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
;
522 struct vmcb
*current_vmcb
;
525 static DEFINE_PER_CPU(struct svm_cpu_data
*, svm_data
);
527 struct svm_init_data
{
532 static const u32 msrpm_ranges
[] = {0, 0xc0000000, 0xc0010000};
534 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
535 #define MSRS_RANGE_SIZE 2048
536 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
538 static u32
svm_msrpm_offset(u32 msr
)
543 for (i
= 0; i
< NUM_MSR_MAPS
; i
++) {
544 if (msr
< msrpm_ranges
[i
] ||
545 msr
>= msrpm_ranges
[i
] + MSRS_IN_RANGE
)
548 offset
= (msr
- msrpm_ranges
[i
]) / 4; /* 4 msrs per u8 */
549 offset
+= (i
* MSRS_RANGE_SIZE
); /* add range offset */
551 /* Now we have the u8 offset - but need the u32 offset */
555 /* MSR not in any range */
559 #define MAX_INST_SIZE 15
561 static inline void clgi(void)
563 asm volatile (__ex(SVM_CLGI
));
566 static inline void stgi(void)
568 asm volatile (__ex(SVM_STGI
));
571 static inline void invlpga(unsigned long addr
, u32 asid
)
573 asm volatile (__ex(SVM_INVLPGA
) : : "a"(addr
), "c"(asid
));
576 static int get_npt_level(void)
579 return PT64_ROOT_LEVEL
;
581 return PT32E_ROOT_LEVEL
;
585 static void svm_set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
587 vcpu
->arch
.efer
= efer
;
588 if (!npt_enabled
&& !(efer
& EFER_LMA
))
591 to_svm(vcpu
)->vmcb
->save
.efer
= efer
| EFER_SVME
;
592 mark_dirty(to_svm(vcpu
)->vmcb
, VMCB_CR
);
595 static int is_external_interrupt(u32 info
)
597 info
&= SVM_EVTINJ_TYPE_MASK
| SVM_EVTINJ_VALID
;
598 return info
== (SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
);
601 static u32
svm_get_interrupt_shadow(struct kvm_vcpu
*vcpu
)
603 struct vcpu_svm
*svm
= to_svm(vcpu
);
606 if (svm
->vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
)
607 ret
= KVM_X86_SHADOW_INT_STI
| KVM_X86_SHADOW_INT_MOV_SS
;
611 static void svm_set_interrupt_shadow(struct kvm_vcpu
*vcpu
, int mask
)
613 struct vcpu_svm
*svm
= to_svm(vcpu
);
616 svm
->vmcb
->control
.int_state
&= ~SVM_INTERRUPT_SHADOW_MASK
;
618 svm
->vmcb
->control
.int_state
|= SVM_INTERRUPT_SHADOW_MASK
;
622 static void skip_emulated_instruction(struct kvm_vcpu
*vcpu
)
624 struct vcpu_svm
*svm
= to_svm(vcpu
);
626 if (svm
->vmcb
->control
.next_rip
!= 0) {
627 WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS
));
628 svm
->next_rip
= svm
->vmcb
->control
.next_rip
;
631 if (!svm
->next_rip
) {
632 if (emulate_instruction(vcpu
, EMULTYPE_SKIP
) !=
634 printk(KERN_DEBUG
"%s: NOP\n", __func__
);
637 if (svm
->next_rip
- kvm_rip_read(vcpu
) > MAX_INST_SIZE
)
638 printk(KERN_ERR
"%s: ip 0x%lx next 0x%llx\n",
639 __func__
, kvm_rip_read(vcpu
), svm
->next_rip
);
641 kvm_rip_write(vcpu
, svm
->next_rip
);
642 svm_set_interrupt_shadow(vcpu
, 0);
645 static void svm_queue_exception(struct kvm_vcpu
*vcpu
)
647 struct vcpu_svm
*svm
= to_svm(vcpu
);
648 unsigned nr
= vcpu
->arch
.exception
.nr
;
649 bool has_error_code
= vcpu
->arch
.exception
.has_error_code
;
650 bool reinject
= vcpu
->arch
.exception
.reinject
;
651 u32 error_code
= vcpu
->arch
.exception
.error_code
;
654 * If we are within a nested VM we'd better #VMEXIT and let the guest
655 * handle the exception
658 nested_svm_check_exception(svm
, nr
, has_error_code
, error_code
))
661 if (nr
== BP_VECTOR
&& !static_cpu_has(X86_FEATURE_NRIPS
)) {
662 unsigned long rip
, old_rip
= kvm_rip_read(&svm
->vcpu
);
665 * For guest debugging where we have to reinject #BP if some
666 * INT3 is guest-owned:
667 * Emulate nRIP by moving RIP forward. Will fail if injection
668 * raises a fault that is not intercepted. Still better than
669 * failing in all cases.
671 skip_emulated_instruction(&svm
->vcpu
);
672 rip
= kvm_rip_read(&svm
->vcpu
);
673 svm
->int3_rip
= rip
+ svm
->vmcb
->save
.cs
.base
;
674 svm
->int3_injected
= rip
- old_rip
;
677 svm
->vmcb
->control
.event_inj
= nr
679 | (has_error_code
? SVM_EVTINJ_VALID_ERR
: 0)
680 | SVM_EVTINJ_TYPE_EXEPT
;
681 svm
->vmcb
->control
.event_inj_err
= error_code
;
684 static void svm_init_erratum_383(void)
690 if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH
))
693 /* Use _safe variants to not break nested virtualization */
694 val
= native_read_msr_safe(MSR_AMD64_DC_CFG
, &err
);
700 low
= lower_32_bits(val
);
701 high
= upper_32_bits(val
);
703 native_write_msr_safe(MSR_AMD64_DC_CFG
, low
, high
);
705 erratum_383_found
= true;
708 static void svm_init_osvw(struct kvm_vcpu
*vcpu
)
711 * Guests should see errata 400 and 415 as fixed (assuming that
712 * HLT and IO instructions are intercepted).
714 vcpu
->arch
.osvw
.length
= (osvw_len
>= 3) ? (osvw_len
) : 3;
715 vcpu
->arch
.osvw
.status
= osvw_status
& ~(6ULL);
718 * By increasing VCPU's osvw.length to 3 we are telling the guest that
719 * all osvw.status bits inside that length, including bit 0 (which is
720 * reserved for erratum 298), are valid. However, if host processor's
721 * osvw_len is 0 then osvw_status[0] carries no information. We need to
722 * be conservative here and therefore we tell the guest that erratum 298
723 * is present (because we really don't know).
725 if (osvw_len
== 0 && boot_cpu_data
.x86
== 0x10)
726 vcpu
->arch
.osvw
.status
|= 1;
729 static int has_svm(void)
733 if (!cpu_has_svm(&msg
)) {
734 printk(KERN_INFO
"has_svm: %s\n", msg
);
741 static void svm_hardware_disable(void)
743 /* Make sure we clean up behind us */
744 if (static_cpu_has(X86_FEATURE_TSCRATEMSR
))
745 wrmsrl(MSR_AMD64_TSC_RATIO
, TSC_RATIO_DEFAULT
);
749 amd_pmu_disable_virt();
752 static int svm_hardware_enable(void)
755 struct svm_cpu_data
*sd
;
757 struct desc_struct
*gdt
;
758 int me
= raw_smp_processor_id();
760 rdmsrl(MSR_EFER
, efer
);
761 if (efer
& EFER_SVME
)
765 pr_err("%s: err EOPNOTSUPP on %d\n", __func__
, me
);
768 sd
= per_cpu(svm_data
, me
);
770 pr_err("%s: svm_data is NULL on %d\n", __func__
, me
);
774 sd
->asid_generation
= 1;
775 sd
->max_asid
= cpuid_ebx(SVM_CPUID_FUNC
) - 1;
776 sd
->next_asid
= sd
->max_asid
+ 1;
778 gdt
= get_current_gdt_rw();
779 sd
->tss_desc
= (struct kvm_ldttss_desc
*)(gdt
+ GDT_ENTRY_TSS
);
781 wrmsrl(MSR_EFER
, efer
| EFER_SVME
);
783 wrmsrl(MSR_VM_HSAVE_PA
, page_to_pfn(sd
->save_area
) << PAGE_SHIFT
);
785 if (static_cpu_has(X86_FEATURE_TSCRATEMSR
)) {
786 wrmsrl(MSR_AMD64_TSC_RATIO
, TSC_RATIO_DEFAULT
);
787 __this_cpu_write(current_tsc_ratio
, TSC_RATIO_DEFAULT
);
794 * Note that it is possible to have a system with mixed processor
795 * revisions and therefore different OSVW bits. If bits are not the same
796 * on different processors then choose the worst case (i.e. if erratum
797 * is present on one processor and not on another then assume that the
798 * erratum is present everywhere).
800 if (cpu_has(&boot_cpu_data
, X86_FEATURE_OSVW
)) {
801 uint64_t len
, status
= 0;
804 len
= native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH
, &err
);
806 status
= native_read_msr_safe(MSR_AMD64_OSVW_STATUS
,
810 osvw_status
= osvw_len
= 0;
814 osvw_status
|= status
;
815 osvw_status
&= (1ULL << osvw_len
) - 1;
818 osvw_status
= osvw_len
= 0;
820 svm_init_erratum_383();
822 amd_pmu_enable_virt();
827 static void svm_cpu_uninit(int cpu
)
829 struct svm_cpu_data
*sd
= per_cpu(svm_data
, raw_smp_processor_id());
834 per_cpu(svm_data
, raw_smp_processor_id()) = NULL
;
835 __free_page(sd
->save_area
);
839 static int svm_cpu_init(int cpu
)
841 struct svm_cpu_data
*sd
;
844 sd
= kzalloc(sizeof(struct svm_cpu_data
), GFP_KERNEL
);
848 sd
->save_area
= alloc_page(GFP_KERNEL
);
853 per_cpu(svm_data
, cpu
) = sd
;
863 static bool valid_msr_intercept(u32 index
)
867 for (i
= 0; direct_access_msrs
[i
].index
!= MSR_INVALID
; i
++)
868 if (direct_access_msrs
[i
].index
== index
)
874 static void set_msr_interception(u32
*msrpm
, unsigned msr
,
877 u8 bit_read
, bit_write
;
882 * If this warning triggers extend the direct_access_msrs list at the
883 * beginning of the file
885 WARN_ON(!valid_msr_intercept(msr
));
887 offset
= svm_msrpm_offset(msr
);
888 bit_read
= 2 * (msr
& 0x0f);
889 bit_write
= 2 * (msr
& 0x0f) + 1;
892 BUG_ON(offset
== MSR_INVALID
);
894 read
? clear_bit(bit_read
, &tmp
) : set_bit(bit_read
, &tmp
);
895 write
? clear_bit(bit_write
, &tmp
) : set_bit(bit_write
, &tmp
);
900 static void svm_vcpu_init_msrpm(u32
*msrpm
)
904 memset(msrpm
, 0xff, PAGE_SIZE
* (1 << MSRPM_ALLOC_ORDER
));
906 for (i
= 0; direct_access_msrs
[i
].index
!= MSR_INVALID
; i
++) {
907 if (!direct_access_msrs
[i
].always
)
910 set_msr_interception(msrpm
, direct_access_msrs
[i
].index
, 1, 1);
914 static void add_msr_offset(u32 offset
)
918 for (i
= 0; i
< MSRPM_OFFSETS
; ++i
) {
920 /* Offset already in list? */
921 if (msrpm_offsets
[i
] == offset
)
924 /* Slot used by another offset? */
925 if (msrpm_offsets
[i
] != MSR_INVALID
)
928 /* Add offset to list */
929 msrpm_offsets
[i
] = offset
;
935 * If this BUG triggers the msrpm_offsets table has an overflow. Just
936 * increase MSRPM_OFFSETS in this case.
941 static void init_msrpm_offsets(void)
945 memset(msrpm_offsets
, 0xff, sizeof(msrpm_offsets
));
947 for (i
= 0; direct_access_msrs
[i
].index
!= MSR_INVALID
; i
++) {
950 offset
= svm_msrpm_offset(direct_access_msrs
[i
].index
);
951 BUG_ON(offset
== MSR_INVALID
);
953 add_msr_offset(offset
);
957 static void svm_enable_lbrv(struct vcpu_svm
*svm
)
959 u32
*msrpm
= svm
->msrpm
;
961 svm
->vmcb
->control
.virt_ext
|= LBR_CTL_ENABLE_MASK
;
962 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHFROMIP
, 1, 1);
963 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHTOIP
, 1, 1);
964 set_msr_interception(msrpm
, MSR_IA32_LASTINTFROMIP
, 1, 1);
965 set_msr_interception(msrpm
, MSR_IA32_LASTINTTOIP
, 1, 1);
968 static void svm_disable_lbrv(struct vcpu_svm
*svm
)
970 u32
*msrpm
= svm
->msrpm
;
972 svm
->vmcb
->control
.virt_ext
&= ~LBR_CTL_ENABLE_MASK
;
973 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHFROMIP
, 0, 0);
974 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHTOIP
, 0, 0);
975 set_msr_interception(msrpm
, MSR_IA32_LASTINTFROMIP
, 0, 0);
976 set_msr_interception(msrpm
, MSR_IA32_LASTINTTOIP
, 0, 0);
979 static void disable_nmi_singlestep(struct vcpu_svm
*svm
)
981 svm
->nmi_singlestep
= false;
982 if (!(svm
->vcpu
.guest_debug
& KVM_GUESTDBG_SINGLESTEP
)) {
983 /* Clear our flags if they were not set by the guest */
984 if (!(svm
->nmi_singlestep_guest_rflags
& X86_EFLAGS_TF
))
985 svm
->vmcb
->save
.rflags
&= ~X86_EFLAGS_TF
;
986 if (!(svm
->nmi_singlestep_guest_rflags
& X86_EFLAGS_RF
))
987 svm
->vmcb
->save
.rflags
&= ~X86_EFLAGS_RF
;
992 * This hash table is used to map VM_ID to a struct kvm_arch,
993 * when handling AMD IOMMU GALOG notification to schedule in
996 #define SVM_VM_DATA_HASH_BITS 8
997 static DEFINE_HASHTABLE(svm_vm_data_hash
, SVM_VM_DATA_HASH_BITS
);
998 static DEFINE_SPINLOCK(svm_vm_data_hash_lock
);
1001 * This function is called from IOMMU driver to notify
1002 * SVM to schedule in a particular vCPU of a particular VM.
1004 static int avic_ga_log_notifier(u32 ga_tag
)
1006 unsigned long flags
;
1007 struct kvm_arch
*ka
= NULL
;
1008 struct kvm_vcpu
*vcpu
= NULL
;
1009 u32 vm_id
= AVIC_GATAG_TO_VMID(ga_tag
);
1010 u32 vcpu_id
= AVIC_GATAG_TO_VCPUID(ga_tag
);
1012 pr_debug("SVM: %s: vm_id=%#x, vcpu_id=%#x\n", __func__
, vm_id
, vcpu_id
);
1014 spin_lock_irqsave(&svm_vm_data_hash_lock
, flags
);
1015 hash_for_each_possible(svm_vm_data_hash
, ka
, hnode
, vm_id
) {
1016 struct kvm
*kvm
= container_of(ka
, struct kvm
, arch
);
1017 struct kvm_arch
*vm_data
= &kvm
->arch
;
1019 if (vm_data
->avic_vm_id
!= vm_id
)
1021 vcpu
= kvm_get_vcpu_by_id(kvm
, vcpu_id
);
1024 spin_unlock_irqrestore(&svm_vm_data_hash_lock
, flags
);
1030 * At this point, the IOMMU should have already set the pending
1031 * bit in the vAPIC backing page. So, we just need to schedule
1034 if (vcpu
->mode
== OUTSIDE_GUEST_MODE
)
1035 kvm_vcpu_wake_up(vcpu
);
1040 static __init
int svm_hardware_setup(void)
1043 struct page
*iopm_pages
;
1047 iopm_pages
= alloc_pages(GFP_KERNEL
, IOPM_ALLOC_ORDER
);
1052 iopm_va
= page_address(iopm_pages
);
1053 memset(iopm_va
, 0xff, PAGE_SIZE
* (1 << IOPM_ALLOC_ORDER
));
1054 iopm_base
= page_to_pfn(iopm_pages
) << PAGE_SHIFT
;
1056 init_msrpm_offsets();
1058 if (boot_cpu_has(X86_FEATURE_NX
))
1059 kvm_enable_efer_bits(EFER_NX
);
1061 if (boot_cpu_has(X86_FEATURE_FXSR_OPT
))
1062 kvm_enable_efer_bits(EFER_FFXSR
);
1064 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR
)) {
1065 kvm_has_tsc_control
= true;
1066 kvm_max_tsc_scaling_ratio
= TSC_RATIO_MAX
;
1067 kvm_tsc_scaling_ratio_frac_bits
= 32;
1071 printk(KERN_INFO
"kvm: Nested Virtualization enabled\n");
1072 kvm_enable_efer_bits(EFER_SVME
| EFER_LMSLE
);
1075 for_each_possible_cpu(cpu
) {
1076 r
= svm_cpu_init(cpu
);
1081 if (!boot_cpu_has(X86_FEATURE_NPT
))
1082 npt_enabled
= false;
1084 if (npt_enabled
&& !npt
) {
1085 printk(KERN_INFO
"kvm: Nested Paging disabled\n");
1086 npt_enabled
= false;
1090 printk(KERN_INFO
"kvm: Nested Paging enabled\n");
1097 !boot_cpu_has(X86_FEATURE_AVIC
) ||
1098 !IS_ENABLED(CONFIG_X86_LOCAL_APIC
)) {
1101 pr_info("AVIC enabled\n");
1103 amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier
);
1109 !boot_cpu_has(X86_FEATURE_V_VMSAVE_VMLOAD
) ||
1110 !IS_ENABLED(CONFIG_X86_64
)) {
1113 pr_info("Virtual VMLOAD VMSAVE supported\n");
1120 __free_pages(iopm_pages
, IOPM_ALLOC_ORDER
);
1125 static __exit
void svm_hardware_unsetup(void)
1129 for_each_possible_cpu(cpu
)
1130 svm_cpu_uninit(cpu
);
1132 __free_pages(pfn_to_page(iopm_base
>> PAGE_SHIFT
), IOPM_ALLOC_ORDER
);
1136 static void init_seg(struct vmcb_seg
*seg
)
1139 seg
->attrib
= SVM_SELECTOR_P_MASK
| SVM_SELECTOR_S_MASK
|
1140 SVM_SELECTOR_WRITE_MASK
; /* Read/Write Data Segment */
1141 seg
->limit
= 0xffff;
1145 static void init_sys_seg(struct vmcb_seg
*seg
, uint32_t type
)
1148 seg
->attrib
= SVM_SELECTOR_P_MASK
| type
;
1149 seg
->limit
= 0xffff;
1153 static void svm_write_tsc_offset(struct kvm_vcpu
*vcpu
, u64 offset
)
1155 struct vcpu_svm
*svm
= to_svm(vcpu
);
1156 u64 g_tsc_offset
= 0;
1158 if (is_guest_mode(vcpu
)) {
1159 g_tsc_offset
= svm
->vmcb
->control
.tsc_offset
-
1160 svm
->nested
.hsave
->control
.tsc_offset
;
1161 svm
->nested
.hsave
->control
.tsc_offset
= offset
;
1163 trace_kvm_write_tsc_offset(vcpu
->vcpu_id
,
1164 svm
->vmcb
->control
.tsc_offset
,
1167 svm
->vmcb
->control
.tsc_offset
= offset
+ g_tsc_offset
;
1169 mark_dirty(svm
->vmcb
, VMCB_INTERCEPTS
);
1172 static void avic_init_vmcb(struct vcpu_svm
*svm
)
1174 struct vmcb
*vmcb
= svm
->vmcb
;
1175 struct kvm_arch
*vm_data
= &svm
->vcpu
.kvm
->arch
;
1176 phys_addr_t bpa
= page_to_phys(svm
->avic_backing_page
);
1177 phys_addr_t lpa
= page_to_phys(vm_data
->avic_logical_id_table_page
);
1178 phys_addr_t ppa
= page_to_phys(vm_data
->avic_physical_id_table_page
);
1180 vmcb
->control
.avic_backing_page
= bpa
& AVIC_HPA_MASK
;
1181 vmcb
->control
.avic_logical_id
= lpa
& AVIC_HPA_MASK
;
1182 vmcb
->control
.avic_physical_id
= ppa
& AVIC_HPA_MASK
;
1183 vmcb
->control
.avic_physical_id
|= AVIC_MAX_PHYSICAL_ID_COUNT
;
1184 vmcb
->control
.int_ctl
|= AVIC_ENABLE_MASK
;
1185 svm
->vcpu
.arch
.apicv_active
= true;
1188 static void init_vmcb(struct vcpu_svm
*svm
)
1190 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
1191 struct vmcb_save_area
*save
= &svm
->vmcb
->save
;
1193 svm
->vcpu
.arch
.hflags
= 0;
1195 set_cr_intercept(svm
, INTERCEPT_CR0_READ
);
1196 set_cr_intercept(svm
, INTERCEPT_CR3_READ
);
1197 set_cr_intercept(svm
, INTERCEPT_CR4_READ
);
1198 set_cr_intercept(svm
, INTERCEPT_CR0_WRITE
);
1199 set_cr_intercept(svm
, INTERCEPT_CR3_WRITE
);
1200 set_cr_intercept(svm
, INTERCEPT_CR4_WRITE
);
1201 if (!kvm_vcpu_apicv_active(&svm
->vcpu
))
1202 set_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
1204 set_dr_intercepts(svm
);
1206 set_exception_intercept(svm
, PF_VECTOR
);
1207 set_exception_intercept(svm
, UD_VECTOR
);
1208 set_exception_intercept(svm
, MC_VECTOR
);
1209 set_exception_intercept(svm
, AC_VECTOR
);
1210 set_exception_intercept(svm
, DB_VECTOR
);
1212 set_intercept(svm
, INTERCEPT_INTR
);
1213 set_intercept(svm
, INTERCEPT_NMI
);
1214 set_intercept(svm
, INTERCEPT_SMI
);
1215 set_intercept(svm
, INTERCEPT_SELECTIVE_CR0
);
1216 set_intercept(svm
, INTERCEPT_RDPMC
);
1217 set_intercept(svm
, INTERCEPT_CPUID
);
1218 set_intercept(svm
, INTERCEPT_INVD
);
1219 set_intercept(svm
, INTERCEPT_HLT
);
1220 set_intercept(svm
, INTERCEPT_INVLPG
);
1221 set_intercept(svm
, INTERCEPT_INVLPGA
);
1222 set_intercept(svm
, INTERCEPT_IOIO_PROT
);
1223 set_intercept(svm
, INTERCEPT_MSR_PROT
);
1224 set_intercept(svm
, INTERCEPT_TASK_SWITCH
);
1225 set_intercept(svm
, INTERCEPT_SHUTDOWN
);
1226 set_intercept(svm
, INTERCEPT_VMRUN
);
1227 set_intercept(svm
, INTERCEPT_VMMCALL
);
1228 set_intercept(svm
, INTERCEPT_VMLOAD
);
1229 set_intercept(svm
, INTERCEPT_VMSAVE
);
1230 set_intercept(svm
, INTERCEPT_STGI
);
1231 set_intercept(svm
, INTERCEPT_CLGI
);
1232 set_intercept(svm
, INTERCEPT_SKINIT
);
1233 set_intercept(svm
, INTERCEPT_WBINVD
);
1234 set_intercept(svm
, INTERCEPT_XSETBV
);
1236 if (!kvm_mwait_in_guest()) {
1237 set_intercept(svm
, INTERCEPT_MONITOR
);
1238 set_intercept(svm
, INTERCEPT_MWAIT
);
1241 control
->iopm_base_pa
= iopm_base
;
1242 control
->msrpm_base_pa
= __pa(svm
->msrpm
);
1243 control
->int_ctl
= V_INTR_MASKING_MASK
;
1245 init_seg(&save
->es
);
1246 init_seg(&save
->ss
);
1247 init_seg(&save
->ds
);
1248 init_seg(&save
->fs
);
1249 init_seg(&save
->gs
);
1251 save
->cs
.selector
= 0xf000;
1252 save
->cs
.base
= 0xffff0000;
1253 /* Executable/Readable Code Segment */
1254 save
->cs
.attrib
= SVM_SELECTOR_READ_MASK
| SVM_SELECTOR_P_MASK
|
1255 SVM_SELECTOR_S_MASK
| SVM_SELECTOR_CODE_MASK
;
1256 save
->cs
.limit
= 0xffff;
1258 save
->gdtr
.limit
= 0xffff;
1259 save
->idtr
.limit
= 0xffff;
1261 init_sys_seg(&save
->ldtr
, SEG_TYPE_LDT
);
1262 init_sys_seg(&save
->tr
, SEG_TYPE_BUSY_TSS16
);
1264 svm_set_efer(&svm
->vcpu
, 0);
1265 save
->dr6
= 0xffff0ff0;
1266 kvm_set_rflags(&svm
->vcpu
, 2);
1267 save
->rip
= 0x0000fff0;
1268 svm
->vcpu
.arch
.regs
[VCPU_REGS_RIP
] = save
->rip
;
1271 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1272 * It also updates the guest-visible cr0 value.
1274 svm_set_cr0(&svm
->vcpu
, X86_CR0_NW
| X86_CR0_CD
| X86_CR0_ET
);
1275 kvm_mmu_reset_context(&svm
->vcpu
);
1277 save
->cr4
= X86_CR4_PAE
;
1281 /* Setup VMCB for Nested Paging */
1282 control
->nested_ctl
= 1;
1283 clr_intercept(svm
, INTERCEPT_INVLPG
);
1284 clr_exception_intercept(svm
, PF_VECTOR
);
1285 clr_cr_intercept(svm
, INTERCEPT_CR3_READ
);
1286 clr_cr_intercept(svm
, INTERCEPT_CR3_WRITE
);
1287 save
->g_pat
= svm
->vcpu
.arch
.pat
;
1291 svm
->asid_generation
= 0;
1293 svm
->nested
.vmcb
= 0;
1294 svm
->vcpu
.arch
.hflags
= 0;
1296 if (boot_cpu_has(X86_FEATURE_PAUSEFILTER
)) {
1297 control
->pause_filter_count
= 3000;
1298 set_intercept(svm
, INTERCEPT_PAUSE
);
1302 avic_init_vmcb(svm
);
1305 * If hardware supports Virtual VMLOAD VMSAVE then enable it
1306 * in VMCB and clear intercepts to avoid #VMEXIT.
1309 clr_intercept(svm
, INTERCEPT_VMLOAD
);
1310 clr_intercept(svm
, INTERCEPT_VMSAVE
);
1311 svm
->vmcb
->control
.virt_ext
|= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK
;
1314 mark_all_dirty(svm
->vmcb
);
1320 static u64
*avic_get_physical_id_entry(struct kvm_vcpu
*vcpu
,
1323 u64
*avic_physical_id_table
;
1324 struct kvm_arch
*vm_data
= &vcpu
->kvm
->arch
;
1326 if (index
>= AVIC_MAX_PHYSICAL_ID_COUNT
)
1329 avic_physical_id_table
= page_address(vm_data
->avic_physical_id_table_page
);
1331 return &avic_physical_id_table
[index
];
1336 * AVIC hardware walks the nested page table to check permissions,
1337 * but does not use the SPA address specified in the leaf page
1338 * table entry since it uses address in the AVIC_BACKING_PAGE pointer
1339 * field of the VMCB. Therefore, we set up the
1340 * APIC_ACCESS_PAGE_PRIVATE_MEMSLOT (4KB) here.
1342 static int avic_init_access_page(struct kvm_vcpu
*vcpu
)
1344 struct kvm
*kvm
= vcpu
->kvm
;
1347 if (kvm
->arch
.apic_access_page_done
)
1350 ret
= x86_set_memory_region(kvm
,
1351 APIC_ACCESS_PAGE_PRIVATE_MEMSLOT
,
1352 APIC_DEFAULT_PHYS_BASE
,
1357 kvm
->arch
.apic_access_page_done
= true;
1361 static int avic_init_backing_page(struct kvm_vcpu
*vcpu
)
1364 u64
*entry
, new_entry
;
1365 int id
= vcpu
->vcpu_id
;
1366 struct vcpu_svm
*svm
= to_svm(vcpu
);
1368 ret
= avic_init_access_page(vcpu
);
1372 if (id
>= AVIC_MAX_PHYSICAL_ID_COUNT
)
1375 if (!svm
->vcpu
.arch
.apic
->regs
)
1378 svm
->avic_backing_page
= virt_to_page(svm
->vcpu
.arch
.apic
->regs
);
1380 /* Setting AVIC backing page address in the phy APIC ID table */
1381 entry
= avic_get_physical_id_entry(vcpu
, id
);
1385 new_entry
= READ_ONCE(*entry
);
1386 new_entry
= (page_to_phys(svm
->avic_backing_page
) &
1387 AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK
) |
1388 AVIC_PHYSICAL_ID_ENTRY_VALID_MASK
;
1389 WRITE_ONCE(*entry
, new_entry
);
1391 svm
->avic_physical_id_cache
= entry
;
1396 static inline int avic_get_next_vm_id(void)
1400 spin_lock(&avic_vm_id_lock
);
1402 /* AVIC VM ID is one-based. */
1403 id
= find_next_zero_bit(avic_vm_id_bitmap
, AVIC_VM_ID_NR
, 1);
1404 if (id
<= AVIC_VM_ID_MASK
)
1405 __set_bit(id
, avic_vm_id_bitmap
);
1409 spin_unlock(&avic_vm_id_lock
);
1413 static inline int avic_free_vm_id(int id
)
1415 if (id
<= 0 || id
> AVIC_VM_ID_MASK
)
1418 spin_lock(&avic_vm_id_lock
);
1419 __clear_bit(id
, avic_vm_id_bitmap
);
1420 spin_unlock(&avic_vm_id_lock
);
1424 static void avic_vm_destroy(struct kvm
*kvm
)
1426 unsigned long flags
;
1427 struct kvm_arch
*vm_data
= &kvm
->arch
;
1432 avic_free_vm_id(vm_data
->avic_vm_id
);
1434 if (vm_data
->avic_logical_id_table_page
)
1435 __free_page(vm_data
->avic_logical_id_table_page
);
1436 if (vm_data
->avic_physical_id_table_page
)
1437 __free_page(vm_data
->avic_physical_id_table_page
);
1439 spin_lock_irqsave(&svm_vm_data_hash_lock
, flags
);
1440 hash_del(&vm_data
->hnode
);
1441 spin_unlock_irqrestore(&svm_vm_data_hash_lock
, flags
);
1444 static int avic_vm_init(struct kvm
*kvm
)
1446 unsigned long flags
;
1447 int vm_id
, err
= -ENOMEM
;
1448 struct kvm_arch
*vm_data
= &kvm
->arch
;
1449 struct page
*p_page
;
1450 struct page
*l_page
;
1455 vm_id
= avic_get_next_vm_id();
1458 vm_data
->avic_vm_id
= (u32
)vm_id
;
1460 /* Allocating physical APIC ID table (4KB) */
1461 p_page
= alloc_page(GFP_KERNEL
);
1465 vm_data
->avic_physical_id_table_page
= p_page
;
1466 clear_page(page_address(p_page
));
1468 /* Allocating logical APIC ID table (4KB) */
1469 l_page
= alloc_page(GFP_KERNEL
);
1473 vm_data
->avic_logical_id_table_page
= l_page
;
1474 clear_page(page_address(l_page
));
1476 spin_lock_irqsave(&svm_vm_data_hash_lock
, flags
);
1477 hash_add(svm_vm_data_hash
, &vm_data
->hnode
, vm_data
->avic_vm_id
);
1478 spin_unlock_irqrestore(&svm_vm_data_hash_lock
, flags
);
1483 avic_vm_destroy(kvm
);
1488 avic_update_iommu_vcpu_affinity(struct kvm_vcpu
*vcpu
, int cpu
, bool r
)
1491 unsigned long flags
;
1492 struct amd_svm_iommu_ir
*ir
;
1493 struct vcpu_svm
*svm
= to_svm(vcpu
);
1495 if (!kvm_arch_has_assigned_device(vcpu
->kvm
))
1499 * Here, we go through the per-vcpu ir_list to update all existing
1500 * interrupt remapping table entry targeting this vcpu.
1502 spin_lock_irqsave(&svm
->ir_list_lock
, flags
);
1504 if (list_empty(&svm
->ir_list
))
1507 list_for_each_entry(ir
, &svm
->ir_list
, node
) {
1508 ret
= amd_iommu_update_ga(cpu
, r
, ir
->data
);
1513 spin_unlock_irqrestore(&svm
->ir_list_lock
, flags
);
1517 static void avic_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
1520 /* ID = 0xff (broadcast), ID > 0xff (reserved) */
1521 int h_physical_id
= kvm_cpu_get_apicid(cpu
);
1522 struct vcpu_svm
*svm
= to_svm(vcpu
);
1524 if (!kvm_vcpu_apicv_active(vcpu
))
1527 if (WARN_ON(h_physical_id
>= AVIC_MAX_PHYSICAL_ID_COUNT
))
1530 entry
= READ_ONCE(*(svm
->avic_physical_id_cache
));
1531 WARN_ON(entry
& AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK
);
1533 entry
&= ~AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK
;
1534 entry
|= (h_physical_id
& AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK
);
1536 entry
&= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK
;
1537 if (svm
->avic_is_running
)
1538 entry
|= AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK
;
1540 WRITE_ONCE(*(svm
->avic_physical_id_cache
), entry
);
1541 avic_update_iommu_vcpu_affinity(vcpu
, h_physical_id
,
1542 svm
->avic_is_running
);
1545 static void avic_vcpu_put(struct kvm_vcpu
*vcpu
)
1548 struct vcpu_svm
*svm
= to_svm(vcpu
);
1550 if (!kvm_vcpu_apicv_active(vcpu
))
1553 entry
= READ_ONCE(*(svm
->avic_physical_id_cache
));
1554 if (entry
& AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK
)
1555 avic_update_iommu_vcpu_affinity(vcpu
, -1, 0);
1557 entry
&= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK
;
1558 WRITE_ONCE(*(svm
->avic_physical_id_cache
), entry
);
1562 * This function is called during VCPU halt/unhalt.
1564 static void avic_set_running(struct kvm_vcpu
*vcpu
, bool is_run
)
1566 struct vcpu_svm
*svm
= to_svm(vcpu
);
1568 svm
->avic_is_running
= is_run
;
1570 avic_vcpu_load(vcpu
, vcpu
->cpu
);
1572 avic_vcpu_put(vcpu
);
1575 static void svm_vcpu_reset(struct kvm_vcpu
*vcpu
, bool init_event
)
1577 struct vcpu_svm
*svm
= to_svm(vcpu
);
1582 svm
->vcpu
.arch
.apic_base
= APIC_DEFAULT_PHYS_BASE
|
1583 MSR_IA32_APICBASE_ENABLE
;
1584 if (kvm_vcpu_is_reset_bsp(&svm
->vcpu
))
1585 svm
->vcpu
.arch
.apic_base
|= MSR_IA32_APICBASE_BSP
;
1589 kvm_cpuid(vcpu
, &eax
, &dummy
, &dummy
, &dummy
);
1590 kvm_register_write(vcpu
, VCPU_REGS_RDX
, eax
);
1592 if (kvm_vcpu_apicv_active(vcpu
) && !init_event
)
1593 avic_update_vapic_bar(svm
, APIC_DEFAULT_PHYS_BASE
);
1596 static struct kvm_vcpu
*svm_create_vcpu(struct kvm
*kvm
, unsigned int id
)
1598 struct vcpu_svm
*svm
;
1600 struct page
*msrpm_pages
;
1601 struct page
*hsave_page
;
1602 struct page
*nested_msrpm_pages
;
1605 svm
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
1611 err
= kvm_vcpu_init(&svm
->vcpu
, kvm
, id
);
1616 page
= alloc_page(GFP_KERNEL
);
1620 msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
1624 nested_msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
1625 if (!nested_msrpm_pages
)
1628 hsave_page
= alloc_page(GFP_KERNEL
);
1633 err
= avic_init_backing_page(&svm
->vcpu
);
1637 INIT_LIST_HEAD(&svm
->ir_list
);
1638 spin_lock_init(&svm
->ir_list_lock
);
1641 /* We initialize this flag to true to make sure that the is_running
1642 * bit would be set the first time the vcpu is loaded.
1644 svm
->avic_is_running
= true;
1646 svm
->nested
.hsave
= page_address(hsave_page
);
1648 svm
->msrpm
= page_address(msrpm_pages
);
1649 svm_vcpu_init_msrpm(svm
->msrpm
);
1651 svm
->nested
.msrpm
= page_address(nested_msrpm_pages
);
1652 svm_vcpu_init_msrpm(svm
->nested
.msrpm
);
1654 svm
->vmcb
= page_address(page
);
1655 clear_page(svm
->vmcb
);
1656 svm
->vmcb_pa
= page_to_pfn(page
) << PAGE_SHIFT
;
1657 svm
->asid_generation
= 0;
1660 svm_init_osvw(&svm
->vcpu
);
1665 __free_page(hsave_page
);
1667 __free_pages(nested_msrpm_pages
, MSRPM_ALLOC_ORDER
);
1669 __free_pages(msrpm_pages
, MSRPM_ALLOC_ORDER
);
1673 kvm_vcpu_uninit(&svm
->vcpu
);
1675 kmem_cache_free(kvm_vcpu_cache
, svm
);
1677 return ERR_PTR(err
);
1680 static void svm_free_vcpu(struct kvm_vcpu
*vcpu
)
1682 struct vcpu_svm
*svm
= to_svm(vcpu
);
1684 __free_page(pfn_to_page(svm
->vmcb_pa
>> PAGE_SHIFT
));
1685 __free_pages(virt_to_page(svm
->msrpm
), MSRPM_ALLOC_ORDER
);
1686 __free_page(virt_to_page(svm
->nested
.hsave
));
1687 __free_pages(virt_to_page(svm
->nested
.msrpm
), MSRPM_ALLOC_ORDER
);
1688 kvm_vcpu_uninit(vcpu
);
1689 kmem_cache_free(kvm_vcpu_cache
, svm
);
1692 * The VMCB could be recycled, causing a false negative in svm_vcpu_load;
1693 * block speculative execution.
1696 wrmsrl(MSR_IA32_PRED_CMD
, FEATURE_SET_IBPB
);
1699 static void svm_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
1701 struct vcpu_svm
*svm
= to_svm(vcpu
);
1702 struct svm_cpu_data
*sd
= per_cpu(svm_data
, cpu
);
1705 if (unlikely(cpu
!= vcpu
->cpu
)) {
1706 svm
->asid_generation
= 0;
1707 mark_all_dirty(svm
->vmcb
);
1710 #ifdef CONFIG_X86_64
1711 rdmsrl(MSR_GS_BASE
, to_svm(vcpu
)->host
.gs_base
);
1713 savesegment(fs
, svm
->host
.fs
);
1714 savesegment(gs
, svm
->host
.gs
);
1715 svm
->host
.ldt
= kvm_read_ldt();
1717 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
1718 rdmsrl(host_save_user_msrs
[i
], svm
->host_user_msrs
[i
]);
1720 if (static_cpu_has(X86_FEATURE_TSCRATEMSR
)) {
1721 u64 tsc_ratio
= vcpu
->arch
.tsc_scaling_ratio
;
1722 if (tsc_ratio
!= __this_cpu_read(current_tsc_ratio
)) {
1723 __this_cpu_write(current_tsc_ratio
, tsc_ratio
);
1724 wrmsrl(MSR_AMD64_TSC_RATIO
, tsc_ratio
);
1727 /* This assumes that the kernel never uses MSR_TSC_AUX */
1728 if (static_cpu_has(X86_FEATURE_RDTSCP
))
1729 wrmsrl(MSR_TSC_AUX
, svm
->tsc_aux
);
1731 if (sd
->current_vmcb
!= svm
->vmcb
) {
1732 sd
->current_vmcb
= svm
->vmcb
;
1734 wrmsrl(MSR_IA32_PRED_CMD
, FEATURE_SET_IBPB
);
1737 avic_vcpu_load(vcpu
, cpu
);
1740 static void svm_vcpu_put(struct kvm_vcpu
*vcpu
)
1742 struct vcpu_svm
*svm
= to_svm(vcpu
);
1745 avic_vcpu_put(vcpu
);
1747 ++vcpu
->stat
.host_state_reload
;
1748 kvm_load_ldt(svm
->host
.ldt
);
1749 #ifdef CONFIG_X86_64
1750 loadsegment(fs
, svm
->host
.fs
);
1751 wrmsrl(MSR_KERNEL_GS_BASE
, current
->thread
.gsbase
);
1752 load_gs_index(svm
->host
.gs
);
1754 #ifdef CONFIG_X86_32_LAZY_GS
1755 loadsegment(gs
, svm
->host
.gs
);
1758 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
1759 wrmsrl(host_save_user_msrs
[i
], svm
->host_user_msrs
[i
]);
1762 static void svm_vcpu_blocking(struct kvm_vcpu
*vcpu
)
1764 avic_set_running(vcpu
, false);
1767 static void svm_vcpu_unblocking(struct kvm_vcpu
*vcpu
)
1769 avic_set_running(vcpu
, true);
1772 static unsigned long svm_get_rflags(struct kvm_vcpu
*vcpu
)
1774 struct vcpu_svm
*svm
= to_svm(vcpu
);
1775 unsigned long rflags
= svm
->vmcb
->save
.rflags
;
1777 if (svm
->nmi_singlestep
) {
1778 /* Hide our flags if they were not set by the guest */
1779 if (!(svm
->nmi_singlestep_guest_rflags
& X86_EFLAGS_TF
))
1780 rflags
&= ~X86_EFLAGS_TF
;
1781 if (!(svm
->nmi_singlestep_guest_rflags
& X86_EFLAGS_RF
))
1782 rflags
&= ~X86_EFLAGS_RF
;
1787 static void svm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
1789 if (to_svm(vcpu
)->nmi_singlestep
)
1790 rflags
|= (X86_EFLAGS_TF
| X86_EFLAGS_RF
);
1793 * Any change of EFLAGS.VM is accompanied by a reload of SS
1794 * (caused by either a task switch or an inter-privilege IRET),
1795 * so we do not need to update the CPL here.
1797 to_svm(vcpu
)->vmcb
->save
.rflags
= rflags
;
1800 static void svm_cache_reg(struct kvm_vcpu
*vcpu
, enum kvm_reg reg
)
1803 case VCPU_EXREG_PDPTR
:
1804 BUG_ON(!npt_enabled
);
1805 load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, kvm_read_cr3(vcpu
));
1812 static void svm_set_vintr(struct vcpu_svm
*svm
)
1814 set_intercept(svm
, INTERCEPT_VINTR
);
1817 static void svm_clear_vintr(struct vcpu_svm
*svm
)
1819 clr_intercept(svm
, INTERCEPT_VINTR
);
1822 static struct vmcb_seg
*svm_seg(struct kvm_vcpu
*vcpu
, int seg
)
1824 struct vmcb_save_area
*save
= &to_svm(vcpu
)->vmcb
->save
;
1827 case VCPU_SREG_CS
: return &save
->cs
;
1828 case VCPU_SREG_DS
: return &save
->ds
;
1829 case VCPU_SREG_ES
: return &save
->es
;
1830 case VCPU_SREG_FS
: return &save
->fs
;
1831 case VCPU_SREG_GS
: return &save
->gs
;
1832 case VCPU_SREG_SS
: return &save
->ss
;
1833 case VCPU_SREG_TR
: return &save
->tr
;
1834 case VCPU_SREG_LDTR
: return &save
->ldtr
;
1840 static u64
svm_get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
1842 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
1847 static void svm_get_segment(struct kvm_vcpu
*vcpu
,
1848 struct kvm_segment
*var
, int seg
)
1850 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
1852 var
->base
= s
->base
;
1853 var
->limit
= s
->limit
;
1854 var
->selector
= s
->selector
;
1855 var
->type
= s
->attrib
& SVM_SELECTOR_TYPE_MASK
;
1856 var
->s
= (s
->attrib
>> SVM_SELECTOR_S_SHIFT
) & 1;
1857 var
->dpl
= (s
->attrib
>> SVM_SELECTOR_DPL_SHIFT
) & 3;
1858 var
->present
= (s
->attrib
>> SVM_SELECTOR_P_SHIFT
) & 1;
1859 var
->avl
= (s
->attrib
>> SVM_SELECTOR_AVL_SHIFT
) & 1;
1860 var
->l
= (s
->attrib
>> SVM_SELECTOR_L_SHIFT
) & 1;
1861 var
->db
= (s
->attrib
>> SVM_SELECTOR_DB_SHIFT
) & 1;
1864 * AMD CPUs circa 2014 track the G bit for all segments except CS.
1865 * However, the SVM spec states that the G bit is not observed by the
1866 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
1867 * So let's synthesize a legal G bit for all segments, this helps
1868 * running KVM nested. It also helps cross-vendor migration, because
1869 * Intel's vmentry has a check on the 'G' bit.
1871 var
->g
= s
->limit
> 0xfffff;
1874 * AMD's VMCB does not have an explicit unusable field, so emulate it
1875 * for cross vendor migration purposes by "not present"
1877 var
->unusable
= !var
->present
;
1882 * Work around a bug where the busy flag in the tr selector
1892 * The accessed bit must always be set in the segment
1893 * descriptor cache, although it can be cleared in the
1894 * descriptor, the cached bit always remains at 1. Since
1895 * Intel has a check on this, set it here to support
1896 * cross-vendor migration.
1903 * On AMD CPUs sometimes the DB bit in the segment
1904 * descriptor is left as 1, although the whole segment has
1905 * been made unusable. Clear it here to pass an Intel VMX
1906 * entry check when cross vendor migrating.
1910 /* This is symmetric with svm_set_segment() */
1911 var
->dpl
= to_svm(vcpu
)->vmcb
->save
.cpl
;
1916 static int svm_get_cpl(struct kvm_vcpu
*vcpu
)
1918 struct vmcb_save_area
*save
= &to_svm(vcpu
)->vmcb
->save
;
1923 static void svm_get_idt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1925 struct vcpu_svm
*svm
= to_svm(vcpu
);
1927 dt
->size
= svm
->vmcb
->save
.idtr
.limit
;
1928 dt
->address
= svm
->vmcb
->save
.idtr
.base
;
1931 static void svm_set_idt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1933 struct vcpu_svm
*svm
= to_svm(vcpu
);
1935 svm
->vmcb
->save
.idtr
.limit
= dt
->size
;
1936 svm
->vmcb
->save
.idtr
.base
= dt
->address
;
1937 mark_dirty(svm
->vmcb
, VMCB_DT
);
1940 static void svm_get_gdt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1942 struct vcpu_svm
*svm
= to_svm(vcpu
);
1944 dt
->size
= svm
->vmcb
->save
.gdtr
.limit
;
1945 dt
->address
= svm
->vmcb
->save
.gdtr
.base
;
1948 static void svm_set_gdt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1950 struct vcpu_svm
*svm
= to_svm(vcpu
);
1952 svm
->vmcb
->save
.gdtr
.limit
= dt
->size
;
1953 svm
->vmcb
->save
.gdtr
.base
= dt
->address
;
1954 mark_dirty(svm
->vmcb
, VMCB_DT
);
1957 static void svm_decache_cr0_guest_bits(struct kvm_vcpu
*vcpu
)
1961 static void svm_decache_cr3(struct kvm_vcpu
*vcpu
)
1965 static void svm_decache_cr4_guest_bits(struct kvm_vcpu
*vcpu
)
1969 static void update_cr0_intercept(struct vcpu_svm
*svm
)
1971 ulong gcr0
= svm
->vcpu
.arch
.cr0
;
1972 u64
*hcr0
= &svm
->vmcb
->save
.cr0
;
1974 *hcr0
= (*hcr0
& ~SVM_CR0_SELECTIVE_MASK
)
1975 | (gcr0
& SVM_CR0_SELECTIVE_MASK
);
1977 mark_dirty(svm
->vmcb
, VMCB_CR
);
1979 if (gcr0
== *hcr0
) {
1980 clr_cr_intercept(svm
, INTERCEPT_CR0_READ
);
1981 clr_cr_intercept(svm
, INTERCEPT_CR0_WRITE
);
1983 set_cr_intercept(svm
, INTERCEPT_CR0_READ
);
1984 set_cr_intercept(svm
, INTERCEPT_CR0_WRITE
);
1988 static void svm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
1990 struct vcpu_svm
*svm
= to_svm(vcpu
);
1992 #ifdef CONFIG_X86_64
1993 if (vcpu
->arch
.efer
& EFER_LME
) {
1994 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
1995 vcpu
->arch
.efer
|= EFER_LMA
;
1996 svm
->vmcb
->save
.efer
|= EFER_LMA
| EFER_LME
;
1999 if (is_paging(vcpu
) && !(cr0
& X86_CR0_PG
)) {
2000 vcpu
->arch
.efer
&= ~EFER_LMA
;
2001 svm
->vmcb
->save
.efer
&= ~(EFER_LMA
| EFER_LME
);
2005 vcpu
->arch
.cr0
= cr0
;
2008 cr0
|= X86_CR0_PG
| X86_CR0_WP
;
2011 * re-enable caching here because the QEMU bios
2012 * does not do it - this results in some delay at
2015 if (kvm_check_has_quirk(vcpu
->kvm
, KVM_X86_QUIRK_CD_NW_CLEARED
))
2016 cr0
&= ~(X86_CR0_CD
| X86_CR0_NW
);
2017 svm
->vmcb
->save
.cr0
= cr0
;
2018 mark_dirty(svm
->vmcb
, VMCB_CR
);
2019 update_cr0_intercept(svm
);
2022 static int svm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
2024 unsigned long host_cr4_mce
= cr4_read_shadow() & X86_CR4_MCE
;
2025 unsigned long old_cr4
= to_svm(vcpu
)->vmcb
->save
.cr4
;
2027 if (cr4
& X86_CR4_VMXE
)
2030 if (npt_enabled
&& ((old_cr4
^ cr4
) & X86_CR4_PGE
))
2031 svm_flush_tlb(vcpu
);
2033 vcpu
->arch
.cr4
= cr4
;
2036 cr4
|= host_cr4_mce
;
2037 to_svm(vcpu
)->vmcb
->save
.cr4
= cr4
;
2038 mark_dirty(to_svm(vcpu
)->vmcb
, VMCB_CR
);
2042 static void svm_set_segment(struct kvm_vcpu
*vcpu
,
2043 struct kvm_segment
*var
, int seg
)
2045 struct vcpu_svm
*svm
= to_svm(vcpu
);
2046 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
2048 s
->base
= var
->base
;
2049 s
->limit
= var
->limit
;
2050 s
->selector
= var
->selector
;
2051 s
->attrib
= (var
->type
& SVM_SELECTOR_TYPE_MASK
);
2052 s
->attrib
|= (var
->s
& 1) << SVM_SELECTOR_S_SHIFT
;
2053 s
->attrib
|= (var
->dpl
& 3) << SVM_SELECTOR_DPL_SHIFT
;
2054 s
->attrib
|= ((var
->present
& 1) && !var
->unusable
) << SVM_SELECTOR_P_SHIFT
;
2055 s
->attrib
|= (var
->avl
& 1) << SVM_SELECTOR_AVL_SHIFT
;
2056 s
->attrib
|= (var
->l
& 1) << SVM_SELECTOR_L_SHIFT
;
2057 s
->attrib
|= (var
->db
& 1) << SVM_SELECTOR_DB_SHIFT
;
2058 s
->attrib
|= (var
->g
& 1) << SVM_SELECTOR_G_SHIFT
;
2061 * This is always accurate, except if SYSRET returned to a segment
2062 * with SS.DPL != 3. Intel does not have this quirk, and always
2063 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
2064 * would entail passing the CPL to userspace and back.
2066 if (seg
== VCPU_SREG_SS
)
2067 /* This is symmetric with svm_get_segment() */
2068 svm
->vmcb
->save
.cpl
= (var
->dpl
& 3);
2070 mark_dirty(svm
->vmcb
, VMCB_SEG
);
2073 static void update_bp_intercept(struct kvm_vcpu
*vcpu
)
2075 struct vcpu_svm
*svm
= to_svm(vcpu
);
2077 clr_exception_intercept(svm
, BP_VECTOR
);
2079 if (vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
) {
2080 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_SW_BP
)
2081 set_exception_intercept(svm
, BP_VECTOR
);
2083 vcpu
->guest_debug
= 0;
2086 static void new_asid(struct vcpu_svm
*svm
, struct svm_cpu_data
*sd
)
2088 if (sd
->next_asid
> sd
->max_asid
) {
2089 ++sd
->asid_generation
;
2091 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ALL_ASID
;
2094 svm
->asid_generation
= sd
->asid_generation
;
2095 svm
->vmcb
->control
.asid
= sd
->next_asid
++;
2097 mark_dirty(svm
->vmcb
, VMCB_ASID
);
2100 static u64
svm_get_dr6(struct kvm_vcpu
*vcpu
)
2102 return to_svm(vcpu
)->vmcb
->save
.dr6
;
2105 static void svm_set_dr6(struct kvm_vcpu
*vcpu
, unsigned long value
)
2107 struct vcpu_svm
*svm
= to_svm(vcpu
);
2109 svm
->vmcb
->save
.dr6
= value
;
2110 mark_dirty(svm
->vmcb
, VMCB_DR
);
2113 static void svm_sync_dirty_debug_regs(struct kvm_vcpu
*vcpu
)
2115 struct vcpu_svm
*svm
= to_svm(vcpu
);
2117 get_debugreg(vcpu
->arch
.db
[0], 0);
2118 get_debugreg(vcpu
->arch
.db
[1], 1);
2119 get_debugreg(vcpu
->arch
.db
[2], 2);
2120 get_debugreg(vcpu
->arch
.db
[3], 3);
2121 vcpu
->arch
.dr6
= svm_get_dr6(vcpu
);
2122 vcpu
->arch
.dr7
= svm
->vmcb
->save
.dr7
;
2124 vcpu
->arch
.switch_db_regs
&= ~KVM_DEBUGREG_WONT_EXIT
;
2125 set_dr_intercepts(svm
);
2128 static void svm_set_dr7(struct kvm_vcpu
*vcpu
, unsigned long value
)
2130 struct vcpu_svm
*svm
= to_svm(vcpu
);
2132 svm
->vmcb
->save
.dr7
= value
;
2133 mark_dirty(svm
->vmcb
, VMCB_DR
);
2136 static int pf_interception(struct vcpu_svm
*svm
)
2138 u64 fault_address
= svm
->vmcb
->control
.exit_info_2
;
2139 u64 error_code
= svm
->vmcb
->control
.exit_info_1
;
2141 return kvm_handle_page_fault(&svm
->vcpu
, error_code
, fault_address
,
2142 svm
->vmcb
->control
.insn_bytes
,
2143 svm
->vmcb
->control
.insn_len
, !npt_enabled
);
2146 static int db_interception(struct vcpu_svm
*svm
)
2148 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
2150 if (!(svm
->vcpu
.guest_debug
&
2151 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
)) &&
2152 !svm
->nmi_singlestep
) {
2153 kvm_queue_exception(&svm
->vcpu
, DB_VECTOR
);
2157 if (svm
->nmi_singlestep
) {
2158 disable_nmi_singlestep(svm
);
2161 if (svm
->vcpu
.guest_debug
&
2162 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
)) {
2163 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
2164 kvm_run
->debug
.arch
.pc
=
2165 svm
->vmcb
->save
.cs
.base
+ svm
->vmcb
->save
.rip
;
2166 kvm_run
->debug
.arch
.exception
= DB_VECTOR
;
2173 static int bp_interception(struct vcpu_svm
*svm
)
2175 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
2177 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
2178 kvm_run
->debug
.arch
.pc
= svm
->vmcb
->save
.cs
.base
+ svm
->vmcb
->save
.rip
;
2179 kvm_run
->debug
.arch
.exception
= BP_VECTOR
;
2183 static int ud_interception(struct vcpu_svm
*svm
)
2187 er
= emulate_instruction(&svm
->vcpu
, EMULTYPE_TRAP_UD
);
2188 if (er
!= EMULATE_DONE
)
2189 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
2193 static int ac_interception(struct vcpu_svm
*svm
)
2195 kvm_queue_exception_e(&svm
->vcpu
, AC_VECTOR
, 0);
2199 static bool is_erratum_383(void)
2204 if (!erratum_383_found
)
2207 value
= native_read_msr_safe(MSR_IA32_MC0_STATUS
, &err
);
2211 /* Bit 62 may or may not be set for this mce */
2212 value
&= ~(1ULL << 62);
2214 if (value
!= 0xb600000000010015ULL
)
2217 /* Clear MCi_STATUS registers */
2218 for (i
= 0; i
< 6; ++i
)
2219 native_write_msr_safe(MSR_IA32_MCx_STATUS(i
), 0, 0);
2221 value
= native_read_msr_safe(MSR_IA32_MCG_STATUS
, &err
);
2225 value
&= ~(1ULL << 2);
2226 low
= lower_32_bits(value
);
2227 high
= upper_32_bits(value
);
2229 native_write_msr_safe(MSR_IA32_MCG_STATUS
, low
, high
);
2232 /* Flush tlb to evict multi-match entries */
2238 static void svm_handle_mce(struct vcpu_svm
*svm
)
2240 if (is_erratum_383()) {
2242 * Erratum 383 triggered. Guest state is corrupt so kill the
2245 pr_err("KVM: Guest triggered AMD Erratum 383\n");
2247 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, &svm
->vcpu
);
2253 * On an #MC intercept the MCE handler is not called automatically in
2254 * the host. So do it by hand here.
2258 /* not sure if we ever come back to this point */
2263 static int mc_interception(struct vcpu_svm
*svm
)
2268 static int shutdown_interception(struct vcpu_svm
*svm
)
2270 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
2273 * VMCB is undefined after a SHUTDOWN intercept
2274 * so reinitialize it.
2276 clear_page(svm
->vmcb
);
2279 kvm_run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
2283 static int io_interception(struct vcpu_svm
*svm
)
2285 struct kvm_vcpu
*vcpu
= &svm
->vcpu
;
2286 u32 io_info
= svm
->vmcb
->control
.exit_info_1
; /* address size bug? */
2287 int size
, in
, string
, ret
;
2290 ++svm
->vcpu
.stat
.io_exits
;
2291 string
= (io_info
& SVM_IOIO_STR_MASK
) != 0;
2292 in
= (io_info
& SVM_IOIO_TYPE_MASK
) != 0;
2294 return emulate_instruction(vcpu
, 0) == EMULATE_DONE
;
2296 port
= io_info
>> 16;
2297 size
= (io_info
& SVM_IOIO_SIZE_MASK
) >> SVM_IOIO_SIZE_SHIFT
;
2298 svm
->next_rip
= svm
->vmcb
->control
.exit_info_2
;
2299 ret
= kvm_skip_emulated_instruction(&svm
->vcpu
);
2302 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
2303 * KVM_EXIT_DEBUG here.
2306 return kvm_fast_pio_in(vcpu
, size
, port
) && ret
;
2308 return kvm_fast_pio_out(vcpu
, size
, port
) && ret
;
2311 static int nmi_interception(struct vcpu_svm
*svm
)
2316 static int intr_interception(struct vcpu_svm
*svm
)
2318 ++svm
->vcpu
.stat
.irq_exits
;
2322 static int nop_on_interception(struct vcpu_svm
*svm
)
2327 static int halt_interception(struct vcpu_svm
*svm
)
2329 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 1;
2330 return kvm_emulate_halt(&svm
->vcpu
);
2333 static int vmmcall_interception(struct vcpu_svm
*svm
)
2335 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2336 return kvm_emulate_hypercall(&svm
->vcpu
);
2339 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu
*vcpu
)
2341 struct vcpu_svm
*svm
= to_svm(vcpu
);
2343 return svm
->nested
.nested_cr3
;
2346 static u64
nested_svm_get_tdp_pdptr(struct kvm_vcpu
*vcpu
, int index
)
2348 struct vcpu_svm
*svm
= to_svm(vcpu
);
2349 u64 cr3
= svm
->nested
.nested_cr3
;
2353 ret
= kvm_vcpu_read_guest_page(vcpu
, gpa_to_gfn(cr3
), &pdpte
,
2354 offset_in_page(cr3
) + index
* 8, 8);
2360 static void nested_svm_set_tdp_cr3(struct kvm_vcpu
*vcpu
,
2363 struct vcpu_svm
*svm
= to_svm(vcpu
);
2365 svm
->vmcb
->control
.nested_cr3
= root
;
2366 mark_dirty(svm
->vmcb
, VMCB_NPT
);
2367 svm_flush_tlb(vcpu
);
2370 static void nested_svm_inject_npf_exit(struct kvm_vcpu
*vcpu
,
2371 struct x86_exception
*fault
)
2373 struct vcpu_svm
*svm
= to_svm(vcpu
);
2375 if (svm
->vmcb
->control
.exit_code
!= SVM_EXIT_NPF
) {
2377 * TODO: track the cause of the nested page fault, and
2378 * correctly fill in the high bits of exit_info_1.
2380 svm
->vmcb
->control
.exit_code
= SVM_EXIT_NPF
;
2381 svm
->vmcb
->control
.exit_code_hi
= 0;
2382 svm
->vmcb
->control
.exit_info_1
= (1ULL << 32);
2383 svm
->vmcb
->control
.exit_info_2
= fault
->address
;
2386 svm
->vmcb
->control
.exit_info_1
&= ~0xffffffffULL
;
2387 svm
->vmcb
->control
.exit_info_1
|= fault
->error_code
;
2390 * The present bit is always zero for page structure faults on real
2393 if (svm
->vmcb
->control
.exit_info_1
& (2ULL << 32))
2394 svm
->vmcb
->control
.exit_info_1
&= ~1;
2396 nested_svm_vmexit(svm
);
2399 static void nested_svm_init_mmu_context(struct kvm_vcpu
*vcpu
)
2401 WARN_ON(mmu_is_nested(vcpu
));
2402 kvm_init_shadow_mmu(vcpu
);
2403 vcpu
->arch
.mmu
.set_cr3
= nested_svm_set_tdp_cr3
;
2404 vcpu
->arch
.mmu
.get_cr3
= nested_svm_get_tdp_cr3
;
2405 vcpu
->arch
.mmu
.get_pdptr
= nested_svm_get_tdp_pdptr
;
2406 vcpu
->arch
.mmu
.inject_page_fault
= nested_svm_inject_npf_exit
;
2407 vcpu
->arch
.mmu
.shadow_root_level
= get_npt_level();
2408 reset_shadow_zero_bits_mask(vcpu
, &vcpu
->arch
.mmu
);
2409 vcpu
->arch
.walk_mmu
= &vcpu
->arch
.nested_mmu
;
2412 static void nested_svm_uninit_mmu_context(struct kvm_vcpu
*vcpu
)
2414 vcpu
->arch
.walk_mmu
= &vcpu
->arch
.mmu
;
2417 static int nested_svm_check_permissions(struct vcpu_svm
*svm
)
2419 if (!(svm
->vcpu
.arch
.efer
& EFER_SVME
) ||
2420 !is_paging(&svm
->vcpu
)) {
2421 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
2425 if (svm
->vmcb
->save
.cpl
) {
2426 kvm_inject_gp(&svm
->vcpu
, 0);
2433 static int nested_svm_check_exception(struct vcpu_svm
*svm
, unsigned nr
,
2434 bool has_error_code
, u32 error_code
)
2438 if (!is_guest_mode(&svm
->vcpu
))
2441 vmexit
= nested_svm_intercept(svm
);
2442 if (vmexit
!= NESTED_EXIT_DONE
)
2445 svm
->vmcb
->control
.exit_code
= SVM_EXIT_EXCP_BASE
+ nr
;
2446 svm
->vmcb
->control
.exit_code_hi
= 0;
2447 svm
->vmcb
->control
.exit_info_1
= error_code
;
2450 * FIXME: we should not write CR2 when L1 intercepts an L2 #PF exception.
2451 * The fix is to add the ancillary datum (CR2 or DR6) to structs
2452 * kvm_queued_exception and kvm_vcpu_events, so that CR2 and DR6 can be
2453 * written only when inject_pending_event runs (DR6 would written here
2454 * too). This should be conditional on a new capability---if the
2455 * capability is disabled, kvm_multiple_exception would write the
2456 * ancillary information to CR2 or DR6, for backwards ABI-compatibility.
2458 if (svm
->vcpu
.arch
.exception
.nested_apf
)
2459 svm
->vmcb
->control
.exit_info_2
= svm
->vcpu
.arch
.apf
.nested_apf_token
;
2461 svm
->vmcb
->control
.exit_info_2
= svm
->vcpu
.arch
.cr2
;
2463 svm
->nested
.exit_required
= true;
2467 /* This function returns true if it is save to enable the irq window */
2468 static inline bool nested_svm_intr(struct vcpu_svm
*svm
)
2470 if (!is_guest_mode(&svm
->vcpu
))
2473 if (!(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
))
2476 if (!(svm
->vcpu
.arch
.hflags
& HF_HIF_MASK
))
2480 * if vmexit was already requested (by intercepted exception
2481 * for instance) do not overwrite it with "external interrupt"
2484 if (svm
->nested
.exit_required
)
2487 svm
->vmcb
->control
.exit_code
= SVM_EXIT_INTR
;
2488 svm
->vmcb
->control
.exit_info_1
= 0;
2489 svm
->vmcb
->control
.exit_info_2
= 0;
2491 if (svm
->nested
.intercept
& 1ULL) {
2493 * The #vmexit can't be emulated here directly because this
2494 * code path runs with irqs and preemption disabled. A
2495 * #vmexit emulation might sleep. Only signal request for
2498 svm
->nested
.exit_required
= true;
2499 trace_kvm_nested_intr_vmexit(svm
->vmcb
->save
.rip
);
2506 /* This function returns true if it is save to enable the nmi window */
2507 static inline bool nested_svm_nmi(struct vcpu_svm
*svm
)
2509 if (!is_guest_mode(&svm
->vcpu
))
2512 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_NMI
)))
2515 svm
->vmcb
->control
.exit_code
= SVM_EXIT_NMI
;
2516 svm
->nested
.exit_required
= true;
2521 static void *nested_svm_map(struct vcpu_svm
*svm
, u64 gpa
, struct page
**_page
)
2527 page
= kvm_vcpu_gfn_to_page(&svm
->vcpu
, gpa
>> PAGE_SHIFT
);
2528 if (is_error_page(page
))
2536 kvm_inject_gp(&svm
->vcpu
, 0);
2541 static void nested_svm_unmap(struct page
*page
)
2544 kvm_release_page_dirty(page
);
2547 static int nested_svm_intercept_ioio(struct vcpu_svm
*svm
)
2549 unsigned port
, size
, iopm_len
;
2554 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_IOIO_PROT
)))
2555 return NESTED_EXIT_HOST
;
2557 port
= svm
->vmcb
->control
.exit_info_1
>> 16;
2558 size
= (svm
->vmcb
->control
.exit_info_1
& SVM_IOIO_SIZE_MASK
) >>
2559 SVM_IOIO_SIZE_SHIFT
;
2560 gpa
= svm
->nested
.vmcb_iopm
+ (port
/ 8);
2561 start_bit
= port
% 8;
2562 iopm_len
= (start_bit
+ size
> 8) ? 2 : 1;
2563 mask
= (0xf >> (4 - size
)) << start_bit
;
2566 if (kvm_vcpu_read_guest(&svm
->vcpu
, gpa
, &val
, iopm_len
))
2567 return NESTED_EXIT_DONE
;
2569 return (val
& mask
) ? NESTED_EXIT_DONE
: NESTED_EXIT_HOST
;
2572 static int nested_svm_exit_handled_msr(struct vcpu_svm
*svm
)
2574 u32 offset
, msr
, value
;
2577 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_MSR_PROT
)))
2578 return NESTED_EXIT_HOST
;
2580 msr
= svm
->vcpu
.arch
.regs
[VCPU_REGS_RCX
];
2581 offset
= svm_msrpm_offset(msr
);
2582 write
= svm
->vmcb
->control
.exit_info_1
& 1;
2583 mask
= 1 << ((2 * (msr
& 0xf)) + write
);
2585 if (offset
== MSR_INVALID
)
2586 return NESTED_EXIT_DONE
;
2588 /* Offset is in 32 bit units but need in 8 bit units */
2591 if (kvm_vcpu_read_guest(&svm
->vcpu
, svm
->nested
.vmcb_msrpm
+ offset
, &value
, 4))
2592 return NESTED_EXIT_DONE
;
2594 return (value
& mask
) ? NESTED_EXIT_DONE
: NESTED_EXIT_HOST
;
2597 /* DB exceptions for our internal use must not cause vmexit */
2598 static int nested_svm_intercept_db(struct vcpu_svm
*svm
)
2602 /* if we're not singlestepping, it's not ours */
2603 if (!svm
->nmi_singlestep
)
2604 return NESTED_EXIT_DONE
;
2606 /* if it's not a singlestep exception, it's not ours */
2607 if (kvm_get_dr(&svm
->vcpu
, 6, &dr6
))
2608 return NESTED_EXIT_DONE
;
2609 if (!(dr6
& DR6_BS
))
2610 return NESTED_EXIT_DONE
;
2612 /* if the guest is singlestepping, it should get the vmexit */
2613 if (svm
->nmi_singlestep_guest_rflags
& X86_EFLAGS_TF
) {
2614 disable_nmi_singlestep(svm
);
2615 return NESTED_EXIT_DONE
;
2618 /* it's ours, the nested hypervisor must not see this one */
2619 return NESTED_EXIT_HOST
;
2622 static int nested_svm_exit_special(struct vcpu_svm
*svm
)
2624 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
2626 switch (exit_code
) {
2629 case SVM_EXIT_EXCP_BASE
+ MC_VECTOR
:
2630 return NESTED_EXIT_HOST
;
2632 /* For now we are always handling NPFs when using them */
2634 return NESTED_EXIT_HOST
;
2636 case SVM_EXIT_EXCP_BASE
+ PF_VECTOR
:
2637 /* When we're shadowing, trap PFs, but not async PF */
2638 if (!npt_enabled
&& svm
->vcpu
.arch
.apf
.host_apf_reason
== 0)
2639 return NESTED_EXIT_HOST
;
2645 return NESTED_EXIT_CONTINUE
;
2649 * If this function returns true, this #vmexit was already handled
2651 static int nested_svm_intercept(struct vcpu_svm
*svm
)
2653 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
2654 int vmexit
= NESTED_EXIT_HOST
;
2656 switch (exit_code
) {
2658 vmexit
= nested_svm_exit_handled_msr(svm
);
2661 vmexit
= nested_svm_intercept_ioio(svm
);
2663 case SVM_EXIT_READ_CR0
... SVM_EXIT_WRITE_CR8
: {
2664 u32 bit
= 1U << (exit_code
- SVM_EXIT_READ_CR0
);
2665 if (svm
->nested
.intercept_cr
& bit
)
2666 vmexit
= NESTED_EXIT_DONE
;
2669 case SVM_EXIT_READ_DR0
... SVM_EXIT_WRITE_DR7
: {
2670 u32 bit
= 1U << (exit_code
- SVM_EXIT_READ_DR0
);
2671 if (svm
->nested
.intercept_dr
& bit
)
2672 vmexit
= NESTED_EXIT_DONE
;
2675 case SVM_EXIT_EXCP_BASE
... SVM_EXIT_EXCP_BASE
+ 0x1f: {
2676 u32 excp_bits
= 1 << (exit_code
- SVM_EXIT_EXCP_BASE
);
2677 if (svm
->nested
.intercept_exceptions
& excp_bits
) {
2678 if (exit_code
== SVM_EXIT_EXCP_BASE
+ DB_VECTOR
)
2679 vmexit
= nested_svm_intercept_db(svm
);
2681 vmexit
= NESTED_EXIT_DONE
;
2683 /* async page fault always cause vmexit */
2684 else if ((exit_code
== SVM_EXIT_EXCP_BASE
+ PF_VECTOR
) &&
2685 svm
->vcpu
.arch
.exception
.nested_apf
!= 0)
2686 vmexit
= NESTED_EXIT_DONE
;
2689 case SVM_EXIT_ERR
: {
2690 vmexit
= NESTED_EXIT_DONE
;
2694 u64 exit_bits
= 1ULL << (exit_code
- SVM_EXIT_INTR
);
2695 if (svm
->nested
.intercept
& exit_bits
)
2696 vmexit
= NESTED_EXIT_DONE
;
2703 static int nested_svm_exit_handled(struct vcpu_svm
*svm
)
2707 vmexit
= nested_svm_intercept(svm
);
2709 if (vmexit
== NESTED_EXIT_DONE
)
2710 nested_svm_vmexit(svm
);
2715 static inline void copy_vmcb_control_area(struct vmcb
*dst_vmcb
, struct vmcb
*from_vmcb
)
2717 struct vmcb_control_area
*dst
= &dst_vmcb
->control
;
2718 struct vmcb_control_area
*from
= &from_vmcb
->control
;
2720 dst
->intercept_cr
= from
->intercept_cr
;
2721 dst
->intercept_dr
= from
->intercept_dr
;
2722 dst
->intercept_exceptions
= from
->intercept_exceptions
;
2723 dst
->intercept
= from
->intercept
;
2724 dst
->iopm_base_pa
= from
->iopm_base_pa
;
2725 dst
->msrpm_base_pa
= from
->msrpm_base_pa
;
2726 dst
->tsc_offset
= from
->tsc_offset
;
2727 dst
->asid
= from
->asid
;
2728 dst
->tlb_ctl
= from
->tlb_ctl
;
2729 dst
->int_ctl
= from
->int_ctl
;
2730 dst
->int_vector
= from
->int_vector
;
2731 dst
->int_state
= from
->int_state
;
2732 dst
->exit_code
= from
->exit_code
;
2733 dst
->exit_code_hi
= from
->exit_code_hi
;
2734 dst
->exit_info_1
= from
->exit_info_1
;
2735 dst
->exit_info_2
= from
->exit_info_2
;
2736 dst
->exit_int_info
= from
->exit_int_info
;
2737 dst
->exit_int_info_err
= from
->exit_int_info_err
;
2738 dst
->nested_ctl
= from
->nested_ctl
;
2739 dst
->event_inj
= from
->event_inj
;
2740 dst
->event_inj_err
= from
->event_inj_err
;
2741 dst
->nested_cr3
= from
->nested_cr3
;
2742 dst
->virt_ext
= from
->virt_ext
;
2745 static int nested_svm_vmexit(struct vcpu_svm
*svm
)
2747 struct vmcb
*nested_vmcb
;
2748 struct vmcb
*hsave
= svm
->nested
.hsave
;
2749 struct vmcb
*vmcb
= svm
->vmcb
;
2752 trace_kvm_nested_vmexit_inject(vmcb
->control
.exit_code
,
2753 vmcb
->control
.exit_info_1
,
2754 vmcb
->control
.exit_info_2
,
2755 vmcb
->control
.exit_int_info
,
2756 vmcb
->control
.exit_int_info_err
,
2759 nested_vmcb
= nested_svm_map(svm
, svm
->nested
.vmcb
, &page
);
2763 /* Exit Guest-Mode */
2764 leave_guest_mode(&svm
->vcpu
);
2765 svm
->nested
.vmcb
= 0;
2767 /* Give the current vmcb to the guest */
2770 nested_vmcb
->save
.es
= vmcb
->save
.es
;
2771 nested_vmcb
->save
.cs
= vmcb
->save
.cs
;
2772 nested_vmcb
->save
.ss
= vmcb
->save
.ss
;
2773 nested_vmcb
->save
.ds
= vmcb
->save
.ds
;
2774 nested_vmcb
->save
.gdtr
= vmcb
->save
.gdtr
;
2775 nested_vmcb
->save
.idtr
= vmcb
->save
.idtr
;
2776 nested_vmcb
->save
.efer
= svm
->vcpu
.arch
.efer
;
2777 nested_vmcb
->save
.cr0
= kvm_read_cr0(&svm
->vcpu
);
2778 nested_vmcb
->save
.cr3
= kvm_read_cr3(&svm
->vcpu
);
2779 nested_vmcb
->save
.cr2
= vmcb
->save
.cr2
;
2780 nested_vmcb
->save
.cr4
= svm
->vcpu
.arch
.cr4
;
2781 nested_vmcb
->save
.rflags
= kvm_get_rflags(&svm
->vcpu
);
2782 nested_vmcb
->save
.rip
= vmcb
->save
.rip
;
2783 nested_vmcb
->save
.rsp
= vmcb
->save
.rsp
;
2784 nested_vmcb
->save
.rax
= vmcb
->save
.rax
;
2785 nested_vmcb
->save
.dr7
= vmcb
->save
.dr7
;
2786 nested_vmcb
->save
.dr6
= vmcb
->save
.dr6
;
2787 nested_vmcb
->save
.cpl
= vmcb
->save
.cpl
;
2789 nested_vmcb
->control
.int_ctl
= vmcb
->control
.int_ctl
;
2790 nested_vmcb
->control
.int_vector
= vmcb
->control
.int_vector
;
2791 nested_vmcb
->control
.int_state
= vmcb
->control
.int_state
;
2792 nested_vmcb
->control
.exit_code
= vmcb
->control
.exit_code
;
2793 nested_vmcb
->control
.exit_code_hi
= vmcb
->control
.exit_code_hi
;
2794 nested_vmcb
->control
.exit_info_1
= vmcb
->control
.exit_info_1
;
2795 nested_vmcb
->control
.exit_info_2
= vmcb
->control
.exit_info_2
;
2796 nested_vmcb
->control
.exit_int_info
= vmcb
->control
.exit_int_info
;
2797 nested_vmcb
->control
.exit_int_info_err
= vmcb
->control
.exit_int_info_err
;
2799 if (svm
->nrips_enabled
)
2800 nested_vmcb
->control
.next_rip
= vmcb
->control
.next_rip
;
2803 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2804 * to make sure that we do not lose injected events. So check event_inj
2805 * here and copy it to exit_int_info if it is valid.
2806 * Exit_int_info and event_inj can't be both valid because the case
2807 * below only happens on a VMRUN instruction intercept which has
2808 * no valid exit_int_info set.
2810 if (vmcb
->control
.event_inj
& SVM_EVTINJ_VALID
) {
2811 struct vmcb_control_area
*nc
= &nested_vmcb
->control
;
2813 nc
->exit_int_info
= vmcb
->control
.event_inj
;
2814 nc
->exit_int_info_err
= vmcb
->control
.event_inj_err
;
2817 nested_vmcb
->control
.tlb_ctl
= 0;
2818 nested_vmcb
->control
.event_inj
= 0;
2819 nested_vmcb
->control
.event_inj_err
= 0;
2821 /* We always set V_INTR_MASKING and remember the old value in hflags */
2822 if (!(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
))
2823 nested_vmcb
->control
.int_ctl
&= ~V_INTR_MASKING_MASK
;
2825 /* Restore the original control entries */
2826 copy_vmcb_control_area(vmcb
, hsave
);
2828 kvm_clear_exception_queue(&svm
->vcpu
);
2829 kvm_clear_interrupt_queue(&svm
->vcpu
);
2831 svm
->nested
.nested_cr3
= 0;
2833 /* Restore selected save entries */
2834 svm
->vmcb
->save
.es
= hsave
->save
.es
;
2835 svm
->vmcb
->save
.cs
= hsave
->save
.cs
;
2836 svm
->vmcb
->save
.ss
= hsave
->save
.ss
;
2837 svm
->vmcb
->save
.ds
= hsave
->save
.ds
;
2838 svm
->vmcb
->save
.gdtr
= hsave
->save
.gdtr
;
2839 svm
->vmcb
->save
.idtr
= hsave
->save
.idtr
;
2840 kvm_set_rflags(&svm
->vcpu
, hsave
->save
.rflags
);
2841 svm_set_efer(&svm
->vcpu
, hsave
->save
.efer
);
2842 svm_set_cr0(&svm
->vcpu
, hsave
->save
.cr0
| X86_CR0_PE
);
2843 svm_set_cr4(&svm
->vcpu
, hsave
->save
.cr4
);
2845 svm
->vmcb
->save
.cr3
= hsave
->save
.cr3
;
2846 svm
->vcpu
.arch
.cr3
= hsave
->save
.cr3
;
2848 (void)kvm_set_cr3(&svm
->vcpu
, hsave
->save
.cr3
);
2850 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
, hsave
->save
.rax
);
2851 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RSP
, hsave
->save
.rsp
);
2852 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RIP
, hsave
->save
.rip
);
2853 svm
->vmcb
->save
.dr7
= 0;
2854 svm
->vmcb
->save
.cpl
= 0;
2855 svm
->vmcb
->control
.exit_int_info
= 0;
2857 mark_all_dirty(svm
->vmcb
);
2859 nested_svm_unmap(page
);
2861 nested_svm_uninit_mmu_context(&svm
->vcpu
);
2862 kvm_mmu_reset_context(&svm
->vcpu
);
2863 kvm_mmu_load(&svm
->vcpu
);
2868 static bool nested_svm_vmrun_msrpm(struct vcpu_svm
*svm
)
2871 * This function merges the msr permission bitmaps of kvm and the
2872 * nested vmcb. It is optimized in that it only merges the parts where
2873 * the kvm msr permission bitmap may contain zero bits
2877 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_MSR_PROT
)))
2880 for (i
= 0; i
< MSRPM_OFFSETS
; i
++) {
2884 if (msrpm_offsets
[i
] == 0xffffffff)
2887 p
= msrpm_offsets
[i
];
2888 offset
= svm
->nested
.vmcb_msrpm
+ (p
* 4);
2890 if (kvm_vcpu_read_guest(&svm
->vcpu
, offset
, &value
, 4))
2893 svm
->nested
.msrpm
[p
] = svm
->msrpm
[p
] | value
;
2896 svm
->vmcb
->control
.msrpm_base_pa
= __pa(svm
->nested
.msrpm
);
2901 static bool nested_vmcb_checks(struct vmcb
*vmcb
)
2903 if ((vmcb
->control
.intercept
& (1ULL << INTERCEPT_VMRUN
)) == 0)
2906 if (vmcb
->control
.asid
== 0)
2909 if (vmcb
->control
.nested_ctl
&& !npt_enabled
)
2915 static bool nested_svm_vmrun(struct vcpu_svm
*svm
)
2917 struct vmcb
*nested_vmcb
;
2918 struct vmcb
*hsave
= svm
->nested
.hsave
;
2919 struct vmcb
*vmcb
= svm
->vmcb
;
2923 vmcb_gpa
= svm
->vmcb
->save
.rax
;
2925 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, &page
);
2929 if (!nested_vmcb_checks(nested_vmcb
)) {
2930 nested_vmcb
->control
.exit_code
= SVM_EXIT_ERR
;
2931 nested_vmcb
->control
.exit_code_hi
= 0;
2932 nested_vmcb
->control
.exit_info_1
= 0;
2933 nested_vmcb
->control
.exit_info_2
= 0;
2935 nested_svm_unmap(page
);
2940 trace_kvm_nested_vmrun(svm
->vmcb
->save
.rip
, vmcb_gpa
,
2941 nested_vmcb
->save
.rip
,
2942 nested_vmcb
->control
.int_ctl
,
2943 nested_vmcb
->control
.event_inj
,
2944 nested_vmcb
->control
.nested_ctl
);
2946 trace_kvm_nested_intercepts(nested_vmcb
->control
.intercept_cr
& 0xffff,
2947 nested_vmcb
->control
.intercept_cr
>> 16,
2948 nested_vmcb
->control
.intercept_exceptions
,
2949 nested_vmcb
->control
.intercept
);
2951 /* Clear internal status */
2952 kvm_clear_exception_queue(&svm
->vcpu
);
2953 kvm_clear_interrupt_queue(&svm
->vcpu
);
2956 * Save the old vmcb, so we don't need to pick what we save, but can
2957 * restore everything when a VMEXIT occurs
2959 hsave
->save
.es
= vmcb
->save
.es
;
2960 hsave
->save
.cs
= vmcb
->save
.cs
;
2961 hsave
->save
.ss
= vmcb
->save
.ss
;
2962 hsave
->save
.ds
= vmcb
->save
.ds
;
2963 hsave
->save
.gdtr
= vmcb
->save
.gdtr
;
2964 hsave
->save
.idtr
= vmcb
->save
.idtr
;
2965 hsave
->save
.efer
= svm
->vcpu
.arch
.efer
;
2966 hsave
->save
.cr0
= kvm_read_cr0(&svm
->vcpu
);
2967 hsave
->save
.cr4
= svm
->vcpu
.arch
.cr4
;
2968 hsave
->save
.rflags
= kvm_get_rflags(&svm
->vcpu
);
2969 hsave
->save
.rip
= kvm_rip_read(&svm
->vcpu
);
2970 hsave
->save
.rsp
= vmcb
->save
.rsp
;
2971 hsave
->save
.rax
= vmcb
->save
.rax
;
2973 hsave
->save
.cr3
= vmcb
->save
.cr3
;
2975 hsave
->save
.cr3
= kvm_read_cr3(&svm
->vcpu
);
2977 copy_vmcb_control_area(hsave
, vmcb
);
2979 if (kvm_get_rflags(&svm
->vcpu
) & X86_EFLAGS_IF
)
2980 svm
->vcpu
.arch
.hflags
|= HF_HIF_MASK
;
2982 svm
->vcpu
.arch
.hflags
&= ~HF_HIF_MASK
;
2984 if (nested_vmcb
->control
.nested_ctl
) {
2985 kvm_mmu_unload(&svm
->vcpu
);
2986 svm
->nested
.nested_cr3
= nested_vmcb
->control
.nested_cr3
;
2987 nested_svm_init_mmu_context(&svm
->vcpu
);
2990 /* Load the nested guest state */
2991 svm
->vmcb
->save
.es
= nested_vmcb
->save
.es
;
2992 svm
->vmcb
->save
.cs
= nested_vmcb
->save
.cs
;
2993 svm
->vmcb
->save
.ss
= nested_vmcb
->save
.ss
;
2994 svm
->vmcb
->save
.ds
= nested_vmcb
->save
.ds
;
2995 svm
->vmcb
->save
.gdtr
= nested_vmcb
->save
.gdtr
;
2996 svm
->vmcb
->save
.idtr
= nested_vmcb
->save
.idtr
;
2997 kvm_set_rflags(&svm
->vcpu
, nested_vmcb
->save
.rflags
);
2998 svm_set_efer(&svm
->vcpu
, nested_vmcb
->save
.efer
);
2999 svm_set_cr0(&svm
->vcpu
, nested_vmcb
->save
.cr0
);
3000 svm_set_cr4(&svm
->vcpu
, nested_vmcb
->save
.cr4
);
3002 svm
->vmcb
->save
.cr3
= nested_vmcb
->save
.cr3
;
3003 svm
->vcpu
.arch
.cr3
= nested_vmcb
->save
.cr3
;
3005 (void)kvm_set_cr3(&svm
->vcpu
, nested_vmcb
->save
.cr3
);
3007 /* Guest paging mode is active - reset mmu */
3008 kvm_mmu_reset_context(&svm
->vcpu
);
3010 svm
->vmcb
->save
.cr2
= svm
->vcpu
.arch
.cr2
= nested_vmcb
->save
.cr2
;
3011 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
, nested_vmcb
->save
.rax
);
3012 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RSP
, nested_vmcb
->save
.rsp
);
3013 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RIP
, nested_vmcb
->save
.rip
);
3015 /* In case we don't even reach vcpu_run, the fields are not updated */
3016 svm
->vmcb
->save
.rax
= nested_vmcb
->save
.rax
;
3017 svm
->vmcb
->save
.rsp
= nested_vmcb
->save
.rsp
;
3018 svm
->vmcb
->save
.rip
= nested_vmcb
->save
.rip
;
3019 svm
->vmcb
->save
.dr7
= nested_vmcb
->save
.dr7
;
3020 svm
->vmcb
->save
.dr6
= nested_vmcb
->save
.dr6
;
3021 svm
->vmcb
->save
.cpl
= nested_vmcb
->save
.cpl
;
3023 svm
->nested
.vmcb_msrpm
= nested_vmcb
->control
.msrpm_base_pa
& ~0x0fffULL
;
3024 svm
->nested
.vmcb_iopm
= nested_vmcb
->control
.iopm_base_pa
& ~0x0fffULL
;
3026 /* cache intercepts */
3027 svm
->nested
.intercept_cr
= nested_vmcb
->control
.intercept_cr
;
3028 svm
->nested
.intercept_dr
= nested_vmcb
->control
.intercept_dr
;
3029 svm
->nested
.intercept_exceptions
= nested_vmcb
->control
.intercept_exceptions
;
3030 svm
->nested
.intercept
= nested_vmcb
->control
.intercept
;
3032 svm_flush_tlb(&svm
->vcpu
);
3033 svm
->vmcb
->control
.int_ctl
= nested_vmcb
->control
.int_ctl
| V_INTR_MASKING_MASK
;
3034 if (nested_vmcb
->control
.int_ctl
& V_INTR_MASKING_MASK
)
3035 svm
->vcpu
.arch
.hflags
|= HF_VINTR_MASK
;
3037 svm
->vcpu
.arch
.hflags
&= ~HF_VINTR_MASK
;
3039 if (svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
) {
3040 /* We only want the cr8 intercept bits of the guest */
3041 clr_cr_intercept(svm
, INTERCEPT_CR8_READ
);
3042 clr_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
3045 /* We don't want to see VMMCALLs from a nested guest */
3046 clr_intercept(svm
, INTERCEPT_VMMCALL
);
3048 svm
->vmcb
->control
.virt_ext
= nested_vmcb
->control
.virt_ext
;
3049 svm
->vmcb
->control
.int_vector
= nested_vmcb
->control
.int_vector
;
3050 svm
->vmcb
->control
.int_state
= nested_vmcb
->control
.int_state
;
3051 svm
->vmcb
->control
.tsc_offset
+= nested_vmcb
->control
.tsc_offset
;
3052 svm
->vmcb
->control
.event_inj
= nested_vmcb
->control
.event_inj
;
3053 svm
->vmcb
->control
.event_inj_err
= nested_vmcb
->control
.event_inj_err
;
3055 nested_svm_unmap(page
);
3057 /* Enter Guest-Mode */
3058 enter_guest_mode(&svm
->vcpu
);
3061 * Merge guest and host intercepts - must be called with vcpu in
3062 * guest-mode to take affect here
3064 recalc_intercepts(svm
);
3066 svm
->nested
.vmcb
= vmcb_gpa
;
3070 mark_all_dirty(svm
->vmcb
);
3075 static void nested_svm_vmloadsave(struct vmcb
*from_vmcb
, struct vmcb
*to_vmcb
)
3077 to_vmcb
->save
.fs
= from_vmcb
->save
.fs
;
3078 to_vmcb
->save
.gs
= from_vmcb
->save
.gs
;
3079 to_vmcb
->save
.tr
= from_vmcb
->save
.tr
;
3080 to_vmcb
->save
.ldtr
= from_vmcb
->save
.ldtr
;
3081 to_vmcb
->save
.kernel_gs_base
= from_vmcb
->save
.kernel_gs_base
;
3082 to_vmcb
->save
.star
= from_vmcb
->save
.star
;
3083 to_vmcb
->save
.lstar
= from_vmcb
->save
.lstar
;
3084 to_vmcb
->save
.cstar
= from_vmcb
->save
.cstar
;
3085 to_vmcb
->save
.sfmask
= from_vmcb
->save
.sfmask
;
3086 to_vmcb
->save
.sysenter_cs
= from_vmcb
->save
.sysenter_cs
;
3087 to_vmcb
->save
.sysenter_esp
= from_vmcb
->save
.sysenter_esp
;
3088 to_vmcb
->save
.sysenter_eip
= from_vmcb
->save
.sysenter_eip
;
3091 static int vmload_interception(struct vcpu_svm
*svm
)
3093 struct vmcb
*nested_vmcb
;
3097 if (nested_svm_check_permissions(svm
))
3100 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, &page
);
3104 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
3105 ret
= kvm_skip_emulated_instruction(&svm
->vcpu
);
3107 nested_svm_vmloadsave(nested_vmcb
, svm
->vmcb
);
3108 nested_svm_unmap(page
);
3113 static int vmsave_interception(struct vcpu_svm
*svm
)
3115 struct vmcb
*nested_vmcb
;
3119 if (nested_svm_check_permissions(svm
))
3122 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, &page
);
3126 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
3127 ret
= kvm_skip_emulated_instruction(&svm
->vcpu
);
3129 nested_svm_vmloadsave(svm
->vmcb
, nested_vmcb
);
3130 nested_svm_unmap(page
);
3135 static int vmrun_interception(struct vcpu_svm
*svm
)
3137 if (nested_svm_check_permissions(svm
))
3140 /* Save rip after vmrun instruction */
3141 kvm_rip_write(&svm
->vcpu
, kvm_rip_read(&svm
->vcpu
) + 3);
3143 if (!nested_svm_vmrun(svm
))
3146 if (!nested_svm_vmrun_msrpm(svm
))
3153 svm
->vmcb
->control
.exit_code
= SVM_EXIT_ERR
;
3154 svm
->vmcb
->control
.exit_code_hi
= 0;
3155 svm
->vmcb
->control
.exit_info_1
= 0;
3156 svm
->vmcb
->control
.exit_info_2
= 0;
3158 nested_svm_vmexit(svm
);
3163 static int stgi_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
);
3172 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
3179 static int clgi_interception(struct vcpu_svm
*svm
)
3183 if (nested_svm_check_permissions(svm
))
3186 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
3187 ret
= kvm_skip_emulated_instruction(&svm
->vcpu
);
3191 /* After a CLGI no interrupts should come */
3192 if (!kvm_vcpu_apicv_active(&svm
->vcpu
)) {
3193 svm_clear_vintr(svm
);
3194 svm
->vmcb
->control
.int_ctl
&= ~V_IRQ_MASK
;
3195 mark_dirty(svm
->vmcb
, VMCB_INTR
);
3201 static int invlpga_interception(struct vcpu_svm
*svm
)
3203 struct kvm_vcpu
*vcpu
= &svm
->vcpu
;
3205 trace_kvm_invlpga(svm
->vmcb
->save
.rip
, kvm_register_read(&svm
->vcpu
, VCPU_REGS_RCX
),
3206 kvm_register_read(&svm
->vcpu
, VCPU_REGS_RAX
));
3208 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
3209 kvm_mmu_invlpg(vcpu
, kvm_register_read(&svm
->vcpu
, VCPU_REGS_RAX
));
3211 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
3212 return kvm_skip_emulated_instruction(&svm
->vcpu
);
3215 static int skinit_interception(struct vcpu_svm
*svm
)
3217 trace_kvm_skinit(svm
->vmcb
->save
.rip
, kvm_register_read(&svm
->vcpu
, VCPU_REGS_RAX
));
3219 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
3223 static int wbinvd_interception(struct vcpu_svm
*svm
)
3225 return kvm_emulate_wbinvd(&svm
->vcpu
);
3228 static int xsetbv_interception(struct vcpu_svm
*svm
)
3230 u64 new_bv
= kvm_read_edx_eax(&svm
->vcpu
);
3231 u32 index
= kvm_register_read(&svm
->vcpu
, VCPU_REGS_RCX
);
3233 if (kvm_set_xcr(&svm
->vcpu
, index
, new_bv
) == 0) {
3234 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
3235 return kvm_skip_emulated_instruction(&svm
->vcpu
);
3241 static int task_switch_interception(struct vcpu_svm
*svm
)
3245 int int_type
= svm
->vmcb
->control
.exit_int_info
&
3246 SVM_EXITINTINFO_TYPE_MASK
;
3247 int int_vec
= svm
->vmcb
->control
.exit_int_info
& SVM_EVTINJ_VEC_MASK
;
3249 svm
->vmcb
->control
.exit_int_info
& SVM_EXITINTINFO_TYPE_MASK
;
3251 svm
->vmcb
->control
.exit_int_info
& SVM_EXITINTINFO_VALID
;
3252 bool has_error_code
= false;
3255 tss_selector
= (u16
)svm
->vmcb
->control
.exit_info_1
;
3257 if (svm
->vmcb
->control
.exit_info_2
&
3258 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET
))
3259 reason
= TASK_SWITCH_IRET
;
3260 else if (svm
->vmcb
->control
.exit_info_2
&
3261 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP
))
3262 reason
= TASK_SWITCH_JMP
;
3264 reason
= TASK_SWITCH_GATE
;
3266 reason
= TASK_SWITCH_CALL
;
3268 if (reason
== TASK_SWITCH_GATE
) {
3270 case SVM_EXITINTINFO_TYPE_NMI
:
3271 svm
->vcpu
.arch
.nmi_injected
= false;
3273 case SVM_EXITINTINFO_TYPE_EXEPT
:
3274 if (svm
->vmcb
->control
.exit_info_2
&
3275 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE
)) {
3276 has_error_code
= true;
3278 (u32
)svm
->vmcb
->control
.exit_info_2
;
3280 kvm_clear_exception_queue(&svm
->vcpu
);
3282 case SVM_EXITINTINFO_TYPE_INTR
:
3283 kvm_clear_interrupt_queue(&svm
->vcpu
);
3290 if (reason
!= TASK_SWITCH_GATE
||
3291 int_type
== SVM_EXITINTINFO_TYPE_SOFT
||
3292 (int_type
== SVM_EXITINTINFO_TYPE_EXEPT
&&
3293 (int_vec
== OF_VECTOR
|| int_vec
== BP_VECTOR
)))
3294 skip_emulated_instruction(&svm
->vcpu
);
3296 if (int_type
!= SVM_EXITINTINFO_TYPE_SOFT
)
3299 if (kvm_task_switch(&svm
->vcpu
, tss_selector
, int_vec
, reason
,
3300 has_error_code
, error_code
) == EMULATE_FAIL
) {
3301 svm
->vcpu
.run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
3302 svm
->vcpu
.run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
3303 svm
->vcpu
.run
->internal
.ndata
= 0;
3309 static int cpuid_interception(struct vcpu_svm
*svm
)
3311 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
3312 return kvm_emulate_cpuid(&svm
->vcpu
);
3315 static int iret_interception(struct vcpu_svm
*svm
)
3317 ++svm
->vcpu
.stat
.nmi_window_exits
;
3318 clr_intercept(svm
, INTERCEPT_IRET
);
3319 svm
->vcpu
.arch
.hflags
|= HF_IRET_MASK
;
3320 svm
->nmi_iret_rip
= kvm_rip_read(&svm
->vcpu
);
3321 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
3325 static int invlpg_interception(struct vcpu_svm
*svm
)
3327 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS
))
3328 return emulate_instruction(&svm
->vcpu
, 0) == EMULATE_DONE
;
3330 kvm_mmu_invlpg(&svm
->vcpu
, svm
->vmcb
->control
.exit_info_1
);
3331 return kvm_skip_emulated_instruction(&svm
->vcpu
);
3334 static int emulate_on_interception(struct vcpu_svm
*svm
)
3336 return emulate_instruction(&svm
->vcpu
, 0) == EMULATE_DONE
;
3339 static int rdpmc_interception(struct vcpu_svm
*svm
)
3343 if (!static_cpu_has(X86_FEATURE_NRIPS
))
3344 return emulate_on_interception(svm
);
3346 err
= kvm_rdpmc(&svm
->vcpu
);
3347 return kvm_complete_insn_gp(&svm
->vcpu
, err
);
3350 static bool check_selective_cr0_intercepted(struct vcpu_svm
*svm
,
3353 unsigned long cr0
= svm
->vcpu
.arch
.cr0
;
3357 intercept
= svm
->nested
.intercept
;
3359 if (!is_guest_mode(&svm
->vcpu
) ||
3360 (!(intercept
& (1ULL << INTERCEPT_SELECTIVE_CR0
))))
3363 cr0
&= ~SVM_CR0_SELECTIVE_MASK
;
3364 val
&= ~SVM_CR0_SELECTIVE_MASK
;
3367 svm
->vmcb
->control
.exit_code
= SVM_EXIT_CR0_SEL_WRITE
;
3368 ret
= (nested_svm_exit_handled(svm
) == NESTED_EXIT_DONE
);
3374 #define CR_VALID (1ULL << 63)
3376 static int cr_interception(struct vcpu_svm
*svm
)
3382 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS
))
3383 return emulate_on_interception(svm
);
3385 if (unlikely((svm
->vmcb
->control
.exit_info_1
& CR_VALID
) == 0))
3386 return emulate_on_interception(svm
);
3388 reg
= svm
->vmcb
->control
.exit_info_1
& SVM_EXITINFO_REG_MASK
;
3389 if (svm
->vmcb
->control
.exit_code
== SVM_EXIT_CR0_SEL_WRITE
)
3390 cr
= SVM_EXIT_WRITE_CR0
- SVM_EXIT_READ_CR0
;
3392 cr
= svm
->vmcb
->control
.exit_code
- SVM_EXIT_READ_CR0
;
3395 if (cr
>= 16) { /* mov to cr */
3397 val
= kvm_register_read(&svm
->vcpu
, reg
);
3400 if (!check_selective_cr0_intercepted(svm
, val
))
3401 err
= kvm_set_cr0(&svm
->vcpu
, val
);
3407 err
= kvm_set_cr3(&svm
->vcpu
, val
);
3410 err
= kvm_set_cr4(&svm
->vcpu
, val
);
3413 err
= kvm_set_cr8(&svm
->vcpu
, val
);
3416 WARN(1, "unhandled write to CR%d", cr
);
3417 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
3420 } else { /* mov from cr */
3423 val
= kvm_read_cr0(&svm
->vcpu
);
3426 val
= svm
->vcpu
.arch
.cr2
;
3429 val
= kvm_read_cr3(&svm
->vcpu
);
3432 val
= kvm_read_cr4(&svm
->vcpu
);
3435 val
= kvm_get_cr8(&svm
->vcpu
);
3438 WARN(1, "unhandled read from CR%d", cr
);
3439 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
3442 kvm_register_write(&svm
->vcpu
, reg
, val
);
3444 return kvm_complete_insn_gp(&svm
->vcpu
, err
);
3447 static int dr_interception(struct vcpu_svm
*svm
)
3452 if (svm
->vcpu
.guest_debug
== 0) {
3454 * No more DR vmexits; force a reload of the debug registers
3455 * and reenter on this instruction. The next vmexit will
3456 * retrieve the full state of the debug registers.
3458 clr_dr_intercepts(svm
);
3459 svm
->vcpu
.arch
.switch_db_regs
|= KVM_DEBUGREG_WONT_EXIT
;
3463 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS
))
3464 return emulate_on_interception(svm
);
3466 reg
= svm
->vmcb
->control
.exit_info_1
& SVM_EXITINFO_REG_MASK
;
3467 dr
= svm
->vmcb
->control
.exit_code
- SVM_EXIT_READ_DR0
;
3469 if (dr
>= 16) { /* mov to DRn */
3470 if (!kvm_require_dr(&svm
->vcpu
, dr
- 16))
3472 val
= kvm_register_read(&svm
->vcpu
, reg
);
3473 kvm_set_dr(&svm
->vcpu
, dr
- 16, val
);
3475 if (!kvm_require_dr(&svm
->vcpu
, dr
))
3477 kvm_get_dr(&svm
->vcpu
, dr
, &val
);
3478 kvm_register_write(&svm
->vcpu
, reg
, val
);
3481 return kvm_skip_emulated_instruction(&svm
->vcpu
);
3484 static int cr8_write_interception(struct vcpu_svm
*svm
)
3486 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
3489 u8 cr8_prev
= kvm_get_cr8(&svm
->vcpu
);
3490 /* instruction emulation calls kvm_set_cr8() */
3491 r
= cr_interception(svm
);
3492 if (lapic_in_kernel(&svm
->vcpu
))
3494 if (cr8_prev
<= kvm_get_cr8(&svm
->vcpu
))
3496 kvm_run
->exit_reason
= KVM_EXIT_SET_TPR
;
3500 static int svm_get_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
3502 struct vcpu_svm
*svm
= to_svm(vcpu
);
3504 switch (msr_info
->index
) {
3505 case MSR_IA32_TSC
: {
3506 msr_info
->data
= svm
->vmcb
->control
.tsc_offset
+
3507 kvm_scale_tsc(vcpu
, rdtsc());
3512 msr_info
->data
= svm
->vmcb
->save
.star
;
3514 #ifdef CONFIG_X86_64
3516 msr_info
->data
= svm
->vmcb
->save
.lstar
;
3519 msr_info
->data
= svm
->vmcb
->save
.cstar
;
3521 case MSR_KERNEL_GS_BASE
:
3522 msr_info
->data
= svm
->vmcb
->save
.kernel_gs_base
;
3524 case MSR_SYSCALL_MASK
:
3525 msr_info
->data
= svm
->vmcb
->save
.sfmask
;
3528 case MSR_IA32_SYSENTER_CS
:
3529 msr_info
->data
= svm
->vmcb
->save
.sysenter_cs
;
3531 case MSR_IA32_SYSENTER_EIP
:
3532 msr_info
->data
= svm
->sysenter_eip
;
3534 case MSR_IA32_SYSENTER_ESP
:
3535 msr_info
->data
= svm
->sysenter_esp
;
3538 if (!boot_cpu_has(X86_FEATURE_RDTSCP
))
3540 msr_info
->data
= svm
->tsc_aux
;
3543 * Nobody will change the following 5 values in the VMCB so we can
3544 * safely return them on rdmsr. They will always be 0 until LBRV is
3547 case MSR_IA32_DEBUGCTLMSR
:
3548 msr_info
->data
= svm
->vmcb
->save
.dbgctl
;
3550 case MSR_IA32_LASTBRANCHFROMIP
:
3551 msr_info
->data
= svm
->vmcb
->save
.br_from
;
3553 case MSR_IA32_LASTBRANCHTOIP
:
3554 msr_info
->data
= svm
->vmcb
->save
.br_to
;
3556 case MSR_IA32_LASTINTFROMIP
:
3557 msr_info
->data
= svm
->vmcb
->save
.last_excp_from
;
3559 case MSR_IA32_LASTINTTOIP
:
3560 msr_info
->data
= svm
->vmcb
->save
.last_excp_to
;
3562 case MSR_VM_HSAVE_PA
:
3563 msr_info
->data
= svm
->nested
.hsave_msr
;
3566 msr_info
->data
= svm
->nested
.vm_cr_msr
;
3568 case MSR_IA32_SPEC_CTRL
:
3569 msr_info
->data
= svm
->spec_ctrl
;
3571 case MSR_IA32_UCODE_REV
:
3572 msr_info
->data
= 0x01000065;
3574 case MSR_F15H_IC_CFG
: {
3578 family
= guest_cpuid_family(vcpu
);
3579 model
= guest_cpuid_model(vcpu
);
3581 if (family
< 0 || model
< 0)
3582 return kvm_get_msr_common(vcpu
, msr_info
);
3586 if (family
== 0x15 &&
3587 (model
>= 0x2 && model
< 0x20))
3588 msr_info
->data
= 0x1E;
3592 return kvm_get_msr_common(vcpu
, msr_info
);
3597 static int rdmsr_interception(struct vcpu_svm
*svm
)
3599 u32 ecx
= kvm_register_read(&svm
->vcpu
, VCPU_REGS_RCX
);
3600 struct msr_data msr_info
;
3602 msr_info
.index
= ecx
;
3603 msr_info
.host_initiated
= false;
3604 if (svm_get_msr(&svm
->vcpu
, &msr_info
)) {
3605 trace_kvm_msr_read_ex(ecx
);
3606 kvm_inject_gp(&svm
->vcpu
, 0);
3609 trace_kvm_msr_read(ecx
, msr_info
.data
);
3611 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
,
3612 msr_info
.data
& 0xffffffff);
3613 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RDX
,
3614 msr_info
.data
>> 32);
3615 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
3616 return kvm_skip_emulated_instruction(&svm
->vcpu
);
3620 static int svm_set_vm_cr(struct kvm_vcpu
*vcpu
, u64 data
)
3622 struct vcpu_svm
*svm
= to_svm(vcpu
);
3623 int svm_dis
, chg_mask
;
3625 if (data
& ~SVM_VM_CR_VALID_MASK
)
3628 chg_mask
= SVM_VM_CR_VALID_MASK
;
3630 if (svm
->nested
.vm_cr_msr
& SVM_VM_CR_SVM_DIS_MASK
)
3631 chg_mask
&= ~(SVM_VM_CR_SVM_LOCK_MASK
| SVM_VM_CR_SVM_DIS_MASK
);
3633 svm
->nested
.vm_cr_msr
&= ~chg_mask
;
3634 svm
->nested
.vm_cr_msr
|= (data
& chg_mask
);
3636 svm_dis
= svm
->nested
.vm_cr_msr
& SVM_VM_CR_SVM_DIS_MASK
;
3638 /* check for svm_disable while efer.svme is set */
3639 if (svm_dis
&& (vcpu
->arch
.efer
& EFER_SVME
))
3645 static int svm_set_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
3647 struct vcpu_svm
*svm
= to_svm(vcpu
);
3649 u32 ecx
= msr
->index
;
3650 u64 data
= msr
->data
;
3653 kvm_write_tsc(vcpu
, msr
);
3656 svm
->vmcb
->save
.star
= data
;
3658 #ifdef CONFIG_X86_64
3660 svm
->vmcb
->save
.lstar
= data
;
3663 svm
->vmcb
->save
.cstar
= data
;
3665 case MSR_KERNEL_GS_BASE
:
3666 svm
->vmcb
->save
.kernel_gs_base
= data
;
3668 case MSR_SYSCALL_MASK
:
3669 svm
->vmcb
->save
.sfmask
= data
;
3672 case MSR_IA32_SYSENTER_CS
:
3673 svm
->vmcb
->save
.sysenter_cs
= data
;
3675 case MSR_IA32_SYSENTER_EIP
:
3676 svm
->sysenter_eip
= data
;
3677 svm
->vmcb
->save
.sysenter_eip
= data
;
3679 case MSR_IA32_SYSENTER_ESP
:
3680 svm
->sysenter_esp
= data
;
3681 svm
->vmcb
->save
.sysenter_esp
= data
;
3684 if (!boot_cpu_has(X86_FEATURE_RDTSCP
))
3688 * This is rare, so we update the MSR here instead of using
3689 * direct_access_msrs. Doing that would require a rdmsr in
3692 svm
->tsc_aux
= data
;
3693 wrmsrl(MSR_TSC_AUX
, svm
->tsc_aux
);
3695 case MSR_IA32_DEBUGCTLMSR
:
3696 if (!boot_cpu_has(X86_FEATURE_LBRV
)) {
3697 vcpu_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3701 if (data
& DEBUGCTL_RESERVED_BITS
)
3704 svm
->vmcb
->save
.dbgctl
= data
;
3705 mark_dirty(svm
->vmcb
, VMCB_LBR
);
3706 if (data
& (1ULL<<0))
3707 svm_enable_lbrv(svm
);
3709 svm_disable_lbrv(svm
);
3711 case MSR_VM_HSAVE_PA
:
3712 svm
->nested
.hsave_msr
= data
;
3715 return svm_set_vm_cr(vcpu
, data
);
3717 vcpu_unimpl(vcpu
, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx
, data
);
3719 case MSR_IA32_SPEC_CTRL
:
3720 svm
->spec_ctrl
= data
;
3722 case MSR_IA32_APICBASE
:
3723 if (kvm_vcpu_apicv_active(vcpu
))
3724 avic_update_vapic_bar(to_svm(vcpu
), data
);
3725 /* Follow through */
3727 return kvm_set_msr_common(vcpu
, msr
);
3732 static int wrmsr_interception(struct vcpu_svm
*svm
)
3734 struct msr_data msr
;
3735 u32 ecx
= kvm_register_read(&svm
->vcpu
, VCPU_REGS_RCX
);
3736 u64 data
= kvm_read_edx_eax(&svm
->vcpu
);
3740 msr
.host_initiated
= false;
3742 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
3743 if (kvm_set_msr(&svm
->vcpu
, &msr
)) {
3744 trace_kvm_msr_write_ex(ecx
, data
);
3745 kvm_inject_gp(&svm
->vcpu
, 0);
3748 trace_kvm_msr_write(ecx
, data
);
3749 return kvm_skip_emulated_instruction(&svm
->vcpu
);
3753 static int msr_interception(struct vcpu_svm
*svm
)
3755 if (svm
->vmcb
->control
.exit_info_1
)
3756 return wrmsr_interception(svm
);
3758 return rdmsr_interception(svm
);
3761 static int interrupt_window_interception(struct vcpu_svm
*svm
)
3763 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
3764 svm_clear_vintr(svm
);
3765 svm
->vmcb
->control
.int_ctl
&= ~V_IRQ_MASK
;
3766 mark_dirty(svm
->vmcb
, VMCB_INTR
);
3767 ++svm
->vcpu
.stat
.irq_window_exits
;
3771 static int pause_interception(struct vcpu_svm
*svm
)
3773 kvm_vcpu_on_spin(&(svm
->vcpu
));
3777 static int nop_interception(struct vcpu_svm
*svm
)
3779 return kvm_skip_emulated_instruction(&(svm
->vcpu
));
3782 static int monitor_interception(struct vcpu_svm
*svm
)
3784 printk_once(KERN_WARNING
"kvm: MONITOR instruction emulated as NOP!\n");
3785 return nop_interception(svm
);
3788 static int mwait_interception(struct vcpu_svm
*svm
)
3790 printk_once(KERN_WARNING
"kvm: MWAIT instruction emulated as NOP!\n");
3791 return nop_interception(svm
);
3794 enum avic_ipi_failure_cause
{
3795 AVIC_IPI_FAILURE_INVALID_INT_TYPE
,
3796 AVIC_IPI_FAILURE_TARGET_NOT_RUNNING
,
3797 AVIC_IPI_FAILURE_INVALID_TARGET
,
3798 AVIC_IPI_FAILURE_INVALID_BACKING_PAGE
,
3801 static int avic_incomplete_ipi_interception(struct vcpu_svm
*svm
)
3803 u32 icrh
= svm
->vmcb
->control
.exit_info_1
>> 32;
3804 u32 icrl
= svm
->vmcb
->control
.exit_info_1
;
3805 u32 id
= svm
->vmcb
->control
.exit_info_2
>> 32;
3806 u32 index
= svm
->vmcb
->control
.exit_info_2
& 0xFF;
3807 struct kvm_lapic
*apic
= svm
->vcpu
.arch
.apic
;
3809 trace_kvm_avic_incomplete_ipi(svm
->vcpu
.vcpu_id
, icrh
, icrl
, id
, index
);
3812 case AVIC_IPI_FAILURE_INVALID_INT_TYPE
:
3814 * AVIC hardware handles the generation of
3815 * IPIs when the specified Message Type is Fixed
3816 * (also known as fixed delivery mode) and
3817 * the Trigger Mode is edge-triggered. The hardware
3818 * also supports self and broadcast delivery modes
3819 * specified via the Destination Shorthand(DSH)
3820 * field of the ICRL. Logical and physical APIC ID
3821 * formats are supported. All other IPI types cause
3822 * a #VMEXIT, which needs to emulated.
3824 kvm_lapic_reg_write(apic
, APIC_ICR2
, icrh
);
3825 kvm_lapic_reg_write(apic
, APIC_ICR
, icrl
);
3827 case AVIC_IPI_FAILURE_TARGET_NOT_RUNNING
: {
3829 struct kvm_vcpu
*vcpu
;
3830 struct kvm
*kvm
= svm
->vcpu
.kvm
;
3831 struct kvm_lapic
*apic
= svm
->vcpu
.arch
.apic
;
3834 * At this point, we expect that the AVIC HW has already
3835 * set the appropriate IRR bits on the valid target
3836 * vcpus. So, we just need to kick the appropriate vcpu.
3838 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
3839 bool m
= kvm_apic_match_dest(vcpu
, apic
,
3840 icrl
& KVM_APIC_SHORT_MASK
,
3841 GET_APIC_DEST_FIELD(icrh
),
3842 icrl
& KVM_APIC_DEST_MASK
);
3844 if (m
&& !avic_vcpu_is_running(vcpu
))
3845 kvm_vcpu_wake_up(vcpu
);
3849 case AVIC_IPI_FAILURE_INVALID_TARGET
:
3851 case AVIC_IPI_FAILURE_INVALID_BACKING_PAGE
:
3852 WARN_ONCE(1, "Invalid backing page\n");
3855 pr_err("Unknown IPI interception\n");
3861 static u32
*avic_get_logical_id_entry(struct kvm_vcpu
*vcpu
, u32 ldr
, bool flat
)
3863 struct kvm_arch
*vm_data
= &vcpu
->kvm
->arch
;
3865 u32
*logical_apic_id_table
;
3866 int dlid
= GET_APIC_LOGICAL_ID(ldr
);
3871 if (flat
) { /* flat */
3872 index
= ffs(dlid
) - 1;
3875 } else { /* cluster */
3876 int cluster
= (dlid
& 0xf0) >> 4;
3877 int apic
= ffs(dlid
& 0x0f) - 1;
3879 if ((apic
< 0) || (apic
> 7) ||
3882 index
= (cluster
<< 2) + apic
;
3885 logical_apic_id_table
= (u32
*) page_address(vm_data
->avic_logical_id_table_page
);
3887 return &logical_apic_id_table
[index
];
3890 static int avic_ldr_write(struct kvm_vcpu
*vcpu
, u8 g_physical_id
, u32 ldr
,
3894 u32
*entry
, new_entry
;
3896 flat
= kvm_lapic_get_reg(vcpu
->arch
.apic
, APIC_DFR
) == APIC_DFR_FLAT
;
3897 entry
= avic_get_logical_id_entry(vcpu
, ldr
, flat
);
3901 new_entry
= READ_ONCE(*entry
);
3902 new_entry
&= ~AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK
;
3903 new_entry
|= (g_physical_id
& AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK
);
3905 new_entry
|= AVIC_LOGICAL_ID_ENTRY_VALID_MASK
;
3907 new_entry
&= ~AVIC_LOGICAL_ID_ENTRY_VALID_MASK
;
3908 WRITE_ONCE(*entry
, new_entry
);
3913 static int avic_handle_ldr_update(struct kvm_vcpu
*vcpu
)
3916 struct vcpu_svm
*svm
= to_svm(vcpu
);
3917 u32 ldr
= kvm_lapic_get_reg(vcpu
->arch
.apic
, APIC_LDR
);
3922 ret
= avic_ldr_write(vcpu
, vcpu
->vcpu_id
, ldr
, true);
3923 if (ret
&& svm
->ldr_reg
) {
3924 avic_ldr_write(vcpu
, 0, svm
->ldr_reg
, false);
3932 static int avic_handle_apic_id_update(struct kvm_vcpu
*vcpu
)
3935 struct vcpu_svm
*svm
= to_svm(vcpu
);
3936 u32 apic_id_reg
= kvm_lapic_get_reg(vcpu
->arch
.apic
, APIC_ID
);
3937 u32 id
= (apic_id_reg
>> 24) & 0xff;
3939 if (vcpu
->vcpu_id
== id
)
3942 old
= avic_get_physical_id_entry(vcpu
, vcpu
->vcpu_id
);
3943 new = avic_get_physical_id_entry(vcpu
, id
);
3947 /* We need to move physical_id_entry to new offset */
3950 to_svm(vcpu
)->avic_physical_id_cache
= new;
3953 * Also update the guest physical APIC ID in the logical
3954 * APIC ID table entry if already setup the LDR.
3957 avic_handle_ldr_update(vcpu
);
3962 static int avic_handle_dfr_update(struct kvm_vcpu
*vcpu
)
3964 struct vcpu_svm
*svm
= to_svm(vcpu
);
3965 struct kvm_arch
*vm_data
= &vcpu
->kvm
->arch
;
3966 u32 dfr
= kvm_lapic_get_reg(vcpu
->arch
.apic
, APIC_DFR
);
3967 u32 mod
= (dfr
>> 28) & 0xf;
3970 * We assume that all local APICs are using the same type.
3971 * If this changes, we need to flush the AVIC logical
3974 if (vm_data
->ldr_mode
== mod
)
3977 clear_page(page_address(vm_data
->avic_logical_id_table_page
));
3978 vm_data
->ldr_mode
= mod
;
3981 avic_handle_ldr_update(vcpu
);
3985 static int avic_unaccel_trap_write(struct vcpu_svm
*svm
)
3987 struct kvm_lapic
*apic
= svm
->vcpu
.arch
.apic
;
3988 u32 offset
= svm
->vmcb
->control
.exit_info_1
&
3989 AVIC_UNACCEL_ACCESS_OFFSET_MASK
;
3993 if (avic_handle_apic_id_update(&svm
->vcpu
))
3997 if (avic_handle_ldr_update(&svm
->vcpu
))
4001 avic_handle_dfr_update(&svm
->vcpu
);
4007 kvm_lapic_reg_write(apic
, offset
, kvm_lapic_get_reg(apic
, offset
));
4012 static bool is_avic_unaccelerated_access_trap(u32 offset
)
4041 static int avic_unaccelerated_access_interception(struct vcpu_svm
*svm
)
4044 u32 offset
= svm
->vmcb
->control
.exit_info_1
&
4045 AVIC_UNACCEL_ACCESS_OFFSET_MASK
;
4046 u32 vector
= svm
->vmcb
->control
.exit_info_2
&
4047 AVIC_UNACCEL_ACCESS_VECTOR_MASK
;
4048 bool write
= (svm
->vmcb
->control
.exit_info_1
>> 32) &
4049 AVIC_UNACCEL_ACCESS_WRITE_MASK
;
4050 bool trap
= is_avic_unaccelerated_access_trap(offset
);
4052 trace_kvm_avic_unaccelerated_access(svm
->vcpu
.vcpu_id
, offset
,
4053 trap
, write
, vector
);
4056 WARN_ONCE(!write
, "svm: Handling trap read.\n");
4057 ret
= avic_unaccel_trap_write(svm
);
4059 /* Handling Fault */
4060 ret
= (emulate_instruction(&svm
->vcpu
, 0) == EMULATE_DONE
);
4066 static int (*const svm_exit_handlers
[])(struct vcpu_svm
*svm
) = {
4067 [SVM_EXIT_READ_CR0
] = cr_interception
,
4068 [SVM_EXIT_READ_CR3
] = cr_interception
,
4069 [SVM_EXIT_READ_CR4
] = cr_interception
,
4070 [SVM_EXIT_READ_CR8
] = cr_interception
,
4071 [SVM_EXIT_CR0_SEL_WRITE
] = cr_interception
,
4072 [SVM_EXIT_WRITE_CR0
] = cr_interception
,
4073 [SVM_EXIT_WRITE_CR3
] = cr_interception
,
4074 [SVM_EXIT_WRITE_CR4
] = cr_interception
,
4075 [SVM_EXIT_WRITE_CR8
] = cr8_write_interception
,
4076 [SVM_EXIT_READ_DR0
] = dr_interception
,
4077 [SVM_EXIT_READ_DR1
] = dr_interception
,
4078 [SVM_EXIT_READ_DR2
] = dr_interception
,
4079 [SVM_EXIT_READ_DR3
] = dr_interception
,
4080 [SVM_EXIT_READ_DR4
] = dr_interception
,
4081 [SVM_EXIT_READ_DR5
] = dr_interception
,
4082 [SVM_EXIT_READ_DR6
] = dr_interception
,
4083 [SVM_EXIT_READ_DR7
] = dr_interception
,
4084 [SVM_EXIT_WRITE_DR0
] = dr_interception
,
4085 [SVM_EXIT_WRITE_DR1
] = dr_interception
,
4086 [SVM_EXIT_WRITE_DR2
] = dr_interception
,
4087 [SVM_EXIT_WRITE_DR3
] = dr_interception
,
4088 [SVM_EXIT_WRITE_DR4
] = dr_interception
,
4089 [SVM_EXIT_WRITE_DR5
] = dr_interception
,
4090 [SVM_EXIT_WRITE_DR6
] = dr_interception
,
4091 [SVM_EXIT_WRITE_DR7
] = dr_interception
,
4092 [SVM_EXIT_EXCP_BASE
+ DB_VECTOR
] = db_interception
,
4093 [SVM_EXIT_EXCP_BASE
+ BP_VECTOR
] = bp_interception
,
4094 [SVM_EXIT_EXCP_BASE
+ UD_VECTOR
] = ud_interception
,
4095 [SVM_EXIT_EXCP_BASE
+ PF_VECTOR
] = pf_interception
,
4096 [SVM_EXIT_EXCP_BASE
+ MC_VECTOR
] = mc_interception
,
4097 [SVM_EXIT_EXCP_BASE
+ AC_VECTOR
] = ac_interception
,
4098 [SVM_EXIT_INTR
] = intr_interception
,
4099 [SVM_EXIT_NMI
] = nmi_interception
,
4100 [SVM_EXIT_SMI
] = nop_on_interception
,
4101 [SVM_EXIT_INIT
] = nop_on_interception
,
4102 [SVM_EXIT_VINTR
] = interrupt_window_interception
,
4103 [SVM_EXIT_RDPMC
] = rdpmc_interception
,
4104 [SVM_EXIT_CPUID
] = cpuid_interception
,
4105 [SVM_EXIT_IRET
] = iret_interception
,
4106 [SVM_EXIT_INVD
] = emulate_on_interception
,
4107 [SVM_EXIT_PAUSE
] = pause_interception
,
4108 [SVM_EXIT_HLT
] = halt_interception
,
4109 [SVM_EXIT_INVLPG
] = invlpg_interception
,
4110 [SVM_EXIT_INVLPGA
] = invlpga_interception
,
4111 [SVM_EXIT_IOIO
] = io_interception
,
4112 [SVM_EXIT_MSR
] = msr_interception
,
4113 [SVM_EXIT_TASK_SWITCH
] = task_switch_interception
,
4114 [SVM_EXIT_SHUTDOWN
] = shutdown_interception
,
4115 [SVM_EXIT_VMRUN
] = vmrun_interception
,
4116 [SVM_EXIT_VMMCALL
] = vmmcall_interception
,
4117 [SVM_EXIT_VMLOAD
] = vmload_interception
,
4118 [SVM_EXIT_VMSAVE
] = vmsave_interception
,
4119 [SVM_EXIT_STGI
] = stgi_interception
,
4120 [SVM_EXIT_CLGI
] = clgi_interception
,
4121 [SVM_EXIT_SKINIT
] = skinit_interception
,
4122 [SVM_EXIT_WBINVD
] = wbinvd_interception
,
4123 [SVM_EXIT_MONITOR
] = monitor_interception
,
4124 [SVM_EXIT_MWAIT
] = mwait_interception
,
4125 [SVM_EXIT_XSETBV
] = xsetbv_interception
,
4126 [SVM_EXIT_NPF
] = pf_interception
,
4127 [SVM_EXIT_RSM
] = emulate_on_interception
,
4128 [SVM_EXIT_AVIC_INCOMPLETE_IPI
] = avic_incomplete_ipi_interception
,
4129 [SVM_EXIT_AVIC_UNACCELERATED_ACCESS
] = avic_unaccelerated_access_interception
,
4132 static void dump_vmcb(struct kvm_vcpu
*vcpu
)
4134 struct vcpu_svm
*svm
= to_svm(vcpu
);
4135 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
4136 struct vmcb_save_area
*save
= &svm
->vmcb
->save
;
4138 pr_err("VMCB Control Area:\n");
4139 pr_err("%-20s%04x\n", "cr_read:", control
->intercept_cr
& 0xffff);
4140 pr_err("%-20s%04x\n", "cr_write:", control
->intercept_cr
>> 16);
4141 pr_err("%-20s%04x\n", "dr_read:", control
->intercept_dr
& 0xffff);
4142 pr_err("%-20s%04x\n", "dr_write:", control
->intercept_dr
>> 16);
4143 pr_err("%-20s%08x\n", "exceptions:", control
->intercept_exceptions
);
4144 pr_err("%-20s%016llx\n", "intercepts:", control
->intercept
);
4145 pr_err("%-20s%d\n", "pause filter count:", control
->pause_filter_count
);
4146 pr_err("%-20s%016llx\n", "iopm_base_pa:", control
->iopm_base_pa
);
4147 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control
->msrpm_base_pa
);
4148 pr_err("%-20s%016llx\n", "tsc_offset:", control
->tsc_offset
);
4149 pr_err("%-20s%d\n", "asid:", control
->asid
);
4150 pr_err("%-20s%d\n", "tlb_ctl:", control
->tlb_ctl
);
4151 pr_err("%-20s%08x\n", "int_ctl:", control
->int_ctl
);
4152 pr_err("%-20s%08x\n", "int_vector:", control
->int_vector
);
4153 pr_err("%-20s%08x\n", "int_state:", control
->int_state
);
4154 pr_err("%-20s%08x\n", "exit_code:", control
->exit_code
);
4155 pr_err("%-20s%016llx\n", "exit_info1:", control
->exit_info_1
);
4156 pr_err("%-20s%016llx\n", "exit_info2:", control
->exit_info_2
);
4157 pr_err("%-20s%08x\n", "exit_int_info:", control
->exit_int_info
);
4158 pr_err("%-20s%08x\n", "exit_int_info_err:", control
->exit_int_info_err
);
4159 pr_err("%-20s%lld\n", "nested_ctl:", control
->nested_ctl
);
4160 pr_err("%-20s%016llx\n", "nested_cr3:", control
->nested_cr3
);
4161 pr_err("%-20s%016llx\n", "avic_vapic_bar:", control
->avic_vapic_bar
);
4162 pr_err("%-20s%08x\n", "event_inj:", control
->event_inj
);
4163 pr_err("%-20s%08x\n", "event_inj_err:", control
->event_inj_err
);
4164 pr_err("%-20s%lld\n", "virt_ext:", control
->virt_ext
);
4165 pr_err("%-20s%016llx\n", "next_rip:", control
->next_rip
);
4166 pr_err("%-20s%016llx\n", "avic_backing_page:", control
->avic_backing_page
);
4167 pr_err("%-20s%016llx\n", "avic_logical_id:", control
->avic_logical_id
);
4168 pr_err("%-20s%016llx\n", "avic_physical_id:", control
->avic_physical_id
);
4169 pr_err("VMCB State Save Area:\n");
4170 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4172 save
->es
.selector
, save
->es
.attrib
,
4173 save
->es
.limit
, save
->es
.base
);
4174 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4176 save
->cs
.selector
, save
->cs
.attrib
,
4177 save
->cs
.limit
, save
->cs
.base
);
4178 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4180 save
->ss
.selector
, save
->ss
.attrib
,
4181 save
->ss
.limit
, save
->ss
.base
);
4182 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4184 save
->ds
.selector
, save
->ds
.attrib
,
4185 save
->ds
.limit
, save
->ds
.base
);
4186 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4188 save
->fs
.selector
, save
->fs
.attrib
,
4189 save
->fs
.limit
, save
->fs
.base
);
4190 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4192 save
->gs
.selector
, save
->gs
.attrib
,
4193 save
->gs
.limit
, save
->gs
.base
);
4194 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4196 save
->gdtr
.selector
, save
->gdtr
.attrib
,
4197 save
->gdtr
.limit
, save
->gdtr
.base
);
4198 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4200 save
->ldtr
.selector
, save
->ldtr
.attrib
,
4201 save
->ldtr
.limit
, save
->ldtr
.base
);
4202 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4204 save
->idtr
.selector
, save
->idtr
.attrib
,
4205 save
->idtr
.limit
, save
->idtr
.base
);
4206 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4208 save
->tr
.selector
, save
->tr
.attrib
,
4209 save
->tr
.limit
, save
->tr
.base
);
4210 pr_err("cpl: %d efer: %016llx\n",
4211 save
->cpl
, save
->efer
);
4212 pr_err("%-15s %016llx %-13s %016llx\n",
4213 "cr0:", save
->cr0
, "cr2:", save
->cr2
);
4214 pr_err("%-15s %016llx %-13s %016llx\n",
4215 "cr3:", save
->cr3
, "cr4:", save
->cr4
);
4216 pr_err("%-15s %016llx %-13s %016llx\n",
4217 "dr6:", save
->dr6
, "dr7:", save
->dr7
);
4218 pr_err("%-15s %016llx %-13s %016llx\n",
4219 "rip:", save
->rip
, "rflags:", save
->rflags
);
4220 pr_err("%-15s %016llx %-13s %016llx\n",
4221 "rsp:", save
->rsp
, "rax:", save
->rax
);
4222 pr_err("%-15s %016llx %-13s %016llx\n",
4223 "star:", save
->star
, "lstar:", save
->lstar
);
4224 pr_err("%-15s %016llx %-13s %016llx\n",
4225 "cstar:", save
->cstar
, "sfmask:", save
->sfmask
);
4226 pr_err("%-15s %016llx %-13s %016llx\n",
4227 "kernel_gs_base:", save
->kernel_gs_base
,
4228 "sysenter_cs:", save
->sysenter_cs
);
4229 pr_err("%-15s %016llx %-13s %016llx\n",
4230 "sysenter_esp:", save
->sysenter_esp
,
4231 "sysenter_eip:", save
->sysenter_eip
);
4232 pr_err("%-15s %016llx %-13s %016llx\n",
4233 "gpat:", save
->g_pat
, "dbgctl:", save
->dbgctl
);
4234 pr_err("%-15s %016llx %-13s %016llx\n",
4235 "br_from:", save
->br_from
, "br_to:", save
->br_to
);
4236 pr_err("%-15s %016llx %-13s %016llx\n",
4237 "excp_from:", save
->last_excp_from
,
4238 "excp_to:", save
->last_excp_to
);
4241 static void svm_get_exit_info(struct kvm_vcpu
*vcpu
, u64
*info1
, u64
*info2
)
4243 struct vmcb_control_area
*control
= &to_svm(vcpu
)->vmcb
->control
;
4245 *info1
= control
->exit_info_1
;
4246 *info2
= control
->exit_info_2
;
4249 static int handle_exit(struct kvm_vcpu
*vcpu
)
4251 struct vcpu_svm
*svm
= to_svm(vcpu
);
4252 struct kvm_run
*kvm_run
= vcpu
->run
;
4253 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
4255 trace_kvm_exit(exit_code
, vcpu
, KVM_ISA_SVM
);
4257 vcpu
->arch
.gpa_available
= (exit_code
== SVM_EXIT_NPF
);
4259 if (!is_cr_intercept(svm
, INTERCEPT_CR0_WRITE
))
4260 vcpu
->arch
.cr0
= svm
->vmcb
->save
.cr0
;
4262 vcpu
->arch
.cr3
= svm
->vmcb
->save
.cr3
;
4264 if (unlikely(svm
->nested
.exit_required
)) {
4265 nested_svm_vmexit(svm
);
4266 svm
->nested
.exit_required
= false;
4271 if (is_guest_mode(vcpu
)) {
4274 trace_kvm_nested_vmexit(svm
->vmcb
->save
.rip
, exit_code
,
4275 svm
->vmcb
->control
.exit_info_1
,
4276 svm
->vmcb
->control
.exit_info_2
,
4277 svm
->vmcb
->control
.exit_int_info
,
4278 svm
->vmcb
->control
.exit_int_info_err
,
4281 vmexit
= nested_svm_exit_special(svm
);
4283 if (vmexit
== NESTED_EXIT_CONTINUE
)
4284 vmexit
= nested_svm_exit_handled(svm
);
4286 if (vmexit
== NESTED_EXIT_DONE
)
4290 svm_complete_interrupts(svm
);
4292 if (svm
->vmcb
->control
.exit_code
== SVM_EXIT_ERR
) {
4293 kvm_run
->exit_reason
= KVM_EXIT_FAIL_ENTRY
;
4294 kvm_run
->fail_entry
.hardware_entry_failure_reason
4295 = svm
->vmcb
->control
.exit_code
;
4296 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
4301 if (is_external_interrupt(svm
->vmcb
->control
.exit_int_info
) &&
4302 exit_code
!= SVM_EXIT_EXCP_BASE
+ PF_VECTOR
&&
4303 exit_code
!= SVM_EXIT_NPF
&& exit_code
!= SVM_EXIT_TASK_SWITCH
&&
4304 exit_code
!= SVM_EXIT_INTR
&& exit_code
!= SVM_EXIT_NMI
)
4305 printk(KERN_ERR
"%s: unexpected exit_int_info 0x%x "
4307 __func__
, svm
->vmcb
->control
.exit_int_info
,
4310 if (exit_code
>= ARRAY_SIZE(svm_exit_handlers
)
4311 || !svm_exit_handlers
[exit_code
]) {
4312 WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code
);
4313 kvm_queue_exception(vcpu
, UD_VECTOR
);
4317 return svm_exit_handlers
[exit_code
](svm
);
4320 static void reload_tss(struct kvm_vcpu
*vcpu
)
4322 int cpu
= raw_smp_processor_id();
4324 struct svm_cpu_data
*sd
= per_cpu(svm_data
, cpu
);
4325 sd
->tss_desc
->type
= 9; /* available 32/64-bit TSS */
4329 static void pre_svm_run(struct vcpu_svm
*svm
)
4331 int cpu
= raw_smp_processor_id();
4333 struct svm_cpu_data
*sd
= per_cpu(svm_data
, cpu
);
4335 /* FIXME: handle wraparound of asid_generation */
4336 if (svm
->asid_generation
!= sd
->asid_generation
)
4340 static void svm_inject_nmi(struct kvm_vcpu
*vcpu
)
4342 struct vcpu_svm
*svm
= to_svm(vcpu
);
4344 svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_NMI
;
4345 vcpu
->arch
.hflags
|= HF_NMI_MASK
;
4346 set_intercept(svm
, INTERCEPT_IRET
);
4347 ++vcpu
->stat
.nmi_injections
;
4350 static inline void svm_inject_irq(struct vcpu_svm
*svm
, int irq
)
4352 struct vmcb_control_area
*control
;
4354 /* The following fields are ignored when AVIC is enabled */
4355 control
= &svm
->vmcb
->control
;
4356 control
->int_vector
= irq
;
4357 control
->int_ctl
&= ~V_INTR_PRIO_MASK
;
4358 control
->int_ctl
|= V_IRQ_MASK
|
4359 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT
);
4360 mark_dirty(svm
->vmcb
, VMCB_INTR
);
4363 static void svm_set_irq(struct kvm_vcpu
*vcpu
)
4365 struct vcpu_svm
*svm
= to_svm(vcpu
);
4367 BUG_ON(!(gif_set(svm
)));
4369 trace_kvm_inj_virq(vcpu
->arch
.interrupt
.nr
);
4370 ++vcpu
->stat
.irq_injections
;
4372 svm
->vmcb
->control
.event_inj
= vcpu
->arch
.interrupt
.nr
|
4373 SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
;
4376 static inline bool svm_nested_virtualize_tpr(struct kvm_vcpu
*vcpu
)
4378 return is_guest_mode(vcpu
) && (vcpu
->arch
.hflags
& HF_VINTR_MASK
);
4381 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
, int tpr
, int irr
)
4383 struct vcpu_svm
*svm
= to_svm(vcpu
);
4385 if (svm_nested_virtualize_tpr(vcpu
) ||
4386 kvm_vcpu_apicv_active(vcpu
))
4389 clr_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
4395 set_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
4398 static void svm_set_virtual_x2apic_mode(struct kvm_vcpu
*vcpu
, bool set
)
4403 static bool svm_get_enable_apicv(void)
4408 static void svm_hwapic_irr_update(struct kvm_vcpu
*vcpu
, int max_irr
)
4412 static void svm_hwapic_isr_update(struct kvm_vcpu
*vcpu
, int max_isr
)
4416 /* Note: Currently only used by Hyper-V. */
4417 static void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu
*vcpu
)
4419 struct vcpu_svm
*svm
= to_svm(vcpu
);
4420 struct vmcb
*vmcb
= svm
->vmcb
;
4425 vmcb
->control
.int_ctl
&= ~AVIC_ENABLE_MASK
;
4426 mark_dirty(vmcb
, VMCB_INTR
);
4429 static void svm_load_eoi_exitmap(struct kvm_vcpu
*vcpu
, u64
*eoi_exit_bitmap
)
4434 static void svm_deliver_avic_intr(struct kvm_vcpu
*vcpu
, int vec
)
4436 kvm_lapic_set_irr(vec
, vcpu
->arch
.apic
);
4437 smp_mb__after_atomic();
4439 if (avic_vcpu_is_running(vcpu
))
4440 wrmsrl(SVM_AVIC_DOORBELL
,
4441 kvm_cpu_get_apicid(vcpu
->cpu
));
4443 kvm_vcpu_wake_up(vcpu
);
4446 static void svm_ir_list_del(struct vcpu_svm
*svm
, struct amd_iommu_pi_data
*pi
)
4448 unsigned long flags
;
4449 struct amd_svm_iommu_ir
*cur
;
4451 spin_lock_irqsave(&svm
->ir_list_lock
, flags
);
4452 list_for_each_entry(cur
, &svm
->ir_list
, node
) {
4453 if (cur
->data
!= pi
->ir_data
)
4455 list_del(&cur
->node
);
4459 spin_unlock_irqrestore(&svm
->ir_list_lock
, flags
);
4462 static int svm_ir_list_add(struct vcpu_svm
*svm
, struct amd_iommu_pi_data
*pi
)
4465 unsigned long flags
;
4466 struct amd_svm_iommu_ir
*ir
;
4469 * In some cases, the existing irte is updaed and re-set,
4470 * so we need to check here if it's already been * added
4473 if (pi
->ir_data
&& (pi
->prev_ga_tag
!= 0)) {
4474 struct kvm
*kvm
= svm
->vcpu
.kvm
;
4475 u32 vcpu_id
= AVIC_GATAG_TO_VCPUID(pi
->prev_ga_tag
);
4476 struct kvm_vcpu
*prev_vcpu
= kvm_get_vcpu_by_id(kvm
, vcpu_id
);
4477 struct vcpu_svm
*prev_svm
;
4484 prev_svm
= to_svm(prev_vcpu
);
4485 svm_ir_list_del(prev_svm
, pi
);
4489 * Allocating new amd_iommu_pi_data, which will get
4490 * add to the per-vcpu ir_list.
4492 ir
= kzalloc(sizeof(struct amd_svm_iommu_ir
), GFP_KERNEL
);
4497 ir
->data
= pi
->ir_data
;
4499 spin_lock_irqsave(&svm
->ir_list_lock
, flags
);
4500 list_add(&ir
->node
, &svm
->ir_list
);
4501 spin_unlock_irqrestore(&svm
->ir_list_lock
, flags
);
4508 * The HW cannot support posting multicast/broadcast
4509 * interrupts to a vCPU. So, we still use legacy interrupt
4510 * remapping for these kind of interrupts.
4512 * For lowest-priority interrupts, we only support
4513 * those with single CPU as the destination, e.g. user
4514 * configures the interrupts via /proc/irq or uses
4515 * irqbalance to make the interrupts single-CPU.
4518 get_pi_vcpu_info(struct kvm
*kvm
, struct kvm_kernel_irq_routing_entry
*e
,
4519 struct vcpu_data
*vcpu_info
, struct vcpu_svm
**svm
)
4521 struct kvm_lapic_irq irq
;
4522 struct kvm_vcpu
*vcpu
= NULL
;
4524 kvm_set_msi_irq(kvm
, e
, &irq
);
4526 if (!kvm_intr_is_single_vcpu(kvm
, &irq
, &vcpu
)) {
4527 pr_debug("SVM: %s: use legacy intr remap mode for irq %u\n",
4528 __func__
, irq
.vector
);
4532 pr_debug("SVM: %s: use GA mode for irq %u\n", __func__
,
4534 *svm
= to_svm(vcpu
);
4535 vcpu_info
->pi_desc_addr
= page_to_phys((*svm
)->avic_backing_page
);
4536 vcpu_info
->vector
= irq
.vector
;
4542 * svm_update_pi_irte - set IRTE for Posted-Interrupts
4545 * @host_irq: host irq of the interrupt
4546 * @guest_irq: gsi of the interrupt
4547 * @set: set or unset PI
4548 * returns 0 on success, < 0 on failure
4550 static int svm_update_pi_irte(struct kvm
*kvm
, unsigned int host_irq
,
4551 uint32_t guest_irq
, bool set
)
4553 struct kvm_kernel_irq_routing_entry
*e
;
4554 struct kvm_irq_routing_table
*irq_rt
;
4555 int idx
, ret
= -EINVAL
;
4557 if (!kvm_arch_has_assigned_device(kvm
) ||
4558 !irq_remapping_cap(IRQ_POSTING_CAP
))
4561 pr_debug("SVM: %s: host_irq=%#x, guest_irq=%#x, set=%#x\n",
4562 __func__
, host_irq
, guest_irq
, set
);
4564 idx
= srcu_read_lock(&kvm
->irq_srcu
);
4565 irq_rt
= srcu_dereference(kvm
->irq_routing
, &kvm
->irq_srcu
);
4566 WARN_ON(guest_irq
>= irq_rt
->nr_rt_entries
);
4568 hlist_for_each_entry(e
, &irq_rt
->map
[guest_irq
], link
) {
4569 struct vcpu_data vcpu_info
;
4570 struct vcpu_svm
*svm
= NULL
;
4572 if (e
->type
!= KVM_IRQ_ROUTING_MSI
)
4576 * Here, we setup with legacy mode in the following cases:
4577 * 1. When cannot target interrupt to a specific vcpu.
4578 * 2. Unsetting posted interrupt.
4579 * 3. APIC virtialization is disabled for the vcpu.
4581 if (!get_pi_vcpu_info(kvm
, e
, &vcpu_info
, &svm
) && set
&&
4582 kvm_vcpu_apicv_active(&svm
->vcpu
)) {
4583 struct amd_iommu_pi_data pi
;
4585 /* Try to enable guest_mode in IRTE */
4586 pi
.base
= page_to_phys(svm
->avic_backing_page
) & AVIC_HPA_MASK
;
4587 pi
.ga_tag
= AVIC_GATAG(kvm
->arch
.avic_vm_id
,
4589 pi
.is_guest_mode
= true;
4590 pi
.vcpu_data
= &vcpu_info
;
4591 ret
= irq_set_vcpu_affinity(host_irq
, &pi
);
4594 * Here, we successfully setting up vcpu affinity in
4595 * IOMMU guest mode. Now, we need to store the posted
4596 * interrupt information in a per-vcpu ir_list so that
4597 * we can reference to them directly when we update vcpu
4598 * scheduling information in IOMMU irte.
4600 if (!ret
&& pi
.is_guest_mode
)
4601 svm_ir_list_add(svm
, &pi
);
4603 /* Use legacy mode in IRTE */
4604 struct amd_iommu_pi_data pi
;
4607 * Here, pi is used to:
4608 * - Tell IOMMU to use legacy mode for this interrupt.
4609 * - Retrieve ga_tag of prior interrupt remapping data.
4611 pi
.is_guest_mode
= false;
4612 ret
= irq_set_vcpu_affinity(host_irq
, &pi
);
4615 * Check if the posted interrupt was previously
4616 * setup with the guest_mode by checking if the ga_tag
4617 * was cached. If so, we need to clean up the per-vcpu
4620 if (!ret
&& pi
.prev_ga_tag
) {
4621 int id
= AVIC_GATAG_TO_VCPUID(pi
.prev_ga_tag
);
4622 struct kvm_vcpu
*vcpu
;
4624 vcpu
= kvm_get_vcpu_by_id(kvm
, id
);
4626 svm_ir_list_del(to_svm(vcpu
), &pi
);
4631 trace_kvm_pi_irte_update(svm
->vcpu
.vcpu_id
,
4634 vcpu_info
.pi_desc_addr
, set
);
4638 pr_err("%s: failed to update PI IRTE\n", __func__
);
4645 srcu_read_unlock(&kvm
->irq_srcu
, idx
);
4649 static int svm_nmi_allowed(struct kvm_vcpu
*vcpu
)
4651 struct vcpu_svm
*svm
= to_svm(vcpu
);
4652 struct vmcb
*vmcb
= svm
->vmcb
;
4654 ret
= !(vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
) &&
4655 !(svm
->vcpu
.arch
.hflags
& HF_NMI_MASK
);
4656 ret
= ret
&& gif_set(svm
) && nested_svm_nmi(svm
);
4661 static bool svm_get_nmi_mask(struct kvm_vcpu
*vcpu
)
4663 struct vcpu_svm
*svm
= to_svm(vcpu
);
4665 return !!(svm
->vcpu
.arch
.hflags
& HF_NMI_MASK
);
4668 static void svm_set_nmi_mask(struct kvm_vcpu
*vcpu
, bool masked
)
4670 struct vcpu_svm
*svm
= to_svm(vcpu
);
4673 svm
->vcpu
.arch
.hflags
|= HF_NMI_MASK
;
4674 set_intercept(svm
, INTERCEPT_IRET
);
4676 svm
->vcpu
.arch
.hflags
&= ~HF_NMI_MASK
;
4677 clr_intercept(svm
, INTERCEPT_IRET
);
4681 static int svm_interrupt_allowed(struct kvm_vcpu
*vcpu
)
4683 struct vcpu_svm
*svm
= to_svm(vcpu
);
4684 struct vmcb
*vmcb
= svm
->vmcb
;
4687 if (!gif_set(svm
) ||
4688 (vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
))
4691 ret
= !!(kvm_get_rflags(vcpu
) & X86_EFLAGS_IF
);
4693 if (is_guest_mode(vcpu
))
4694 return ret
&& !(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
);
4699 static void enable_irq_window(struct kvm_vcpu
*vcpu
)
4701 struct vcpu_svm
*svm
= to_svm(vcpu
);
4703 if (kvm_vcpu_apicv_active(vcpu
))
4707 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
4708 * 1, because that's a separate STGI/VMRUN intercept. The next time we
4709 * get that intercept, this function will be called again though and
4710 * we'll get the vintr intercept.
4712 if (gif_set(svm
) && nested_svm_intr(svm
)) {
4714 svm_inject_irq(svm
, 0x0);
4718 static void enable_nmi_window(struct kvm_vcpu
*vcpu
)
4720 struct vcpu_svm
*svm
= to_svm(vcpu
);
4722 if ((svm
->vcpu
.arch
.hflags
& (HF_NMI_MASK
| HF_IRET_MASK
))
4724 return; /* IRET will cause a vm exit */
4726 if ((svm
->vcpu
.arch
.hflags
& HF_GIF_MASK
) == 0)
4727 return; /* STGI will cause a vm exit */
4729 if (svm
->nested
.exit_required
)
4730 return; /* we're not going to run the guest yet */
4733 * Something prevents NMI from been injected. Single step over possible
4734 * problem (IRET or exception injection or interrupt shadow)
4736 svm
->nmi_singlestep_guest_rflags
= svm_get_rflags(vcpu
);
4737 svm
->nmi_singlestep
= true;
4738 svm
->vmcb
->save
.rflags
|= (X86_EFLAGS_TF
| X86_EFLAGS_RF
);
4741 static int svm_set_tss_addr(struct kvm
*kvm
, unsigned int addr
)
4746 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
)
4748 struct vcpu_svm
*svm
= to_svm(vcpu
);
4750 if (static_cpu_has(X86_FEATURE_FLUSHBYASID
))
4751 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ASID
;
4753 svm
->asid_generation
--;
4756 static void svm_prepare_guest_switch(struct kvm_vcpu
*vcpu
)
4760 static inline void sync_cr8_to_lapic(struct kvm_vcpu
*vcpu
)
4762 struct vcpu_svm
*svm
= to_svm(vcpu
);
4764 if (svm_nested_virtualize_tpr(vcpu
))
4767 if (!is_cr_intercept(svm
, INTERCEPT_CR8_WRITE
)) {
4768 int cr8
= svm
->vmcb
->control
.int_ctl
& V_TPR_MASK
;
4769 kvm_set_cr8(vcpu
, cr8
);
4773 static inline void sync_lapic_to_cr8(struct kvm_vcpu
*vcpu
)
4775 struct vcpu_svm
*svm
= to_svm(vcpu
);
4778 if (svm_nested_virtualize_tpr(vcpu
) ||
4779 kvm_vcpu_apicv_active(vcpu
))
4782 cr8
= kvm_get_cr8(vcpu
);
4783 svm
->vmcb
->control
.int_ctl
&= ~V_TPR_MASK
;
4784 svm
->vmcb
->control
.int_ctl
|= cr8
& V_TPR_MASK
;
4787 static void svm_complete_interrupts(struct vcpu_svm
*svm
)
4791 u32 exitintinfo
= svm
->vmcb
->control
.exit_int_info
;
4792 unsigned int3_injected
= svm
->int3_injected
;
4794 svm
->int3_injected
= 0;
4797 * If we've made progress since setting HF_IRET_MASK, we've
4798 * executed an IRET and can allow NMI injection.
4800 if ((svm
->vcpu
.arch
.hflags
& HF_IRET_MASK
)
4801 && kvm_rip_read(&svm
->vcpu
) != svm
->nmi_iret_rip
) {
4802 svm
->vcpu
.arch
.hflags
&= ~(HF_NMI_MASK
| HF_IRET_MASK
);
4803 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
4806 svm
->vcpu
.arch
.nmi_injected
= false;
4807 kvm_clear_exception_queue(&svm
->vcpu
);
4808 kvm_clear_interrupt_queue(&svm
->vcpu
);
4810 if (!(exitintinfo
& SVM_EXITINTINFO_VALID
))
4813 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
4815 vector
= exitintinfo
& SVM_EXITINTINFO_VEC_MASK
;
4816 type
= exitintinfo
& SVM_EXITINTINFO_TYPE_MASK
;
4819 case SVM_EXITINTINFO_TYPE_NMI
:
4820 svm
->vcpu
.arch
.nmi_injected
= true;
4822 case SVM_EXITINTINFO_TYPE_EXEPT
:
4824 * In case of software exceptions, do not reinject the vector,
4825 * but re-execute the instruction instead. Rewind RIP first
4826 * if we emulated INT3 before.
4828 if (kvm_exception_is_soft(vector
)) {
4829 if (vector
== BP_VECTOR
&& int3_injected
&&
4830 kvm_is_linear_rip(&svm
->vcpu
, svm
->int3_rip
))
4831 kvm_rip_write(&svm
->vcpu
,
4832 kvm_rip_read(&svm
->vcpu
) -
4836 if (exitintinfo
& SVM_EXITINTINFO_VALID_ERR
) {
4837 u32 err
= svm
->vmcb
->control
.exit_int_info_err
;
4838 kvm_requeue_exception_e(&svm
->vcpu
, vector
, err
);
4841 kvm_requeue_exception(&svm
->vcpu
, vector
);
4843 case SVM_EXITINTINFO_TYPE_INTR
:
4844 kvm_queue_interrupt(&svm
->vcpu
, vector
, false);
4851 static void svm_cancel_injection(struct kvm_vcpu
*vcpu
)
4853 struct vcpu_svm
*svm
= to_svm(vcpu
);
4854 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
4856 control
->exit_int_info
= control
->event_inj
;
4857 control
->exit_int_info_err
= control
->event_inj_err
;
4858 control
->event_inj
= 0;
4859 svm_complete_interrupts(svm
);
4862 static void svm_vcpu_run(struct kvm_vcpu
*vcpu
)
4864 struct vcpu_svm
*svm
= to_svm(vcpu
);
4866 svm
->vmcb
->save
.rax
= vcpu
->arch
.regs
[VCPU_REGS_RAX
];
4867 svm
->vmcb
->save
.rsp
= vcpu
->arch
.regs
[VCPU_REGS_RSP
];
4868 svm
->vmcb
->save
.rip
= vcpu
->arch
.regs
[VCPU_REGS_RIP
];
4871 * A vmexit emulation is required before the vcpu can be executed
4874 if (unlikely(svm
->nested
.exit_required
))
4878 * Disable singlestep if we're injecting an interrupt/exception.
4879 * We don't want our modified rflags to be pushed on the stack where
4880 * we might not be able to easily reset them if we disabled NMI
4883 if (svm
->nmi_singlestep
&& svm
->vmcb
->control
.event_inj
) {
4885 * Event injection happens before external interrupts cause a
4886 * vmexit and interrupts are disabled here, so smp_send_reschedule
4887 * is enough to force an immediate vmexit.
4889 disable_nmi_singlestep(svm
);
4890 smp_send_reschedule(vcpu
->cpu
);
4895 sync_lapic_to_cr8(vcpu
);
4897 svm
->vmcb
->save
.cr2
= vcpu
->arch
.cr2
;
4903 if (ibrs_inuse
&& (svm
->spec_ctrl
!= FEATURE_ENABLE_IBRS
))
4904 wrmsrl(MSR_IA32_SPEC_CTRL
, svm
->spec_ctrl
);
4907 "push %%" _ASM_BP
"; \n\t"
4908 "mov %c[rbx](%[svm]), %%" _ASM_BX
" \n\t"
4909 "mov %c[rcx](%[svm]), %%" _ASM_CX
" \n\t"
4910 "mov %c[rdx](%[svm]), %%" _ASM_DX
" \n\t"
4911 "mov %c[rsi](%[svm]), %%" _ASM_SI
" \n\t"
4912 "mov %c[rdi](%[svm]), %%" _ASM_DI
" \n\t"
4913 "mov %c[rbp](%[svm]), %%" _ASM_BP
" \n\t"
4914 #ifdef CONFIG_X86_64
4915 "mov %c[r8](%[svm]), %%r8 \n\t"
4916 "mov %c[r9](%[svm]), %%r9 \n\t"
4917 "mov %c[r10](%[svm]), %%r10 \n\t"
4918 "mov %c[r11](%[svm]), %%r11 \n\t"
4919 "mov %c[r12](%[svm]), %%r12 \n\t"
4920 "mov %c[r13](%[svm]), %%r13 \n\t"
4921 "mov %c[r14](%[svm]), %%r14 \n\t"
4922 "mov %c[r15](%[svm]), %%r15 \n\t"
4925 /* Enter guest mode */
4926 "push %%" _ASM_AX
" \n\t"
4927 "mov %c[vmcb](%[svm]), %%" _ASM_AX
" \n\t"
4928 __ex(SVM_VMLOAD
) "\n\t"
4929 __ex(SVM_VMRUN
) "\n\t"
4930 __ex(SVM_VMSAVE
) "\n\t"
4931 "pop %%" _ASM_AX
" \n\t"
4933 /* Save guest registers, load host registers */
4934 "mov %%" _ASM_BX
", %c[rbx](%[svm]) \n\t"
4935 "mov %%" _ASM_CX
", %c[rcx](%[svm]) \n\t"
4936 "mov %%" _ASM_DX
", %c[rdx](%[svm]) \n\t"
4937 "mov %%" _ASM_SI
", %c[rsi](%[svm]) \n\t"
4938 "mov %%" _ASM_DI
", %c[rdi](%[svm]) \n\t"
4939 "mov %%" _ASM_BP
", %c[rbp](%[svm]) \n\t"
4940 #ifdef CONFIG_X86_64
4941 "mov %%r8, %c[r8](%[svm]) \n\t"
4942 "mov %%r9, %c[r9](%[svm]) \n\t"
4943 "mov %%r10, %c[r10](%[svm]) \n\t"
4944 "mov %%r11, %c[r11](%[svm]) \n\t"
4945 "mov %%r12, %c[r12](%[svm]) \n\t"
4946 "mov %%r13, %c[r13](%[svm]) \n\t"
4947 "mov %%r14, %c[r14](%[svm]) \n\t"
4948 "mov %%r15, %c[r15](%[svm]) \n\t"
4953 [vmcb
]"i"(offsetof(struct vcpu_svm
, vmcb_pa
)),
4954 [rbx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RBX
])),
4955 [rcx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RCX
])),
4956 [rdx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RDX
])),
4957 [rsi
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RSI
])),
4958 [rdi
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RDI
])),
4959 [rbp
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RBP
]))
4960 #ifdef CONFIG_X86_64
4961 , [r8
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R8
])),
4962 [r9
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R9
])),
4963 [r10
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R10
])),
4964 [r11
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R11
])),
4965 [r12
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R12
])),
4966 [r13
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R13
])),
4967 [r14
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R14
])),
4968 [r15
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R15
]))
4971 #ifdef CONFIG_X86_64
4972 , "rbx", "rcx", "rdx", "rsi", "rdi"
4973 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
4975 , "ebx", "ecx", "edx", "esi", "edi"
4980 rdmsrl(MSR_IA32_SPEC_CTRL
, svm
->spec_ctrl
);
4981 if (svm
->spec_ctrl
!= FEATURE_ENABLE_IBRS
)
4982 wrmsrl(MSR_IA32_SPEC_CTRL
, FEATURE_ENABLE_IBRS
);
4987 #ifdef CONFIG_X86_64
4988 wrmsrl(MSR_GS_BASE
, svm
->host
.gs_base
);
4990 loadsegment(fs
, svm
->host
.fs
);
4991 #ifndef CONFIG_X86_32_LAZY_GS
4992 loadsegment(gs
, svm
->host
.gs
);
4998 local_irq_disable();
5000 vcpu
->arch
.cr2
= svm
->vmcb
->save
.cr2
;
5001 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = svm
->vmcb
->save
.rax
;
5002 vcpu
->arch
.regs
[VCPU_REGS_RSP
] = svm
->vmcb
->save
.rsp
;
5003 vcpu
->arch
.regs
[VCPU_REGS_RIP
] = svm
->vmcb
->save
.rip
;
5005 if (unlikely(svm
->vmcb
->control
.exit_code
== SVM_EXIT_NMI
))
5006 kvm_before_handle_nmi(&svm
->vcpu
);
5010 /* Any pending NMI will happen here */
5012 if (unlikely(svm
->vmcb
->control
.exit_code
== SVM_EXIT_NMI
))
5013 kvm_after_handle_nmi(&svm
->vcpu
);
5015 sync_cr8_to_lapic(vcpu
);
5019 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_DO_NOTHING
;
5021 /* if exit due to PF check for async PF */
5022 if (svm
->vmcb
->control
.exit_code
== SVM_EXIT_EXCP_BASE
+ PF_VECTOR
)
5023 svm
->vcpu
.arch
.apf
.host_apf_reason
= kvm_read_and_reset_pf_reason();
5026 vcpu
->arch
.regs_avail
&= ~(1 << VCPU_EXREG_PDPTR
);
5027 vcpu
->arch
.regs_dirty
&= ~(1 << VCPU_EXREG_PDPTR
);
5031 * We need to handle MC intercepts here before the vcpu has a chance to
5032 * change the physical cpu
5034 if (unlikely(svm
->vmcb
->control
.exit_code
==
5035 SVM_EXIT_EXCP_BASE
+ MC_VECTOR
))
5036 svm_handle_mce(svm
);
5038 mark_all_clean(svm
->vmcb
);
5040 STACK_FRAME_NON_STANDARD(svm_vcpu_run
);
5042 static void svm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
5044 struct vcpu_svm
*svm
= to_svm(vcpu
);
5046 svm
->vmcb
->save
.cr3
= root
;
5047 mark_dirty(svm
->vmcb
, VMCB_CR
);
5048 svm_flush_tlb(vcpu
);
5051 static void set_tdp_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
5053 struct vcpu_svm
*svm
= to_svm(vcpu
);
5055 svm
->vmcb
->control
.nested_cr3
= root
;
5056 mark_dirty(svm
->vmcb
, VMCB_NPT
);
5058 /* Also sync guest cr3 here in case we live migrate */
5059 svm
->vmcb
->save
.cr3
= kvm_read_cr3(vcpu
);
5060 mark_dirty(svm
->vmcb
, VMCB_CR
);
5062 svm_flush_tlb(vcpu
);
5065 static int is_disabled(void)
5069 rdmsrl(MSR_VM_CR
, vm_cr
);
5070 if (vm_cr
& (1 << SVM_VM_CR_SVM_DISABLE
))
5077 svm_patch_hypercall(struct kvm_vcpu
*vcpu
, unsigned char *hypercall
)
5080 * Patch in the VMMCALL instruction:
5082 hypercall
[0] = 0x0f;
5083 hypercall
[1] = 0x01;
5084 hypercall
[2] = 0xd9;
5087 static void svm_check_processor_compat(void *rtn
)
5092 static bool svm_cpu_has_accelerated_tpr(void)
5097 static bool svm_has_high_real_mode_segbase(void)
5102 static u64
svm_get_mt_mask(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool is_mmio
)
5107 static void svm_cpuid_update(struct kvm_vcpu
*vcpu
)
5109 struct vcpu_svm
*svm
= to_svm(vcpu
);
5110 struct kvm_cpuid_entry2
*entry
;
5112 /* Update nrips enabled cache */
5113 svm
->nrips_enabled
= !!guest_cpuid_has_nrips(&svm
->vcpu
);
5115 if (!kvm_vcpu_apicv_active(vcpu
))
5118 entry
= kvm_find_cpuid_entry(vcpu
, 1, 0);
5120 entry
->ecx
&= ~bit(X86_FEATURE_X2APIC
);
5123 static void svm_set_supported_cpuid(u32 func
, struct kvm_cpuid_entry2
*entry
)
5128 entry
->ecx
&= ~bit(X86_FEATURE_X2APIC
);
5132 entry
->ecx
|= (1 << 2); /* Set SVM bit */
5135 entry
->eax
= 1; /* SVM revision 1 */
5136 entry
->ebx
= 8; /* Lets support 8 ASIDs in case we add proper
5137 ASID emulation to nested SVM */
5138 entry
->ecx
= 0; /* Reserved */
5139 entry
->edx
= 0; /* Per default do not support any
5140 additional features */
5142 /* Support next_rip if host supports it */
5143 if (boot_cpu_has(X86_FEATURE_NRIPS
))
5144 entry
->edx
|= SVM_FEATURE_NRIP
;
5146 /* Support NPT for the guest if enabled */
5148 entry
->edx
|= SVM_FEATURE_NPT
;
5154 static int svm_get_lpage_level(void)
5156 return PT_PDPE_LEVEL
;
5159 static bool svm_rdtscp_supported(void)
5161 return boot_cpu_has(X86_FEATURE_RDTSCP
);
5164 static bool svm_invpcid_supported(void)
5169 static bool svm_mpx_supported(void)
5174 static bool svm_xsaves_supported(void)
5179 static bool svm_has_wbinvd_exit(void)
5184 #define PRE_EX(exit) { .exit_code = (exit), \
5185 .stage = X86_ICPT_PRE_EXCEPT, }
5186 #define POST_EX(exit) { .exit_code = (exit), \
5187 .stage = X86_ICPT_POST_EXCEPT, }
5188 #define POST_MEM(exit) { .exit_code = (exit), \
5189 .stage = X86_ICPT_POST_MEMACCESS, }
5191 static const struct __x86_intercept
{
5193 enum x86_intercept_stage stage
;
5194 } x86_intercept_map
[] = {
5195 [x86_intercept_cr_read
] = POST_EX(SVM_EXIT_READ_CR0
),
5196 [x86_intercept_cr_write
] = POST_EX(SVM_EXIT_WRITE_CR0
),
5197 [x86_intercept_clts
] = POST_EX(SVM_EXIT_WRITE_CR0
),
5198 [x86_intercept_lmsw
] = POST_EX(SVM_EXIT_WRITE_CR0
),
5199 [x86_intercept_smsw
] = POST_EX(SVM_EXIT_READ_CR0
),
5200 [x86_intercept_dr_read
] = POST_EX(SVM_EXIT_READ_DR0
),
5201 [x86_intercept_dr_write
] = POST_EX(SVM_EXIT_WRITE_DR0
),
5202 [x86_intercept_sldt
] = POST_EX(SVM_EXIT_LDTR_READ
),
5203 [x86_intercept_str
] = POST_EX(SVM_EXIT_TR_READ
),
5204 [x86_intercept_lldt
] = POST_EX(SVM_EXIT_LDTR_WRITE
),
5205 [x86_intercept_ltr
] = POST_EX(SVM_EXIT_TR_WRITE
),
5206 [x86_intercept_sgdt
] = POST_EX(SVM_EXIT_GDTR_READ
),
5207 [x86_intercept_sidt
] = POST_EX(SVM_EXIT_IDTR_READ
),
5208 [x86_intercept_lgdt
] = POST_EX(SVM_EXIT_GDTR_WRITE
),
5209 [x86_intercept_lidt
] = POST_EX(SVM_EXIT_IDTR_WRITE
),
5210 [x86_intercept_vmrun
] = POST_EX(SVM_EXIT_VMRUN
),
5211 [x86_intercept_vmmcall
] = POST_EX(SVM_EXIT_VMMCALL
),
5212 [x86_intercept_vmload
] = POST_EX(SVM_EXIT_VMLOAD
),
5213 [x86_intercept_vmsave
] = POST_EX(SVM_EXIT_VMSAVE
),
5214 [x86_intercept_stgi
] = POST_EX(SVM_EXIT_STGI
),
5215 [x86_intercept_clgi
] = POST_EX(SVM_EXIT_CLGI
),
5216 [x86_intercept_skinit
] = POST_EX(SVM_EXIT_SKINIT
),
5217 [x86_intercept_invlpga
] = POST_EX(SVM_EXIT_INVLPGA
),
5218 [x86_intercept_rdtscp
] = POST_EX(SVM_EXIT_RDTSCP
),
5219 [x86_intercept_monitor
] = POST_MEM(SVM_EXIT_MONITOR
),
5220 [x86_intercept_mwait
] = POST_EX(SVM_EXIT_MWAIT
),
5221 [x86_intercept_invlpg
] = POST_EX(SVM_EXIT_INVLPG
),
5222 [x86_intercept_invd
] = POST_EX(SVM_EXIT_INVD
),
5223 [x86_intercept_wbinvd
] = POST_EX(SVM_EXIT_WBINVD
),
5224 [x86_intercept_wrmsr
] = POST_EX(SVM_EXIT_MSR
),
5225 [x86_intercept_rdtsc
] = POST_EX(SVM_EXIT_RDTSC
),
5226 [x86_intercept_rdmsr
] = POST_EX(SVM_EXIT_MSR
),
5227 [x86_intercept_rdpmc
] = POST_EX(SVM_EXIT_RDPMC
),
5228 [x86_intercept_cpuid
] = PRE_EX(SVM_EXIT_CPUID
),
5229 [x86_intercept_rsm
] = PRE_EX(SVM_EXIT_RSM
),
5230 [x86_intercept_pause
] = PRE_EX(SVM_EXIT_PAUSE
),
5231 [x86_intercept_pushf
] = PRE_EX(SVM_EXIT_PUSHF
),
5232 [x86_intercept_popf
] = PRE_EX(SVM_EXIT_POPF
),
5233 [x86_intercept_intn
] = PRE_EX(SVM_EXIT_SWINT
),
5234 [x86_intercept_iret
] = PRE_EX(SVM_EXIT_IRET
),
5235 [x86_intercept_icebp
] = PRE_EX(SVM_EXIT_ICEBP
),
5236 [x86_intercept_hlt
] = POST_EX(SVM_EXIT_HLT
),
5237 [x86_intercept_in
] = POST_EX(SVM_EXIT_IOIO
),
5238 [x86_intercept_ins
] = POST_EX(SVM_EXIT_IOIO
),
5239 [x86_intercept_out
] = POST_EX(SVM_EXIT_IOIO
),
5240 [x86_intercept_outs
] = POST_EX(SVM_EXIT_IOIO
),
5247 static int svm_check_intercept(struct kvm_vcpu
*vcpu
,
5248 struct x86_instruction_info
*info
,
5249 enum x86_intercept_stage stage
)
5251 struct vcpu_svm
*svm
= to_svm(vcpu
);
5252 int vmexit
, ret
= X86EMUL_CONTINUE
;
5253 struct __x86_intercept icpt_info
;
5254 struct vmcb
*vmcb
= svm
->vmcb
;
5256 if (info
->intercept
>= ARRAY_SIZE(x86_intercept_map
))
5259 icpt_info
= x86_intercept_map
[info
->intercept
];
5261 if (stage
!= icpt_info
.stage
)
5264 switch (icpt_info
.exit_code
) {
5265 case SVM_EXIT_READ_CR0
:
5266 if (info
->intercept
== x86_intercept_cr_read
)
5267 icpt_info
.exit_code
+= info
->modrm_reg
;
5269 case SVM_EXIT_WRITE_CR0
: {
5270 unsigned long cr0
, val
;
5273 if (info
->intercept
== x86_intercept_cr_write
)
5274 icpt_info
.exit_code
+= info
->modrm_reg
;
5276 if (icpt_info
.exit_code
!= SVM_EXIT_WRITE_CR0
||
5277 info
->intercept
== x86_intercept_clts
)
5280 intercept
= svm
->nested
.intercept
;
5282 if (!(intercept
& (1ULL << INTERCEPT_SELECTIVE_CR0
)))
5285 cr0
= vcpu
->arch
.cr0
& ~SVM_CR0_SELECTIVE_MASK
;
5286 val
= info
->src_val
& ~SVM_CR0_SELECTIVE_MASK
;
5288 if (info
->intercept
== x86_intercept_lmsw
) {
5291 /* lmsw can't clear PE - catch this here */
5292 if (cr0
& X86_CR0_PE
)
5297 icpt_info
.exit_code
= SVM_EXIT_CR0_SEL_WRITE
;
5301 case SVM_EXIT_READ_DR0
:
5302 case SVM_EXIT_WRITE_DR0
:
5303 icpt_info
.exit_code
+= info
->modrm_reg
;
5306 if (info
->intercept
== x86_intercept_wrmsr
)
5307 vmcb
->control
.exit_info_1
= 1;
5309 vmcb
->control
.exit_info_1
= 0;
5311 case SVM_EXIT_PAUSE
:
5313 * We get this for NOP only, but pause
5314 * is rep not, check this here
5316 if (info
->rep_prefix
!= REPE_PREFIX
)
5318 case SVM_EXIT_IOIO
: {
5322 if (info
->intercept
== x86_intercept_in
||
5323 info
->intercept
== x86_intercept_ins
) {
5324 exit_info
= ((info
->src_val
& 0xffff) << 16) |
5326 bytes
= info
->dst_bytes
;
5328 exit_info
= (info
->dst_val
& 0xffff) << 16;
5329 bytes
= info
->src_bytes
;
5332 if (info
->intercept
== x86_intercept_outs
||
5333 info
->intercept
== x86_intercept_ins
)
5334 exit_info
|= SVM_IOIO_STR_MASK
;
5336 if (info
->rep_prefix
)
5337 exit_info
|= SVM_IOIO_REP_MASK
;
5339 bytes
= min(bytes
, 4u);
5341 exit_info
|= bytes
<< SVM_IOIO_SIZE_SHIFT
;
5343 exit_info
|= (u32
)info
->ad_bytes
<< (SVM_IOIO_ASIZE_SHIFT
- 1);
5345 vmcb
->control
.exit_info_1
= exit_info
;
5346 vmcb
->control
.exit_info_2
= info
->next_rip
;
5354 /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
5355 if (static_cpu_has(X86_FEATURE_NRIPS
))
5356 vmcb
->control
.next_rip
= info
->next_rip
;
5357 vmcb
->control
.exit_code
= icpt_info
.exit_code
;
5358 vmexit
= nested_svm_exit_handled(svm
);
5360 ret
= (vmexit
== NESTED_EXIT_DONE
) ? X86EMUL_INTERCEPTED
5367 static void svm_handle_external_intr(struct kvm_vcpu
*vcpu
)
5371 * We must have an instruction with interrupts enabled, so
5372 * the timer interrupt isn't delayed by the interrupt shadow.
5375 local_irq_disable();
5378 static void svm_sched_in(struct kvm_vcpu
*vcpu
, int cpu
)
5382 static inline void avic_post_state_restore(struct kvm_vcpu
*vcpu
)
5384 if (avic_handle_apic_id_update(vcpu
) != 0)
5386 if (avic_handle_dfr_update(vcpu
) != 0)
5388 avic_handle_ldr_update(vcpu
);
5391 static void svm_setup_mce(struct kvm_vcpu
*vcpu
)
5393 /* [63:9] are reserved. */
5394 vcpu
->arch
.mcg_cap
&= 0x1ff;
5397 static struct kvm_x86_ops svm_x86_ops __ro_after_init
= {
5398 .cpu_has_kvm_support
= has_svm
,
5399 .disabled_by_bios
= is_disabled
,
5400 .hardware_setup
= svm_hardware_setup
,
5401 .hardware_unsetup
= svm_hardware_unsetup
,
5402 .check_processor_compatibility
= svm_check_processor_compat
,
5403 .hardware_enable
= svm_hardware_enable
,
5404 .hardware_disable
= svm_hardware_disable
,
5405 .cpu_has_accelerated_tpr
= svm_cpu_has_accelerated_tpr
,
5406 .cpu_has_high_real_mode_segbase
= svm_has_high_real_mode_segbase
,
5408 .vcpu_create
= svm_create_vcpu
,
5409 .vcpu_free
= svm_free_vcpu
,
5410 .vcpu_reset
= svm_vcpu_reset
,
5412 .vm_init
= avic_vm_init
,
5413 .vm_destroy
= avic_vm_destroy
,
5415 .prepare_guest_switch
= svm_prepare_guest_switch
,
5416 .vcpu_load
= svm_vcpu_load
,
5417 .vcpu_put
= svm_vcpu_put
,
5418 .vcpu_blocking
= svm_vcpu_blocking
,
5419 .vcpu_unblocking
= svm_vcpu_unblocking
,
5421 .update_bp_intercept
= update_bp_intercept
,
5422 .get_msr
= svm_get_msr
,
5423 .set_msr
= svm_set_msr
,
5424 .get_segment_base
= svm_get_segment_base
,
5425 .get_segment
= svm_get_segment
,
5426 .set_segment
= svm_set_segment
,
5427 .get_cpl
= svm_get_cpl
,
5428 .get_cs_db_l_bits
= kvm_get_cs_db_l_bits
,
5429 .decache_cr0_guest_bits
= svm_decache_cr0_guest_bits
,
5430 .decache_cr3
= svm_decache_cr3
,
5431 .decache_cr4_guest_bits
= svm_decache_cr4_guest_bits
,
5432 .set_cr0
= svm_set_cr0
,
5433 .set_cr3
= svm_set_cr3
,
5434 .set_cr4
= svm_set_cr4
,
5435 .set_efer
= svm_set_efer
,
5436 .get_idt
= svm_get_idt
,
5437 .set_idt
= svm_set_idt
,
5438 .get_gdt
= svm_get_gdt
,
5439 .set_gdt
= svm_set_gdt
,
5440 .get_dr6
= svm_get_dr6
,
5441 .set_dr6
= svm_set_dr6
,
5442 .set_dr7
= svm_set_dr7
,
5443 .sync_dirty_debug_regs
= svm_sync_dirty_debug_regs
,
5444 .cache_reg
= svm_cache_reg
,
5445 .get_rflags
= svm_get_rflags
,
5446 .set_rflags
= svm_set_rflags
,
5448 .tlb_flush
= svm_flush_tlb
,
5450 .run
= svm_vcpu_run
,
5451 .handle_exit
= handle_exit
,
5452 .skip_emulated_instruction
= skip_emulated_instruction
,
5453 .set_interrupt_shadow
= svm_set_interrupt_shadow
,
5454 .get_interrupt_shadow
= svm_get_interrupt_shadow
,
5455 .patch_hypercall
= svm_patch_hypercall
,
5456 .set_irq
= svm_set_irq
,
5457 .set_nmi
= svm_inject_nmi
,
5458 .queue_exception
= svm_queue_exception
,
5459 .cancel_injection
= svm_cancel_injection
,
5460 .interrupt_allowed
= svm_interrupt_allowed
,
5461 .nmi_allowed
= svm_nmi_allowed
,
5462 .get_nmi_mask
= svm_get_nmi_mask
,
5463 .set_nmi_mask
= svm_set_nmi_mask
,
5464 .enable_nmi_window
= enable_nmi_window
,
5465 .enable_irq_window
= enable_irq_window
,
5466 .update_cr8_intercept
= update_cr8_intercept
,
5467 .set_virtual_x2apic_mode
= svm_set_virtual_x2apic_mode
,
5468 .get_enable_apicv
= svm_get_enable_apicv
,
5469 .refresh_apicv_exec_ctrl
= svm_refresh_apicv_exec_ctrl
,
5470 .load_eoi_exitmap
= svm_load_eoi_exitmap
,
5471 .hwapic_irr_update
= svm_hwapic_irr_update
,
5472 .hwapic_isr_update
= svm_hwapic_isr_update
,
5473 .apicv_post_state_restore
= avic_post_state_restore
,
5475 .set_tss_addr
= svm_set_tss_addr
,
5476 .get_tdp_level
= get_npt_level
,
5477 .get_mt_mask
= svm_get_mt_mask
,
5479 .get_exit_info
= svm_get_exit_info
,
5481 .get_lpage_level
= svm_get_lpage_level
,
5483 .cpuid_update
= svm_cpuid_update
,
5485 .rdtscp_supported
= svm_rdtscp_supported
,
5486 .invpcid_supported
= svm_invpcid_supported
,
5487 .mpx_supported
= svm_mpx_supported
,
5488 .xsaves_supported
= svm_xsaves_supported
,
5490 .set_supported_cpuid
= svm_set_supported_cpuid
,
5492 .has_wbinvd_exit
= svm_has_wbinvd_exit
,
5494 .write_tsc_offset
= svm_write_tsc_offset
,
5496 .set_tdp_cr3
= set_tdp_cr3
,
5498 .check_intercept
= svm_check_intercept
,
5499 .handle_external_intr
= svm_handle_external_intr
,
5501 .sched_in
= svm_sched_in
,
5503 .pmu_ops
= &amd_pmu_ops
,
5504 .deliver_posted_interrupt
= svm_deliver_avic_intr
,
5505 .update_pi_irte
= svm_update_pi_irte
,
5506 .setup_mce
= svm_setup_mce
,
5509 static int __init
svm_init(void)
5511 return kvm_init(&svm_x86_ops
, sizeof(struct vcpu_svm
),
5512 __alignof__(struct vcpu_svm
), THIS_MODULE
);
5515 static void __exit
svm_exit(void)
5520 module_init(svm_init
)
5521 module_exit(svm_exit
)