status
= native_read_msr_safe(MSR_AMD64_OSVW_STATUS
,
808 osvw_status
= osvw_len
= 0;
812 osvw_status
|= status
;
813 osvw_status
&= (1ULL << osvw_len
) - 1;
816 osvw_status
= osvw_len
= 0;
818 svm_init_erratum_383();
820 amd_pmu_enable_virt();
825 static void svm_cpu_uninit(int cpu
)
827 struct svm_cpu_data
*sd
= per_cpu(svm_data
, raw_smp_processor_id());
832 per_cpu(svm_data
, raw_smp_processor_id()) = NULL
;
833 __free_page(sd
->save_area
);
837 static int svm_cpu_init(int cpu
)
839 struct svm_cpu_data
*sd
;
842 sd
= kzalloc(sizeof(struct svm_cpu_data
), GFP_KERNEL
);
846 sd
->save_area
= alloc_page(GFP_KERNEL
);
851 per_cpu(svm_data
, cpu
) = sd
;
861 static bool valid_msr_intercept(u32 index
)
865 for (i
= 0; direct_access_msrs
[i
].index
!= MSR_INVALID
; i
++)
866 if (direct_access_msrs
[i
].index
== index
)
872 static void set_msr_interception(u32
*msrpm
, unsigned msr
,
875 u8 bit_read
, bit_write
;
880 * If this warning triggers extend the direct_access_msrs list at the
881 * beginning of the file
883 WARN_ON(!valid_msr_intercept(msr
));
885 offset
= svm_msrpm_offset(msr
);
886 bit_read
= 2 * (msr
& 0x0f);
887 bit_write
= 2 * (msr
& 0x0f) + 1;
890 BUG_ON(offset
== MSR_INVALID
);
892 read
? clear_bit(bit_read
, &tmp
) : set_bit(bit_read
, &tmp
);
893 write
? clear_bit(bit_write
, &tmp
) : set_bit(bit_write
, &tmp
);
898 static void svm_vcpu_init_msrpm(u32
*msrpm
)
902 memset(msrpm
, 0xff, PAGE_SIZE
* (1 << MSRPM_ALLOC_ORDER
));
904 for (i
= 0; direct_access_msrs
[i
].index
!= MSR_INVALID
; i
++) {
905 if (!direct_access_msrs
[i
].always
)
908 set_msr_interception(msrpm
, direct_access_msrs
[i
].index
, 1, 1);
912 static void add_msr_offset(u32 offset
)
916 for (i
= 0; i
< MSRPM_OFFSETS
; ++i
) {
918 /* Offset already in list? */
919 if (msrpm_offsets
[i
] == offset
)
922 /* Slot used by another offset? */
923 if (msrpm_offsets
[i
] != MSR_INVALID
)
926 /* Add offset to list */
927 msrpm_offsets
[i
] = offset
;
933 * If this BUG triggers the msrpm_offsets table has an overflow. Just
934 * increase MSRPM_OFFSETS in this case.
939 static void init_msrpm_offsets(void)
943 memset(msrpm_offsets
, 0xff, sizeof(msrpm_offsets
));
945 for (i
= 0; direct_access_msrs
[i
].index
!= MSR_INVALID
; i
++) {
948 offset
= svm_msrpm_offset(direct_access_msrs
[i
].index
);
949 BUG_ON(offset
== MSR_INVALID
);
951 add_msr_offset(offset
);
955 static void svm_enable_lbrv(struct vcpu_svm
*svm
)
957 u32
*msrpm
= svm
->msrpm
;
959 svm
->vmcb
->control
.virt_ext
|= LBR_CTL_ENABLE_MASK
;
960 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHFROMIP
, 1, 1);
961 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHTOIP
, 1, 1);
962 set_msr_interception(msrpm
, MSR_IA32_LASTINTFROMIP
, 1, 1);
963 set_msr_interception(msrpm
, MSR_IA32_LASTINTTOIP
, 1, 1);
966 static void svm_disable_lbrv(struct vcpu_svm
*svm
)
968 u32
*msrpm
= svm
->msrpm
;
970 svm
->vmcb
->control
.virt_ext
&= ~LBR_CTL_ENABLE_MASK
;
971 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHFROMIP
, 0, 0);
972 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHTOIP
, 0, 0);
973 set_msr_interception(msrpm
, MSR_IA32_LASTINTFROMIP
, 0, 0);
974 set_msr_interception(msrpm
, MSR_IA32_LASTINTTOIP
, 0, 0);
977 static void disable_nmi_singlestep(struct vcpu_svm
*svm
)
979 svm
->nmi_singlestep
= false;
980 if (!(svm
->vcpu
.guest_debug
& KVM_GUESTDBG_SINGLESTEP
)) {
981 /* Clear our flags if they were not set by the guest */
982 if (!(svm
->nmi_singlestep_guest_rflags
& X86_EFLAGS_TF
))
983 svm
->vmcb
->save
.rflags
&= ~X86_EFLAGS_TF
;
984 if (!(svm
->nmi_singlestep_guest_rflags
& X86_EFLAGS_RF
))
985 svm
->vmcb
->save
.rflags
&= ~X86_EFLAGS_RF
;
990 * This hash table is used to map VM_ID to a struct kvm_arch,
991 * when handling AMD IOMMU GALOG notification to schedule in
994 #define SVM_VM_DATA_HASH_BITS 8
995 static DEFINE_HASHTABLE(svm_vm_data_hash
, SVM_VM_DATA_HASH_BITS
);
996 static DEFINE_SPINLOCK(svm_vm_data_hash_lock
);
999 * This function is called from IOMMU driver to notify
1000 * SVM to schedule in a particular vCPU of a particular VM.
1002 static int avic_ga_log_notifier(u32 ga_tag
)
1004 unsigned long flags
;
1005 struct kvm_arch
*ka
= NULL
;
1006 struct kvm_vcpu
*vcpu
= NULL
;
1007 u32 vm_id
= AVIC_GATAG_TO_VMID(ga_tag
);
1008 u32 vcpu_id
= AVIC_GATAG_TO_VCPUID(ga_tag
);
1010 pr_debug("SVM: %s: vm_id=%#x, vcpu_id=%#x\n", __func__
, vm_id
, vcpu_id
);
1012 spin_lock_irqsave(&svm_vm_data_hash_lock
, flags
);
1013 hash_for_each_possible(svm_vm_data_hash
, ka
, hnode
, vm_id
) {
1014 struct kvm
*kvm
= container_of(ka
, struct kvm
, arch
);
1015 struct kvm_arch
*vm_data
= &kvm
->arch
;
1017 if (vm_data
->avic_vm_id
!= vm_id
)
1019 vcpu
= kvm_get_vcpu_by_id(kvm
, vcpu_id
);
1022 spin_unlock_irqrestore(&svm_vm_data_hash_lock
, flags
);
1028 * At this point, the IOMMU should have already set the pending
1029 * bit in the vAPIC backing page. So, we just need to schedule
1032 if (vcpu
->mode
== OUTSIDE_GUEST_MODE
)
1033 kvm_vcpu_wake_up(vcpu
);
1038 static __init
int svm_hardware_setup(void)
1041 struct page
*iopm_pages
;
1045 iopm_pages
= alloc_pages(GFP_KERNEL
, IOPM_ALLOC_ORDER
);
1050 iopm_va
= page_address(iopm_pages
);
1051 memset(iopm_va
, 0xff, PAGE_SIZE
* (1 << IOPM_ALLOC_ORDER
));
1052 iopm_base
= page_to_pfn(iopm_pages
) << PAGE_SHIFT
;
1054 init_msrpm_offsets();
1056 if (boot_cpu_has(X86_FEATURE_NX
))
1057 kvm_enable_efer_bits(EFER_NX
);
1059 if (boot_cpu_has(X86_FEATURE_FXSR_OPT
))
1060 kvm_enable_efer_bits(EFER_FFXSR
);
1062 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR
)) {
1063 kvm_has_tsc_control
= true;
1064 kvm_max_tsc_scaling_ratio
= TSC_RATIO_MAX
;
1065 kvm_tsc_scaling_ratio_frac_bits
= 32;
1069 printk(KERN_INFO
"kvm: Nested Virtualization enabled\n");
1070 kvm_enable_efer_bits(EFER_SVME
| EFER_LMSLE
);
1073 for_each_possible_cpu(cpu
) {
1074 r
= svm_cpu_init(cpu
);
1079 if (!boot_cpu_has(X86_FEATURE_NPT
))
1080 npt_enabled
= false;
1082 if (npt_enabled
&& !npt
) {
1083 printk(KERN_INFO
"kvm: Nested Paging disabled\n");
1084 npt_enabled
= false;
1088 printk(KERN_INFO
"kvm: Nested Paging enabled\n");
1095 !boot_cpu_has(X86_FEATURE_AVIC
) ||
1096 !IS_ENABLED(CONFIG_X86_LOCAL_APIC
)) {
1099 pr_info("AVIC enabled\n");
1101 amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier
);
1107 !boot_cpu_has(X86_FEATURE_V_VMSAVE_VMLOAD
) ||
1108 !IS_ENABLED(CONFIG_X86_64
)) {
1111 pr_info("Virtual VMLOAD VMSAVE supported\n");
1118 __free_pages(iopm_pages
, IOPM_ALLOC_ORDER
);
1123 static __exit
void svm_hardware_unsetup(void)
1127 for_each_possible_cpu(cpu
)
1128 svm_cpu_uninit(cpu
);
1130 __free_pages(pfn_to_page(iopm_base
>> PAGE_SHIFT
), IOPM_ALLOC_ORDER
);
1134 static void init_seg(struct vmcb_seg
*seg
)
1137 seg
->attrib
= SVM_SELECTOR_P_MASK
| SVM_SELECTOR_S_MASK
|
1138 SVM_SELECTOR_WRITE_MASK
; /* Read/Write Data Segment */
1139 seg
->limit
= 0xffff;
1143 static void init_sys_seg(struct vmcb_seg
*seg
, uint32_t type
)
1146 seg
->attrib
= SVM_SELECTOR_P_MASK
| type
;
1147 seg
->limit
= 0xffff;
1151 static void svm_write_tsc_offset(struct kvm_vcpu
*vcpu
, u64 offset
)
1153 struct vcpu_svm
*svm
= to_svm(vcpu
);
1154 u64 g_tsc_offset
= 0;
1156 if (is_guest_mode(vcpu
)) {
1157 g_tsc_offset
= svm
->vmcb
->control
.tsc_offset
-
1158 svm
->nested
.hsave
->control
.tsc_offset
;
1159 svm
->nested
.hsave
->control
.tsc_offset
= offset
;
1161 trace_kvm_write_tsc_offset(vcpu
->vcpu_id
,
1162 svm
->vmcb
->control
.tsc_offset
,
1165 svm
->vmcb
->control
.tsc_offset
= offset
+ g_tsc_offset
;
1167 mark_dirty(svm
->vmcb
, VMCB_INTERCEPTS
);
1170 static void avic_init_vmcb(struct vcpu_svm
*svm
)
1172 struct vmcb
*vmcb
= svm
->vmcb
;
1173 struct kvm_arch
*vm_data
= &svm
->vcpu
.kvm
->arch
;
1174 phys_addr_t bpa
= page_to_phys(svm
->avic_backing_page
);
1175 phys_addr_t lpa
= page_to_phys(vm_data
->avic_logical_id_table_page
);
1176 phys_addr_t ppa
= page_to_phys(vm_data
->avic_physical_id_table_page
);
1178 vmcb
->control
.avic_backing_page
= bpa
& AVIC_HPA_MASK
;
1179 vmcb
->control
.avic_logical_id
= lpa
& AVIC_HPA_MASK
;
1180 vmcb
->control
.avic_physical_id
= ppa
& AVIC_HPA_MASK
;
1181 vmcb
->control
.avic_physical_id
|= AVIC_MAX_PHYSICAL_ID_COUNT
;
1182 vmcb
->control
.int_ctl
|= AVIC_ENABLE_MASK
;
1183 svm
->vcpu
.arch
.apicv_active
= true;
1186 static void init_vmcb(struct vcpu_svm
*svm
)
1188 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
1189 struct vmcb_save_area
*save
= &svm
->vmcb
->save
;
1191 svm
->vcpu
.arch
.hflags
= 0;
1193 set_cr_intercept(svm
, INTERCEPT_CR0_READ
);
1194 set_cr_intercept(svm
, INTERCEPT_CR3_READ
);
1195 set_cr_intercept(svm
, INTERCEPT_CR4_READ
);
1196 set_cr_intercept(svm
, INTERCEPT_CR0_WRITE
);
1197 set_cr_intercept(svm
, INTERCEPT_CR3_WRITE
);
1198 set_cr_intercept(svm
, INTERCEPT_CR4_WRITE
);
1199 if (!kvm_vcpu_apicv_active(&svm
->vcpu
))
1200 set_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
1202 set_dr_intercepts(svm
);
1204 set_exception_intercept(svm
, PF_VECTOR
);
1205 set_exception_intercept(svm
, UD_VECTOR
);
1206 set_exception_intercept(svm
, MC_VECTOR
);
1207 set_exception_intercept(svm
, AC_VECTOR
);
1208 set_exception_intercept(svm
, DB_VECTOR
);
1210 set_intercept(svm
, INTERCEPT_INTR
);
1211 set_intercept(svm
, INTERCEPT_NMI
);
1212 set_intercept(svm
, INTERCEPT_SMI
);
1213 set_intercept(svm
, INTERCEPT_SELECTIVE_CR0
);
1214 set_intercept(svm
, INTERCEPT_RDPMC
);
1215 set_intercept(svm
, INTERCEPT_CPUID
);
1216 set_intercept(svm
, INTERCEPT_INVD
);
1217 set_intercept(svm
, INTERCEPT_HLT
);
1218 set_intercept(svm
, INTERCEPT_INVLPG
);
1219 set_intercept(svm
, INTERCEPT_INVLPGA
);
1220 set_intercept(svm
, INTERCEPT_IOIO_PROT
);
1221 set_intercept(svm
, INTERCEPT_MSR_PROT
);
1222 set_intercept(svm
, INTERCEPT_TASK_SWITCH
);
1223 set_intercept(svm
, INTERCEPT_SHUTDOWN
);
1224 set_intercept(svm
, INTERCEPT_VMRUN
);
1225 set_intercept(svm
, INTERCEPT_VMMCALL
);
1226 set_intercept(svm
, INTERCEPT_VMLOAD
);
1227 set_intercept(svm
, INTERCEPT_VMSAVE
);
1228 set_intercept(svm
, INTERCEPT_STGI
);
1229 set_intercept(svm
, INTERCEPT_CLGI
);
1230 set_intercept(svm
, INTERCEPT_SKINIT
);
1231 set_intercept(svm
, INTERCEPT_WBINVD
);
1232 set_intercept(svm
, INTERCEPT_XSETBV
);
1234 if (!kvm_mwait_in_guest()) {
1235 set_intercept(svm
, INTERCEPT_MONITOR
);
1236 set_intercept(svm
, INTERCEPT_MWAIT
);
1239 control
->iopm_base_pa
= iopm_base
;
1240 control
->msrpm_base_pa
= __pa(svm
->msrpm
);
1241 control
->int_ctl
= V_INTR_MASKING_MASK
;
1243 init_seg(&save
->es
);
1244 init_seg(&save
->ss
);
1245 init_seg(&save
->ds
);
1246 init_seg(&save
->fs
);
1247 init_seg(&save
->gs
);
1249 save
->cs
.selector
= 0xf000;
1250 save
->cs
.base
= 0xffff0000;
1251 /* Executable/Readable Code Segment */
1252 save
->cs
.attrib
= SVM_SELECTOR_READ_MASK
| SVM_SELECTOR_P_MASK
|
1253 SVM_SELECTOR_S_MASK
| SVM_SELECTOR_CODE_MASK
;
1254 save
->cs
.limit
= 0xffff;
1256 save
->gdtr
.limit
= 0xffff;
1257 save
->idtr
.limit
= 0xffff;
1259 init_sys_seg(&save
->ldtr
, SEG_TYPE_LDT
);
1260 init_sys_seg(&save
->tr
, SEG_TYPE_BUSY_TSS16
);
1262 svm_set_efer(&svm
->vcpu
, 0);
1263 save
->dr6
= 0xffff0ff0;
1264 kvm_set_rflags(&svm
->vcpu
, 2);
1265 save
->rip
= 0x0000fff0;
1266 svm
->vcpu
.arch
.regs
[VCPU_REGS_RIP
] = save
->rip
;
1269 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1270 * It also updates the guest-visible cr0 value.
1272 svm_set_cr0(&svm
->vcpu
, X86_CR0_NW
| X86_CR0_CD
| X86_CR0_ET
);
1273 kvm_mmu_reset_context(&svm
->vcpu
);
1275 save
->cr4
= X86_CR4_PAE
;
1279 /* Setup VMCB for Nested Paging */
1280 control
->nested_ctl
= 1;
1281 clr_intercept(svm
, INTERCEPT_INVLPG
);
1282 clr_exception_intercept(svm
, PF_VECTOR
);
1283 clr_cr_intercept(svm
, INTERCEPT_CR3_READ
);
1284 clr_cr_intercept(svm
, INTERCEPT_CR3_WRITE
);
1285 save
->g_pat
= svm
->vcpu
.arch
.pat
;
1289 svm
->asid_generation
= 0;
1291 svm
->nested
.vmcb
= 0;
1292 svm
->vcpu
.arch
.hflags
= 0;
1294 if (boot_cpu_has(X86_FEATURE_PAUSEFILTER
)) {
1295 control
->pause_filter_count
= 3000;
1296 set_intercept(svm
, INTERCEPT_PAUSE
);
1300 avic_init_vmcb(svm
);
1303 * If hardware supports Virtual VMLOAD VMSAVE then enable it
1304 * in VMCB and clear intercepts to avoid #VMEXIT.
1307 clr_intercept(svm
, INTERCEPT_VMLOAD
);
1308 clr_intercept(svm
, INTERCEPT_VMSAVE
);
1309 svm
->vmcb
->control
.virt_ext
|= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK
;
1312 mark_all_dirty(svm
->vmcb
);
1318 static u64
*avic_get_physical_id_entry(struct kvm_vcpu
*vcpu
,
1321 u64
*avic_physical_id_table
;
1322 struct kvm_arch
*vm_data
= &vcpu
->kvm
->arch
;
1324 if (index
>= AVIC_MAX_PHYSICAL_ID_COUNT
)
1327 avic_physical_id_table
= page_address(vm_data
->avic_physical_id_table_page
);
1329 return &avic_physical_id_table
[index
];
1334 * AVIC hardware walks the nested page table to check permissions,
1335 * but does not use the SPA address specified in the leaf page
1336 * table entry since it uses address in the AVIC_BACKING_PAGE pointer
1337 * field of the VMCB. Therefore, we set up the
1338 * APIC_ACCESS_PAGE_PRIVATE_MEMSLOT (4KB) here.
1340 static int avic_init_access_page(struct kvm_vcpu
*vcpu
)
1342 struct kvm
*kvm
= vcpu
->kvm
;
1345 if (kvm
->arch
.apic_access_page_done
)
1348 ret
= x86_set_memory_region(kvm
,
1349 APIC_ACCESS_PAGE_PRIVATE_MEMSLOT
,
1350 APIC_DEFAULT_PHYS_BASE
,
1355 kvm
->arch
.apic_access_page_done
= true;
1359 static int avic_init_backing_page(struct kvm_vcpu
*vcpu
)
1362 u64
*entry
, new_entry
;
1363 int id
= vcpu
->vcpu_id
;
1364 struct vcpu_svm
*svm
= to_svm(vcpu
);
1366 ret
= avic_init_access_page(vcpu
);
1370 if (id
>= AVIC_MAX_PHYSICAL_ID_COUNT
)
1373 if (!svm
->vcpu
.arch
.apic
->regs
)
1376 svm
->avic_backing_page
= virt_to_page(svm
->vcpu
.arch
.apic
->regs
);
1378 /* Setting AVIC backing page address in the phy APIC ID table */
1379 entry
= avic_get_physical_id_entry(vcpu
, id
);
1383 new_entry
= READ_ONCE(*entry
);
1384 new_entry
= (page_to_phys(svm
->avic_backing_page
) &
1385 AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK
) |
1386 AVIC_PHYSICAL_ID_ENTRY_VALID_MASK
;
1387 WRITE_ONCE(*entry
, new_entry
);
1389 svm
->avic_physical_id_cache
= entry
;
1394 static inline int avic_get_next_vm_id(void)
1398 spin_lock(&avic_vm_id_lock
);
1400 /* AVIC VM ID is one-based. */
1401 id
= find_next_zero_bit(avic_vm_id_bitmap
, AVIC_VM_ID_NR
, 1);
1402 if (id
<= AVIC_VM_ID_MASK
)
1403 __set_bit(id
, avic_vm_id_bitmap
);
1407 spin_unlock(&avic_vm_id_lock
);
1411 static inline int avic_free_vm_id(int id
)
1413 if (id
<= 0 || id
> AVIC_VM_ID_MASK
)
1416 spin_lock(&avic_vm_id_lock
);
1417 __clear_bit(id
, avic_vm_id_bitmap
);
1418 spin_unlock(&avic_vm_id_lock
);
1422 static void avic_vm_destroy(struct kvm
*kvm
)
1424 unsigned long flags
;
1425 struct kvm_arch
*vm_data
= &kvm
->arch
;
1430 avic_free_vm_id(vm_data
->avic_vm_id
);
1432 if (vm_data
->avic_logical_id_table_page
)
1433 __free_page(vm_data
->avic_logical_id_table_page
);
1434 if (vm_data
->avic_physical_id_table_page
)
1435 __free_page(vm_data
->avic_physical_id_table_page
);
1437 spin_lock_irqsave(&svm_vm_data_hash_lock
, flags
);
1438 hash_del(&vm_data
->hnode
);
1439 spin_unlock_irqrestore(&svm_vm_data_hash_lock
, flags
);
1442 static int avic_vm_init(struct kvm
*kvm
)
1444 unsigned long flags
;
1445 int vm_id
, err
= -ENOMEM
;
1446 struct kvm_arch
*vm_data
= &kvm
->arch
;
1447 struct page
*p_page
;
1448 struct page
*l_page
;
1453 vm_id
= avic_get_next_vm_id();
1456 vm_data
->avic_vm_id
= (u32
)vm_id
;
1458 /* Allocating physical APIC ID table (4KB) */
1459 p_page
= alloc_page(GFP_KERNEL
);
1463 vm_data
->avic_physical_id_table_page
= p_page
;
1464 clear_page(page_address(p_page
));
1466 /* Allocating logical APIC ID table (4KB) */
1467 l_page
= alloc_page(GFP_KERNEL
);
1471 vm_data
->avic_logical_id_table_page
= l_page
;
1472 clear_page(page_address(l_page
));
1474 spin_lock_irqsave(&svm_vm_data_hash_lock
, flags
);
1475 hash_add(svm_vm_data_hash
, &vm_data
->hnode
, vm_data
->avic_vm_id
);
1476 spin_unlock_irqrestore(&svm_vm_data_hash_lock
, flags
);
1481 avic_vm_destroy(kvm
);
1486 avic_update_iommu_vcpu_affinity(struct kvm_vcpu
*vcpu
, int cpu
, bool r
)
1489 unsigned long flags
;
1490 struct amd_svm_iommu_ir
*ir
;
1491 struct vcpu_svm
*svm
= to_svm(vcpu
);
1493 if (!kvm_arch_has_assigned_device(vcpu
->kvm
))
1497 * Here, we go through the per-vcpu ir_list to update all existing
1498 * interrupt remapping table entry targeting this vcpu.
1500 spin_lock_irqsave(&svm
->ir_list_lock
, flags
);
1502 if (list_empty(&svm
->ir_list
))
1505 list_for_each_entry(ir
, &svm
->ir_list
, node
) {
1506 ret
= amd_iommu_update_ga(cpu
, r
, ir
->data
);
1511 spin_unlock_irqrestore(&svm
->ir_list_lock
, flags
);
1515 static void avic_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
1518 /* ID = 0xff (broadcast), ID > 0xff (reserved) */
1519 int h_physical_id
= kvm_cpu_get_apicid(cpu
);
1520 struct vcpu_svm
*svm
= to_svm(vcpu
);
1522 if (!kvm_vcpu_apicv_active(vcpu
))
1525 if (WARN_ON(h_physical_id
>= AVIC_MAX_PHYSICAL_ID_COUNT
))
1528 entry
= READ_ONCE(*(svm
->avic_physical_id_cache
));
1529 WARN_ON(entry
& AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK
);
1531 entry
&= ~AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK
;
1532 entry
|= (h_physical_id
& AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK
);
1534 entry
&= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK
;
1535 if (svm
->avic_is_running
)
1536 entry
|= AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK
;
1538 WRITE_ONCE(*(svm
->avic_physical_id_cache
), entry
);
1539 avic_update_iommu_vcpu_affinity(vcpu
, h_physical_id
,
1540 svm
->avic_is_running
);
1543 static void avic_vcpu_put(struct kvm_vcpu
*vcpu
)
1546 struct vcpu_svm
*svm
= to_svm(vcpu
);
1548 if (!kvm_vcpu_apicv_active(vcpu
))
1551 entry
= READ_ONCE(*(svm
->avic_physical_id_cache
));
1552 if (entry
& AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK
)
1553 avic_update_iommu_vcpu_affinity(vcpu
, -1, 0);
1555 entry
&= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK
;
1556 WRITE_ONCE(*(svm
->avic_physical_id_cache
), entry
);
1560 * This function is called during VCPU halt/unhalt.
1562 static void avic_set_running(struct kvm_vcpu
*vcpu
, bool is_run
)
1564 struct vcpu_svm
*svm
= to_svm(vcpu
);
1566 svm
->avic_is_running
= is_run
;
1568 avic_vcpu_load(vcpu
, vcpu
->cpu
);
1570 avic_vcpu_put(vcpu
);
1573 static void svm_vcpu_reset(struct kvm_vcpu
*vcpu
, bool init_event
)
1575 struct vcpu_svm
*svm
= to_svm(vcpu
);
1580 svm
->vcpu
.arch
.apic_base
= APIC_DEFAULT_PHYS_BASE
|
1581 MSR_IA32_APICBASE_ENABLE
;
1582 if (kvm_vcpu_is_reset_bsp(&svm
->vcpu
))
1583 svm
->vcpu
.arch
.apic_base
|= MSR_IA32_APICBASE_BSP
;
1587 kvm_cpuid(vcpu
, &eax
, &dummy
, &dummy
, &dummy
);
1588 kvm_register_write(vcpu
, VCPU_REGS_RDX
, eax
);
1590 if (kvm_vcpu_apicv_active(vcpu
) && !init_event
)
1591 avic_update_vapic_bar(svm
, APIC_DEFAULT_PHYS_BASE
);
1594 static struct kvm_vcpu
*svm_create_vcpu(struct kvm
*kvm
, unsigned int id
)
1596 struct vcpu_svm
*svm
;
1598 struct page
*msrpm_pages
;
1599 struct page
*hsave_page
;
1600 struct page
*nested_msrpm_pages
;
1603 svm
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
1609 err
= kvm_vcpu_init(&svm
->vcpu
, kvm
, id
);
1614 page
= alloc_page(GFP_KERNEL
);
1618 msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
1622 nested_msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
1623 if (!nested_msrpm_pages
)
1626 hsave_page
= alloc_page(GFP_KERNEL
);
1631 err
= avic_init_backing_page(&svm
->vcpu
);
1635 INIT_LIST_HEAD(&svm
->ir_list
);
1636 spin_lock_init(&svm
->ir_list_lock
);
1639 /* We initialize this flag to true to make sure that the is_running
1640 * bit would be set the first time the vcpu is loaded.
1642 svm
->avic_is_running
= true;
1644 svm
->nested
.hsave
= page_address(hsave_page
);
1646 svm
->msrpm
= page_address(msrpm_pages
);
1647 svm_vcpu_init_msrpm(svm
->msrpm
);
1649 svm
->nested
.msrpm
= page_address(nested_msrpm_pages
);
1650 svm_vcpu_init_msrpm(svm
->nested
.msrpm
);
1652 svm
->vmcb
= page_address(page
);
1653 clear_page(svm
->vmcb
);
1654 svm
->vmcb_pa
= page_to_pfn(page
) << PAGE_SHIFT
;
1655 svm
->asid_generation
= 0;
1658 svm_init_osvw(&svm
->vcpu
);
1663 __free_page(hsave_page
);
1665 __free_pages(nested_msrpm_pages
, MSRPM_ALLOC_ORDER
);
1667 __free_pages(msrpm_pages
, MSRPM_ALLOC_ORDER
);
1671 kvm_vcpu_uninit(&svm
->vcpu
);
1673 kmem_cache_free(kvm_vcpu_cache
, svm
);
1675 return ERR_PTR(err
);
1678 static void svm_free_vcpu(struct kvm_vcpu
*vcpu
)
1680 struct vcpu_svm
*svm
= to_svm(vcpu
);
1682 __free_page(pfn_to_page(svm
->vmcb_pa
>> PAGE_SHIFT
));
1683 __free_pages(virt_to_page(svm
->msrpm
), MSRPM_ALLOC_ORDER
);
1684 __free_page(virt_to_page(svm
->nested
.hsave
));
1685 __free_pages(virt_to_page(svm
->nested
.msrpm
), MSRPM_ALLOC_ORDER
);
1686 kvm_vcpu_uninit(vcpu
);
1687 kmem_cache_free(kvm_vcpu_cache
, svm
);
1690 static void svm_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
1692 struct vcpu_svm
*svm
= to_svm(vcpu
);
1695 if (unlikely(cpu
!= vcpu
->cpu
)) {
1696 svm
->asid_generation
= 0;
1697 mark_all_dirty(svm
->vmcb
);
1700 #ifdef CONFIG_X86_64
1701 rdmsrl(MSR_GS_BASE
, to_svm(vcpu
)->host
.gs_base
);
1703 savesegment(fs
, svm
->host
.fs
);
1704 savesegment(gs
, svm
->host
.gs
);
1705 svm
->host
.ldt
= kvm_read_ldt();
1707 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
1708 rdmsrl(host_save_user_msrs
[i
], svm
->host_user_msrs
[i
]);
1710 if (static_cpu_has(X86_FEATURE_TSCRATEMSR
)) {
1711 u64 tsc_ratio
= vcpu
->arch
.tsc_scaling_ratio
;
1712 if (tsc_ratio
!= __this_cpu_read(current_tsc_ratio
)) {
1713 __this_cpu_write(current_tsc_ratio
, tsc_ratio
);
1714 wrmsrl(MSR_AMD64_TSC_RATIO
, tsc_ratio
);
1717 /* This assumes that the kernel never uses MSR_TSC_AUX */
1718 if (static_cpu_has(X86_FEATURE_RDTSCP
))
1719 wrmsrl(MSR_TSC_AUX
, svm
->tsc_aux
);
1721 avic_vcpu_load(vcpu
, cpu
);
1724 static void svm_vcpu_put(struct kvm_vcpu
*vcpu
)
1726 struct vcpu_svm
*svm
= to_svm(vcpu
);
1729 avic_vcpu_put(vcpu
);
1731 ++vcpu
->stat
.host_state_reload
;
1732 kvm_load_ldt(svm
->host
.ldt
);
1733 #ifdef CONFIG_X86_64
1734 loadsegment(fs
, svm
->host
.fs
);
1735 wrmsrl(MSR_KERNEL_GS_BASE
, current
->thread
.gsbase
);
1736 load_gs_index(svm
->host
.gs
);
1738 #ifdef CONFIG_X86_32_LAZY_GS
1739 loadsegment(gs
, svm
->host
.gs
);
1742 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
1743 wrmsrl(host_save_user_msrs
[i
], svm
->host_user_msrs
[i
]);
1746 static void svm_vcpu_blocking(struct kvm_vcpu
*vcpu
)
1748 avic_set_running(vcpu
, false);
1751 static void svm_vcpu_unblocking(struct kvm_vcpu
*vcpu
)
1753 avic_set_running(vcpu
, true);
1756 static unsigned long svm_get_rflags(struct kvm_vcpu
*vcpu
)
1758 struct vcpu_svm
*svm
= to_svm(vcpu
);
1759 unsigned long rflags
= svm
->vmcb
->save
.rflags
;
1761 if (svm
->nmi_singlestep
) {
1762 /* Hide our flags if they were not set by the guest */
1763 if (!(svm
->nmi_singlestep_guest_rflags
& X86_EFLAGS_TF
))
1764 rflags
&= ~X86_EFLAGS_TF
;
1765 if (!(svm
->nmi_singlestep_guest_rflags
& X86_EFLAGS_RF
))
1766 rflags
&= ~X86_EFLAGS_RF
;
1771 static void svm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
1773 if (to_svm(vcpu
)->nmi_singlestep
)
1774 rflags
|= (X86_EFLAGS_TF
| X86_EFLAGS_RF
);
1777 * Any change of EFLAGS.VM is accompanied by a reload of SS
1778 * (caused by either a task switch or an inter-privilege IRET),
1779 * so we do not need to update the CPL here.
1781 to_svm(vcpu
)->vmcb
->save
.rflags
= rflags
;
1784 static void svm_cache_reg(struct kvm_vcpu
*vcpu
, enum kvm_reg reg
)
1787 case VCPU_EXREG_PDPTR
:
1788 BUG_ON(!npt_enabled
);
1789 load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, kvm_read_cr3(vcpu
));
1796 static void svm_set_vintr(struct vcpu_svm
*svm
)
1798 set_intercept(svm
, INTERCEPT_VINTR
);
1801 static void svm_clear_vintr(struct vcpu_svm
*svm
)
1803 clr_intercept(svm
, INTERCEPT_VINTR
);
1806 static struct vmcb_seg
*svm_seg(struct kvm_vcpu
*vcpu
, int seg
)
1808 struct vmcb_save_area
*save
= &to_svm(vcpu
)->vmcb
->save
;
1811 case VCPU_SREG_CS
: return &save
->cs
;
1812 case VCPU_SREG_DS
: return &save
->ds
;
1813 case VCPU_SREG_ES
: return &save
->es
;
1814 case VCPU_SREG_FS
: return &save
->fs
;
1815 case VCPU_SREG_GS
: return &save
->gs
;
1816 case VCPU_SREG_SS
: return &save
->ss
;
1817 case VCPU_SREG_TR
: return &save
->tr
;
1818 case VCPU_SREG_LDTR
: return &save
->ldtr
;
1824 static u64
svm_get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
1826 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
1831 static void svm_get_segment(struct kvm_vcpu
*vcpu
,
1832 struct kvm_segment
*var
, int seg
)
1834 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
1836 var
->base
= s
->base
;
1837 var
->limit
= s
->limit
;
1838 var
->selector
= s
->selector
;
1839 var
->type
= s
->attrib
& SVM_SELECTOR_TYPE_MASK
;
1840 var
->s
= (s
->attrib
>> SVM_SELECTOR_S_SHIFT
) & 1;
1841 var
->dpl
= (s
->attrib
>> SVM_SELECTOR_DPL_SHIFT
) & 3;
1842 var
->present
= (s
->attrib
>> SVM_SELECTOR_P_SHIFT
) & 1;
1843 var
->avl
= (s
->attrib
>> SVM_SELECTOR_AVL_SHIFT
) & 1;
1844 var
->l
= (s
->attrib
>> SVM_SELECTOR_L_SHIFT
) & 1;
1845 var
->db
= (s
->attrib
>> SVM_SELECTOR_DB_SHIFT
) & 1;
1848 * AMD CPUs circa 2014 track the G bit for all segments except CS.
1849 * However, the SVM spec states that the G bit is not observed by the
1850 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
1851 * So let's synthesize a legal G bit for all segments, this helps
1852 * running KVM nested. It also helps cross-vendor migration, because
1853 * Intel's vmentry has a check on the 'G' bit.
1855 var
->g
= s
->limit
> 0xfffff;
1858 * AMD's VMCB does not have an explicit unusable field, so emulate it
1859 * for cross vendor migration purposes by "not present"
1861 var
->unusable
= !var
->present
;
1866 * Work around a bug where the busy flag in the tr selector
1876 * The accessed bit must always be set in the segment
1877 * descriptor cache, although it can be cleared in the
1878 * descriptor, the cached bit always remains at 1. Since
1879 * Intel has a check on this, set it here to support
1880 * cross-vendor migration.
1887 * On AMD CPUs sometimes the DB bit in the segment
1888 * descriptor is left as 1, although the whole segment has
1889 * been made unusable. Clear it here to pass an Intel VMX
1890 * entry check when cross vendor migrating.
1894 /* This is symmetric with svm_set_segment() */
1895 var
->dpl
= to_svm(vcpu
)->vmcb
->save
.cpl
;
1900 static int svm_get_cpl(struct kvm_vcpu
*vcpu
)
1902 struct vmcb_save_area
*save
= &to_svm(vcpu
)->vmcb
->save
;
1907 static void svm_get_idt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1909 struct vcpu_svm
*svm
= to_svm(vcpu
);
1911 dt
->size
= svm
->vmcb
->save
.idtr
.limit
;
1912 dt
->address
= svm
->vmcb
->save
.idtr
.base
;
1915 static void svm_set_idt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1917 struct vcpu_svm
*svm
= to_svm(vcpu
);
1919 svm
->vmcb
->save
.idtr
.limit
= dt
->size
;
1920 svm
->vmcb
->save
.idtr
.base
= dt
->address
;
1921 mark_dirty(svm
->vmcb
, VMCB_DT
);
1924 static void svm_get_gdt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1926 struct vcpu_svm
*svm
= to_svm(vcpu
);
1928 dt
->size
= svm
->vmcb
->save
.gdtr
.limit
;
1929 dt
->address
= svm
->vmcb
->save
.gdtr
.base
;
1932 static void svm_set_gdt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1934 struct vcpu_svm
*svm
= to_svm(vcpu
);
1936 svm
->vmcb
->save
.gdtr
.limit
= dt
->size
;
1937 svm
->vmcb
->save
.gdtr
.base
= dt
->address
;
1938 mark_dirty(svm
->vmcb
, VMCB_DT
);
1941 static void svm_decache_cr0_guest_bits(struct kvm_vcpu
*vcpu
)
1945 static void svm_decache_cr3(struct kvm_vcpu
*vcpu
)
1949 static void svm_decache_cr4_guest_bits(struct kvm_vcpu
*vcpu
)
1953 static void update_cr0_intercept(struct vcpu_svm
*svm
)
1955 ulong gcr0
= svm
->vcpu
.arch
.cr0
;
1956 u64
*hcr0
= &svm
->vmcb
->save
.cr0
;
1958 *hcr0
= (*hcr0
& ~SVM_CR0_SELECTIVE_MASK
)
1959 | (gcr0
& SVM_CR0_SELECTIVE_MASK
);
1961 mark_dirty(svm
->vmcb
, VMCB_CR
);
1963 if (gcr0
== *hcr0
) {
1964 clr_cr_intercept(svm
, INTERCEPT_CR0_READ
);
1965 clr_cr_intercept(svm
, INTERCEPT_CR0_WRITE
);
1967 set_cr_intercept(svm
, INTERCEPT_CR0_READ
);
1968 set_cr_intercept(svm
, INTERCEPT_CR0_WRITE
);
1972 static void svm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
1974 struct vcpu_svm
*svm
= to_svm(vcpu
);
1976 #ifdef CONFIG_X86_64
1977 if (vcpu
->arch
.efer
& EFER_LME
) {
1978 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
1979 vcpu
->arch
.efer
|= EFER_LMA
;
1980 svm
->vmcb
->save
.efer
|= EFER_LMA
| EFER_LME
;
1983 if (is_paging(vcpu
) && !(cr0
& X86_CR0_PG
)) {
1984 vcpu
->arch
.efer
&= ~EFER_LMA
;
1985 svm
->vmcb
->save
.efer
&= ~(EFER_LMA
| EFER_LME
);
1989 vcpu
->arch
.cr0
= cr0
;
1992 cr0
|= X86_CR0_PG
| X86_CR0_WP
;
1995 * re-enable caching here because the QEMU bios
1996 * does not do it - this results in some delay at
1999 if (kvm_check_has_quirk(vcpu
->kvm
, KVM_X86_QUIRK_CD_NW_CLEARED
))
2000 cr0
&= ~(X86_CR0_CD
| X86_CR0_NW
);
2001 svm
->vmcb
->save
.cr0
= cr0
;
2002 mark_dirty(svm
->vmcb
, VMCB_CR
);
2003 update_cr0_intercept(svm
);
2006 static int svm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
2008 unsigned long host_cr4_mce
= cr4_read_shadow() & X86_CR4_MCE
;
2009 unsigned long old_cr4
= to_svm(vcpu
)->vmcb
->save
.cr4
;
2011 if (cr4
& X86_CR4_VMXE
)
2014 if (npt_enabled
&& ((old_cr4
^ cr4
) & X86_CR4_PGE
))
2015 svm_flush_tlb(vcpu
);
2017 vcpu
->arch
.cr4
= cr4
;
2020 cr4
|= host_cr4_mce
;
2021 to_svm(vcpu
)->vmcb
->save
.cr4
= cr4
;
2022 mark_dirty(to_svm(vcpu
)->vmcb
, VMCB_CR
);
2026 static void svm_set_segment(struct kvm_vcpu
*vcpu
,
2027 struct kvm_segment
*var
, int seg
)
2029 struct vcpu_svm
*svm
= to_svm(vcpu
);
2030 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
2032 s
->base
= var
->base
;
2033 s
->limit
= var
->limit
;
2034 s
->selector
= var
->selector
;
2035 s
->attrib
= (var
->type
& SVM_SELECTOR_TYPE_MASK
);
2036 s
->attrib
|= (var
->s
& 1) << SVM_SELECTOR_S_SHIFT
;
2037 s
->attrib
|= (var
->dpl
& 3) << SVM_SELECTOR_DPL_SHIFT
;
2038 s
->attrib
|= ((var
->present
& 1) && !var
->unusable
) << SVM_SELECTOR_P_SHIFT
;
2039 s
->attrib
|= (var
->avl
& 1) << SVM_SELECTOR_AVL_SHIFT
;
2040 s
->attrib
|= (var
->l
& 1) << SVM_SELECTOR_L_SHIFT
;
2041 s
->attrib
|= (var
->db
& 1) << SVM_SELECTOR_DB_SHIFT
;
2042 s
->attrib
|= (var
->g
& 1) << SVM_SELECTOR_G_SHIFT
;
2045 * This is always accurate, except if SYSRET returned to a segment
2046 * with SS.DPL != 3. Intel does not have this quirk, and always
2047 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
2048 * would entail passing the CPL to userspace and back.
2050 if (seg
== VCPU_SREG_SS
)
2051 /* This is symmetric with svm_get_segment() */
2052 svm
->vmcb
->save
.cpl
= (var
->dpl
& 3);
2054 mark_dirty(svm
->vmcb
, VMCB_SEG
);
2057 static void update_bp_intercept(struct kvm_vcpu
*vcpu
)
2059 struct vcpu_svm
*svm
= to_svm(vcpu
);
2061 clr_exception_intercept(svm
, BP_VECTOR
);
2063 if (vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
) {
2064 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_SW_BP
)
2065 set_exception_intercept(svm
, BP_VECTOR
);
2067 vcpu
->guest_debug
= 0;
2070 static void new_asid(struct vcpu_svm
*svm
, struct svm_cpu_data
*sd
)
2072 if (sd
->next_asid
> sd
->max_asid
) {
2073 ++sd
->asid_generation
;
2075 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ALL_ASID
;
2078 svm
->asid_generation
= sd
->asid_generation
;
2079 svm
->vmcb
->control
.asid
= sd
->next_asid
++;
2081 mark_dirty(svm
->vmcb
, VMCB_ASID
);
2084 static u64
svm_get_dr6(struct kvm_vcpu
*vcpu
)
2086 return to_svm(vcpu
)->vmcb
->save
.dr6
;
2089 static void svm_set_dr6(struct kvm_vcpu
*vcpu
, unsigned long value
)
2091 struct vcpu_svm
*svm
= to_svm(vcpu
);
2093 svm
->vmcb
->save
.dr6
= value
;
2094 mark_dirty(svm
->vmcb
, VMCB_DR
);
2097 static void svm_sync_dirty_debug_regs(struct kvm_vcpu
*vcpu
)
2099 struct vcpu_svm
*svm
= to_svm(vcpu
);
2101 get_debugreg(vcpu
->arch
.db
[0], 0);
2102 get_debugreg(vcpu
->arch
.db
[1], 1);
2103 get_debugreg(vcpu
->arch
.db
[2], 2);
2104 get_debugreg(vcpu
->arch
.db
[3], 3);
2105 vcpu
->arch
.dr6
= svm_get_dr6(vcpu
);
2106 vcpu
->arch
.dr7
= svm
->vmcb
->save
.dr7
;
2108 vcpu
->arch
.switch_db_regs
&= ~KVM_DEBUGREG_WONT_EXIT
;
2109 set_dr_intercepts(svm
);
2112 static void svm_set_dr7(struct kvm_vcpu
*vcpu
, unsigned long value
)
2114 struct vcpu_svm
*svm
= to_svm(vcpu
);
2116 svm
->vmcb
->save
.dr7
= value
;
2117 mark_dirty(svm
->vmcb
, VMCB_DR
);
2120 static int pf_interception(struct vcpu_svm
*svm
)
2122 u64 fault_address
= svm
->vmcb
->control
.exit_info_2
;
2123 u64 error_code
= svm
->vmcb
->control
.exit_info_1
;
2125 return kvm_handle_page_fault(&svm
->vcpu
, error_code
, fault_address
,
2126 svm
->vmcb
->control
.insn_bytes
,
2127 svm
->vmcb
->control
.insn_len
, !npt_enabled
);
2130 static int db_interception(struct vcpu_svm
*svm
)
2132 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
2134 if (!(svm
->vcpu
.guest_debug
&
2135 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
)) &&
2136 !svm
->nmi_singlestep
) {
2137 kvm_queue_exception(&svm
->vcpu
, DB_VECTOR
);
2141 if (svm
->nmi_singlestep
) {
2142 disable_nmi_singlestep(svm
);
2145 if (svm
->vcpu
.guest_debug
&
2146 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
)) {
2147 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
2148 kvm_run
->debug
.arch
.pc
=
2149 svm
->vmcb
->save
.cs
.base
+ svm
->vmcb
->save
.rip
;
2150 kvm_run
->debug
.arch
.exception
= DB_VECTOR
;
2157 static int bp_interception(struct vcpu_svm
*svm
)
2159 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
2161 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
2162 kvm_run
->debug
.arch
.pc
= svm
->vmcb
->save
.cs
.base
+ svm
->vmcb
->save
.rip
;
2163 kvm_run
->debug
.arch
.exception
= BP_VECTOR
;
2167 static int ud_interception(struct vcpu_svm
*svm
)
2171 er
= emulate_instruction(&svm
->vcpu
, EMULTYPE_TRAP_UD
);
2172 if (er
!= EMULATE_DONE
)
2173 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
2177 static int ac_interception(struct vcpu_svm
*svm
)
2179 kvm_queue_exception_e(&svm
->vcpu
, AC_VECTOR
, 0);
2183 static bool is_erratum_383(void)
2188 if (!erratum_383_found
)
2191 value
= native_read_msr_safe(MSR_IA32_MC0_STATUS
, &err
);
2195 /* Bit 62 may or may not be set for this mce */
2196 value
&= ~(1ULL << 62);
2198 if (value
!= 0xb600000000010015ULL
)
2201 /* Clear MCi_STATUS registers */
2202 for (i
= 0; i
< 6; ++i
)
2203 native_write_msr_safe(MSR_IA32_MCx_STATUS(i
), 0, 0);
2205 value
= native_read_msr_safe(MSR_IA32_MCG_STATUS
, &err
);
2209 value
&= ~(1ULL << 2);
2210 low
= lower_32_bits(value
);
2211 high
= upper_32_bits(value
);
2213 native_write_msr_safe(MSR_IA32_MCG_STATUS
, low
, high
);
2216 /* Flush tlb to evict multi-match entries */
2222 static void svm_handle_mce(struct vcpu_svm
*svm
)
2224 if (is_erratum_383()) {
2226 * Erratum 383 triggered. Guest state is corrupt so kill the
2229 pr_err("KVM: Guest triggered AMD Erratum 383\n");
2231 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, &svm
->vcpu
);
2237 * On an #MC intercept the MCE handler is not called automatically in
2238 * the host. So do it by hand here.
2242 /* not sure if we ever come back to this point */
2247 static int mc_interception(struct vcpu_svm
*svm
)
2252 static int shutdown_interception(struct vcpu_svm
*svm
)
2254 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
2257 * VMCB is undefined after a SHUTDOWN intercept
2258 * so reinitialize it.
2260 clear_page(svm
->vmcb
);
2263 kvm_run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
2267 static int io_interception(struct vcpu_svm
*svm
)
2269 struct kvm_vcpu
*vcpu
= &svm
->vcpu
;
2270 u32 io_info
= svm
->vmcb
->control
.exit_info_1
; /* address size bug? */
2271 int size
, in
, string
, ret
;
2274 ++svm
->vcpu
.stat
.io_exits
;
2275 string
= (io_info
& SVM_IOIO_STR_MASK
) != 0;
2276 in
= (io_info
& SVM_IOIO_TYPE_MASK
) != 0;
2278 return emulate_instruction(vcpu
, 0) == EMULATE_DONE
;
2280 port
= io_info
>> 16;
2281 size
= (io_info
& SVM_IOIO_SIZE_MASK
) >> SVM_IOIO_SIZE_SHIFT
;
2282 svm
->next_rip
= svm
->vmcb
->control
.exit_info_2
;
2283 ret
= kvm_skip_emulated_instruction(&svm
->vcpu
);
2286 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
2287 * KVM_EXIT_DEBUG here.
2290 return kvm_fast_pio_in(vcpu
, size
, port
) && ret
;
2292 return kvm_fast_pio_out(vcpu
, size
, port
) && ret
;
2295 static int nmi_interception(struct vcpu_svm
*svm
)
2300 static int intr_interception(struct vcpu_svm
*svm
)
2302 ++svm
->vcpu
.stat
.irq_exits
;
2306 static int nop_on_interception(struct vcpu_svm
*svm
)
2311 static int halt_interception(struct vcpu_svm
*svm
)
2313 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 1;
2314 return kvm_emulate_halt(&svm
->vcpu
);
2317 static int vmmcall_interception(struct vcpu_svm
*svm
)
2319 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2320 return kvm_emulate_hypercall(&svm
->vcpu
);
2323 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu
*vcpu
)
2325 struct vcpu_svm
*svm
= to_svm(vcpu
);
2327 return svm
->nested
.nested_cr3
;
2330 static u64
nested_svm_get_tdp_pdptr(struct kvm_vcpu
*vcpu
, int index
)
2332 struct vcpu_svm
*svm
= to_svm(vcpu
);
2333 u64 cr3
= svm
->nested
.nested_cr3
;
2337 ret
= kvm_vcpu_read_guest_page(vcpu
, gpa_to_gfn(cr3
), &pdpte
,
2338 offset_in_page(cr3
) + index
* 8, 8);
2344 static void nested_svm_set_tdp_cr3(struct kvm_vcpu
*vcpu
,
2347 struct vcpu_svm
*svm
= to_svm(vcpu
);
2349 svm
->vmcb
->control
.nested_cr3
= root
;
2350 mark_dirty(svm
->vmcb
, VMCB_NPT
);
2351 svm_flush_tlb(vcpu
);
2354 static void nested_svm_inject_npf_exit(struct kvm_vcpu
*vcpu
,
2355 struct x86_exception
*fault
)
2357 struct vcpu_svm
*svm
= to_svm(vcpu
);
2359 if (svm
->vmcb
->control
.exit_code
!= SVM_EXIT_NPF
) {
2361 * TODO: track the cause of the nested page fault, and
2362 * correctly fill in the high bits of exit_info_1.
2364 svm
->vmcb
->control
.exit_code
= SVM_EXIT_NPF
;
2365 svm
->vmcb
->control
.exit_code_hi
= 0;
2366 svm
->vmcb
->control
.exit_info_1
= (1ULL << 32);
2367 svm
->vmcb
->control
.exit_info_2
= fault
->address
;
2370 svm
->vmcb
->control
.exit_info_1
&= ~0xffffffffULL
;
2371 svm
->vmcb
->control
.exit_info_1
|= fault
->error_code
;
2374 * The present bit is always zero for page structure faults on real
2377 if (svm
->vmcb
->control
.exit_info_1
& (2ULL << 32))
2378 svm
->vmcb
->control
.exit_info_1
&= ~1;
2380 nested_svm_vmexit(svm
);
2383 static void nested_svm_init_mmu_context(struct kvm_vcpu
*vcpu
)
2385 WARN_ON(mmu_is_nested(vcpu
));
2386 kvm_init_shadow_mmu(vcpu
);
2387 vcpu
->arch
.mmu
.set_cr3
= nested_svm_set_tdp_cr3
;
2388 vcpu
->arch
.mmu
.get_cr3
= nested_svm_get_tdp_cr3
;
2389 vcpu
->arch
.mmu
.get_pdptr
= nested_svm_get_tdp_pdptr
;
2390 vcpu
->arch
.mmu
.inject_page_fault
= nested_svm_inject_npf_exit
;
2391 vcpu
->arch
.mmu
.shadow_root_level
= get_npt_level();
2392 reset_shadow_zero_bits_mask(vcpu
, &vcpu
->arch
.mmu
);
2393 vcpu
->arch
.walk_mmu
= &vcpu
->arch
.nested_mmu
;
2396 static void nested_svm_uninit_mmu_context(struct kvm_vcpu
*vcpu
)
2398 vcpu
->arch
.walk_mmu
= &vcpu
->arch
.mmu
;
2401 static int nested_svm_check_permissions(struct vcpu_svm
*svm
)
2403 if (!(svm
->vcpu
.arch
.efer
& EFER_SVME
) ||
2404 !is_paging(&svm
->vcpu
)) {
2405 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
2409 if (svm
->vmcb
->save
.cpl
) {
2410 kvm_inject_gp(&svm
->vcpu
, 0);
2417 static int nested_svm_check_exception(struct vcpu_svm
*svm
, unsigned nr
,
2418 bool has_error_code
, u32 error_code
)
2422 if (!is_guest_mode(&svm
->vcpu
))
2425 vmexit
= nested_svm_intercept(svm
);
2426 if (vmexit
!= NESTED_EXIT_DONE
)
2429 svm
->vmcb
->control
.exit_code
= SVM_EXIT_EXCP_BASE
+ nr
;
2430 svm
->vmcb
->control
.exit_code_hi
= 0;
2431 svm
->vmcb
->control
.exit_info_1
= error_code
;
2434 * FIXME: we should not write CR2 when L1 intercepts an L2 #PF exception.
2435 * The fix is to add the ancillary datum (CR2 or DR6) to structs
2436 * kvm_queued_exception and kvm_vcpu_events, so that CR2 and DR6 can be
2437 * written only when inject_pending_event runs (DR6 would written here
2438 * too). This should be conditional on a new capability---if the
2439 * capability is disabled, kvm_multiple_exception would write the
2440 * ancillary information to CR2 or DR6, for backwards ABI-compatibility.
2442 if (svm
->vcpu
.arch
.exception
.nested_apf
)
2443 svm
->vmcb
->control
.exit_info_2
= svm
->vcpu
.arch
.apf
.nested_apf_token
;
2445 svm
->vmcb
->control
.exit_info_2
= svm
->vcpu
.arch
.cr2
;
2447 svm
->nested
.exit_required
= true;
2451 /* This function returns true if it is save to enable the irq window */
2452 static inline bool nested_svm_intr(struct vcpu_svm
*svm
)
2454 if (!is_guest_mode(&svm
->vcpu
))
2457 if (!(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
))
2460 if (!(svm
->vcpu
.arch
.hflags
& HF_HIF_MASK
))
2464 * if vmexit was already requested (by intercepted exception
2465 * for instance) do not overwrite it with "external interrupt"
2468 if (svm
->nested
.exit_required
)
2471 svm
->vmcb
->control
.exit_code
= SVM_EXIT_INTR
;
2472 svm
->vmcb
->control
.exit_info_1
= 0;
2473 svm
->vmcb
->control
.exit_info_2
= 0;
2475 if (svm
->nested
.intercept
& 1ULL) {
2477 * The #vmexit can't be emulated here directly because this
2478 * code path runs with irqs and preemption disabled. A
2479 * #vmexit emulation might sleep. Only signal request for
2482 svm
->nested
.exit_required
= true;
2483 trace_kvm_nested_intr_vmexit(svm
->vmcb
->save
.rip
);
2490 /* This function returns true if it is save to enable the nmi window */
2491 static inline bool nested_svm_nmi(struct vcpu_svm
*svm
)
2493 if (!is_guest_mode(&svm
->vcpu
))
2496 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_NMI
)))
2499 svm
->vmcb
->control
.exit_code
= SVM_EXIT_NMI
;
2500 svm
->nested
.exit_required
= true;
2505 static void *nested_svm_map(struct vcpu_svm
*svm
, u64 gpa
, struct page
**_page
)
2511 page
= kvm_vcpu_gfn_to_page(&svm
->vcpu
, gpa
>> PAGE_SHIFT
);
2512 if (is_error_page(page
))
2520 kvm_inject_gp(&svm
->vcpu
, 0);
2525 static void nested_svm_unmap(struct page
*page
)
2528 kvm_release_page_dirty(page
);
2531 static int nested_svm_intercept_ioio(struct vcpu_svm
*svm
)
2533 unsigned port
, size
, iopm_len
;
2538 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_IOIO_PROT
)))
2539 return NESTED_EXIT_HOST
;
2541 port
= svm
->vmcb
->control
.exit_info_1
>> 16;
2542 size
= (svm
->vmcb
->control
.exit_info_1
& SVM_IOIO_SIZE_MASK
) >>
2543 SVM_IOIO_SIZE_SHIFT
;
2544 gpa
= svm
->nested
.vmcb_iopm
+ (port
/ 8);
2545 start_bit
= port
% 8;
2546 iopm_len
= (start_bit
+ size
> 8) ? 2 : 1;
2547 mask
= (0xf >> (4 - size
)) << start_bit
;
2550 if (kvm_vcpu_read_guest(&svm
->vcpu
, gpa
, &val
, iopm_len
))
2551 return NESTED_EXIT_DONE
;
2553 return (val
& mask
) ? NESTED_EXIT_DONE
: NESTED_EXIT_HOST
;
2556 static int nested_svm_exit_handled_msr(struct vcpu_svm
*svm
)
2558 u32 offset
, msr
, value
;
2561 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_MSR_PROT
)))
2562 return NESTED_EXIT_HOST
;
2564 msr
= svm
->vcpu
.arch
.regs
[VCPU_REGS_RCX
];
2565 offset
= svm_msrpm_offset(msr
);
2566 write
= svm
->vmcb
->control
.exit_info_1
& 1;
2567 mask
= 1 << ((2 * (msr
& 0xf)) + write
);
2569 if (offset
== MSR_INVALID
)
2570 return NESTED_EXIT_DONE
;
2572 /* Offset is in 32 bit units but need in 8 bit units */
2575 if (kvm_vcpu_read_guest(&svm
->vcpu
, svm
->nested
.vmcb_msrpm
+ offset
, &value
, 4))
2576 return NESTED_EXIT_DONE
;
2578 return (value
& mask
) ? NESTED_EXIT_DONE
: NESTED_EXIT_HOST
;
2581 /* DB exceptions for our internal use must not cause vmexit */
2582 static int nested_svm_intercept_db(struct vcpu_svm
*svm
)
2586 /* if we're not singlestepping, it's not ours */
2587 if (!svm
->nmi_singlestep
)
2588 return NESTED_EXIT_DONE
;
2590 /* if it's not a singlestep exception, it's not ours */
2591 if (kvm_get_dr(&svm
->vcpu
, 6, &dr6
))
2592 return NESTED_EXIT_DONE
;
2593 if (!(dr6
& DR6_BS
))
2594 return NESTED_EXIT_DONE
;
2596 /* if the guest is singlestepping, it should get the vmexit */
2597 if (svm
->nmi_singlestep_guest_rflags
& X86_EFLAGS_TF
) {
2598 disable_nmi_singlestep(svm
);
2599 return NESTED_EXIT_DONE
;
2602 /* it's ours, the nested hypervisor must not see this one */
2603 return NESTED_EXIT_HOST
;
2606 static int nested_svm_exit_special(struct vcpu_svm
*svm
)
2608 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
2610 switch (exit_code
) {
2613 case SVM_EXIT_EXCP_BASE
+ MC_VECTOR
:
2614 return NESTED_EXIT_HOST
;
2616 /* For now we are always handling NPFs when using them */
2618 return NESTED_EXIT_HOST
;
2620 case SVM_EXIT_EXCP_BASE
+ PF_VECTOR
:
2621 /* When we're shadowing, trap PFs, but not async PF */
2622 if (!npt_enabled
&& svm
->vcpu
.arch
.apf
.host_apf_reason
== 0)
2623 return NESTED_EXIT_HOST
;
2629 return NESTED_EXIT_CONTINUE
;
2633 * If this function returns true, this #vmexit was already handled
2635 static int nested_svm_intercept(struct vcpu_svm
*svm
)
2637 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
2638 int vmexit
= NESTED_EXIT_HOST
;
2640 switch (exit_code
) {
2642 vmexit
= nested_svm_exit_handled_msr(svm
);
2645 vmexit
= nested_svm_intercept_ioio(svm
);
2647 case SVM_EXIT_READ_CR0
... SVM_EXIT_WRITE_CR8
: {
2648 u32 bit
= 1U << (exit_code
- SVM_EXIT_READ_CR0
);
2649 if (svm
->nested
.intercept_cr
& bit
)
2650 vmexit
= NESTED_EXIT_DONE
;
2653 case SVM_EXIT_READ_DR0
... SVM_EXIT_WRITE_DR7
: {
2654 u32 bit
= 1U << (exit_code
- SVM_EXIT_READ_DR0
);
2655 if (svm
->nested
.intercept_dr
& bit
)
2656 vmexit
= NESTED_EXIT_DONE
;
2659 case SVM_EXIT_EXCP_BASE
... SVM_EXIT_EXCP_BASE
+ 0x1f: {
2660 u32 excp_bits
= 1 << (exit_code
- SVM_EXIT_EXCP_BASE
);
2661 if (svm
->nested
.intercept_exceptions
& excp_bits
) {
2662 if (exit_code
== SVM_EXIT_EXCP_BASE
+ DB_VECTOR
)
2663 vmexit
= nested_svm_intercept_db(svm
);
2665 vmexit
= NESTED_EXIT_DONE
;
2667 /* async page fault always cause vmexit */
2668 else if ((exit_code
== SVM_EXIT_EXCP_BASE
+ PF_VECTOR
) &&
2669 svm
->vcpu
.arch
.exception
.nested_apf
!= 0)
2670 vmexit
= NESTED_EXIT_DONE
;
2673 case SVM_EXIT_ERR
: {
2674 vmexit
= NESTED_EXIT_DONE
;
2678 u64 exit_bits
= 1ULL << (exit_code
- SVM_EXIT_INTR
);
2679 if (svm
->nested
.intercept
& exit_bits
)
2680 vmexit
= NESTED_EXIT_DONE
;
2687 static int nested_svm_exit_handled(struct vcpu_svm
*svm
)
2691 vmexit
= nested_svm_intercept(svm
);
2693 if (vmexit
== NESTED_EXIT_DONE
)
2694 nested_svm_vmexit(svm
);
2699 static inline void copy_vmcb_control_area(struct vmcb
*dst_vmcb
, struct vmcb
*from_vmcb
)
2701 struct vmcb_control_area
*dst
= &dst_vmcb
->control
;
2702 struct vmcb_control_area
*from
= &from_vmcb
->control
;
2704 dst
->intercept_cr
= from
->intercept_cr
;
2705 dst
->intercept_dr
= from
->intercept_dr
;
2706 dst
->intercept_exceptions
= from
->intercept_exceptions
;
2707 dst
->intercept
= from
->intercept
;
2708 dst
->iopm_base_pa
= from
->iopm_base_pa
;
2709 dst
->msrpm_base_pa
= from
->msrpm_base_pa
;
2710 dst
->tsc_offset
= from
->tsc_offset
;
2711 dst
->asid
= from
->asid
;
2712 dst
->tlb_ctl
= from
->tlb_ctl
;
2713 dst
->int_ctl
= from
->int_ctl
;
2714 dst
->int_vector
= from
->int_vector
;
2715 dst
->int_state
= from
->int_state
;
2716 dst
->exit_code
= from
->exit_code
;
2717 dst
->exit_code_hi
= from
->exit_code_hi
;
2718 dst
->exit_info_1
= from
->exit_info_1
;
2719 dst
->exit_info_2
= from
->exit_info_2
;
2720 dst
->exit_int_info
= from
->exit_int_info
;
2721 dst
->exit_int_info_err
= from
->exit_int_info_err
;
2722 dst
->nested_ctl
= from
->nested_ctl
;
2723 dst
->event_inj
= from
->event_inj
;
2724 dst
->event_inj_err
= from
->event_inj_err
;
2725 dst
->nested_cr3
= from
->nested_cr3
;
2726 dst
->virt_ext
= from
->virt_ext
;
2729 static int nested_svm_vmexit(struct vcpu_svm
*svm
)
2731 struct vmcb
*nested_vmcb
;
2732 struct vmcb
*hsave
= svm
->nested
.hsave
;
2733 struct vmcb
*vmcb
= svm
->vmcb
;
2736 trace_kvm_nested_vmexit_inject(vmcb
->control
.exit_code
,
2737 vmcb
->control
.exit_info_1
,
2738 vmcb
->control
.exit_info_2
,
2739 vmcb
->control
.exit_int_info
,
2740 vmcb
->control
.exit_int_info_err
,
2743 nested_vmcb
= nested_svm_map(svm
, svm
->nested
.vmcb
, &page
);
2747 /* Exit Guest-Mode */
2748 leave_guest_mode(&svm
->vcpu
);
2749 svm
->nested
.vmcb
= 0;
2751 /* Give the current vmcb to the guest */
2754 nested_vmcb
->save
.es
= vmcb
->save
.es
;
2755 nested_vmcb
->save
.cs
= vmcb
->save
.cs
;
2756 nested_vmcb
->save
.ss
= vmcb
->save
.ss
;
2757 nested_vmcb
->save
.ds
= vmcb
->save
.ds
;
2758 nested_vmcb
->save
.gdtr
= vmcb
->save
.gdtr
;
2759 nested_vmcb
->save
.idtr
= vmcb
->save
.idtr
;
2760 nested_vmcb
->save
.efer
= svm
->vcpu
.arch
.efer
;
2761 nested_vmcb
->save
.cr0
= kvm_read_cr0(&svm
->vcpu
);
2762 nested_vmcb
->save
.cr3
= kvm_read_cr3(&svm
->vcpu
);
2763 nested_vmcb
->save
.cr2
= vmcb
->save
.cr2
;
2764 nested_vmcb
->save
.cr4
= svm
->vcpu
.arch
.cr4
;
2765 nested_vmcb
->save
.rflags
= kvm_get_rflags(&svm
->vcpu
);
2766 nested_vmcb
->save
.rip
= vmcb
->save
.rip
;
2767 nested_vmcb
->save
.rsp
= vmcb
->save
.rsp
;
2768 nested_vmcb
->save
.rax
= vmcb
->save
.rax
;
2769 nested_vmcb
->save
.dr7
= vmcb
->save
.dr7
;
2770 nested_vmcb
->save
.dr6
= vmcb
->save
.dr6
;
2771 nested_vmcb
->save
.cpl
= vmcb
->save
.cpl
;
2773 nested_vmcb
->control
.int_ctl
= vmcb
->control
.int_ctl
;
2774 nested_vmcb
->control
.int_vector
= vmcb
->control
.int_vector
;
2775 nested_vmcb
->control
.int_state
= vmcb
->control
.int_state
;
2776 nested_vmcb
->control
.exit_code
= vmcb
->control
.exit_code
;
2777 nested_vmcb
->control
.exit_code_hi
= vmcb
->control
.exit_code_hi
;
2778 nested_vmcb
->control
.exit_info_1
= vmcb
->control
.exit_info_1
;
2779 nested_vmcb
->control
.exit_info_2
= vmcb
->control
.exit_info_2
;
2780 nested_vmcb
->control
.exit_int_info
= vmcb
->control
.exit_int_info
;
2781 nested_vmcb
->control
.exit_int_info_err
= vmcb
->control
.exit_int_info_err
;
2783 if (svm
->nrips_enabled
)
2784 nested_vmcb
->control
.next_rip
= vmcb
->control
.next_rip
;
2787 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2788 * to make sure that we do not lose injected events. So check event_inj
2789 * here and copy it to exit_int_info if it is valid.
2790 * Exit_int_info and event_inj can't be both valid because the case
2791 * below only happens on a VMRUN instruction intercept which has
2792 * no valid exit_int_info set.
2794 if (vmcb
->control
.event_inj
& SVM_EVTINJ_VALID
) {
2795 struct vmcb_control_area
*nc
= &nested_vmcb
->control
;
2797 nc
->exit_int_info
= vmcb
->control
.event_inj
;
2798 nc
->exit_int_info_err
= vmcb
->control
.event_inj_err
;
2801 nested_vmcb
->control
.tlb_ctl
= 0;
2802 nested_vmcb
->control
.event_inj
= 0;
2803 nested_vmcb
->control
.event_inj_err
= 0;
2805 /* We always set V_INTR_MASKING and remember the old value in hflags */
2806 if (!(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
))
2807 nested_vmcb
->control
.int_ctl
&= ~V_INTR_MASKING_MASK
;
2809 /* Restore the original control entries */
2810 copy_vmcb_control_area(vmcb
, hsave
);
2812 kvm_clear_exception_queue(&svm
->vcpu
);
2813 kvm_clear_interrupt_queue(&svm
->vcpu
);
2815 svm
->nested
.nested_cr3
= 0;
2817 /* Restore selected save entries */
2818 svm
->vmcb
->save
.es
= hsave
->save
.es
;
2819 svm
->vmcb
->save
.cs
= hsave
->save
.cs
;
2820 svm
->vmcb
->save
.ss
= hsave
->save
.ss
;
2821 svm
->vmcb
->save
.ds
= hsave
->save
.ds
;
2822 svm
->vmcb
->save
.gdtr
= hsave
->save
.gdtr
;
2823 svm
->vmcb
->save
.idtr
= hsave
->save
.idtr
;
2824 kvm_set_rflags(&svm
->vcpu
, hsave
->save
.rflags
);
2825 svm_set_efer(&svm
->vcpu
, hsave
->save
.efer
);
2826 svm_set_cr0(&svm
->vcpu
, hsave
->save
.cr0
| X86_CR0_PE
);
2827 svm_set_cr4(&svm
->vcpu
, hsave
->save
.cr4
);
2829 svm
->vmcb
->save
.cr3
= hsave
->save
.cr3
;
2830 svm
->vcpu
.arch
.cr3
= hsave
->save
.cr3
;
2832 (void)kvm_set_cr3(&svm
->vcpu
, hsave
->save
.cr3
);
2834 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
, hsave
->save
.rax
);
2835 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RSP
, hsave
->save
.rsp
);
2836 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RIP
, hsave
->save
.rip
);
2837 svm
->vmcb
->save
.dr7
= 0;
2838 svm
->vmcb
->save
.cpl
= 0;
2839 svm
->vmcb
->control
.exit_int_info
= 0;
2841 mark_all_dirty(svm
->vmcb
);
2843 nested_svm_unmap(page
);
2845 nested_svm_uninit_mmu_context(&svm
->vcpu
);
2846 kvm_mmu_reset_context(&svm
->vcpu
);
2847 kvm_mmu_load(&svm
->vcpu
);
2852 static bool nested_svm_vmrun_msrpm(struct vcpu_svm
*svm
)
2855 * This function merges the msr permission bitmaps of kvm and the
2856 * nested vmcb. It is optimized in that it only merges the parts where
2857 * the kvm msr permission bitmap may contain zero bits
2861 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_MSR_PROT
)))
2864 for (i
= 0; i
< MSRPM_OFFSETS
; i
++) {
2868 if (msrpm_offsets
[i
] == 0xffffffff)
2871 p
= msrpm_offsets
[i
];
2872 offset
= svm
->nested
.vmcb_msrpm
+ (p
* 4);
2874 if (kvm_vcpu_read_guest(&svm
->vcpu
, offset
, &value
, 4))
2877 svm
->nested
.msrpm
[p
] = svm
->msrpm
[p
] | value
;
2880 svm
->vmcb
->control
.msrpm_base_pa
= __pa(svm
->nested
.msrpm
);
2885 static bool nested_vmcb_checks(struct vmcb
*vmcb
)
2887 if ((vmcb
->control
.intercept
& (1ULL << INTERCEPT_VMRUN
)) == 0)
2890 if (vmcb
->control
.asid
== 0)
2893 if (vmcb
->control
.nested_ctl
&& !npt_enabled
)
2899 static bool nested_svm_vmrun(struct vcpu_svm
*svm
)
2901 struct vmcb
*nested_vmcb
;
2902 struct vmcb
*hsave
= svm
->nested
.hsave
;
2903 struct vmcb
*vmcb
= svm
->vmcb
;
2907 vmcb_gpa
= svm
->vmcb
->save
.rax
;
2909 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, &page
);
2913 if (!nested_vmcb_checks(nested_vmcb
)) {
2914 nested_vmcb
->control
.exit_code
= SVM_EXIT_ERR
;
2915 nested_vmcb
->control
.exit_code_hi
= 0;
2916 nested_vmcb
->control
.exit_info_1
= 0;
2917 nested_vmcb
->control
.exit_info_2
= 0;
2919 nested_svm_unmap(page
);
2924 trace_kvm_nested_vmrun(svm
->vmcb
->save
.rip
, vmcb_gpa
,
2925 nested_vmcb
->save
.rip
,
2926 nested_vmcb
->control
.int_ctl
,
2927 nested_vmcb
->control
.event_inj
,
2928 nested_vmcb
->control
.nested_ctl
);
2930 trace_kvm_nested_intercepts(nested_vmcb
->control
.intercept_cr
& 0xffff,
2931 nested_vmcb
->control
.intercept_cr
>> 16,
2932 nested_vmcb
->control
.intercept_exceptions
,
2933 nested_vmcb
->control
.intercept
);
2935 /* Clear internal status */
2936 kvm_clear_exception_queue(&svm
->vcpu
);
2937 kvm_clear_interrupt_queue(&svm
->vcpu
);
2940 * Save the old vmcb, so we don't need to pick what we save, but can
2941 * restore everything when a VMEXIT occurs
2943 hsave
->save
.es
= vmcb
->save
.es
;
2944 hsave
->save
.cs
= vmcb
->save
.cs
;
2945 hsave
->save
.ss
= vmcb
->save
.ss
;
2946 hsave
->save
.ds
= vmcb
->save
.ds
;
2947 hsave
->save
.gdtr
= vmcb
->save
.gdtr
;
2948 hsave
->save
.idtr
= vmcb
->save
.idtr
;
2949 hsave
->save
.efer
= svm
->vcpu
.arch
.efer
;
2950 hsave
->save
.cr0
= kvm_read_cr0(&svm
->vcpu
);
2951 hsave
->save
.cr4
= svm
->vcpu
.arch
.cr4
;
2952 hsave
->save
.rflags
= kvm_get_rflags(&svm
->vcpu
);
2953 hsave
->save
.rip
= kvm_rip_read(&svm
->vcpu
);
2954 hsave
->save
.rsp
= vmcb
->save
.rsp
;
2955 hsave
->save
.rax
= vmcb
->save
.rax
;
2957 hsave
->save
.cr3
= vmcb
->save
.cr3
;
2959 hsave
->save
.cr3
= kvm_read_cr3(&svm
->vcpu
);
2961 copy_vmcb_control_area(hsave
, vmcb
);
2963 if (kvm_get_rflags(&svm
->vcpu
) & X86_EFLAGS_IF
)
2964 svm
->vcpu
.arch
.hflags
|= HF_HIF_MASK
;
2966 svm
->vcpu
.arch
.hflags
&= ~HF_HIF_MASK
;
2968 if (nested_vmcb
->control
.nested_ctl
) {
2969 kvm_mmu_unload(&svm
->vcpu
);
2970 svm
->nested
.nested_cr3
= nested_vmcb
->control
.nested_cr3
;
2971 nested_svm_init_mmu_context(&svm
->vcpu
);
2974 /* Load the nested guest state */
2975 svm
->vmcb
->save
.es
= nested_vmcb
->save
.es
;
2976 svm
->vmcb
->save
.cs
= nested_vmcb
->save
.cs
;
2977 svm
->vmcb
->save
.ss
= nested_vmcb
->save
.ss
;
2978 svm
->vmcb
->save
.ds
= nested_vmcb
->save
.ds
;
2979 svm
->vmcb
->save
.gdtr
= nested_vmcb
->save
.gdtr
;
2980 svm
->vmcb
->save
.idtr
= nested_vmcb
->save
.idtr
;
2981 kvm_set_rflags(&svm
->vcpu
, nested_vmcb
->save
.rflags
);
2982 svm_set_efer(&svm
->vcpu
, nested_vmcb
->save
.efer
);
2983 svm_set_cr0(&svm
->vcpu
, nested_vmcb
->save
.cr0
);
2984 svm_set_cr4(&svm
->vcpu
, nested_vmcb
->save
.cr4
);
2986 svm
->vmcb
->save
.cr3
= nested_vmcb
->save
.cr3
;
2987 svm
->vcpu
.arch
.cr3
= nested_vmcb
->save
.cr3
;
2989 (void)kvm_set_cr3(&svm
->vcpu
, nested_vmcb
->save
.cr3
);
2991 /* Guest paging mode is active - reset mmu */
2992 kvm_mmu_reset_context(&svm
->vcpu
);
2994 svm
->vmcb
->save
.cr2
= svm
->vcpu
.arch
.cr2
= nested_vmcb
->save
.cr2
;
2995 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
, nested_vmcb
->save
.rax
);
2996 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RSP
, nested_vmcb
->save
.rsp
);
2997 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RIP
, nested_vmcb
->save
.rip
);
2999 /* In case we don't even reach vcpu_run, the fields are not updated */
3000 svm
->vmcb
->save
.rax
= nested_vmcb
->save
.rax
;
3001 svm
->vmcb
->save
.rsp
= nested_vmcb
->save
.rsp
;
3002 svm
->vmcb
->save
.rip
= nested_vmcb
->save
.rip
;
3003 svm
->vmcb
->save
.dr7
= nested_vmcb
->save
.dr7
;
3004 svm
->vmcb
->save
.dr6
= nested_vmcb
->save
.dr6
;
3005 svm
->vmcb
->save
.cpl
= nested_vmcb
->save
.cpl
;
3007 svm
->nested
.vmcb_msrpm
= nested_vmcb
->control
.msrpm_base_pa
& ~0x0fffULL
;
3008 svm
->nested
.vmcb_iopm
= nested_vmcb
->control
.iopm_base_pa
& ~0x0fffULL
;
3010 /* cache intercepts */
3011 svm
->nested
.intercept_cr
= nested_vmcb
->control
.intercept_cr
;
3012 svm
->nested
.intercept_dr
= nested_vmcb
->control
.intercept_dr
;
3013 svm
->nested
.intercept_exceptions
= nested_vmcb
->control
.intercept_exceptions
;
3014 svm
->nested
.intercept
= nested_vmcb
->control
.intercept
;
3016 svm_flush_tlb(&svm
->vcpu
);
3017 svm
->vmcb
->control
.int_ctl
= nested_vmcb
->control
.int_ctl
| V_INTR_MASKING_MASK
;
3018 if (nested_vmcb
->control
.int_ctl
& V_INTR_MASKING_MASK
)
3019 svm
->vcpu
.arch
.hflags
|= HF_VINTR_MASK
;
3021 svm
->vcpu
.arch
.hflags
&= ~HF_VINTR_MASK
;
3023 if (svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
) {
3024 /* We only want the cr8 intercept bits of the guest */
3025 clr_cr_intercept(svm
, INTERCEPT_CR8_READ
);
3026 clr_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
3029 /* We don't want to see VMMCALLs from a nested guest */
3030 clr_intercept(svm
, INTERCEPT_VMMCALL
);
3032 svm
->vmcb
->control
.virt_ext
= nested_vmcb
->control
.virt_ext
;
3033 svm
->vmcb
->control
.int_vector
= nested_vmcb
->control
.int_vector
;
3034 svm
->vmcb
->control
.int_state
= nested_vmcb
->control
.int_state
;
3035 svm
->vmcb
->control
.tsc_offset
+= nested_vmcb
->control
.tsc_offset
;
3036 svm
->vmcb
->control
.event_inj
= nested_vmcb
->control
.event_inj
;
3037 svm
->vmcb
->control
.event_inj_err
= nested_vmcb
->control
.event_inj_err
;
3039 nested_svm_unmap(page
);
3041 /* Enter Guest-Mode */
3042 enter_guest_mode(&svm
->vcpu
);
3045 * Merge guest and host intercepts - must be called with vcpu in
3046 * guest-mode to take affect here
3048 recalc_intercepts(svm
);
3050 svm
->nested
.vmcb
= vmcb_gpa
;
3054 mark_all_dirty(svm
->vmcb
);
3059 static void nested_svm_vmloadsave(struct vmcb
*from_vmcb
, struct vmcb
*to_vmcb
)
3061 to_vmcb
->save
.fs
= from_vmcb
->save
.fs
;
3062 to_vmcb
->save
.gs
= from_vmcb
->save
.gs
;
3063 to_vmcb
->save
.tr
= from_vmcb
->save
.tr
;
3064 to_vmcb
->save
.ldtr
= from_vmcb
->save
.ldtr
;
3065 to_vmcb
->save
.kernel_gs_base
= from_vmcb
->save
.kernel_gs_base
;
3066 to_vmcb
->save
.star
= from_vmcb
->save
.star
;
3067 to_vmcb
->save
.lstar
= from_vmcb
->save
.lstar
;
3068 to_vmcb
->save
.cstar
= from_vmcb
->save
.cstar
;
3069 to_vmcb
->save
.sfmask
= from_vmcb
->save
.sfmask
;
3070 to_vmcb
->save
.sysenter_cs
= from_vmcb
->save
.sysenter_cs
;
3071 to_vmcb
->save
.sysenter_esp
= from_vmcb
->save
.sysenter_esp
;
3072 to_vmcb
->save
.sysenter_eip
= from_vmcb
->save
.sysenter_eip
;
3075 static int vmload_interception(struct vcpu_svm
*svm
)
3077 struct vmcb
*nested_vmcb
;
3081 if (nested_svm_check_permissions(svm
))
3084 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, &page
);
3088 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
3089 ret
= kvm_skip_emulated_instruction(&svm
->vcpu
);
3091 nested_svm_vmloadsave(nested_vmcb
, svm
->vmcb
);
3092 nested_svm_unmap(page
);
3097 static int vmsave_interception(struct vcpu_svm
*svm
)
3099 struct vmcb
*nested_vmcb
;
3103 if (nested_svm_check_permissions(svm
))
3106 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, &page
);
3110 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
3111 ret
= kvm_skip_emulated_instruction(&svm
->vcpu
);
3113 nested_svm_vmloadsave(svm
->vmcb
, nested_vmcb
);
3114 nested_svm_unmap(page
);
3119 static int vmrun_interception(struct vcpu_svm
*svm
)
3121 if (nested_svm_check_permissions(svm
))
3124 /* Save rip after vmrun instruction */
3125 kvm_rip_write(&svm
->vcpu
, kvm_rip_read(&svm
->vcpu
) + 3);
3127 if (!nested_svm_vmrun(svm
))
3130 if (!nested_svm_vmrun_msrpm(svm
))
3137 svm
->vmcb
->control
.exit_code
= SVM_EXIT_ERR
;
3138 svm
->vmcb
->control
.exit_code_hi
= 0;
3139 svm
->vmcb
->control
.exit_info_1
= 0;
3140 svm
->vmcb
->control
.exit_info_2
= 0;
3142 nested_svm_vmexit(svm
);
3147 static int stgi_interception(struct vcpu_svm
*svm
)
3151 if (nested_svm_check_permissions(svm
))
3154 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
3155 ret
= kvm_skip_emulated_instruction(&svm
->vcpu
);
3156 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
3163 static int clgi_interception(struct vcpu_svm
*svm
)
3167 if (nested_svm_check_permissions(svm
))
3170 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
3171 ret
= kvm_skip_emulated_instruction(&svm
->vcpu
);
3175 /* After a CLGI no interrupts should come */
3176 if (!kvm_vcpu_apicv_active(&svm
->vcpu
)) {
3177 svm_clear_vintr(svm
);
3178 svm
->vmcb
->control
.int_ctl
&= ~V_IRQ_MASK
;
3179 mark_dirty(svm
->vmcb
, VMCB_INTR
);
3185 static int invlpga_interception(struct vcpu_svm
*svm
)
3187 struct kvm_vcpu
*vcpu
= &svm
->vcpu
;
3189 trace_kvm_invlpga(svm
->vmcb
->save
.rip
, kvm_register_read(&svm
->vcpu
, VCPU_REGS_RCX
),
3190 kvm_register_read(&svm
->vcpu
, VCPU_REGS_RAX
));
3192 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
3193 kvm_mmu_invlpg(vcpu
, kvm_register_read(&svm
->vcpu
, VCPU_REGS_RAX
));
3195 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
3196 return kvm_skip_emulated_instruction(&svm
->vcpu
);
3199 static int skinit_interception(struct vcpu_svm
*svm
)
3201 trace_kvm_skinit(svm
->vmcb
->save
.rip
, kvm_register_read(&svm
->vcpu
, VCPU_REGS_RAX
));
3203 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
3207 static int wbinvd_interception(struct vcpu_svm
*svm
)
3209 return kvm_emulate_wbinvd(&svm
->vcpu
);
3212 static int xsetbv_interception(struct vcpu_svm
*svm
)
3214 u64 new_bv
= kvm_read_edx_eax(&svm
->vcpu
);
3215 u32 index
= kvm_register_read(&svm
->vcpu
, VCPU_REGS_RCX
);
3217 if (kvm_set_xcr(&svm
->vcpu
, index
, new_bv
) == 0) {
3218 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
3219 return kvm_skip_emulated_instruction(&svm
->vcpu
);
3225 static int task_switch_interception(struct vcpu_svm
*svm
)
3229 int int_type
= svm
->vmcb
->control
.exit_int_info
&
3230 SVM_EXITINTINFO_TYPE_MASK
;
3231 int int_vec
= svm
->vmcb
->control
.exit_int_info
& SVM_EVTINJ_VEC_MASK
;
3233 svm
->vmcb
->control
.exit_int_info
& SVM_EXITINTINFO_TYPE_MASK
;
3235 svm
->vmcb
->control
.exit_int_info
& SVM_EXITINTINFO_VALID
;
3236 bool has_error_code
= false;
3239 tss_selector
= (u16
)svm
->vmcb
->control
.exit_info_1
;
3241 if (svm
->vmcb
->control
.exit_info_2
&
3242 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET
))
3243 reason
= TASK_SWITCH_IRET
;
3244 else if (svm
->vmcb
->control
.exit_info_2
&
3245 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP
))
3246 reason
= TASK_SWITCH_JMP
;
3248 reason
= TASK_SWITCH_GATE
;
3250 reason
= TASK_SWITCH_CALL
;
3252 if (reason
== TASK_SWITCH_GATE
) {
3254 case SVM_EXITINTINFO_TYPE_NMI
:
3255 svm
->vcpu
.arch
.nmi_injected
= false;
3257 case SVM_EXITINTINFO_TYPE_EXEPT
:
3258 if (svm
->vmcb
->control
.exit_info_2
&
3259 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE
)) {
3260 has_error_code
= true;
3262 (u32
)svm
->vmcb
->control
.exit_info_2
;
3264 kvm_clear_exception_queue(&svm
->vcpu
);
3266 case SVM_EXITINTINFO_TYPE_INTR
:
3267 kvm_clear_interrupt_queue(&svm
->vcpu
);
3274 if (reason
!= TASK_SWITCH_GATE
||
3275 int_type
== SVM_EXITINTINFO_TYPE_SOFT
||
3276 (int_type
== SVM_EXITINTINFO_TYPE_EXEPT
&&
3277 (int_vec
== OF_VECTOR
|| int_vec
== BP_VECTOR
)))
3278 skip_emulated_instruction(&svm
->vcpu
);
3280 if (int_type
!= SVM_EXITINTINFO_TYPE_SOFT
)
3283 if (kvm_task_switch(&svm
->vcpu
, tss_selector
, int_vec
, reason
,
3284 has_error_code
, error_code
) == EMULATE_FAIL
) {
3285 svm
->vcpu
.run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
3286 svm
->vcpu
.run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
3287 svm
->vcpu
.run
->internal
.ndata
= 0;
3293 static int cpuid_interception(struct vcpu_svm
*svm
)
3295 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
3296 return kvm_emulate_cpuid(&svm
->vcpu
);
3299 static int iret_interception(struct vcpu_svm
*svm
)
3301 ++svm
->vcpu
.stat
.nmi_window_exits
;
3302 clr_intercept(svm
, INTERCEPT_IRET
);
3303 svm
->vcpu
.arch
.hflags
|= HF_IRET_MASK
;
3304 svm
->nmi_iret_rip
= kvm_rip_read(&svm
->vcpu
);
3305 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
3309 static int invlpg_interception(struct vcpu_svm
*svm
)
3311 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS
))
3312 return emulate_instruction(&svm
->vcpu
, 0) == EMULATE_DONE
;
3314 kvm_mmu_invlpg(&svm
->vcpu
, svm
->vmcb
->control
.exit_info_1
);
3315 return kvm_skip_emulated_instruction(&svm
->vcpu
);
3318 static int emulate_on_interception(struct vcpu_svm
*svm
)
3320 return emulate_instruction(&svm
->vcpu
, 0) == EMULATE_DONE
;
3323 static int rdpmc_interception(struct vcpu_svm
*svm
)
3327 if (!static_cpu_has(X86_FEATURE_NRIPS
))
3328 return emulate_on_interception(svm
);
3330 err
= kvm_rdpmc(&svm
->vcpu
);
3331 return kvm_complete_insn_gp(&svm
->vcpu
, err
);
3334 static bool check_selective_cr0_intercepted(struct vcpu_svm
*svm
,
3337 unsigned long cr0
= svm
->vcpu
.arch
.cr0
;
3341 intercept
= svm
->nested
.intercept
;
3343 if (!is_guest_mode(&svm
->vcpu
) ||
3344 (!(intercept
& (1ULL << INTERCEPT_SELECTIVE_CR0
))))
3347 cr0
&= ~SVM_CR0_SELECTIVE_MASK
;
3348 val
&= ~SVM_CR0_SELECTIVE_MASK
;
3351 svm
->vmcb
->control
.exit_code
= SVM_EXIT_CR0_SEL_WRITE
;
3352 ret
= (nested_svm_exit_handled(svm
) == NESTED_EXIT_DONE
);
3358 #define CR_VALID (1ULL << 63)
3360 static int cr_interception(struct vcpu_svm
*svm
)
3366 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS
))
3367 return emulate_on_interception(svm
);
3369 if (unlikely((svm
->vmcb
->control
.exit_info_1
& CR_VALID
) == 0))
3370 return emulate_on_interception(svm
);
3372 reg
= svm
->vmcb
->control
.exit_info_1
& SVM_EXITINFO_REG_MASK
;
3373 if (svm
->vmcb
->control
.exit_code
== SVM_EXIT_CR0_SEL_WRITE
)
3374 cr
= SVM_EXIT_WRITE_CR0
- SVM_EXIT_READ_CR0
;
3376 cr
= svm
->vmcb
->control
.exit_code
- SVM_EXIT_READ_CR0
;
3379 if (cr
>= 16) { /* mov to cr */
3381 val
= kvm_register_read(&svm
->vcpu
, reg
);
3384 if (!check_selective_cr0_intercepted(svm
, val
))
3385 err
= kvm_set_cr0(&svm
->vcpu
, val
);
3391 err
= kvm_set_cr3(&svm
->vcpu
, val
);
3394 err
= kvm_set_cr4(&svm
->vcpu
, val
);
3397 err
= kvm_set_cr8(&svm
->vcpu
, val
);
3400 WARN(1, "unhandled write to CR%d", cr
);
3401 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
3404 } else { /* mov from cr */
3407 val
= kvm_read_cr0(&svm
->vcpu
);
3410 val
= svm
->vcpu
.arch
.cr2
;
3413 val
= kvm_read_cr3(&svm
->vcpu
);
3416 val
= kvm_read_cr4(&svm
->vcpu
);
3419 val
= kvm_get_cr8(&svm
->vcpu
);
3422 WARN(1, "unhandled read from CR%d", cr
);
3423 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
3426 kvm_register_write(&svm
->vcpu
, reg
, val
);
3428 return kvm_complete_insn_gp(&svm
->vcpu
, err
);
3431 static int dr_interception(struct vcpu_svm
*svm
)
3436 if (svm
->vcpu
.guest_debug
== 0) {
3438 * No more DR vmexits; force a reload of the debug registers
3439 * and reenter on this instruction. The next vmexit will
3440 * retrieve the full state of the debug registers.
3442 clr_dr_intercepts(svm
);
3443 svm
->vcpu
.arch
.switch_db_regs
|= KVM_DEBUGREG_WONT_EXIT
;
3447 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS
))
3448 return emulate_on_interception(svm
);
3450 reg
= svm
->vmcb
->control
.exit_info_1
& SVM_EXITINFO_REG_MASK
;
3451 dr
= svm
->vmcb
->control
.exit_code
- SVM_EXIT_READ_DR0
;
3453 if (dr
>= 16) { /* mov to DRn */
3454 if (!kvm_require_dr(&svm
->vcpu
, dr
- 16))
3456 val
= kvm_register_read(&svm
->vcpu
, reg
);
3457 kvm_set_dr(&svm
->vcpu
, dr
- 16, val
);
3459 if (!kvm_require_dr(&svm
->vcpu
, dr
))
3461 kvm_get_dr(&svm
->vcpu
, dr
, &val
);
3462 kvm_register_write(&svm
->vcpu
, reg
, val
);
3465 return kvm_skip_emulated_instruction(&svm
->vcpu
);
3468 static int cr8_write_interception(struct vcpu_svm
*svm
)
3470 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
3473 u8 cr8_prev
= kvm_get_cr8(&svm
->vcpu
);
3474 /* instruction emulation calls kvm_set_cr8() */
3475 r
= cr_interception(svm
);
3476 if (lapic_in_kernel(&svm
->vcpu
))
3478 if (cr8_prev
<= kvm_get_cr8(&svm
->vcpu
))
3480 kvm_run
->exit_reason
= KVM_EXIT_SET_TPR
;
3484 static int svm_get_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
3486 struct vcpu_svm
*svm
= to_svm(vcpu
);
3488 switch (msr_info
->index
) {
3489 case MSR_IA32_TSC
: {
3490 msr_info
->data
= svm
->vmcb
->control
.tsc_offset
+
3491 kvm_scale_tsc(vcpu
, rdtsc());
3496 msr_info
->data
= svm
->vmcb
->save
.star
;
3498 #ifdef CONFIG_X86_64
3500 msr_info
->data
= svm
->vmcb
->save
.lstar
;
3503 msr_info
->data
= svm
->vmcb
->save
.cstar
;
3505 case MSR_KERNEL_GS_BASE
:
3506 msr_info
->data
= svm
->vmcb
->save
.kernel_gs_base
;
3508 case MSR_SYSCALL_MASK
:
3509 msr_info
->data
= svm
->vmcb
->save
.sfmask
;
3512 case MSR_IA32_SYSENTER_CS
:
3513 msr_info
->data
= svm
->vmcb
->save
.sysenter_cs
;
3515 case MSR_IA32_SYSENTER_EIP
:
3516 msr_info
->data
= svm
->sysenter_eip
;
3518 case MSR_IA32_SYSENTER_ESP
:
3519 msr_info
->data
= svm
->sysenter_esp
;
3522 if (!boot_cpu_has(X86_FEATURE_RDTSCP
))
3524 msr_info
->data
= svm
->tsc_aux
;
3527 * Nobody will change the following 5 values in the VMCB so we can
3528 * safely return them on rdmsr. They will always be 0 until LBRV is
3531 case MSR_IA32_DEBUGCTLMSR
:
3532 msr_info
->data
= svm
->vmcb
->save
.dbgctl
;
3534 case MSR_IA32_LASTBRANCHFROMIP
:
3535 msr_info
->data
= svm
->vmcb
->save
.br_from
;
3537 case MSR_IA32_LASTBRANCHTOIP
:
3538 msr_info
->data
= svm
->vmcb
->save
.br_to
;
3540 case MSR_IA32_LASTINTFROMIP
:
3541 msr_info
->data
= svm
->vmcb
->save
.last_excp_from
;
3543 case MSR_IA32_LASTINTTOIP
:
3544 msr_info
->data
= svm
->vmcb
->save
.last_excp_to
;
3546 case MSR_VM_HSAVE_PA
:
3547 msr_info
->data
= svm
->nested
.hsave_msr
;
3550 msr_info
->data
= svm
->nested
.vm_cr_msr
;
3552 case MSR_IA32_SPEC_CTRL
:
3553 msr_info
->data
= svm
->spec_ctrl
;
3555 case MSR_IA32_UCODE_REV
:
3556 msr_info
->data
= 0x01000065;
3558 case MSR_F15H_IC_CFG
: {
3562 family
= guest_cpuid_family(vcpu
);
3563 model
= guest_cpuid_model(vcpu
);
3565 if (family
< 0 || model
< 0)
3566 return kvm_get_msr_common(vcpu
, msr_info
);
3570 if (family
== 0x15 &&
3571 (model
>= 0x2 && model
< 0x20))
3572 msr_info
->data
= 0x1E;
3576 return kvm_get_msr_common(vcpu
, msr_info
);
3581 static int rdmsr_interception(struct vcpu_svm
*svm
)
3583 u32 ecx
= kvm_register_read(&svm
->vcpu
, VCPU_REGS_RCX
);
3584 struct msr_data msr_info
;
3586 msr_info
.index
= ecx
;
3587 msr_info
.host_initiated
= false;
3588 if (svm_get_msr(&svm
->vcpu
, &msr_info
)) {
3589 trace_kvm_msr_read_ex(ecx
);
3590 kvm_inject_gp(&svm
->vcpu
, 0);
3593 trace_kvm_msr_read(ecx
, msr_info
.data
);
3595 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
,
3596 msr_info
.data
& 0xffffffff);
3597 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RDX
,
3598 msr_info
.data
>> 32);
3599 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
3600 return kvm_skip_emulated_instruction(&svm
->vcpu
);
3604 static int svm_set_vm_cr(struct kvm_vcpu
*vcpu
, u64 data
)
3606 struct vcpu_svm
*svm
= to_svm(vcpu
);
3607 int svm_dis
, chg_mask
;
3609 if (data
& ~SVM_VM_CR_VALID_MASK
)
3612 chg_mask
= SVM_VM_CR_VALID_MASK
;
3614 if (svm
->nested
.vm_cr_msr
& SVM_VM_CR_SVM_DIS_MASK
)
3615 chg_mask
&= ~(SVM_VM_CR_SVM_LOCK_MASK
| SVM_VM_CR_SVM_DIS_MASK
);
3617 svm
->nested
.vm_cr_msr
&= ~chg_mask
;
3618 svm
->nested
.vm_cr_msr
|= (data
& chg_mask
);
3620 svm_dis
= svm
->nested
.vm_cr_msr
& SVM_VM_CR_SVM_DIS_MASK
;
3622 /* check for svm_disable while efer.svme is set */
3623 if (svm_dis
&& (vcpu
->arch
.efer
& EFER_SVME
))
3629 static int svm_set_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
3631 struct vcpu_svm
*svm
= to_svm(vcpu
);
3633 u32 ecx
= msr
->index
;
3634 u64 data
= msr
->data
;
3637 kvm_write_tsc(vcpu
, msr
);
3640 svm
->vmcb
->save
.star
= data
;
3642 #ifdef CONFIG_X86_64
3644 svm
->vmcb
->save
.lstar
= data
;
3647 svm
->vmcb
->save
.cstar
= data
;
3649 case MSR_KERNEL_GS_BASE
:
3650 svm
->vmcb
->save
.kernel_gs_base
= data
;
3652 case MSR_SYSCALL_MASK
:
3653 svm
->vmcb
->save
.sfmask
= data
;
3656 case MSR_IA32_SYSENTER_CS
:
3657 svm
->vmcb
->save
.sysenter_cs
= data
;
3659 case MSR_IA32_SYSENTER_EIP
:
3660 svm
->sysenter_eip
= data
;
3661 svm
->vmcb
->save
.sysenter_eip
= data
;
3663 case MSR_IA32_SYSENTER_ESP
:
3664 svm
->sysenter_esp
= data
;
3665 svm
->vmcb
->save
.sysenter_esp
= data
;
3668 if (!boot_cpu_has(X86_FEATURE_RDTSCP
))
3672 * This is rare, so we update the MSR here instead of using
3673 * direct_access_msrs. Doing that would require a rdmsr in
3676 svm
->tsc_aux
= data
;
3677 wrmsrl(MSR_TSC_AUX
, svm
->tsc_aux
);
3679 case MSR_IA32_DEBUGCTLMSR
:
3680 if (!boot_cpu_has(X86_FEATURE_LBRV
)) {
3681 vcpu_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3685 if (data
& DEBUGCTL_RESERVED_BITS
)
3688 svm
->vmcb
->save
.dbgctl
= data
;
3689 mark_dirty(svm
->vmcb
, VMCB_LBR
);
3690 if (data
& (1ULL<<0))
3691 svm_enable_lbrv(svm
);
3693 svm_disable_lbrv(svm
);
3695 case MSR_VM_HSAVE_PA
:
3696 svm
->nested
.hsave_msr
= data
;
3699 return svm_set_vm_cr(vcpu
, data
);
3701 vcpu_unimpl(vcpu
, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx
, data
);
3703 case MSR_IA32_SPEC_CTRL
:
3704 svm
->spec_ctrl
= data
;
3706 case MSR_IA32_APICBASE
:
3707 if (kvm_vcpu_apicv_active(vcpu
))
3708 avic_update_vapic_bar(to_svm(vcpu
), data
);
3709 /* Follow through */
3711 return kvm_set_msr_common(vcpu
, msr
);
3716 static int wrmsr_interception(struct vcpu_svm
*svm
)
3718 struct msr_data msr
;
3719 u32 ecx
= kvm_register_read(&svm
->vcpu
, VCPU_REGS_RCX
);
3720 u64 data
= kvm_read_edx_eax(&svm
->vcpu
);
3724 msr
.host_initiated
= false;
3726 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
3727 if (kvm_set_msr(&svm
->vcpu
, &msr
)) {
3728 trace_kvm_msr_write_ex(ecx
, data
);
3729 kvm_inject_gp(&svm
->vcpu
, 0);
3732 trace_kvm_msr_write(ecx
, data
);
3733 return kvm_skip_emulated_instruction(&svm
->vcpu
);
3737 static int msr_interception(struct vcpu_svm
*svm
)
3739 if (svm
->vmcb
->control
.exit_info_1
)
3740 return wrmsr_interception(svm
);
3742 return rdmsr_interception(svm
);
3745 static int interrupt_window_interception(struct vcpu_svm
*svm
)
3747 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
3748 svm_clear_vintr(svm
);
3749 svm
->vmcb
->control
.int_ctl
&= ~V_IRQ_MASK
;
3750 mark_dirty(svm
->vmcb
, VMCB_INTR
);
3751 ++svm
->vcpu
.stat
.irq_window_exits
;
3755 static int pause_interception(struct vcpu_svm
*svm
)
3757 kvm_vcpu_on_spin(&(svm
->vcpu
));
3761 static int nop_interception(struct vcpu_svm
*svm
)
3763 return kvm_skip_emulated_instruction(&(svm
->vcpu
));
3766 static int monitor_interception(struct vcpu_svm
*svm
)
3768 printk_once(KERN_WARNING
"kvm: MONITOR instruction emulated as NOP!\n");
3769 return nop_interception(svm
);
3772 static int mwait_interception(struct vcpu_svm
*svm
)
3774 printk_once(KERN_WARNING
"kvm: MWAIT instruction emulated as NOP!\n");
3775 return nop_interception(svm
);
3778 enum avic_ipi_failure_cause
{
3779 AVIC_IPI_FAILURE_INVALID_INT_TYPE
,
3780 AVIC_IPI_FAILURE_TARGET_NOT_RUNNING
,
3781 AVIC_IPI_FAILURE_INVALID_TARGET
,
3782 AVIC_IPI_FAILURE_INVALID_BACKING_PAGE
,
3785 static int avic_incomplete_ipi_interception(struct vcpu_svm
*svm
)
3787 u32 icrh
= svm
->vmcb
->control
.exit_info_1
>> 32;
3788 u32 icrl
= svm
->vmcb
->control
.exit_info_1
;
3789 u32 id
= svm
->vmcb
->control
.exit_info_2
>> 32;
3790 u32 index
= svm
->vmcb
->control
.exit_info_2
& 0xFF;
3791 struct kvm_lapic
*apic
= svm
->vcpu
.arch
.apic
;
3793 trace_kvm_avic_incomplete_ipi(svm
->vcpu
.vcpu_id
, icrh
, icrl
, id
, index
);
3796 case AVIC_IPI_FAILURE_INVALID_INT_TYPE
:
3798 * AVIC hardware handles the generation of
3799 * IPIs when the specified Message Type is Fixed
3800 * (also known as fixed delivery mode) and
3801 * the Trigger Mode is edge-triggered. The hardware
3802 * also supports self and broadcast delivery modes
3803 * specified via the Destination Shorthand(DSH)
3804 * field of the ICRL. Logical and physical APIC ID
3805 * formats are supported. All other IPI types cause
3806 * a #VMEXIT, which needs to emulated.
3808 kvm_lapic_reg_write(apic
, APIC_ICR2
, icrh
);
3809 kvm_lapic_reg_write(apic
, APIC_ICR
, icrl
);
3811 case AVIC_IPI_FAILURE_TARGET_NOT_RUNNING
: {
3813 struct kvm_vcpu
*vcpu
;
3814 struct kvm
*kvm
= svm
->vcpu
.kvm
;
3815 struct kvm_lapic
*apic
= svm
->vcpu
.arch
.apic
;
3818 * At this point, we expect that the AVIC HW has already
3819 * set the appropriate IRR bits on the valid target
3820 * vcpus. So, we just need to kick the appropriate vcpu.
3822 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
3823 bool m
= kvm_apic_match_dest(vcpu
, apic
,
3824 icrl
& KVM_APIC_SHORT_MASK
,
3825 GET_APIC_DEST_FIELD(icrh
),
3826 icrl
& KVM_APIC_DEST_MASK
);
3828 if (m
&& !avic_vcpu_is_running(vcpu
))
3829 kvm_vcpu_wake_up(vcpu
);
3833 case AVIC_IPI_FAILURE_INVALID_TARGET
:
3835 case AVIC_IPI_FAILURE_INVALID_BACKING_PAGE
:
3836 WARN_ONCE(1, "Invalid backing page\n");
3839 pr_err("Unknown IPI interception\n");
3845 static u32
*avic_get_logical_id_entry(struct kvm_vcpu
*vcpu
, u32 ldr
, bool flat
)
3847 struct kvm_arch
*vm_data
= &vcpu
->kvm
->arch
;
3849 u32
*logical_apic_id_table
;
3850 int dlid
= GET_APIC_LOGICAL_ID(ldr
);
3855 if (flat
) { /* flat */
3856 index
= ffs(dlid
) - 1;
3859 } else { /* cluster */
3860 int cluster
= (dlid
& 0xf0) >> 4;
3861 int apic
= ffs(dlid
& 0x0f) - 1;
3863 if ((apic
< 0) || (apic
> 7) ||
3866 index
= (cluster
<< 2) + apic
;
3869 logical_apic_id_table
= (u32
*) page_address(vm_data
->avic_logical_id_table_page
);
3871 return &logical_apic_id_table
[index
];
3874 static int avic_ldr_write(struct kvm_vcpu
*vcpu
, u8 g_physical_id
, u32 ldr
,
3878 u32
*entry
, new_entry
;
3880 flat
= kvm_lapic_get_reg(vcpu
->arch
.apic
, APIC_DFR
) == APIC_DFR_FLAT
;
3881 entry
= avic_get_logical_id_entry(vcpu
, ldr
, flat
);
3885 new_entry
= READ_ONCE(*entry
);
3886 new_entry
&= ~AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK
;
3887 new_entry
|= (g_physical_id
& AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK
);
3889 new_entry
|= AVIC_LOGICAL_ID_ENTRY_VALID_MASK
;
3891 new_entry
&= ~AVIC_LOGICAL_ID_ENTRY_VALID_MASK
;
3892 WRITE_ONCE(*entry
, new_entry
);
3897 static int avic_handle_ldr_update(struct kvm_vcpu
*vcpu
)
3900 struct vcpu_svm
*svm
= to_svm(vcpu
);
3901 u32 ldr
= kvm_lapic_get_reg(vcpu
->arch
.apic
, APIC_LDR
);
3906 ret
= avic_ldr_write(vcpu
, vcpu
->vcpu_id
, ldr
, true);
3907 if (ret
&& svm
->ldr_reg
) {
3908 avic_ldr_write(vcpu
, 0, svm
->ldr_reg
, false);
3916 static int avic_handle_apic_id_update(struct kvm_vcpu
*vcpu
)
3919 struct vcpu_svm
*svm
= to_svm(vcpu
);
3920 u32 apic_id_reg
= kvm_lapic_get_reg(vcpu
->arch
.apic
, APIC_ID
);
3921 u32 id
= (apic_id_reg
>> 24) & 0xff;
3923 if (vcpu
->vcpu_id
== id
)
3926 old
= avic_get_physical_id_entry(vcpu
, vcpu
->vcpu_id
);
3927 new = avic_get_physical_id_entry(vcpu
, id
);
3931 /* We need to move physical_id_entry to new offset */
3934 to_svm(vcpu
)->avic_physical_id_cache
= new;
3937 * Also update the guest physical APIC ID in the logical
3938 * APIC ID table entry if already setup the LDR.
3941 avic_handle_ldr_update(vcpu
);
3946 static int avic_handle_dfr_update(struct kvm_vcpu
*vcpu
)
3948 struct vcpu_svm
*svm
= to_svm(vcpu
);
3949 struct kvm_arch
*vm_data
= &vcpu
->kvm
->arch
;
3950 u32 dfr
= kvm_lapic_get_reg(vcpu
->arch
.apic
, APIC_DFR
);
3951 u32 mod
= (dfr
>> 28) & 0xf;
3954 * We assume that all local APICs are using the same type.
3955 * If this changes, we need to flush the AVIC logical
3958 if (vm_data
->ldr_mode
== mod
)
3961 clear_page(page_address(vm_data
->avic_logical_id_table_page
));
3962 vm_data
->ldr_mode
= mod
;
3965 avic_handle_ldr_update(vcpu
);
3969 static int avic_unaccel_trap_write(struct vcpu_svm
*svm
)
3971 struct kvm_lapic
*apic
= svm
->vcpu
.arch
.apic
;
3972 u32 offset
= svm
->vmcb
->control
.exit_info_1
&
3973 AVIC_UNACCEL_ACCESS_OFFSET_MASK
;
3977 if (avic_handle_apic_id_update(&svm
->vcpu
))
3981 if (avic_handle_ldr_update(&svm
->vcpu
))
3985 avic_handle_dfr_update(&svm
->vcpu
);
3991 kvm_lapic_reg_write(apic
, offset
, kvm_lapic_get_reg(apic
, offset
));
3996 static bool is_avic_unaccelerated_access_trap(u32 offset
)
4025 static int avic_unaccelerated_access_interception(struct vcpu_svm
*svm
)
4028 u32 offset
= svm
->vmcb
->control
.exit_info_1
&
4029 AVIC_UNACCEL_ACCESS_OFFSET_MASK
;
4030 u32 vector
= svm
->vmcb
->control
.exit_info_2
&
4031 AVIC_UNACCEL_ACCESS_VECTOR_MASK
;
4032 bool write
= (svm
->vmcb
->control
.exit_info_1
>> 32) &
4033 AVIC_UNACCEL_ACCESS_WRITE_MASK
;
4034 bool trap
= is_avic_unaccelerated_access_trap(offset
);
4036 trace_kvm_avic_unaccelerated_access(svm
->vcpu
.vcpu_id
, offset
,
4037 trap
, write
, vector
);
4040 WARN_ONCE(!write
, "svm: Handling trap read.\n");
4041 ret
= avic_unaccel_trap_write(svm
);
4043 /* Handling Fault */
4044 ret
= (emulate_instruction(&svm
->vcpu
, 0) == EMULATE_DONE
);
4050 static int (*const svm_exit_handlers
[])(struct vcpu_svm
*svm
) = {
4051 [SVM_EXIT_READ_CR0
] = cr_interception
,
4052 [SVM_EXIT_READ_CR3
] = cr_interception
,
4053 [SVM_EXIT_READ_CR4
] = cr_interception
,
4054 [SVM_EXIT_READ_CR8
] = cr_interception
,
4055 [SVM_EXIT_CR0_SEL_WRITE
] = cr_interception
,
4056 [SVM_EXIT_WRITE_CR0
] = cr_interception
,
4057 [SVM_EXIT_WRITE_CR3
] = cr_interception
,
4058 [SVM_EXIT_WRITE_CR4
] = cr_interception
,
4059 [SVM_EXIT_WRITE_CR8
] = cr8_write_interception
,
4060 [SVM_EXIT_READ_DR0
] = dr_interception
,
4061 [SVM_EXIT_READ_DR1
] = dr_interception
,
4062 [SVM_EXIT_READ_DR2
] = dr_interception
,
4063 [SVM_EXIT_READ_DR3
] = dr_interception
,
4064 [SVM_EXIT_READ_DR4
] = dr_interception
,
4065 [SVM_EXIT_READ_DR5
] = dr_interception
,
4066 [SVM_EXIT_READ_DR6
] = dr_interception
,
4067 [SVM_EXIT_READ_DR7
] = dr_interception
,
4068 [SVM_EXIT_WRITE_DR0
] = dr_interception
,
4069 [SVM_EXIT_WRITE_DR1
] = dr_interception
,
4070 [SVM_EXIT_WRITE_DR2
] = dr_interception
,
4071 [SVM_EXIT_WRITE_DR3
] = dr_interception
,
4072 [SVM_EXIT_WRITE_DR4
] = dr_interception
,
4073 [SVM_EXIT_WRITE_DR5
] = dr_interception
,
4074 [SVM_EXIT_WRITE_DR6
] = dr_interception
,
4075 [SVM_EXIT_WRITE_DR7
] = dr_interception
,
4076 [SVM_EXIT_EXCP_BASE
+ DB_VECTOR
] = db_interception
,
4077 [SVM_EXIT_EXCP_BASE
+ BP_VECTOR
] = bp_interception
,
4078 [SVM_EXIT_EXCP_BASE
+ UD_VECTOR
] = ud_interception
,
4079 [SVM_EXIT_EXCP_BASE
+ PF_VECTOR
] = pf_interception
,
4080 [SVM_EXIT_EXCP_BASE
+ MC_VECTOR
] = mc_interception
,
4081 [SVM_EXIT_EXCP_BASE
+ AC_VECTOR
] = ac_interception
,
4082 [SVM_EXIT_INTR
] = intr_interception
,
4083 [SVM_EXIT_NMI
] = nmi_interception
,
4084 [SVM_EXIT_SMI
] = nop_on_interception
,
4085 [SVM_EXIT_INIT
] = nop_on_interception
,
4086 [SVM_EXIT_VINTR
] = interrupt_window_interception
,
4087 [SVM_EXIT_RDPMC
] = rdpmc_interception
,
4088 [SVM_EXIT_CPUID
] = cpuid_interception
,
4089 [SVM_EXIT_IRET
] = iret_interception
,
4090 [SVM_EXIT_INVD
] = emulate_on_interception
,
4091 [SVM_EXIT_PAUSE
] = pause_interception
,
4092 [SVM_EXIT_HLT
] = halt_interception
,
4093 [SVM_EXIT_INVLPG
] = invlpg_interception
,
4094 [SVM_EXIT_INVLPGA
] = invlpga_interception
,
4095 [SVM_EXIT_IOIO
] = io_interception
,
4096 [SVM_EXIT_MSR
] = msr_interception
,
4097 [SVM_EXIT_TASK_SWITCH
] = task_switch_interception
,
4098 [SVM_EXIT_SHUTDOWN
] = shutdown_interception
,
4099 [SVM_EXIT_VMRUN
] = vmrun_interception
,
4100 [SVM_EXIT_VMMCALL
] = vmmcall_interception
,
4101 [SVM_EXIT_VMLOAD
] = vmload_interception
,
4102 [SVM_EXIT_VMSAVE
] = vmsave_interception
,
4103 [SVM_EXIT_STGI
] = stgi_interception
,
4104 [SVM_EXIT_CLGI
] = clgi_interception
,
4105 [SVM_EXIT_SKINIT
] = skinit_interception
,
4106 [SVM_EXIT_WBINVD
] = wbinvd_interception
,
4107 [SVM_EXIT_MONITOR
] = monitor_interception
,
4108 [SVM_EXIT_MWAIT
] = mwait_interception
,
4109 [SVM_EXIT_XSETBV
] = xsetbv_interception
,
4110 [SVM_EXIT_NPF
] = pf_interception
,
4111 [SVM_EXIT_RSM
] = emulate_on_interception
,
4112 [SVM_EXIT_AVIC_INCOMPLETE_IPI
] = avic_incomplete_ipi_interception
,
4113 [SVM_EXIT_AVIC_UNACCELERATED_ACCESS
] = avic_unaccelerated_access_interception
,
4116 static void dump_vmcb(struct kvm_vcpu
*vcpu
)
4118 struct vcpu_svm
*svm
= to_svm(vcpu
);
4119 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
4120 struct vmcb_save_area
*save
= &svm
->vmcb
->save
;
4122 pr_err("VMCB Control Area:\n");
4123 pr_err("%-20s%04x\n", "cr_read:", control
->intercept_cr
& 0xffff);
4124 pr_err("%-20s%04x\n", "cr_write:", control
->intercept_cr
>> 16);
4125 pr_err("%-20s%04x\n", "dr_read:", control
->intercept_dr
& 0xffff);
4126 pr_err("%-20s%04x\n", "dr_write:", control
->intercept_dr
>> 16);
4127 pr_err("%-20s%08x\n", "exceptions:", control
->intercept_exceptions
);
4128 pr_err("%-20s%016llx\n", "intercepts:", control
->intercept
);
4129 pr_err("%-20s%d\n", "pause filter count:", control
->pause_filter_count
);
4130 pr_err("%-20s%016llx\n", "iopm_base_pa:", control
->iopm_base_pa
);
4131 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control
->msrpm_base_pa
);
4132 pr_err("%-20s%016llx\n", "tsc_offset:", control
->tsc_offset
);
4133 pr_err("%-20s%d\n", "asid:", control
->asid
);
4134 pr_err("%-20s%d\n", "tlb_ctl:", control
->tlb_ctl
);
4135 pr_err("%-20s%08x\n", "int_ctl:", control
->int_ctl
);
4136 pr_err("%-20s%08x\n", "int_vector:", control
->int_vector
);
4137 pr_err("%-20s%08x\n", "int_state:", control
->int_state
);
4138 pr_err("%-20s%08x\n", "exit_code:", control
->exit_code
);
4139 pr_err("%-20s%016llx\n", "exit_info1:", control
->exit_info_1
);
4140 pr_err("%-20s%016llx\n", "exit_info2:", control
->exit_info_2
);
4141 pr_err("%-20s%08x\n", "exit_int_info:", control
->exit_int_info
);
4142 pr_err("%-20s%08x\n", "exit_int_info_err:", control
->exit_int_info_err
);
4143 pr_err("%-20s%lld\n", "nested_ctl:", control
->nested_ctl
);
4144 pr_err("%-20s%016llx\n", "nested_cr3:", control
->nested_cr3
);
4145 pr_err("%-20s%016llx\n", "avic_vapic_bar:", control
->avic_vapic_bar
);
4146 pr_err("%-20s%08x\n", "event_inj:", control
->event_inj
);
4147 pr_err("%-20s%08x\n", "event_inj_err:", control
->event_inj_err
);
4148 pr_err("%-20s%lld\n", "virt_ext:", control
->virt_ext
);
4149 pr_err("%-20s%016llx\n", "next_rip:", control
->next_rip
);
4150 pr_err("%-20s%016llx\n", "avic_backing_page:", control
->avic_backing_page
);
4151 pr_err("%-20s%016llx\n", "avic_logical_id:", control
->avic_logical_id
);
4152 pr_err("%-20s%016llx\n", "avic_physical_id:", control
->avic_physical_id
);
4153 pr_err("VMCB State Save Area:\n");
4154 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4156 save
->es
.selector
, save
->es
.attrib
,
4157 save
->es
.limit
, save
->es
.base
);
4158 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4160 save
->cs
.selector
, save
->cs
.attrib
,
4161 save
->cs
.limit
, save
->cs
.base
);
4162 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4164 save
->ss
.selector
, save
->ss
.attrib
,
4165 save
->ss
.limit
, save
->ss
.base
);
4166 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4168 save
->ds
.selector
, save
->ds
.attrib
,
4169 save
->ds
.limit
, save
->ds
.base
);
4170 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4172 save
->fs
.selector
, save
->fs
.attrib
,
4173 save
->fs
.limit
, save
->fs
.base
);
4174 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4176 save
->gs
.selector
, save
->gs
.attrib
,
4177 save
->gs
.limit
, save
->gs
.base
);
4178 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4180 save
->gdtr
.selector
, save
->gdtr
.attrib
,
4181 save
->gdtr
.limit
, save
->gdtr
.base
);
4182 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4184 save
->ldtr
.selector
, save
->ldtr
.attrib
,
4185 save
->ldtr
.limit
, save
->ldtr
.base
);
4186 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4188 save
->idtr
.selector
, save
->idtr
.attrib
,
4189 save
->idtr
.limit
, save
->idtr
.base
);
4190 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4192 save
->tr
.selector
, save
->tr
.attrib
,
4193 save
->tr
.limit
, save
->tr
.base
);
4194 pr_err("cpl: %d efer: %016llx\n",
4195 save
->cpl
, save
->efer
);
4196 pr_err("%-15s %016llx %-13s %016llx\n",
4197 "cr0:", save
->cr0
, "cr2:", save
->cr2
);
4198 pr_err("%-15s %016llx %-13s %016llx\n",
4199 "cr3:", save
->cr3
, "cr4:", save
->cr4
);
4200 pr_err("%-15s %016llx %-13s %016llx\n",
4201 "dr6:", save
->dr6
, "dr7:", save
->dr7
);
4202 pr_err("%-15s %016llx %-13s %016llx\n",
4203 "rip:", save
->rip
, "rflags:", save
->rflags
);
4204 pr_err("%-15s %016llx %-13s %016llx\n",
4205 "rsp:", save
->rsp
, "rax:", save
->rax
);
4206 pr_err("%-15s %016llx %-13s %016llx\n",
4207 "star:", save
->star
, "lstar:", save
->lstar
);
4208 pr_err("%-15s %016llx %-13s %016llx\n",
4209 "cstar:", save
->cstar
, "sfmask:", save
->sfmask
);
4210 pr_err("%-15s %016llx %-13s %016llx\n",
4211 "kernel_gs_base:", save
->kernel_gs_base
,
4212 "sysenter_cs:", save
->sysenter_cs
);
4213 pr_err("%-15s %016llx %-13s %016llx\n",
4214 "sysenter_esp:", save
->sysenter_esp
,
4215 "sysenter_eip:", save
->sysenter_eip
);
4216 pr_err("%-15s %016llx %-13s %016llx\n",
4217 "gpat:", save
->g_pat
, "dbgctl:", save
->dbgctl
);
4218 pr_err("%-15s %016llx %-13s %016llx\n",
4219 "br_from:", save
->br_from
, "br_to:", save
->br_to
);
4220 pr_err("%-15s %016llx %-13s %016llx\n",
4221 "excp_from:", save
->last_excp_from
,
4222 "excp_to:", save
->last_excp_to
);
4225 static void svm_get_exit_info(struct kvm_vcpu
*vcpu
, u64
*info1
, u64
*info2
)
4227 struct vmcb_control_area
*control
= &to_svm(vcpu
)->vmcb
->control
;
4229 *info1
= control
->exit_info_1
;
4230 *info2
= control
->exit_info_2
;
4233 static int handle_exit(struct kvm_vcpu
*vcpu
)
4235 struct vcpu_svm
*svm
= to_svm(vcpu
);
4236 struct kvm_run
*kvm_run
= vcpu
->run
;
4237 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
4239 trace_kvm_exit(exit_code
, vcpu
, KVM_ISA_SVM
);
4241 vcpu
->arch
.gpa_available
= (exit_code
== SVM_EXIT_NPF
);
4243 if (!is_cr_intercept(svm
, INTERCEPT_CR0_WRITE
))
4244 vcpu
->arch
.cr0
= svm
->vmcb
->save
.cr0
;
4246 vcpu
->arch
.cr3
= svm
->vmcb
->save
.cr3
;
4248 if (unlikely(svm
->nested
.exit_required
)) {
4249 nested_svm_vmexit(svm
);
4250 svm
->nested
.exit_required
= false;
4255 if (is_guest_mode(vcpu
)) {
4258 trace_kvm_nested_vmexit(svm
->vmcb
->save
.rip
, exit_code
,
4259 svm
->vmcb
->control
.exit_info_1
,
4260 svm
->vmcb
->control
.exit_info_2
,
4261 svm
->vmcb
->control
.exit_int_info
,
4262 svm
->vmcb
->control
.exit_int_info_err
,
4265 vmexit
= nested_svm_exit_special(svm
);
4267 if (vmexit
== NESTED_EXIT_CONTINUE
)
4268 vmexit
= nested_svm_exit_handled(svm
);
4270 if (vmexit
== NESTED_EXIT_DONE
)
4274 svm_complete_interrupts(svm
);
4276 if (svm
->vmcb
->control
.exit_code
== SVM_EXIT_ERR
) {
4277 kvm_run
->exit_reason
= KVM_EXIT_FAIL_ENTRY
;
4278 kvm_run
->fail_entry
.hardware_entry_failure_reason
4279 = svm
->vmcb
->control
.exit_code
;
4280 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
4285 if (is_external_interrupt(svm
->vmcb
->control
.exit_int_info
) &&
4286 exit_code
!= SVM_EXIT_EXCP_BASE
+ PF_VECTOR
&&
4287 exit_code
!= SVM_EXIT_NPF
&& exit_code
!= SVM_EXIT_TASK_SWITCH
&&
4288 exit_code
!= SVM_EXIT_INTR
&& exit_code
!= SVM_EXIT_NMI
)
4289 printk(KERN_ERR
"%s: unexpected exit_int_info 0x%x "
4291 __func__
, svm
->vmcb
->control
.exit_int_info
,
4294 if (exit_code
>= ARRAY_SIZE(svm_exit_handlers
)
4295 || !svm_exit_handlers
[exit_code
]) {
4296 WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code
);
4297 kvm_queue_exception(vcpu
, UD_VECTOR
);
4301 return svm_exit_handlers
[exit_code
](svm
);
4304 static void reload_tss(struct kvm_vcpu
*vcpu
)
4306 int cpu
= raw_smp_processor_id();
4308 struct svm_cpu_data
*sd
= per_cpu(svm_data
, cpu
);
4309 sd
->tss_desc
->type
= 9; /* available 32/64-bit TSS */
4313 static void pre_svm_run(struct vcpu_svm
*svm
)
4315 int cpu
= raw_smp_processor_id();
4317 struct svm_cpu_data
*sd
= per_cpu(svm_data
, cpu
);
4319 /* FIXME: handle wraparound of asid_generation */
4320 if (svm
->asid_generation
!= sd
->asid_generation
)
4324 static void svm_inject_nmi(struct kvm_vcpu
*vcpu
)
4326 struct vcpu_svm
*svm
= to_svm(vcpu
);
4328 svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_NMI
;
4329 vcpu
->arch
.hflags
|= HF_NMI_MASK
;
4330 set_intercept(svm
, INTERCEPT_IRET
);
4331 ++vcpu
->stat
.nmi_injections
;
4334 static inline void svm_inject_irq(struct vcpu_svm
*svm
, int irq
)
4336 struct vmcb_control_area
*control
;
4338 /* The following fields are ignored when AVIC is enabled */
4339 control
= &svm
->vmcb
->control
;
4340 control
->int_vector
= irq
;
4341 control
->int_ctl
&= ~V_INTR_PRIO_MASK
;
4342 control
->int_ctl
|= V_IRQ_MASK
|
4343 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT
);
4344 mark_dirty(svm
->vmcb
, VMCB_INTR
);
4347 static void svm_set_irq(struct kvm_vcpu
*vcpu
)
4349 struct vcpu_svm
*svm
= to_svm(vcpu
);
4351 BUG_ON(!(gif_set(svm
)));
4353 trace_kvm_inj_virq(vcpu
->arch
.interrupt
.nr
);
4354 ++vcpu
->stat
.irq_injections
;
4356 svm
->vmcb
->control
.event_inj
= vcpu
->arch
.interrupt
.nr
|
4357 SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
;
4360 static inline bool svm_nested_virtualize_tpr(struct kvm_vcpu
*vcpu
)
4362 return is_guest_mode(vcpu
) && (vcpu
->arch
.hflags
& HF_VINTR_MASK
);
4365 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
, int tpr
, int irr
)
4367 struct vcpu_svm
*svm
= to_svm(vcpu
);
4369 if (svm_nested_virtualize_tpr(vcpu
) ||
4370 kvm_vcpu_apicv_active(vcpu
))
4373 clr_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
4379 set_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
4382 static void svm_set_virtual_x2apic_mode(struct kvm_vcpu
*vcpu
, bool set
)
4387 static bool svm_get_enable_apicv(void)
4392 static void svm_hwapic_irr_update(struct kvm_vcpu
*vcpu
, int max_irr
)
4396 static void svm_hwapic_isr_update(struct kvm_vcpu
*vcpu
, int max_isr
)
4400 /* Note: Currently only used by Hyper-V. */
4401 static void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu
*vcpu
)
4403 struct vcpu_svm
*svm
= to_svm(vcpu
);
4404 struct vmcb
*vmcb
= svm
->vmcb
;
4409 vmcb
->control
.int_ctl
&= ~AVIC_ENABLE_MASK
;
4410 mark_dirty(vmcb
, VMCB_INTR
);
4413 static void svm_load_eoi_exitmap(struct kvm_vcpu
*vcpu
, u64
*eoi_exit_bitmap
)
4418 static void svm_deliver_avic_intr(struct kvm_vcpu
*vcpu
, int vec
)
4420 kvm_lapic_set_irr(vec
, vcpu
->arch
.apic
);
4421 smp_mb__after_atomic();
4423 if (avic_vcpu_is_running(vcpu
))
4424 wrmsrl(SVM_AVIC_DOORBELL
,
4425 kvm_cpu_get_apicid(vcpu
->cpu
));
4427 kvm_vcpu_wake_up(vcpu
);
4430 static void svm_ir_list_del(struct vcpu_svm
*svm
, struct amd_iommu_pi_data
*pi
)
4432 unsigned long flags
;
4433 struct amd_svm_iommu_ir
*cur
;
4435 spin_lock_irqsave(&svm
->ir_list_lock
, flags
);
4436 list_for_each_entry(cur
, &svm
->ir_list
, node
) {
4437 if (cur
->data
!= pi
->ir_data
)
4439 list_del(&cur
->node
);
4443 spin_unlock_irqrestore(&svm
->ir_list_lock
, flags
);
4446 static int svm_ir_list_add(struct vcpu_svm
*svm
, struct amd_iommu_pi_data
*pi
)
4449 unsigned long flags
;
4450 struct amd_svm_iommu_ir
*ir
;
4453 * In some cases, the existing irte is updaed and re-set,
4454 * so we need to check here if it's already been * added
4457 if (pi
->ir_data
&& (pi
->prev_ga_tag
!= 0)) {
4458 struct kvm
*kvm
= svm
->vcpu
.kvm
;
4459 u32 vcpu_id
= AVIC_GATAG_TO_VCPUID(pi
->prev_ga_tag
);
4460 struct kvm_vcpu
*prev_vcpu
= kvm_get_vcpu_by_id(kvm
, vcpu_id
);
4461 struct vcpu_svm
*prev_svm
;
4468 prev_svm
= to_svm(prev_vcpu
);
4469 svm_ir_list_del(prev_svm
, pi
);
4473 * Allocating new amd_iommu_pi_data, which will get
4474 * add to the per-vcpu ir_list.
4476 ir
= kzalloc(sizeof(struct amd_svm_iommu_ir
), GFP_KERNEL
);
4481 ir
->data
= pi
->ir_data
;
4483 spin_lock_irqsave(&svm
->ir_list_lock
, flags
);
4484 list_add(&ir
->node
, &svm
->ir_list
);
4485 spin_unlock_irqrestore(&svm
->ir_list_lock
, flags
);
4492 * The HW cannot support posting multicast/broadcast
4493 * interrupts to a vCPU. So, we still use legacy interrupt
4494 * remapping for these kind of interrupts.
4496 * For lowest-priority interrupts, we only support
4497 * those with single CPU as the destination, e.g. user
4498 * configures the interrupts via /proc/irq or uses
4499 * irqbalance to make the interrupts single-CPU.
4502 get_pi_vcpu_info(struct kvm
*kvm
, struct kvm_kernel_irq_routing_entry
*e
,
4503 struct vcpu_data
*vcpu_info
, struct vcpu_svm
**svm
)
4505 struct kvm_lapic_irq irq
;
4506 struct kvm_vcpu
*vcpu
= NULL
;
4508 kvm_set_msi_irq(kvm
, e
, &irq
);
4510 if (!kvm_intr_is_single_vcpu(kvm
, &irq
, &vcpu
)) {
4511 pr_debug("SVM: %s: use legacy intr remap mode for irq %u\n",
4512 __func__
, irq
.vector
);
4516 pr_debug("SVM: %s: use GA mode for irq %u\n", __func__
,
4518 *svm
= to_svm(vcpu
);
4519 vcpu_info
->pi_desc_addr
= page_to_phys((*svm
)->avic_backing_page
);
4520 vcpu_info
->vector
= irq
.vector
;
4526 * svm_update_pi_irte - set IRTE for Posted-Interrupts
4529 * @host_irq: host irq of the interrupt
4530 * @guest_irq: gsi of the interrupt
4531 * @set: set or unset PI
4532 * returns 0 on success, < 0 on failure
4534 static int svm_update_pi_irte(struct kvm
*kvm
, unsigned int host_irq
,
4535 uint32_t guest_irq
, bool set
)
4537 struct kvm_kernel_irq_routing_entry
*e
;
4538 struct kvm_irq_routing_table
*irq_rt
;
4539 int idx
, ret
= -EINVAL
;
4541 if (!kvm_arch_has_assigned_device(kvm
) ||
4542 !irq_remapping_cap(IRQ_POSTING_CAP
))
4545 pr_debug("SVM: %s: host_irq=%#x, guest_irq=%#x, set=%#x\n",
4546 __func__
, host_irq
, guest_irq
, set
);
4548 idx
= srcu_read_lock(&kvm
->irq_srcu
);
4549 irq_rt
= srcu_dereference(kvm
->irq_routing
, &kvm
->irq_srcu
);
4550 WARN_ON(guest_irq
>= irq_rt
->nr_rt_entries
);
4552 hlist_for_each_entry(e
, &irq_rt
->map
[guest_irq
], link
) {
4553 struct vcpu_data vcpu_info
;
4554 struct vcpu_svm
*svm
= NULL
;
4556 if (e
->type
!= KVM_IRQ_ROUTING_MSI
)
4560 * Here, we setup with legacy mode in the following cases:
4561 * 1. When cannot target interrupt to a specific vcpu.
4562 * 2. Unsetting posted interrupt.
4563 * 3. APIC virtialization is disabled for the vcpu.
4565 if (!get_pi_vcpu_info(kvm
, e
, &vcpu_info
, &svm
) && set
&&
4566 kvm_vcpu_apicv_active(&svm
->vcpu
)) {
4567 struct amd_iommu_pi_data pi
;
4569 /* Try to enable guest_mode in IRTE */
4570 pi
.base
= page_to_phys(svm
->avic_backing_page
) & AVIC_HPA_MASK
;
4571 pi
.ga_tag
= AVIC_GATAG(kvm
->arch
.avic_vm_id
,
4573 pi
.is_guest_mode
= true;
4574 pi
.vcpu_data
= &vcpu_info
;
4575 ret
= irq_set_vcpu_affinity(host_irq
, &pi
);
4578 * Here, we successfully setting up vcpu affinity in
4579 * IOMMU guest mode. Now, we need to store the posted
4580 * interrupt information in a per-vcpu ir_list so that
4581 * we can reference to them directly when we update vcpu
4582 * scheduling information in IOMMU irte.
4584 if (!ret
&& pi
.is_guest_mode
)
4585 svm_ir_list_add(svm
, &pi
);
4587 /* Use legacy mode in IRTE */
4588 struct amd_iommu_pi_data pi
;
4591 * Here, pi is used to:
4592 * - Tell IOMMU to use legacy mode for this interrupt.
4593 * - Retrieve ga_tag of prior interrupt remapping data.
4595 pi
.is_guest_mode
= false;
4596 ret
= irq_set_vcpu_affinity(host_irq
, &pi
);
4599 * Check if the posted interrupt was previously
4600 * setup with the guest_mode by checking if the ga_tag
4601 * was cached. If so, we need to clean up the per-vcpu
4604 if (!ret
&& pi
.prev_ga_tag
) {
4605 int id
= AVIC_GATAG_TO_VCPUID(pi
.prev_ga_tag
);
4606 struct kvm_vcpu
*vcpu
;
4608 vcpu
= kvm_get_vcpu_by_id(kvm
, id
);
4610 svm_ir_list_del(to_svm(vcpu
), &pi
);
4615 trace_kvm_pi_irte_update(svm
->vcpu
.vcpu_id
,
4618 vcpu_info
.pi_desc_addr
, set
);
4622 pr_err("%s: failed to update PI IRTE\n", __func__
);
4629 srcu_read_unlock(&kvm
->irq_srcu
, idx
);
4633 static int svm_nmi_allowed(struct kvm_vcpu
*vcpu
)
4635 struct vcpu_svm
*svm
= to_svm(vcpu
);
4636 struct vmcb
*vmcb
= svm
->vmcb
;
4638 ret
= !(vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
) &&
4639 !(svm
->vcpu
.arch
.hflags
& HF_NMI_MASK
);
4640 ret
= ret
&& gif_set(svm
) && nested_svm_nmi(svm
);
4645 static bool svm_get_nmi_mask(struct kvm_vcpu
*vcpu
)
4647 struct vcpu_svm
*svm
= to_svm(vcpu
);
4649 return !!(svm
->vcpu
.arch
.hflags
& HF_NMI_MASK
);
4652 static void svm_set_nmi_mask(struct kvm_vcpu
*vcpu
, bool masked
)
4654 struct vcpu_svm
*svm
= to_svm(vcpu
);
4657 svm
->vcpu
.arch
.hflags
|= HF_NMI_MASK
;
4658 set_intercept(svm
, INTERCEPT_IRET
);
4660 svm
->vcpu
.arch
.hflags
&= ~HF_NMI_MASK
;
4661 clr_intercept(svm
, INTERCEPT_IRET
);
4665 static int svm_interrupt_allowed(struct kvm_vcpu
*vcpu
)
4667 struct vcpu_svm
*svm
= to_svm(vcpu
);
4668 struct vmcb
*vmcb
= svm
->vmcb
;
4671 if (!gif_set(svm
) ||
4672 (vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
))
4675 ret
= !!(kvm_get_rflags(vcpu
) & X86_EFLAGS_IF
);
4677 if (is_guest_mode(vcpu
))
4678 return ret
&& !(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
);
4683 static void enable_irq_window(struct kvm_vcpu
*vcpu
)
4685 struct vcpu_svm
*svm
= to_svm(vcpu
);
4687 if (kvm_vcpu_apicv_active(vcpu
))
4691 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
4692 * 1, because that's a separate STGI/VMRUN intercept. The next time we
4693 * get that intercept, this function will be called again though and
4694 * we'll get the vintr intercept.
4696 if (gif_set(svm
) && nested_svm_intr(svm
)) {
4698 svm_inject_irq(svm
, 0x0);
4702 static void enable_nmi_window(struct kvm_vcpu
*vcpu
)
4704 struct vcpu_svm
*svm
= to_svm(vcpu
);
4706 if ((svm
->vcpu
.arch
.hflags
& (HF_NMI_MASK
| HF_IRET_MASK
))
4708 return; /* IRET will cause a vm exit */
4710 if ((svm
->vcpu
.arch
.hflags
& HF_GIF_MASK
) == 0)
4711 return; /* STGI will cause a vm exit */
4713 if (svm
->nested
.exit_required
)
4714 return; /* we're not going to run the guest yet */
4717 * Something prevents NMI from been injected. Single step over possible
4718 * problem (IRET or exception injection or interrupt shadow)
4720 svm
->nmi_singlestep_guest_rflags
= svm_get_rflags(vcpu
);
4721 svm
->nmi_singlestep
= true;
4722 svm
->vmcb
->save
.rflags
|= (X86_EFLAGS_TF
| X86_EFLAGS_RF
);
4725 static int svm_set_tss_addr(struct kvm
*kvm
, unsigned int addr
)
4730 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
)
4732 struct vcpu_svm
*svm
= to_svm(vcpu
);
4734 if (static_cpu_has(X86_FEATURE_FLUSHBYASID
))
4735 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ASID
;
4737 svm
->asid_generation
--;
4740 static void svm_prepare_guest_switch(struct kvm_vcpu
*vcpu
)
4744 static inline void sync_cr8_to_lapic(struct kvm_vcpu
*vcpu
)
4746 struct vcpu_svm
*svm
= to_svm(vcpu
);
4748 if (svm_nested_virtualize_tpr(vcpu
))
4751 if (!is_cr_intercept(svm
, INTERCEPT_CR8_WRITE
)) {
4752 int cr8
= svm
->vmcb
->control
.int_ctl
& V_TPR_MASK
;
4753 kvm_set_cr8(vcpu
, cr8
);
4757 static inline void sync_lapic_to_cr8(struct kvm_vcpu
*vcpu
)
4759 struct vcpu_svm
*svm
= to_svm(vcpu
);
4762 if (svm_nested_virtualize_tpr(vcpu
) ||
4763 kvm_vcpu_apicv_active(vcpu
))
4766 cr8
= kvm_get_cr8(vcpu
);
4767 svm
->vmcb
->control
.int_ctl
&= ~V_TPR_MASK
;
4768 svm
->vmcb
->control
.int_ctl
|= cr8
& V_TPR_MASK
;
4771 static void svm_complete_interrupts(struct vcpu_svm
*svm
)
4775 u32 exitintinfo
= svm
->vmcb
->control
.exit_int_info
;
4776 unsigned int3_injected
= svm
->int3_injected
;
4778 svm
->int3_injected
= 0;
4781 * If we've made progress since setting HF_IRET_MASK, we've
4782 * executed an IRET and can allow NMI injection.
4784 if ((svm
->vcpu
.arch
.hflags
& HF_IRET_MASK
)
4785 && kvm_rip_read(&svm
->vcpu
) != svm
->nmi_iret_rip
) {
4786 svm
->vcpu
.arch
.hflags
&= ~(HF_NMI_MASK
| HF_IRET_MASK
);
4787 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
4790 svm
->vcpu
.arch
.nmi_injected
= false;
4791 kvm_clear_exception_queue(&svm
->vcpu
);
4792 kvm_clear_interrupt_queue(&svm
->vcpu
);
4794 if (!(exitintinfo
& SVM_EXITINTINFO_VALID
))
4797 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
4799 vector
= exitintinfo
& SVM_EXITINTINFO_VEC_MASK
;
4800 type
= exitintinfo
& SVM_EXITINTINFO_TYPE_MASK
;
4803 case SVM_EXITINTINFO_TYPE_NMI
:
4804 svm
->vcpu
.arch
.nmi_injected
= true;
4806 case SVM_EXITINTINFO_TYPE_EXEPT
:
4808 * In case of software exceptions, do not reinject the vector,
4809 * but re-execute the instruction instead. Rewind RIP first
4810 * if we emulated INT3 before.
4812 if (kvm_exception_is_soft(vector
)) {
4813 if (vector
== BP_VECTOR
&& int3_injected
&&
4814 kvm_is_linear_rip(&svm
->vcpu
, svm
->int3_rip
))
4815 kvm_rip_write(&svm
->vcpu
,
4816 kvm_rip_read(&svm
->vcpu
) -
4820 if (exitintinfo
& SVM_EXITINTINFO_VALID_ERR
) {
4821 u32 err
= svm
->vmcb
->control
.exit_int_info_err
;
4822 kvm_requeue_exception_e(&svm
->vcpu
, vector
, err
);
4825 kvm_requeue_exception(&svm
->vcpu
, vector
);
4827 case SVM_EXITINTINFO_TYPE_INTR
:
4828 kvm_queue_interrupt(&svm
->vcpu
, vector
, false);
4835 static void svm_cancel_injection(struct kvm_vcpu
*vcpu
)
4837 struct vcpu_svm
*svm
= to_svm(vcpu
);
4838 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
4840 control
->exit_int_info
= control
->event_inj
;
4841 control
->exit_int_info_err
= control
->event_inj_err
;
4842 control
->event_inj
= 0;
4843 svm_complete_interrupts(svm
);
4846 static void svm_vcpu_run(struct kvm_vcpu
*vcpu
)
4848 struct vcpu_svm
*svm
= to_svm(vcpu
);
4850 svm
->vmcb
->save
.rax
= vcpu
->arch
.regs
[VCPU_REGS_RAX
];
4851 svm
->vmcb
->save
.rsp
= vcpu
->arch
.regs
[VCPU_REGS_RSP
];
4852 svm
->vmcb
->save
.rip
= vcpu
->arch
.regs
[VCPU_REGS_RIP
];
4855 * A vmexit emulation is required before the vcpu can be executed
4858 if (unlikely(svm
->nested
.exit_required
))
4862 * Disable singlestep if we're injecting an interrupt/exception.
4863 * We don't want our modified rflags to be pushed on the stack where
4864 * we might not be able to easily reset them if we disabled NMI
4867 if (svm
->nmi_singlestep
&& svm
->vmcb
->control
.event_inj
) {
4869 * Event injection happens before external interrupts cause a
4870 * vmexit and interrupts are disabled here, so smp_send_reschedule
4871 * is enough to force an immediate vmexit.
4873 disable_nmi_singlestep(svm
);
4874 smp_send_reschedule(vcpu
->cpu
);
4879 sync_lapic_to_cr8(vcpu
);
4881 svm
->vmcb
->save
.cr2
= vcpu
->arch
.cr2
;
4887 if (ibrs_inuse
&& (svm
->spec_ctrl
!= FEATURE_ENABLE_IBRS
))
4888 wrmsrl(MSR_IA32_SPEC_CTRL
, svm
->spec_ctrl
);
4891 "push %%" _ASM_BP
"; \n\t"
4892 "mov %c[rbx](%[svm]), %%" _ASM_BX
" \n\t"
4893 "mov %c[rcx](%[svm]), %%" _ASM_CX
" \n\t"
4894 "mov %c[rdx](%[svm]), %%" _ASM_DX
" \n\t"
4895 "mov %c[rsi](%[svm]), %%" _ASM_SI
" \n\t"
4896 "mov %c[rdi](%[svm]), %%" _ASM_DI
" \n\t"
4897 "mov %c[rbp](%[svm]), %%" _ASM_BP
" \n\t"
4898 #ifdef CONFIG_X86_64
4899 "mov %c[r8](%[svm]), %%r8 \n\t"
4900 "mov %c[r9](%[svm]), %%r9 \n\t"
4901 "mov %c[r10](%[svm]), %%r10 \n\t"
4902 "mov %c[r11](%[svm]), %%r11 \n\t"
4903 "mov %c[r12](%[svm]), %%r12 \n\t"
4904 "mov %c[r13](%[svm]), %%r13 \n\t"
4905 "mov %c[r14](%[svm]), %%r14 \n\t"
4906 "mov %c[r15](%[svm]), %%r15 \n\t"
4909 /* Enter guest mode */
4910 "push %%" _ASM_AX
" \n\t"
4911 "mov %c[vmcb](%[svm]), %%" _ASM_AX
" \n\t"
4912 __ex(SVM_VMLOAD
) "\n\t"
4913 __ex(SVM_VMRUN
) "\n\t"
4914 __ex(SVM_VMSAVE
) "\n\t"
4915 "pop %%" _ASM_AX
" \n\t"
4917 /* Save guest registers, load host registers */
4918 "mov %%" _ASM_BX
", %c[rbx](%[svm]) \n\t"
4919 "mov %%" _ASM_CX
", %c[rcx](%[svm]) \n\t"
4920 "mov %%" _ASM_DX
", %c[rdx](%[svm]) \n\t"
4921 "mov %%" _ASM_SI
", %c[rsi](%[svm]) \n\t"
4922 "mov %%" _ASM_DI
", %c[rdi](%[svm]) \n\t"
4923 "mov %%" _ASM_BP
", %c[rbp](%[svm]) \n\t"
4924 #ifdef CONFIG_X86_64
4925 "mov %%r8, %c[r8](%[svm]) \n\t"
4926 "mov %%r9, %c[r9](%[svm]) \n\t"
4927 "mov %%r10, %c[r10](%[svm]) \n\t"
4928 "mov %%r11, %c[r11](%[svm]) \n\t"
4929 "mov %%r12, %c[r12](%[svm]) \n\t"
4930 "mov %%r13, %c[r13](%[svm]) \n\t"
4931 "mov %%r14, %c[r14](%[svm]) \n\t"
4932 "mov %%r15, %c[r15](%[svm]) \n\t"
4937 [vmcb
]"i"(offsetof(struct vcpu_svm
, vmcb_pa
)),
4938 [rbx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RBX
])),
4939 [rcx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RCX
])),
4940 [rdx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RDX
])),
4941 [rsi
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RSI
])),
4942 [rdi
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RDI
])),
4943 [rbp
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RBP
]))
4944 #ifdef CONFIG_X86_64
4945 , [r8
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R8
])),
4946 [r9
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R9
])),
4947 [r10
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R10
])),
4948 [r11
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R11
])),
4949 [r12
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R12
])),
4950 [r13
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R13
])),
4951 [r14
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R14
])),
4952 [r15
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R15
]))
4955 #ifdef CONFIG_X86_64
4956 , "rbx", "rcx", "rdx", "rsi", "rdi"
4957 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
4959 , "ebx", "ecx", "edx", "esi", "edi"
4964 rdmsrl(MSR_IA32_SPEC_CTRL
, svm
->spec_ctrl
);
4965 if (svm
->spec_ctrl
!= FEATURE_ENABLE_IBRS
)
4966 wrmsrl(MSR_IA32_SPEC_CTRL
, FEATURE_ENABLE_IBRS
);
4969 #ifdef CONFIG_X86_64
4970 wrmsrl(MSR_GS_BASE
, svm
->host
.gs_base
);
4972 loadsegment(fs
, svm
->host
.fs
);
4973 #ifndef CONFIG_X86_32_LAZY_GS
4974 loadsegment(gs
, svm
->host
.gs
);
4980 local_irq_disable();
4982 vcpu
->arch
.cr2
= svm
->vmcb
->save
.cr2
;
4983 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = svm
->vmcb
->save
.rax
;
4984 vcpu
->arch
.regs
[VCPU_REGS_RSP
] = svm
->vmcb
->save
.rsp
;
4985 vcpu
->arch
.regs
[VCPU_REGS_RIP
] = svm
->vmcb
->save
.rip
;
4987 if (unlikely(svm
->vmcb
->control
.exit_code
== SVM_EXIT_NMI
))
4988 kvm_before_handle_nmi(&svm
->vcpu
);
4992 /* Any pending NMI will happen here */
4994 if (unlikely(svm
->vmcb
->control
.exit_code
== SVM_EXIT_NMI
))
4995 kvm_after_handle_nmi(&svm
->vcpu
);
4997 sync_cr8_to_lapic(vcpu
);
5001 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_DO_NOTHING
;
5003 /* if exit due to PF check for async PF */
5004 if (svm
->vmcb
->control
.exit_code
== SVM_EXIT_EXCP_BASE
+ PF_VECTOR
)
5005 svm
->vcpu
.arch
.apf
.host_apf_reason
= kvm_read_and_reset_pf_reason();
5008 vcpu
->arch
.regs_avail
&= ~(1 << VCPU_EXREG_PDPTR
);
5009 vcpu
->arch
.regs_dirty
&= ~(1 << VCPU_EXREG_PDPTR
);
5013 * We need to handle MC intercepts here before the vcpu has a chance to
5014 * change the physical cpu
5016 if (unlikely(svm
->vmcb
->control
.exit_code
==
5017 SVM_EXIT_EXCP_BASE
+ MC_VECTOR
))
5018 svm_handle_mce(svm
);
5020 mark_all_clean(svm
->vmcb
);
5022 STACK_FRAME_NON_STANDARD(svm_vcpu_run
);
5024 static void svm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
5026 struct vcpu_svm
*svm
= to_svm(vcpu
);
5028 svm
->vmcb
->save
.cr3
= root
;
5029 mark_dirty(svm
->vmcb
, VMCB_CR
);
5030 svm_flush_tlb(vcpu
);
5033 static void set_tdp_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
5035 struct vcpu_svm
*svm
= to_svm(vcpu
);
5037 svm
->vmcb
->control
.nested_cr3
= root
;
5038 mark_dirty(svm
->vmcb
, VMCB_NPT
);
5040 /* Also sync guest cr3 here in case we live migrate */
5041 svm
->vmcb
->save
.cr3
= kvm_read_cr3(vcpu
);
5042 mark_dirty(svm
->vmcb
, VMCB_CR
);
5044 svm_flush_tlb(vcpu
);
5047 static int is_disabled(void)
5051 rdmsrl(MSR_VM_CR
, vm_cr
);
5052 if (vm_cr
& (1 << SVM_VM_CR_SVM_DISABLE
))
5059 svm_patch_hypercall(struct kvm_vcpu
*vcpu
, unsigned char *hypercall
)
5062 * Patch in the VMMCALL instruction:
5064 hypercall
[0] = 0x0f;
5065 hypercall
[1] = 0x01;
5066 hypercall
[2] = 0xd9;
5069 static void svm_check_processor_compat(void *rtn
)
5074 static bool svm_cpu_has_accelerated_tpr(void)
5079 static bool svm_has_high_real_mode_segbase(void)
5084 static u64
svm_get_mt_mask(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool is_mmio
)
5089 static void svm_cpuid_update(struct kvm_vcpu
*vcpu
)
5091 struct vcpu_svm
*svm
= to_svm(vcpu
);
5092 struct kvm_cpuid_entry2
*entry
;
5094 /* Update nrips enabled cache */
5095 svm
->nrips_enabled
= !!guest_cpuid_has_nrips(&svm
->vcpu
);
5097 if (!kvm_vcpu_apicv_active(vcpu
))
5100 entry
= kvm_find_cpuid_entry(vcpu
, 1, 0);
5102 entry
->ecx
&= ~bit(X86_FEATURE_X2APIC
);
5105 static void svm_set_supported_cpuid(u32 func
, struct kvm_cpuid_entry2
*entry
)
5110 entry
->ecx
&= ~bit(X86_FEATURE_X2APIC
);
5114 entry
->ecx
|= (1 << 2); /* Set SVM bit */
5117 entry
->eax
= 1; /* SVM revision 1 */
5118 entry
->ebx
= 8; /* Lets support 8 ASIDs in case we add proper
5119 ASID emulation to nested SVM */
5120 entry
->ecx
= 0; /* Reserved */
5121 entry
->edx
= 0; /* Per default do not support any
5122 additional features */
5124 /* Support next_rip if host supports it */
5125 if (boot_cpu_has(X86_FEATURE_NRIPS
))
5126 entry
->edx
|= SVM_FEATURE_NRIP
;
5128 /* Support NPT for the guest if enabled */
5130 entry
->edx
|= SVM_FEATURE_NPT
;
5136 static int svm_get_lpage_level(void)
5138 return PT_PDPE_LEVEL
;
5141 static bool svm_rdtscp_supported(void)
5143 return boot_cpu_has(X86_FEATURE_RDTSCP
);
5146 static bool svm_invpcid_supported(void)
5151 static bool svm_mpx_supported(void)
5156 static bool svm_xsaves_supported(void)
5161 static bool svm_has_wbinvd_exit(void)
5166 #define PRE_EX(exit) { .exit_code = (exit), \
5167 .stage = X86_ICPT_PRE_EXCEPT, }
5168 #define POST_EX(exit) { .exit_code = (exit), \
5169 .stage = X86_ICPT_POST_EXCEPT, }
5170 #define POST_MEM(exit) { .exit_code = (exit), \
5171 .stage = X86_ICPT_POST_MEMACCESS, }
5173 static const struct __x86_intercept
{
5175 enum x86_intercept_stage stage
;
5176 } x86_intercept_map
[] = {
5177 [x86_intercept_cr_read
] = POST_EX(SVM_EXIT_READ_CR0
),
5178 [x86_intercept_cr_write
] = POST_EX(SVM_EXIT_WRITE_CR0
),
5179 [x86_intercept_clts
] = POST_EX(SVM_EXIT_WRITE_CR0
),
5180 [x86_intercept_lmsw
] = POST_EX(SVM_EXIT_WRITE_CR0
),
5181 [x86_intercept_smsw
] = POST_EX(SVM_EXIT_READ_CR0
),
5182 [x86_intercept_dr_read
] = POST_EX(SVM_EXIT_READ_DR0
),
5183 [x86_intercept_dr_write
] = POST_EX(SVM_EXIT_WRITE_DR0
),
5184 [x86_intercept_sldt
] = POST_EX(SVM_EXIT_LDTR_READ
),
5185 [x86_intercept_str
] = POST_EX(SVM_EXIT_TR_READ
),
5186 [x86_intercept_lldt
] = POST_EX(SVM_EXIT_LDTR_WRITE
),
5187 [x86_intercept_ltr
] = POST_EX(SVM_EXIT_TR_WRITE
),
5188 [x86_intercept_sgdt
] = POST_EX(SVM_EXIT_GDTR_READ
),
5189 [x86_intercept_sidt
] = POST_EX(SVM_EXIT_IDTR_READ
),
5190 [x86_intercept_lgdt
] = POST_EX(SVM_EXIT_GDTR_WRITE
),
5191 [x86_intercept_lidt
] = POST_EX(SVM_EXIT_IDTR_WRITE
),
5192 [x86_intercept_vmrun
] = POST_EX(SVM_EXIT_VMRUN
),
5193 [x86_intercept_vmmcall
] = POST_EX(SVM_EXIT_VMMCALL
),
5194 [x86_intercept_vmload
] = POST_EX(SVM_EXIT_VMLOAD
),
5195 [x86_intercept_vmsave
] = POST_EX(SVM_EXIT_VMSAVE
),
5196 [x86_intercept_stgi
] = POST_EX(SVM_EXIT_STGI
),
5197 [x86_intercept_clgi
] = POST_EX(SVM_EXIT_CLGI
),
5198 [x86_intercept_skinit
] = POST_EX(SVM_EXIT_SKINIT
),
5199 [x86_intercept_invlpga
] = POST_EX(SVM_EXIT_INVLPGA
),
5200 [x86_intercept_rdtscp
] = POST_EX(SVM_EXIT_RDTSCP
),
5201 [x86_intercept_monitor
] = POST_MEM(SVM_EXIT_MONITOR
),
5202 [x86_intercept_mwait
] = POST_EX(SVM_EXIT_MWAIT
),
5203 [x86_intercept_invlpg
] = POST_EX(SVM_EXIT_INVLPG
),
5204 [x86_intercept_invd
] = POST_EX(SVM_EXIT_INVD
),
5205 [x86_intercept_wbinvd
] = POST_EX(SVM_EXIT_WBINVD
),
5206 [x86_intercept_wrmsr
] = POST_EX(SVM_EXIT_MSR
),
5207 [x86_intercept_rdtsc
] = POST_EX(SVM_EXIT_RDTSC
),
5208 [x86_intercept_rdmsr
] = POST_EX(SVM_EXIT_MSR
),
5209 [x86_intercept_rdpmc
] = POST_EX(SVM_EXIT_RDPMC
),
5210 [x86_intercept_cpuid
] = PRE_EX(SVM_EXIT_CPUID
),
5211 [x86_intercept_rsm
] = PRE_EX(SVM_EXIT_RSM
),
5212 [x86_intercept_pause
] = PRE_EX(SVM_EXIT_PAUSE
),
5213 [x86_intercept_pushf
] = PRE_EX(SVM_EXIT_PUSHF
),
5214 [x86_intercept_popf
] = PRE_EX(SVM_EXIT_POPF
),
5215 [x86_intercept_intn
] = PRE_EX(SVM_EXIT_SWINT
),
5216 [x86_intercept_iret
] = PRE_EX(SVM_EXIT_IRET
),
5217 [x86_intercept_icebp
] = PRE_EX(SVM_EXIT_ICEBP
),
5218 [x86_intercept_hlt
] = POST_EX(SVM_EXIT_HLT
),
5219 [x86_intercept_in
] = POST_EX(SVM_EXIT_IOIO
),
5220 [x86_intercept_ins
] = POST_EX(SVM_EXIT_IOIO
),
5221 [x86_intercept_out
] = POST_EX(SVM_EXIT_IOIO
),
5222 [x86_intercept_outs
] = POST_EX(SVM_EXIT_IOIO
),
5229 static int svm_check_intercept(struct kvm_vcpu
*vcpu
,
5230 struct x86_instruction_info
*info
,
5231 enum x86_intercept_stage stage
)
5233 struct vcpu_svm
*svm
= to_svm(vcpu
);
5234 int vmexit
, ret
= X86EMUL_CONTINUE
;
5235 struct __x86_intercept icpt_info
;
5236 struct vmcb
*vmcb
= svm
->vmcb
;
5238 if (info
->intercept
>= ARRAY_SIZE(x86_intercept_map
))
5241 icpt_info
= x86_intercept_map
[info
->intercept
];
5243 if (stage
!= icpt_info
.stage
)
5246 switch (icpt_info
.exit_code
) {
5247 case SVM_EXIT_READ_CR0
:
5248 if (info
->intercept
== x86_intercept_cr_read
)
5249 icpt_info
.exit_code
+= info
->modrm_reg
;
5251 case SVM_EXIT_WRITE_CR0
: {
5252 unsigned long cr0
, val
;
5255 if (info
->intercept
== x86_intercept_cr_write
)
5256 icpt_info
.exit_code
+= info
->modrm_reg
;
5258 if (icpt_info
.exit_code
!= SVM_EXIT_WRITE_CR0
||
5259 info
->intercept
== x86_intercept_clts
)
5262 intercept
= svm
->nested
.intercept
;
5264 if (!(intercept
& (1ULL << INTERCEPT_SELECTIVE_CR0
)))
5267 cr0
= vcpu
->arch
.cr0
& ~SVM_CR0_SELECTIVE_MASK
;
5268 val
= info
->src_val
& ~SVM_CR0_SELECTIVE_MASK
;
5270 if (info
->intercept
== x86_intercept_lmsw
) {
5273 /* lmsw can't clear PE - catch this here */
5274 if (cr0
& X86_CR0_PE
)
5279 icpt_info
.exit_code
= SVM_EXIT_CR0_SEL_WRITE
;
5283 case SVM_EXIT_READ_DR0
:
5284 case SVM_EXIT_WRITE_DR0
:
5285 icpt_info
.exit_code
+= info
->modrm_reg
;
5288 if (info
->intercept
== x86_intercept_wrmsr
)
5289 vmcb
->control
.exit_info_1
= 1;
5291 vmcb
->control
.exit_info_1
= 0;
5293 case SVM_EXIT_PAUSE
:
5295 * We get this for NOP only, but pause
5296 * is rep not, check this here
5298 if (info
->rep_prefix
!= REPE_PREFIX
)
5300 case SVM_EXIT_IOIO
: {
5304 if (info
->intercept
== x86_intercept_in
||
5305 info
->intercept
== x86_intercept_ins
) {
5306 exit_info
= ((info
->src_val
& 0xffff) << 16) |
5308 bytes
= info
->dst_bytes
;
5310 exit_info
= (info
->dst_val
& 0xffff) << 16;
5311 bytes
= info
->src_bytes
;
5314 if (info
->intercept
== x86_intercept_outs
||
5315 info
->intercept
== x86_intercept_ins
)
5316 exit_info
|= SVM_IOIO_STR_MASK
;
5318 if (info
->rep_prefix
)
5319 exit_info
|= SVM_IOIO_REP_MASK
;
5321 bytes
= min(bytes
, 4u);
5323 exit_info
|= bytes
<< SVM_IOIO_SIZE_SHIFT
;
5325 exit_info
|= (u32
)info
->ad_bytes
<< (SVM_IOIO_ASIZE_SHIFT
- 1);
5327 vmcb
->control
.exit_info_1
= exit_info
;
5328 vmcb
->control
.exit_info_2
= info
->next_rip
;
5336 /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
5337 if (static_cpu_has(X86_FEATURE_NRIPS
))
5338 vmcb
->control
.next_rip
= info
->next_rip
;
5339 vmcb
->control
.exit_code
= icpt_info
.exit_code
;
5340 vmexit
= nested_svm_exit_handled(svm
);
5342 ret
= (vmexit
== NESTED_EXIT_DONE
) ? X86EMUL_INTERCEPTED
5349 static void svm_handle_external_intr(struct kvm_vcpu
*vcpu
)
5353 * We must have an instruction with interrupts enabled, so
5354 * the timer interrupt isn't delayed by the interrupt shadow.
5357 local_irq_disable();
5360 static void svm_sched_in(struct kvm_vcpu
*vcpu
, int cpu
)
5364 static inline void avic_post_state_restore(struct kvm_vcpu
*vcpu
)
5366 if (avic_handle_apic_id_update(vcpu
) != 0)
5368 if (avic_handle_dfr_update(vcpu
) != 0)
5370 avic_handle_ldr_update(vcpu
);
5373 static void svm_setup_mce(struct kvm_vcpu
*vcpu
)
5375 /* [63:9] are reserved. */
5376 vcpu
->arch
.mcg_cap
&= 0x1ff;
5379 static struct kvm_x86_ops svm_x86_ops __ro_after_init
= {
5380 .cpu_has_kvm_support
= has_svm
,
5381 .disabled_by_bios
= is_disabled
,
5382 .hardware_setup
= svm_hardware_setup
,
5383 .hardware_unsetup
= svm_hardware_unsetup
,
5384 .check_processor_compatibility
= svm_check_processor_compat
,
5385 .hardware_enable
= svm_hardware_enable
,
5386 .hardware_disable
= svm_hardware_disable
,
5387 .cpu_has_accelerated_tpr
= svm_cpu_has_accelerated_tpr
,
5388 .cpu_has_high_real_mode_segbase
= svm_has_high_real_mode_segbase
,
5390 .vcpu_create
= svm_create_vcpu
,
5391 .vcpu_free
= svm_free_vcpu
,
5392 .vcpu_reset
= svm_vcpu_reset
,
5394 .vm_init
= avic_vm_init
,
5395 .vm_destroy
= avic_vm_destroy
,
5397 .prepare_guest_switch
= svm_prepare_guest_switch
,
5398 .vcpu_load
= svm_vcpu_load
,
5399 .vcpu_put
= svm_vcpu_put
,
5400 .vcpu_blocking
= svm_vcpu_blocking
,
5401 .vcpu_unblocking
= svm_vcpu_unblocking
,
5403 .update_bp_intercept
= update_bp_intercept
,
5404 .get_msr
= svm_get_msr
,
5405 .set_msr
= svm_set_msr
,
5406 .get_segment_base
= svm_get_segment_base
,
5407 .get_segment
= svm_get_segment
,
5408 .set_segment
= svm_set_segment
,
5409 .get_cpl
= svm_get_cpl
,
5410 .get_cs_db_l_bits
= kvm_get_cs_db_l_bits
,
5411 .decache_cr0_guest_bits
= svm_decache_cr0_guest_bits
,
5412 .decache_cr3
= svm_decache_cr3
,
5413 .decache_cr4_guest_bits
= svm_decache_cr4_guest_bits
,
5414 .set_cr0
= svm_set_cr0
,
5415 .set_cr3
= svm_set_cr3
,
5416 .set_cr4
= svm_set_cr4
,
5417 .set_efer
= svm_set_efer
,
5418 .get_idt
= svm_get_idt
,
5419 .set_idt
= svm_set_idt
,
5420 .get_gdt
= svm_get_gdt
,
5421 .set_gdt
= svm_set_gdt
,
5422 .get_dr6
= svm_get_dr6
,
5423 .set_dr6
= svm_set_dr6
,
5424 .set_dr7
= svm_set_dr7
,
5425 .sync_dirty_debug_regs
= svm_sync_dirty_debug_regs
,
5426 .cache_reg
= svm_cache_reg
,
5427 .get_rflags
= svm_get_rflags
,
5428 .set_rflags
= svm_set_rflags
,
5430 .tlb_flush
= svm_flush_tlb
,
5432 .run
= svm_vcpu_run
,
5433 .handle_exit
= handle_exit
,
5434 .skip_emulated_instruction
= skip_emulated_instruction
,
5435 .set_interrupt_shadow
= svm_set_interrupt_shadow
,
5436 .get_interrupt_shadow
= svm_get_interrupt_shadow
,
5437 .patch_hypercall
= svm_patch_hypercall
,
5438 .set_irq
= svm_set_irq
,
5439 .set_nmi
= svm_inject_nmi
,
5440 .queue_exception
= svm_queue_exception
,
5441 .cancel_injection
= svm_cancel_injection
,
5442 .interrupt_allowed
= svm_interrupt_allowed
,
5443 .nmi_allowed
= svm_nmi_allowed
,
5444 .get_nmi_mask
= svm_get_nmi_mask
,
5445 .set_nmi_mask
= svm_set_nmi_mask
,
5446 .enable_nmi_window
= enable_nmi_window
,
5447 .enable_irq_window
= enable_irq_window
,
5448 .update_cr8_intercept
= update_cr8_intercept
,
5449 .set_virtual_x2apic_mode
= svm_set_virtual_x2apic_mode
,
5450 .get_enable_apicv
= svm_get_enable_apicv
,
5451 .refresh_apicv_exec_ctrl
= svm_refresh_apicv_exec_ctrl
,
5452 .load_eoi_exitmap
= svm_load_eoi_exitmap
,
5453 .hwapic_irr_update
= svm_hwapic_irr_update
,
5454 .hwapic_isr_update
= svm_hwapic_isr_update
,
5455 .apicv_post_state_restore
= avic_post_state_restore
,
5457 .set_tss_addr
= svm_set_tss_addr
,
5458 .get_tdp_level
= get_npt_level
,
5459 .get_mt_mask
= svm_get_mt_mask
,
5461 .get_exit_info
= svm_get_exit_info
,
5463 .get_lpage_level
= svm_get_lpage_level
,
5465 .cpuid_update
= svm_cpuid_update
,
5467 .rdtscp_supported
= svm_rdtscp_supported
,
5468 .invpcid_supported
= svm_invpcid_supported
,
5469 .mpx_supported
= svm_mpx_supported
,
5470 .xsaves_supported
= svm_xsaves_supported
,
5472 .set_supported_cpuid
= svm_set_supported_cpuid
,
5474 .has_wbinvd_exit
= svm_has_wbinvd_exit
,
5476 .write_tsc_offset
= svm_write_tsc_offset
,
5478 .set_tdp_cr3
= set_tdp_cr3
,
5480 .check_intercept
= svm_check_intercept
,
5481 .handle_external_intr
= svm_handle_external_intr
,
5483 .sched_in
= svm_sched_in
,
5485 .pmu_ops
= &amd_pmu_ops
,
5486 .deliver_posted_interrupt
= svm_deliver_avic_intr
,
5487 .update_pi_irte
= svm_update_pi_irte
,
5488 .setup_mce
= svm_setup_mce
,
5491 static int __init
svm_init(void)
5493 return kvm_init(&svm_x86_ops
, sizeof(struct vcpu_svm
),
5494 __alignof__(struct vcpu_svm
), THIS_MODULE
);
5497 static void __exit
svm_exit(void)
5502 module_init(svm_init
)
5503 module_exit(svm_exit
)