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>
48 #include <asm/spec-ctrl.h>
50 #include <asm/virtext.h>
53 #define __ex(x) __kvm_handle_fault_on_reboot(x)
55 MODULE_AUTHOR("Qumranet");
56 MODULE_LICENSE("GPL");
58 static const struct x86_cpu_id svm_cpu_id
[] = {
59 X86_FEATURE_MATCH(X86_FEATURE_SVM
),
62 MODULE_DEVICE_TABLE(x86cpu
, svm_cpu_id
);
64 #define IOPM_ALLOC_ORDER 2
65 #define MSRPM_ALLOC_ORDER 1
67 #define SEG_TYPE_LDT 2
68 #define SEG_TYPE_BUSY_TSS16 3
70 #define SVM_FEATURE_NPT (1 << 0)
71 #define SVM_FEATURE_LBRV (1 << 1)
72 #define SVM_FEATURE_SVML (1 << 2)
73 #define SVM_FEATURE_NRIP (1 << 3)
74 #define SVM_FEATURE_TSC_RATE (1 << 4)
75 #define SVM_FEATURE_VMCB_CLEAN (1 << 5)
76 #define SVM_FEATURE_FLUSH_ASID (1 << 6)
77 #define SVM_FEATURE_DECODE_ASSIST (1 << 7)
78 #define SVM_FEATURE_PAUSE_FILTER (1 << 10)
80 #define SVM_AVIC_DOORBELL 0xc001011b
82 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
83 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
84 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
86 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
88 #define TSC_RATIO_RSVD 0xffffff0000000000ULL
89 #define TSC_RATIO_MIN 0x0000000000000001ULL
90 #define TSC_RATIO_MAX 0x000000ffffffffffULL
92 #define AVIC_HPA_MASK ~((0xFFFULL << 52) | 0xFFF)
95 * 0xff is broadcast, so the max index allowed for physical APIC ID
96 * table is 0xfe. APIC IDs above 0xff are reserved.
98 #define AVIC_MAX_PHYSICAL_ID_COUNT 255
100 #define AVIC_UNACCEL_ACCESS_WRITE_MASK 1
101 #define AVIC_UNACCEL_ACCESS_OFFSET_MASK 0xFF0
102 #define AVIC_UNACCEL_ACCESS_VECTOR_MASK 0xFFFFFFFF
104 /* AVIC GATAG is encoded using VM and VCPU IDs */
105 #define AVIC_VCPU_ID_BITS 8
106 #define AVIC_VCPU_ID_MASK ((1 << AVIC_VCPU_ID_BITS) - 1)
108 #define AVIC_VM_ID_BITS 24
109 #define AVIC_VM_ID_NR (1 << AVIC_VM_ID_BITS)
110 #define AVIC_VM_ID_MASK ((1 << AVIC_VM_ID_BITS) - 1)
112 #define AVIC_GATAG(x, y) (((x & AVIC_VM_ID_MASK) << AVIC_VCPU_ID_BITS) | \
113 (y & AVIC_VCPU_ID_MASK))
114 #define AVIC_GATAG_TO_VMID(x) ((x >> AVIC_VCPU_ID_BITS) & AVIC_VM_ID_MASK)
115 #define AVIC_GATAG_TO_VCPUID(x) (x & AVIC_VCPU_ID_MASK)
117 static bool erratum_383_found __read_mostly
;
119 static const u32 host_save_user_msrs
[] = {
121 MSR_STAR
, MSR_LSTAR
, MSR_CSTAR
, MSR_SYSCALL_MASK
, MSR_KERNEL_GS_BASE
,
124 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
128 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
132 struct nested_state
{
138 /* These are the merged vectors */
141 /* gpa pointers to the real vectors */
145 /* A VMEXIT is required but not yet emulated */
148 /* cache for intercepts of the guest */
151 u32 intercept_exceptions
;
154 /* Nested Paging related state */
158 #define MSRPM_OFFSETS 16
159 static u32 msrpm_offsets
[MSRPM_OFFSETS
] __read_mostly
;
162 * Set osvw_len to higher value when updated Revision Guides
163 * are published and we know what the new status bits are
165 static uint64_t osvw_len
= 4, osvw_status
;
168 struct kvm_vcpu vcpu
;
170 unsigned long vmcb_pa
;
171 struct svm_cpu_data
*svm_data
;
172 uint64_t asid_generation
;
173 uint64_t sysenter_esp
;
174 uint64_t sysenter_eip
;
181 * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be
182 * translated into the appropriate L2_CFG bits on the host to
183 * perform speculative control.
187 u64 host_user_msrs
[NR_HOST_SAVE_USER_MSRS
];
199 struct nested_state nested
;
202 u64 nmi_singlestep_guest_rflags
;
204 unsigned int3_injected
;
205 unsigned long int3_rip
;
207 /* cached guest cpuid flags for faster access */
208 bool nrips_enabled
: 1;
211 struct page
*avic_backing_page
;
212 u64
*avic_physical_id_cache
;
213 bool avic_is_running
;
216 * Per-vcpu list of struct amd_svm_iommu_ir:
217 * This is used mainly to store interrupt remapping information used
218 * when update the vcpu affinity. This avoids the need to scan for
219 * IRTE and try to match ga_tag in the IOMMU driver.
221 struct list_head ir_list
;
222 spinlock_t ir_list_lock
;
226 * This is a wrapper of struct amd_iommu_ir_data.
228 struct amd_svm_iommu_ir
{
229 struct list_head node
; /* Used by SVM for per-vcpu ir_list */
230 void *data
; /* Storing pointer to struct amd_ir_data */
233 #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK (0xFF)
234 #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK (1 << 31)
236 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK (0xFFULL)
237 #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12)
238 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62)
239 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63)
241 static DEFINE_PER_CPU(u64
, current_tsc_ratio
);
242 #define TSC_RATIO_DEFAULT 0x0100000000ULL
244 #define MSR_INVALID 0xffffffffU
246 static const struct svm_direct_access_msrs
{
247 u32 index
; /* Index of the MSR */
248 bool always
; /* True if intercept is always on */
249 } direct_access_msrs
[] = {
250 { .index
= MSR_STAR
, .always
= true },
251 { .index
= MSR_IA32_SYSENTER_CS
, .always
= true },
253 { .index
= MSR_GS_BASE
, .always
= true },
254 { .index
= MSR_FS_BASE
, .always
= true },
255 { .index
= MSR_KERNEL_GS_BASE
, .always
= true },
256 { .index
= MSR_LSTAR
, .always
= true },
257 { .index
= MSR_CSTAR
, .always
= true },
258 { .index
= MSR_SYSCALL_MASK
, .always
= true },
260 { .index
= MSR_IA32_SPEC_CTRL
, .always
= true },
261 { .index
= MSR_IA32_PRED_CMD
, .always
= true },
262 { .index
= MSR_IA32_LASTBRANCHFROMIP
, .always
= false },
263 { .index
= MSR_IA32_LASTBRANCHTOIP
, .always
= false },
264 { .index
= MSR_IA32_LASTINTFROMIP
, .always
= false },
265 { .index
= MSR_IA32_LASTINTTOIP
, .always
= false },
266 { .index
= MSR_INVALID
, .always
= false },
269 /* enable NPT for AMD64 and X86 with PAE */
270 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
271 static bool npt_enabled
= true;
273 static bool npt_enabled
;
276 /* allow nested paging (virtualized MMU) for all guests */
277 static int npt
= true;
278 module_param(npt
, int, S_IRUGO
);
280 /* allow nested virtualization in KVM/SVM */
281 static int nested
= true;
282 module_param(nested
, int, S_IRUGO
);
284 /* enable / disable AVIC */
286 #ifdef CONFIG_X86_LOCAL_APIC
287 module_param(avic
, int, S_IRUGO
);
290 /* enable/disable Virtual VMLOAD VMSAVE */
291 static int vls
= true;
292 module_param(vls
, int, 0444);
294 /* AVIC VM ID bit masks and lock */
295 static DECLARE_BITMAP(avic_vm_id_bitmap
, AVIC_VM_ID_NR
);
296 static DEFINE_SPINLOCK(avic_vm_id_lock
);
298 static void svm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
);
299 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
);
300 static void svm_complete_interrupts(struct vcpu_svm
*svm
);
302 static int nested_svm_exit_handled(struct vcpu_svm
*svm
);
303 static int nested_svm_intercept(struct vcpu_svm
*svm
);
304 static int nested_svm_vmexit(struct vcpu_svm
*svm
);
305 static int nested_svm_check_exception(struct vcpu_svm
*svm
, unsigned nr
,
306 bool has_error_code
, u32 error_code
);
309 VMCB_INTERCEPTS
, /* Intercept vectors, TSC offset,
310 pause filter count */
311 VMCB_PERM_MAP
, /* IOPM Base and MSRPM Base */
312 VMCB_ASID
, /* ASID */
313 VMCB_INTR
, /* int_ctl, int_vector */
314 VMCB_NPT
, /* npt_en, nCR3, gPAT */
315 VMCB_CR
, /* CR0, CR3, CR4, EFER */
316 VMCB_DR
, /* DR6, DR7 */
317 VMCB_DT
, /* GDT, IDT */
318 VMCB_SEG
, /* CS, DS, SS, ES, CPL */
319 VMCB_CR2
, /* CR2 only */
320 VMCB_LBR
, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
321 VMCB_AVIC
, /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE,
322 * AVIC PHYSICAL_TABLE pointer,
323 * AVIC LOGICAL_TABLE pointer
328 /* TPR and CR2 are always written before VMRUN */
329 #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2))
331 #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL
333 static inline void mark_all_dirty(struct vmcb
*vmcb
)
335 vmcb
->control
.clean
= 0;
338 static inline void mark_all_clean(struct vmcb
*vmcb
)
340 vmcb
->control
.clean
= ((1 << VMCB_DIRTY_MAX
) - 1)
341 & ~VMCB_ALWAYS_DIRTY_MASK
;
344 static inline void mark_dirty(struct vmcb
*vmcb
, int bit
)
346 vmcb
->control
.clean
&= ~(1 << bit
);
349 static inline struct vcpu_svm
*to_svm(struct kvm_vcpu
*vcpu
)
351 return container_of(vcpu
, struct vcpu_svm
, vcpu
);
354 static inline void avic_update_vapic_bar(struct vcpu_svm
*svm
, u64 data
)
356 svm
->vmcb
->control
.avic_vapic_bar
= data
& VMCB_AVIC_APIC_BAR_MASK
;
357 mark_dirty(svm
->vmcb
, VMCB_AVIC
);
360 static inline bool avic_vcpu_is_running(struct kvm_vcpu
*vcpu
)
362 struct vcpu_svm
*svm
= to_svm(vcpu
);
363 u64
*entry
= svm
->avic_physical_id_cache
;
368 return (READ_ONCE(*entry
) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK
);
371 static void recalc_intercepts(struct vcpu_svm
*svm
)
373 struct vmcb_control_area
*c
, *h
;
374 struct nested_state
*g
;
376 mark_dirty(svm
->vmcb
, VMCB_INTERCEPTS
);
378 if (!is_guest_mode(&svm
->vcpu
))
381 c
= &svm
->vmcb
->control
;
382 h
= &svm
->nested
.hsave
->control
;
385 c
->intercept_cr
= h
->intercept_cr
| g
->intercept_cr
;
386 c
->intercept_dr
= h
->intercept_dr
| g
->intercept_dr
;
387 c
->intercept_exceptions
= h
->intercept_exceptions
| g
->intercept_exceptions
;
388 c
->intercept
= h
->intercept
| g
->intercept
;
391 static inline struct vmcb
*get_host_vmcb(struct vcpu_svm
*svm
)
393 if (is_guest_mode(&svm
->vcpu
))
394 return svm
->nested
.hsave
;
399 static inline void set_cr_intercept(struct vcpu_svm
*svm
, int bit
)
401 struct vmcb
*vmcb
= get_host_vmcb(svm
);
403 vmcb
->control
.intercept_cr
|= (1U << bit
);
405 recalc_intercepts(svm
);
408 static inline void clr_cr_intercept(struct vcpu_svm
*svm
, int bit
)
410 struct vmcb
*vmcb
= get_host_vmcb(svm
);
412 vmcb
->control
.intercept_cr
&= ~(1U << bit
);
414 recalc_intercepts(svm
);
417 static inline bool is_cr_intercept(struct vcpu_svm
*svm
, int bit
)
419 struct vmcb
*vmcb
= get_host_vmcb(svm
);
421 return vmcb
->control
.intercept_cr
& (1U << bit
);
424 static inline void set_dr_intercepts(struct vcpu_svm
*svm
)
426 struct vmcb
*vmcb
= get_host_vmcb(svm
);
428 vmcb
->control
.intercept_dr
= (1 << INTERCEPT_DR0_READ
)
429 | (1 << INTERCEPT_DR1_READ
)
430 | (1 << INTERCEPT_DR2_READ
)
431 | (1 << INTERCEPT_DR3_READ
)
432 | (1 << INTERCEPT_DR4_READ
)
433 | (1 << INTERCEPT_DR5_READ
)
434 | (1 << INTERCEPT_DR6_READ
)
435 | (1 << INTERCEPT_DR7_READ
)
436 | (1 << INTERCEPT_DR0_WRITE
)
437 | (1 << INTERCEPT_DR1_WRITE
)
438 | (1 << INTERCEPT_DR2_WRITE
)
439 | (1 << INTERCEPT_DR3_WRITE
)
440 | (1 << INTERCEPT_DR4_WRITE
)
441 | (1 << INTERCEPT_DR5_WRITE
)
442 | (1 << INTERCEPT_DR6_WRITE
)
443 | (1 << INTERCEPT_DR7_WRITE
);
445 recalc_intercepts(svm
);
448 static inline void clr_dr_intercepts(struct vcpu_svm
*svm
)
450 struct vmcb
*vmcb
= get_host_vmcb(svm
);
452 vmcb
->control
.intercept_dr
= 0;
454 recalc_intercepts(svm
);
457 static inline void set_exception_intercept(struct vcpu_svm
*svm
, int bit
)
459 struct vmcb
*vmcb
= get_host_vmcb(svm
);
461 vmcb
->control
.intercept_exceptions
|= (1U << bit
);
463 recalc_intercepts(svm
);
466 static inline void clr_exception_intercept(struct vcpu_svm
*svm
, int bit
)
468 struct vmcb
*vmcb
= get_host_vmcb(svm
);
470 vmcb
->control
.intercept_exceptions
&= ~(1U << bit
);
472 recalc_intercepts(svm
);
475 static inline void set_intercept(struct vcpu_svm
*svm
, int bit
)
477 struct vmcb
*vmcb
= get_host_vmcb(svm
);
479 vmcb
->control
.intercept
|= (1ULL << bit
);
481 recalc_intercepts(svm
);
484 static inline void clr_intercept(struct vcpu_svm
*svm
, int bit
)
486 struct vmcb
*vmcb
= get_host_vmcb(svm
);
488 vmcb
->control
.intercept
&= ~(1ULL << bit
);
490 recalc_intercepts(svm
);
493 static inline void enable_gif(struct vcpu_svm
*svm
)
495 svm
->vcpu
.arch
.hflags
|= HF_GIF_MASK
;
498 static inline void disable_gif(struct vcpu_svm
*svm
)
500 svm
->vcpu
.arch
.hflags
&= ~HF_GIF_MASK
;
503 static inline bool gif_set(struct vcpu_svm
*svm
)
505 return !!(svm
->vcpu
.arch
.hflags
& HF_GIF_MASK
);
508 static unsigned long iopm_base
;
510 struct kvm_ldttss_desc
{
513 unsigned base1
:8, type
:5, dpl
:2, p
:1;
514 unsigned limit1
:4, zero0
:3, g
:1, base2
:8;
517 } __attribute__((packed
));
519 struct svm_cpu_data
{
525 struct kvm_ldttss_desc
*tss_desc
;
527 struct page
*save_area
;
529 struct vmcb
*current_vmcb
;
532 static DEFINE_PER_CPU(struct svm_cpu_data
*, svm_data
);
534 struct svm_init_data
{
539 static const u32 msrpm_ranges
[] = {0, 0xc0000000, 0xc0010000};
541 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
542 #define MSRS_RANGE_SIZE 2048
543 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
545 static u32
svm_msrpm_offset(u32 msr
)
550 for (i
= 0; i
< NUM_MSR_MAPS
; i
++) {
551 if (msr
< msrpm_ranges
[i
] ||
552 msr
>= msrpm_ranges
[i
] + MSRS_IN_RANGE
)
555 offset
= (msr
- msrpm_ranges
[i
]) / 4; /* 4 msrs per u8 */
556 offset
+= (i
* MSRS_RANGE_SIZE
); /* add range offset */
558 /* Now we have the u8 offset - but need the u32 offset */
562 /* MSR not in any range */
566 #define MAX_INST_SIZE 15
568 static inline void clgi(void)
570 asm volatile (__ex(SVM_CLGI
));
573 static inline void stgi(void)
575 asm volatile (__ex(SVM_STGI
));
578 static inline void invlpga(unsigned long addr
, u32 asid
)
580 asm volatile (__ex(SVM_INVLPGA
) : : "a"(addr
), "c"(asid
));
583 static int get_npt_level(void)
586 return PT64_ROOT_LEVEL
;
588 return PT32E_ROOT_LEVEL
;
592 static void svm_set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
594 vcpu
->arch
.efer
= efer
;
595 if (!npt_enabled
&& !(efer
& EFER_LMA
))
598 to_svm(vcpu
)->vmcb
->save
.efer
= efer
| EFER_SVME
;
599 mark_dirty(to_svm(vcpu
)->vmcb
, VMCB_CR
);
602 static int is_external_interrupt(u32 info
)
604 info
&= SVM_EVTINJ_TYPE_MASK
| SVM_EVTINJ_VALID
;
605 return info
== (SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
);
608 static u32
svm_get_interrupt_shadow(struct kvm_vcpu
*vcpu
)
610 struct vcpu_svm
*svm
= to_svm(vcpu
);
613 if (svm
->vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
)
614 ret
= KVM_X86_SHADOW_INT_STI
| KVM_X86_SHADOW_INT_MOV_SS
;
618 static void svm_set_interrupt_shadow(struct kvm_vcpu
*vcpu
, int mask
)
620 struct vcpu_svm
*svm
= to_svm(vcpu
);
623 svm
->vmcb
->control
.int_state
&= ~SVM_INTERRUPT_SHADOW_MASK
;
625 svm
->vmcb
->control
.int_state
|= SVM_INTERRUPT_SHADOW_MASK
;
629 static void skip_emulated_instruction(struct kvm_vcpu
*vcpu
)
631 struct vcpu_svm
*svm
= to_svm(vcpu
);
633 if (svm
->vmcb
->control
.next_rip
!= 0) {
634 WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS
));
635 svm
->next_rip
= svm
->vmcb
->control
.next_rip
;
638 if (!svm
->next_rip
) {
639 if (emulate_instruction(vcpu
, EMULTYPE_SKIP
) !=
641 printk(KERN_DEBUG
"%s: NOP\n", __func__
);
644 if (svm
->next_rip
- kvm_rip_read(vcpu
) > MAX_INST_SIZE
)
645 printk(KERN_ERR
"%s: ip 0x%lx next 0x%llx\n",
646 __func__
, kvm_rip_read(vcpu
), svm
->next_rip
);
648 kvm_rip_write(vcpu
, svm
->next_rip
);
649 svm_set_interrupt_shadow(vcpu
, 0);
652 static void svm_queue_exception(struct kvm_vcpu
*vcpu
)
654 struct vcpu_svm
*svm
= to_svm(vcpu
);
655 unsigned nr
= vcpu
->arch
.exception
.nr
;
656 bool has_error_code
= vcpu
->arch
.exception
.has_error_code
;
657 bool reinject
= vcpu
->arch
.exception
.reinject
;
658 u32 error_code
= vcpu
->arch
.exception
.error_code
;
661 * If we are within a nested VM we'd better #VMEXIT and let the guest
662 * handle the exception
665 nested_svm_check_exception(svm
, nr
, has_error_code
, error_code
))
668 if (nr
== BP_VECTOR
&& !static_cpu_has(X86_FEATURE_NRIPS
)) {
669 unsigned long rip
, old_rip
= kvm_rip_read(&svm
->vcpu
);
672 * For guest debugging where we have to reinject #BP if some
673 * INT3 is guest-owned:
674 * Emulate nRIP by moving RIP forward. Will fail if injection
675 * raises a fault that is not intercepted. Still better than
676 * failing in all cases.
678 skip_emulated_instruction(&svm
->vcpu
);
679 rip
= kvm_rip_read(&svm
->vcpu
);
680 svm
->int3_rip
= rip
+ svm
->vmcb
->save
.cs
.base
;
681 svm
->int3_injected
= rip
- old_rip
;
684 svm
->vmcb
->control
.event_inj
= nr
686 | (has_error_code
? SVM_EVTINJ_VALID_ERR
: 0)
687 | SVM_EVTINJ_TYPE_EXEPT
;
688 svm
->vmcb
->control
.event_inj_err
= error_code
;
691 static void svm_init_erratum_383(void)
697 if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH
))
700 /* Use _safe variants to not break nested virtualization */
701 val
= native_read_msr_safe(MSR_AMD64_DC_CFG
, &err
);
707 low
= lower_32_bits(val
);
708 high
= upper_32_bits(val
);
710 native_write_msr_safe(MSR_AMD64_DC_CFG
, low
, high
);
712 erratum_383_found
= true;
715 static void svm_init_osvw(struct kvm_vcpu
*vcpu
)
718 * Guests should see errata 400 and 415 as fixed (assuming that
719 * HLT and IO instructions are intercepted).
721 vcpu
->arch
.osvw
.length
= (osvw_len
>= 3) ? (osvw_len
) : 3;
722 vcpu
->arch
.osvw
.status
= osvw_status
& ~(6ULL);
725 * By increasing VCPU's osvw.length to 3 we are telling the guest that
726 * all osvw.status bits inside that length, including bit 0 (which is
727 * reserved for erratum 298), are valid. However, if host processor's
728 * osvw_len is 0 then osvw_status[0] carries no information. We need to
729 * be conservative here and therefore we tell the guest that erratum 298
730 * is present (because we really don't know).
732 if (osvw_len
== 0 && boot_cpu_data
.x86
== 0x10)
733 vcpu
->arch
.osvw
.status
|= 1;
736 static int has_svm(void)
740 if (!cpu_has_svm(&msg
)) {
741 printk(KERN_INFO
"has_svm: %s\n", msg
);
748 static void svm_hardware_disable(void)
750 /* Make sure we clean up behind us */
751 if (static_cpu_has(X86_FEATURE_TSCRATEMSR
))
752 wrmsrl(MSR_AMD64_TSC_RATIO
, TSC_RATIO_DEFAULT
);
756 amd_pmu_disable_virt();
759 static int svm_hardware_enable(void)
762 struct svm_cpu_data
*sd
;
764 struct desc_struct
*gdt
;
765 int me
= raw_smp_processor_id();
767 rdmsrl(MSR_EFER
, efer
);
768 if (efer
& EFER_SVME
)
772 pr_err("%s: err EOPNOTSUPP on %d\n", __func__
, me
);
775 sd
= per_cpu(svm_data
, me
);
777 pr_err("%s: svm_data is NULL on %d\n", __func__
, me
);
781 sd
->asid_generation
= 1;
782 sd
->max_asid
= cpuid_ebx(SVM_CPUID_FUNC
) - 1;
783 sd
->next_asid
= sd
->max_asid
+ 1;
785 gdt
= get_current_gdt_rw();
786 sd
->tss_desc
= (struct kvm_ldttss_desc
*)(gdt
+ GDT_ENTRY_TSS
);
788 wrmsrl(MSR_EFER
, efer
| EFER_SVME
);
790 wrmsrl(MSR_VM_HSAVE_PA
, page_to_pfn(sd
->save_area
) << PAGE_SHIFT
);
792 if (static_cpu_has(X86_FEATURE_TSCRATEMSR
)) {
793 wrmsrl(MSR_AMD64_TSC_RATIO
, TSC_RATIO_DEFAULT
);
794 __this_cpu_write(current_tsc_ratio
, TSC_RATIO_DEFAULT
);
801 * Note that it is possible to have a system with mixed processor
802 * revisions and therefore different OSVW bits. If bits are not the same
803 * on different processors then choose the worst case (i.e. if erratum
804 * is present on one processor and not on another then assume that the
805 * erratum is present everywhere).
807 if (cpu_has(&boot_cpu_data
, X86_FEATURE_OSVW
)) {
808 uint64_t len
, status
= 0;
811 len
= native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH
, &err
);
813 status
= native_read_msr_safe(MSR_AMD64_OSVW_STATUS
,
817 osvw_status
= osvw_len
= 0;
821 osvw_status
|= status
;
822 osvw_status
&= (1ULL << osvw_len
) - 1;
825 osvw_status
= osvw_len
= 0;
827 svm_init_erratum_383();
829 amd_pmu_enable_virt();
834 static void svm_cpu_uninit(int cpu
)
836 struct svm_cpu_data
*sd
= per_cpu(svm_data
, raw_smp_processor_id());
841 per_cpu(svm_data
, raw_smp_processor_id()) = NULL
;
842 __free_page(sd
->save_area
);
846 static int svm_cpu_init(int cpu
)
848 struct svm_cpu_data
*sd
;
851 sd
= kzalloc(sizeof(struct svm_cpu_data
), GFP_KERNEL
);
855 sd
->save_area
= alloc_page(GFP_KERNEL
);
860 per_cpu(svm_data
, cpu
) = sd
;
870 static bool valid_msr_intercept(u32 index
)
874 for (i
= 0; direct_access_msrs
[i
].index
!= MSR_INVALID
; i
++)
875 if (direct_access_msrs
[i
].index
== index
)
881 static void set_msr_interception(u32
*msrpm
, unsigned msr
,
884 u8 bit_read
, bit_write
;
889 * If this warning triggers extend the direct_access_msrs list at the
890 * beginning of the file
892 WARN_ON(!valid_msr_intercept(msr
));
894 offset
= svm_msrpm_offset(msr
);
895 bit_read
= 2 * (msr
& 0x0f);
896 bit_write
= 2 * (msr
& 0x0f) + 1;
899 BUG_ON(offset
== MSR_INVALID
);
901 read
? clear_bit(bit_read
, &tmp
) : set_bit(bit_read
, &tmp
);
902 write
? clear_bit(bit_write
, &tmp
) : set_bit(bit_write
, &tmp
);
907 static void svm_vcpu_init_msrpm(u32
*msrpm
)
911 memset(msrpm
, 0xff, PAGE_SIZE
* (1 << MSRPM_ALLOC_ORDER
));
913 for (i
= 0; direct_access_msrs
[i
].index
!= MSR_INVALID
; i
++) {
914 if (!direct_access_msrs
[i
].always
)
917 set_msr_interception(msrpm
, direct_access_msrs
[i
].index
, 1, 1);
921 static void add_msr_offset(u32 offset
)
925 for (i
= 0; i
< MSRPM_OFFSETS
; ++i
) {
927 /* Offset already in list? */
928 if (msrpm_offsets
[i
] == offset
)
931 /* Slot used by another offset? */
932 if (msrpm_offsets
[i
] != MSR_INVALID
)
935 /* Add offset to list */
936 msrpm_offsets
[i
] = offset
;
942 * If this BUG triggers the msrpm_offsets table has an overflow. Just
943 * increase MSRPM_OFFSETS in this case.
948 static void init_msrpm_offsets(void)
952 memset(msrpm_offsets
, 0xff, sizeof(msrpm_offsets
));
954 for (i
= 0; direct_access_msrs
[i
].index
!= MSR_INVALID
; i
++) {
957 offset
= svm_msrpm_offset(direct_access_msrs
[i
].index
);
958 BUG_ON(offset
== MSR_INVALID
);
960 add_msr_offset(offset
);
964 static void svm_enable_lbrv(struct vcpu_svm
*svm
)
966 u32
*msrpm
= svm
->msrpm
;
968 svm
->vmcb
->control
.virt_ext
|= LBR_CTL_ENABLE_MASK
;
969 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHFROMIP
, 1, 1);
970 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHTOIP
, 1, 1);
971 set_msr_interception(msrpm
, MSR_IA32_LASTINTFROMIP
, 1, 1);
972 set_msr_interception(msrpm
, MSR_IA32_LASTINTTOIP
, 1, 1);
975 static void svm_disable_lbrv(struct vcpu_svm
*svm
)
977 u32
*msrpm
= svm
->msrpm
;
979 svm
->vmcb
->control
.virt_ext
&= ~LBR_CTL_ENABLE_MASK
;
980 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHFROMIP
, 0, 0);
981 set_msr_interception(msrpm
, MSR_IA32_LASTBRANCHTOIP
, 0, 0);
982 set_msr_interception(msrpm
, MSR_IA32_LASTINTFROMIP
, 0, 0);
983 set_msr_interception(msrpm
, MSR_IA32_LASTINTTOIP
, 0, 0);
986 static void disable_nmi_singlestep(struct vcpu_svm
*svm
)
988 svm
->nmi_singlestep
= false;
989 if (!(svm
->vcpu
.guest_debug
& KVM_GUESTDBG_SINGLESTEP
)) {
990 /* Clear our flags if they were not set by the guest */
991 if (!(svm
->nmi_singlestep_guest_rflags
& X86_EFLAGS_TF
))
992 svm
->vmcb
->save
.rflags
&= ~X86_EFLAGS_TF
;
993 if (!(svm
->nmi_singlestep_guest_rflags
& X86_EFLAGS_RF
))
994 svm
->vmcb
->save
.rflags
&= ~X86_EFLAGS_RF
;
999 * This hash table is used to map VM_ID to a struct kvm_arch,
1000 * when handling AMD IOMMU GALOG notification to schedule in
1001 * a particular vCPU.
1003 #define SVM_VM_DATA_HASH_BITS 8
1004 static DEFINE_HASHTABLE(svm_vm_data_hash
, SVM_VM_DATA_HASH_BITS
);
1005 static DEFINE_SPINLOCK(svm_vm_data_hash_lock
);
1008 * This function is called from IOMMU driver to notify
1009 * SVM to schedule in a particular vCPU of a particular VM.
1011 static int avic_ga_log_notifier(u32 ga_tag
)
1013 unsigned long flags
;
1014 struct kvm_arch
*ka
= NULL
;
1015 struct kvm_vcpu
*vcpu
= NULL
;
1016 u32 vm_id
= AVIC_GATAG_TO_VMID(ga_tag
);
1017 u32 vcpu_id
= AVIC_GATAG_TO_VCPUID(ga_tag
);
1019 pr_debug("SVM: %s: vm_id=%#x, vcpu_id=%#x\n", __func__
, vm_id
, vcpu_id
);
1021 spin_lock_irqsave(&svm_vm_data_hash_lock
, flags
);
1022 hash_for_each_possible(svm_vm_data_hash
, ka
, hnode
, vm_id
) {
1023 struct kvm
*kvm
= container_of(ka
, struct kvm
, arch
);
1024 struct kvm_arch
*vm_data
= &kvm
->arch
;
1026 if (vm_data
->avic_vm_id
!= vm_id
)
1028 vcpu
= kvm_get_vcpu_by_id(kvm
, vcpu_id
);
1031 spin_unlock_irqrestore(&svm_vm_data_hash_lock
, flags
);
1037 * At this point, the IOMMU should have already set the pending
1038 * bit in the vAPIC backing page. So, we just need to schedule
1041 if (vcpu
->mode
== OUTSIDE_GUEST_MODE
)
1042 kvm_vcpu_wake_up(vcpu
);
1047 static __init
int svm_hardware_setup(void)
1050 struct page
*iopm_pages
;
1054 iopm_pages
= alloc_pages(GFP_KERNEL
, IOPM_ALLOC_ORDER
);
1059 iopm_va
= page_address(iopm_pages
);
1060 memset(iopm_va
, 0xff, PAGE_SIZE
* (1 << IOPM_ALLOC_ORDER
));
1061 iopm_base
= page_to_pfn(iopm_pages
) << PAGE_SHIFT
;
1063 init_msrpm_offsets();
1065 if (boot_cpu_has(X86_FEATURE_NX
))
1066 kvm_enable_efer_bits(EFER_NX
);
1068 if (boot_cpu_has(X86_FEATURE_FXSR_OPT
))
1069 kvm_enable_efer_bits(EFER_FFXSR
);
1071 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR
)) {
1072 kvm_has_tsc_control
= true;
1073 kvm_max_tsc_scaling_ratio
= TSC_RATIO_MAX
;
1074 kvm_tsc_scaling_ratio_frac_bits
= 32;
1078 printk(KERN_INFO
"kvm: Nested Virtualization enabled\n");
1079 kvm_enable_efer_bits(EFER_SVME
| EFER_LMSLE
);
1082 for_each_possible_cpu(cpu
) {
1083 r
= svm_cpu_init(cpu
);
1088 if (!boot_cpu_has(X86_FEATURE_NPT
))
1089 npt_enabled
= false;
1091 if (npt_enabled
&& !npt
) {
1092 printk(KERN_INFO
"kvm: Nested Paging disabled\n");
1093 npt_enabled
= false;
1097 printk(KERN_INFO
"kvm: Nested Paging enabled\n");
1104 !boot_cpu_has(X86_FEATURE_AVIC
) ||
1105 !IS_ENABLED(CONFIG_X86_LOCAL_APIC
)) {
1108 pr_info("AVIC enabled\n");
1110 amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier
);
1116 !boot_cpu_has(X86_FEATURE_V_VMSAVE_VMLOAD
) ||
1117 !IS_ENABLED(CONFIG_X86_64
)) {
1120 pr_info("Virtual VMLOAD VMSAVE supported\n");
1127 __free_pages(iopm_pages
, IOPM_ALLOC_ORDER
);
1132 static __exit
void svm_hardware_unsetup(void)
1136 for_each_possible_cpu(cpu
)
1137 svm_cpu_uninit(cpu
);
1139 __free_pages(pfn_to_page(iopm_base
>> PAGE_SHIFT
), IOPM_ALLOC_ORDER
);
1143 static void init_seg(struct vmcb_seg
*seg
)
1146 seg
->attrib
= SVM_SELECTOR_P_MASK
| SVM_SELECTOR_S_MASK
|
1147 SVM_SELECTOR_WRITE_MASK
; /* Read/Write Data Segment */
1148 seg
->limit
= 0xffff;
1152 static void init_sys_seg(struct vmcb_seg
*seg
, uint32_t type
)
1155 seg
->attrib
= SVM_SELECTOR_P_MASK
| type
;
1156 seg
->limit
= 0xffff;
1160 static void svm_write_tsc_offset(struct kvm_vcpu
*vcpu
, u64 offset
)
1162 struct vcpu_svm
*svm
= to_svm(vcpu
);
1163 u64 g_tsc_offset
= 0;
1165 if (is_guest_mode(vcpu
)) {
1166 g_tsc_offset
= svm
->vmcb
->control
.tsc_offset
-
1167 svm
->nested
.hsave
->control
.tsc_offset
;
1168 svm
->nested
.hsave
->control
.tsc_offset
= offset
;
1170 trace_kvm_write_tsc_offset(vcpu
->vcpu_id
,
1171 svm
->vmcb
->control
.tsc_offset
,
1174 svm
->vmcb
->control
.tsc_offset
= offset
+ g_tsc_offset
;
1176 mark_dirty(svm
->vmcb
, VMCB_INTERCEPTS
);
1179 static void avic_init_vmcb(struct vcpu_svm
*svm
)
1181 struct vmcb
*vmcb
= svm
->vmcb
;
1182 struct kvm_arch
*vm_data
= &svm
->vcpu
.kvm
->arch
;
1183 phys_addr_t bpa
= page_to_phys(svm
->avic_backing_page
);
1184 phys_addr_t lpa
= page_to_phys(vm_data
->avic_logical_id_table_page
);
1185 phys_addr_t ppa
= page_to_phys(vm_data
->avic_physical_id_table_page
);
1187 vmcb
->control
.avic_backing_page
= bpa
& AVIC_HPA_MASK
;
1188 vmcb
->control
.avic_logical_id
= lpa
& AVIC_HPA_MASK
;
1189 vmcb
->control
.avic_physical_id
= ppa
& AVIC_HPA_MASK
;
1190 vmcb
->control
.avic_physical_id
|= AVIC_MAX_PHYSICAL_ID_COUNT
;
1191 vmcb
->control
.int_ctl
|= AVIC_ENABLE_MASK
;
1192 svm
->vcpu
.arch
.apicv_active
= true;
1195 static void init_vmcb(struct vcpu_svm
*svm
)
1197 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
1198 struct vmcb_save_area
*save
= &svm
->vmcb
->save
;
1200 svm
->vcpu
.arch
.hflags
= 0;
1202 set_cr_intercept(svm
, INTERCEPT_CR0_READ
);
1203 set_cr_intercept(svm
, INTERCEPT_CR3_READ
);
1204 set_cr_intercept(svm
, INTERCEPT_CR4_READ
);
1205 set_cr_intercept(svm
, INTERCEPT_CR0_WRITE
);
1206 set_cr_intercept(svm
, INTERCEPT_CR3_WRITE
);
1207 set_cr_intercept(svm
, INTERCEPT_CR4_WRITE
);
1208 if (!kvm_vcpu_apicv_active(&svm
->vcpu
))
1209 set_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
1211 set_dr_intercepts(svm
);
1213 set_exception_intercept(svm
, PF_VECTOR
);
1214 set_exception_intercept(svm
, UD_VECTOR
);
1215 set_exception_intercept(svm
, MC_VECTOR
);
1216 set_exception_intercept(svm
, AC_VECTOR
);
1217 set_exception_intercept(svm
, DB_VECTOR
);
1219 set_intercept(svm
, INTERCEPT_INTR
);
1220 set_intercept(svm
, INTERCEPT_NMI
);
1221 set_intercept(svm
, INTERCEPT_SMI
);
1222 set_intercept(svm
, INTERCEPT_SELECTIVE_CR0
);
1223 set_intercept(svm
, INTERCEPT_RDPMC
);
1224 set_intercept(svm
, INTERCEPT_CPUID
);
1225 set_intercept(svm
, INTERCEPT_INVD
);
1226 set_intercept(svm
, INTERCEPT_HLT
);
1227 set_intercept(svm
, INTERCEPT_INVLPG
);
1228 set_intercept(svm
, INTERCEPT_INVLPGA
);
1229 set_intercept(svm
, INTERCEPT_IOIO_PROT
);
1230 set_intercept(svm
, INTERCEPT_MSR_PROT
);
1231 set_intercept(svm
, INTERCEPT_TASK_SWITCH
);
1232 set_intercept(svm
, INTERCEPT_SHUTDOWN
);
1233 set_intercept(svm
, INTERCEPT_VMRUN
);
1234 set_intercept(svm
, INTERCEPT_VMMCALL
);
1235 set_intercept(svm
, INTERCEPT_VMLOAD
);
1236 set_intercept(svm
, INTERCEPT_VMSAVE
);
1237 set_intercept(svm
, INTERCEPT_STGI
);
1238 set_intercept(svm
, INTERCEPT_CLGI
);
1239 set_intercept(svm
, INTERCEPT_SKINIT
);
1240 set_intercept(svm
, INTERCEPT_WBINVD
);
1241 set_intercept(svm
, INTERCEPT_XSETBV
);
1243 if (!kvm_mwait_in_guest()) {
1244 set_intercept(svm
, INTERCEPT_MONITOR
);
1245 set_intercept(svm
, INTERCEPT_MWAIT
);
1248 control
->iopm_base_pa
= iopm_base
;
1249 control
->msrpm_base_pa
= __pa(svm
->msrpm
);
1250 control
->int_ctl
= V_INTR_MASKING_MASK
;
1252 init_seg(&save
->es
);
1253 init_seg(&save
->ss
);
1254 init_seg(&save
->ds
);
1255 init_seg(&save
->fs
);
1256 init_seg(&save
->gs
);
1258 save
->cs
.selector
= 0xf000;
1259 save
->cs
.base
= 0xffff0000;
1260 /* Executable/Readable Code Segment */
1261 save
->cs
.attrib
= SVM_SELECTOR_READ_MASK
| SVM_SELECTOR_P_MASK
|
1262 SVM_SELECTOR_S_MASK
| SVM_SELECTOR_CODE_MASK
;
1263 save
->cs
.limit
= 0xffff;
1265 save
->gdtr
.limit
= 0xffff;
1266 save
->idtr
.limit
= 0xffff;
1268 init_sys_seg(&save
->ldtr
, SEG_TYPE_LDT
);
1269 init_sys_seg(&save
->tr
, SEG_TYPE_BUSY_TSS16
);
1271 svm_set_efer(&svm
->vcpu
, 0);
1272 save
->dr6
= 0xffff0ff0;
1273 kvm_set_rflags(&svm
->vcpu
, 2);
1274 save
->rip
= 0x0000fff0;
1275 svm
->vcpu
.arch
.regs
[VCPU_REGS_RIP
] = save
->rip
;
1278 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1279 * It also updates the guest-visible cr0 value.
1281 svm_set_cr0(&svm
->vcpu
, X86_CR0_NW
| X86_CR0_CD
| X86_CR0_ET
);
1282 kvm_mmu_reset_context(&svm
->vcpu
);
1284 save
->cr4
= X86_CR4_PAE
;
1288 /* Setup VMCB for Nested Paging */
1289 control
->nested_ctl
= 1;
1290 clr_intercept(svm
, INTERCEPT_INVLPG
);
1291 clr_exception_intercept(svm
, PF_VECTOR
);
1292 clr_cr_intercept(svm
, INTERCEPT_CR3_READ
);
1293 clr_cr_intercept(svm
, INTERCEPT_CR3_WRITE
);
1294 save
->g_pat
= svm
->vcpu
.arch
.pat
;
1298 svm
->asid_generation
= 0;
1300 svm
->nested
.vmcb
= 0;
1301 svm
->vcpu
.arch
.hflags
= 0;
1303 if (boot_cpu_has(X86_FEATURE_PAUSEFILTER
)) {
1304 control
->pause_filter_count
= 3000;
1305 set_intercept(svm
, INTERCEPT_PAUSE
);
1309 avic_init_vmcb(svm
);
1312 * If hardware supports Virtual VMLOAD VMSAVE then enable it
1313 * in VMCB and clear intercepts to avoid #VMEXIT.
1316 clr_intercept(svm
, INTERCEPT_VMLOAD
);
1317 clr_intercept(svm
, INTERCEPT_VMSAVE
);
1318 svm
->vmcb
->control
.virt_ext
|= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK
;
1321 mark_all_dirty(svm
->vmcb
);
1327 static u64
*avic_get_physical_id_entry(struct kvm_vcpu
*vcpu
,
1330 u64
*avic_physical_id_table
;
1331 struct kvm_arch
*vm_data
= &vcpu
->kvm
->arch
;
1333 if (index
>= AVIC_MAX_PHYSICAL_ID_COUNT
)
1336 avic_physical_id_table
= page_address(vm_data
->avic_physical_id_table_page
);
1338 return &avic_physical_id_table
[index
];
1343 * AVIC hardware walks the nested page table to check permissions,
1344 * but does not use the SPA address specified in the leaf page
1345 * table entry since it uses address in the AVIC_BACKING_PAGE pointer
1346 * field of the VMCB. Therefore, we set up the
1347 * APIC_ACCESS_PAGE_PRIVATE_MEMSLOT (4KB) here.
1349 static int avic_init_access_page(struct kvm_vcpu
*vcpu
)
1351 struct kvm
*kvm
= vcpu
->kvm
;
1354 if (kvm
->arch
.apic_access_page_done
)
1357 ret
= x86_set_memory_region(kvm
,
1358 APIC_ACCESS_PAGE_PRIVATE_MEMSLOT
,
1359 APIC_DEFAULT_PHYS_BASE
,
1364 kvm
->arch
.apic_access_page_done
= true;
1368 static int avic_init_backing_page(struct kvm_vcpu
*vcpu
)
1371 u64
*entry
, new_entry
;
1372 int id
= vcpu
->vcpu_id
;
1373 struct vcpu_svm
*svm
= to_svm(vcpu
);
1375 ret
= avic_init_access_page(vcpu
);
1379 if (id
>= AVIC_MAX_PHYSICAL_ID_COUNT
)
1382 if (!svm
->vcpu
.arch
.apic
->regs
)
1385 svm
->avic_backing_page
= virt_to_page(svm
->vcpu
.arch
.apic
->regs
);
1387 /* Setting AVIC backing page address in the phy APIC ID table */
1388 entry
= avic_get_physical_id_entry(vcpu
, id
);
1392 new_entry
= READ_ONCE(*entry
);
1393 new_entry
= (page_to_phys(svm
->avic_backing_page
) &
1394 AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK
) |
1395 AVIC_PHYSICAL_ID_ENTRY_VALID_MASK
;
1396 WRITE_ONCE(*entry
, new_entry
);
1398 svm
->avic_physical_id_cache
= entry
;
1403 static inline int avic_get_next_vm_id(void)
1407 spin_lock(&avic_vm_id_lock
);
1409 /* AVIC VM ID is one-based. */
1410 id
= find_next_zero_bit(avic_vm_id_bitmap
, AVIC_VM_ID_NR
, 1);
1411 if (id
<= AVIC_VM_ID_MASK
)
1412 __set_bit(id
, avic_vm_id_bitmap
);
1416 spin_unlock(&avic_vm_id_lock
);
1420 static inline int avic_free_vm_id(int id
)
1422 if (id
<= 0 || id
> AVIC_VM_ID_MASK
)
1425 spin_lock(&avic_vm_id_lock
);
1426 __clear_bit(id
, avic_vm_id_bitmap
);
1427 spin_unlock(&avic_vm_id_lock
);
1431 static void avic_vm_destroy(struct kvm
*kvm
)
1433 unsigned long flags
;
1434 struct kvm_arch
*vm_data
= &kvm
->arch
;
1439 avic_free_vm_id(vm_data
->avic_vm_id
);
1441 if (vm_data
->avic_logical_id_table_page
)
1442 __free_page(vm_data
->avic_logical_id_table_page
);
1443 if (vm_data
->avic_physical_id_table_page
)
1444 __free_page(vm_data
->avic_physical_id_table_page
);
1446 spin_lock_irqsave(&svm_vm_data_hash_lock
, flags
);
1447 hash_del(&vm_data
->hnode
);
1448 spin_unlock_irqrestore(&svm_vm_data_hash_lock
, flags
);
1451 static int avic_vm_init(struct kvm
*kvm
)
1453 unsigned long flags
;
1454 int vm_id
, err
= -ENOMEM
;
1455 struct kvm_arch
*vm_data
= &kvm
->arch
;
1456 struct page
*p_page
;
1457 struct page
*l_page
;
1462 vm_id
= avic_get_next_vm_id();
1465 vm_data
->avic_vm_id
= (u32
)vm_id
;
1467 /* Allocating physical APIC ID table (4KB) */
1468 p_page
= alloc_page(GFP_KERNEL
);
1472 vm_data
->avic_physical_id_table_page
= p_page
;
1473 clear_page(page_address(p_page
));
1475 /* Allocating logical APIC ID table (4KB) */
1476 l_page
= alloc_page(GFP_KERNEL
);
1480 vm_data
->avic_logical_id_table_page
= l_page
;
1481 clear_page(page_address(l_page
));
1483 spin_lock_irqsave(&svm_vm_data_hash_lock
, flags
);
1484 hash_add(svm_vm_data_hash
, &vm_data
->hnode
, vm_data
->avic_vm_id
);
1485 spin_unlock_irqrestore(&svm_vm_data_hash_lock
, flags
);
1490 avic_vm_destroy(kvm
);
1495 avic_update_iommu_vcpu_affinity(struct kvm_vcpu
*vcpu
, int cpu
, bool r
)
1498 unsigned long flags
;
1499 struct amd_svm_iommu_ir
*ir
;
1500 struct vcpu_svm
*svm
= to_svm(vcpu
);
1502 if (!kvm_arch_has_assigned_device(vcpu
->kvm
))
1506 * Here, we go through the per-vcpu ir_list to update all existing
1507 * interrupt remapping table entry targeting this vcpu.
1509 spin_lock_irqsave(&svm
->ir_list_lock
, flags
);
1511 if (list_empty(&svm
->ir_list
))
1514 list_for_each_entry(ir
, &svm
->ir_list
, node
) {
1515 ret
= amd_iommu_update_ga(cpu
, r
, ir
->data
);
1520 spin_unlock_irqrestore(&svm
->ir_list_lock
, flags
);
1524 static void avic_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
1527 /* ID = 0xff (broadcast), ID > 0xff (reserved) */
1528 int h_physical_id
= kvm_cpu_get_apicid(cpu
);
1529 struct vcpu_svm
*svm
= to_svm(vcpu
);
1531 if (!kvm_vcpu_apicv_active(vcpu
))
1534 if (WARN_ON(h_physical_id
>= AVIC_MAX_PHYSICAL_ID_COUNT
))
1537 entry
= READ_ONCE(*(svm
->avic_physical_id_cache
));
1538 WARN_ON(entry
& AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK
);
1540 entry
&= ~AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK
;
1541 entry
|= (h_physical_id
& AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK
);
1543 entry
&= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK
;
1544 if (svm
->avic_is_running
)
1545 entry
|= AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK
;
1547 WRITE_ONCE(*(svm
->avic_physical_id_cache
), entry
);
1548 avic_update_iommu_vcpu_affinity(vcpu
, h_physical_id
,
1549 svm
->avic_is_running
);
1552 static void avic_vcpu_put(struct kvm_vcpu
*vcpu
)
1555 struct vcpu_svm
*svm
= to_svm(vcpu
);
1557 if (!kvm_vcpu_apicv_active(vcpu
))
1560 entry
= READ_ONCE(*(svm
->avic_physical_id_cache
));
1561 if (entry
& AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK
)
1562 avic_update_iommu_vcpu_affinity(vcpu
, -1, 0);
1564 entry
&= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK
;
1565 WRITE_ONCE(*(svm
->avic_physical_id_cache
), entry
);
1569 * This function is called during VCPU halt/unhalt.
1571 static void avic_set_running(struct kvm_vcpu
*vcpu
, bool is_run
)
1573 struct vcpu_svm
*svm
= to_svm(vcpu
);
1575 svm
->avic_is_running
= is_run
;
1577 avic_vcpu_load(vcpu
, vcpu
->cpu
);
1579 avic_vcpu_put(vcpu
);
1582 static void svm_vcpu_reset(struct kvm_vcpu
*vcpu
, bool init_event
)
1584 struct vcpu_svm
*svm
= to_svm(vcpu
);
1589 svm
->virt_spec_ctrl
= 0;
1592 svm
->vcpu
.arch
.apic_base
= APIC_DEFAULT_PHYS_BASE
|
1593 MSR_IA32_APICBASE_ENABLE
;
1594 if (kvm_vcpu_is_reset_bsp(&svm
->vcpu
))
1595 svm
->vcpu
.arch
.apic_base
|= MSR_IA32_APICBASE_BSP
;
1599 kvm_cpuid(vcpu
, &eax
, &dummy
, &dummy
, &dummy
);
1600 kvm_register_write(vcpu
, VCPU_REGS_RDX
, eax
);
1602 if (kvm_vcpu_apicv_active(vcpu
) && !init_event
)
1603 avic_update_vapic_bar(svm
, APIC_DEFAULT_PHYS_BASE
);
1606 static struct kvm_vcpu
*svm_create_vcpu(struct kvm
*kvm
, unsigned int id
)
1608 struct vcpu_svm
*svm
;
1610 struct page
*msrpm_pages
;
1611 struct page
*hsave_page
;
1612 struct page
*nested_msrpm_pages
;
1615 svm
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
1621 err
= kvm_vcpu_init(&svm
->vcpu
, kvm
, id
);
1626 page
= alloc_page(GFP_KERNEL
);
1630 msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
1634 nested_msrpm_pages
= alloc_pages(GFP_KERNEL
, MSRPM_ALLOC_ORDER
);
1635 if (!nested_msrpm_pages
)
1638 hsave_page
= alloc_page(GFP_KERNEL
);
1643 err
= avic_init_backing_page(&svm
->vcpu
);
1647 INIT_LIST_HEAD(&svm
->ir_list
);
1648 spin_lock_init(&svm
->ir_list_lock
);
1651 /* We initialize this flag to true to make sure that the is_running
1652 * bit would be set the first time the vcpu is loaded.
1654 svm
->avic_is_running
= true;
1656 svm
->nested
.hsave
= page_address(hsave_page
);
1658 svm
->msrpm
= page_address(msrpm_pages
);
1659 svm_vcpu_init_msrpm(svm
->msrpm
);
1661 svm
->nested
.msrpm
= page_address(nested_msrpm_pages
);
1662 svm_vcpu_init_msrpm(svm
->nested
.msrpm
);
1664 svm
->vmcb
= page_address(page
);
1665 clear_page(svm
->vmcb
);
1666 svm
->vmcb_pa
= page_to_pfn(page
) << PAGE_SHIFT
;
1667 svm
->asid_generation
= 0;
1670 svm_init_osvw(&svm
->vcpu
);
1675 __free_page(hsave_page
);
1677 __free_pages(nested_msrpm_pages
, MSRPM_ALLOC_ORDER
);
1679 __free_pages(msrpm_pages
, MSRPM_ALLOC_ORDER
);
1683 kvm_vcpu_uninit(&svm
->vcpu
);
1685 kmem_cache_free(kvm_vcpu_cache
, svm
);
1687 return ERR_PTR(err
);
1690 static void svm_free_vcpu(struct kvm_vcpu
*vcpu
)
1692 struct vcpu_svm
*svm
= to_svm(vcpu
);
1694 __free_page(pfn_to_page(svm
->vmcb_pa
>> PAGE_SHIFT
));
1695 __free_pages(virt_to_page(svm
->msrpm
), MSRPM_ALLOC_ORDER
);
1696 __free_page(virt_to_page(svm
->nested
.hsave
));
1697 __free_pages(virt_to_page(svm
->nested
.msrpm
), MSRPM_ALLOC_ORDER
);
1698 kvm_vcpu_uninit(vcpu
);
1699 kmem_cache_free(kvm_vcpu_cache
, svm
);
1702 * The VMCB could be recycled, causing a false negative in svm_vcpu_load;
1703 * block speculative execution.
1706 wrmsrl(MSR_IA32_PRED_CMD
, FEATURE_SET_IBPB
);
1709 static void svm_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
1711 struct vcpu_svm
*svm
= to_svm(vcpu
);
1712 struct svm_cpu_data
*sd
= per_cpu(svm_data
, cpu
);
1715 if (unlikely(cpu
!= vcpu
->cpu
)) {
1716 svm
->asid_generation
= 0;
1717 mark_all_dirty(svm
->vmcb
);
1720 #ifdef CONFIG_X86_64
1721 rdmsrl(MSR_GS_BASE
, to_svm(vcpu
)->host
.gs_base
);
1723 savesegment(fs
, svm
->host
.fs
);
1724 savesegment(gs
, svm
->host
.gs
);
1725 svm
->host
.ldt
= kvm_read_ldt();
1727 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
1728 rdmsrl(host_save_user_msrs
[i
], svm
->host_user_msrs
[i
]);
1730 if (static_cpu_has(X86_FEATURE_TSCRATEMSR
)) {
1731 u64 tsc_ratio
= vcpu
->arch
.tsc_scaling_ratio
;
1732 if (tsc_ratio
!= __this_cpu_read(current_tsc_ratio
)) {
1733 __this_cpu_write(current_tsc_ratio
, tsc_ratio
);
1734 wrmsrl(MSR_AMD64_TSC_RATIO
, tsc_ratio
);
1737 /* This assumes that the kernel never uses MSR_TSC_AUX */
1738 if (static_cpu_has(X86_FEATURE_RDTSCP
))
1739 wrmsrl(MSR_TSC_AUX
, svm
->tsc_aux
);
1741 if (sd
->current_vmcb
!= svm
->vmcb
) {
1742 sd
->current_vmcb
= svm
->vmcb
;
1744 wrmsrl(MSR_IA32_PRED_CMD
, FEATURE_SET_IBPB
);
1747 avic_vcpu_load(vcpu
, cpu
);
1750 static void svm_vcpu_put(struct kvm_vcpu
*vcpu
)
1752 struct vcpu_svm
*svm
= to_svm(vcpu
);
1755 avic_vcpu_put(vcpu
);
1757 ++vcpu
->stat
.host_state_reload
;
1758 kvm_load_ldt(svm
->host
.ldt
);
1759 #ifdef CONFIG_X86_64
1760 loadsegment(fs
, svm
->host
.fs
);
1761 wrmsrl(MSR_KERNEL_GS_BASE
, current
->thread
.gsbase
);
1762 load_gs_index(svm
->host
.gs
);
1764 #ifdef CONFIG_X86_32_LAZY_GS
1765 loadsegment(gs
, svm
->host
.gs
);
1768 for (i
= 0; i
< NR_HOST_SAVE_USER_MSRS
; i
++)
1769 wrmsrl(host_save_user_msrs
[i
], svm
->host_user_msrs
[i
]);
1772 static void svm_vcpu_blocking(struct kvm_vcpu
*vcpu
)
1774 avic_set_running(vcpu
, false);
1777 static void svm_vcpu_unblocking(struct kvm_vcpu
*vcpu
)
1779 avic_set_running(vcpu
, true);
1782 static unsigned long svm_get_rflags(struct kvm_vcpu
*vcpu
)
1784 struct vcpu_svm
*svm
= to_svm(vcpu
);
1785 unsigned long rflags
= svm
->vmcb
->save
.rflags
;
1787 if (svm
->nmi_singlestep
) {
1788 /* Hide our flags if they were not set by the guest */
1789 if (!(svm
->nmi_singlestep_guest_rflags
& X86_EFLAGS_TF
))
1790 rflags
&= ~X86_EFLAGS_TF
;
1791 if (!(svm
->nmi_singlestep_guest_rflags
& X86_EFLAGS_RF
))
1792 rflags
&= ~X86_EFLAGS_RF
;
1797 static void svm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
1799 if (to_svm(vcpu
)->nmi_singlestep
)
1800 rflags
|= (X86_EFLAGS_TF
| X86_EFLAGS_RF
);
1803 * Any change of EFLAGS.VM is accompanied by a reload of SS
1804 * (caused by either a task switch or an inter-privilege IRET),
1805 * so we do not need to update the CPL here.
1807 to_svm(vcpu
)->vmcb
->save
.rflags
= rflags
;
1810 static void svm_cache_reg(struct kvm_vcpu
*vcpu
, enum kvm_reg reg
)
1813 case VCPU_EXREG_PDPTR
:
1814 BUG_ON(!npt_enabled
);
1815 load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, kvm_read_cr3(vcpu
));
1822 static void svm_set_vintr(struct vcpu_svm
*svm
)
1824 set_intercept(svm
, INTERCEPT_VINTR
);
1827 static void svm_clear_vintr(struct vcpu_svm
*svm
)
1829 clr_intercept(svm
, INTERCEPT_VINTR
);
1832 static struct vmcb_seg
*svm_seg(struct kvm_vcpu
*vcpu
, int seg
)
1834 struct vmcb_save_area
*save
= &to_svm(vcpu
)->vmcb
->save
;
1837 case VCPU_SREG_CS
: return &save
->cs
;
1838 case VCPU_SREG_DS
: return &save
->ds
;
1839 case VCPU_SREG_ES
: return &save
->es
;
1840 case VCPU_SREG_FS
: return &save
->fs
;
1841 case VCPU_SREG_GS
: return &save
->gs
;
1842 case VCPU_SREG_SS
: return &save
->ss
;
1843 case VCPU_SREG_TR
: return &save
->tr
;
1844 case VCPU_SREG_LDTR
: return &save
->ldtr
;
1850 static u64
svm_get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
1852 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
1857 static void svm_get_segment(struct kvm_vcpu
*vcpu
,
1858 struct kvm_segment
*var
, int seg
)
1860 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
1862 var
->base
= s
->base
;
1863 var
->limit
= s
->limit
;
1864 var
->selector
= s
->selector
;
1865 var
->type
= s
->attrib
& SVM_SELECTOR_TYPE_MASK
;
1866 var
->s
= (s
->attrib
>> SVM_SELECTOR_S_SHIFT
) & 1;
1867 var
->dpl
= (s
->attrib
>> SVM_SELECTOR_DPL_SHIFT
) & 3;
1868 var
->present
= (s
->attrib
>> SVM_SELECTOR_P_SHIFT
) & 1;
1869 var
->avl
= (s
->attrib
>> SVM_SELECTOR_AVL_SHIFT
) & 1;
1870 var
->l
= (s
->attrib
>> SVM_SELECTOR_L_SHIFT
) & 1;
1871 var
->db
= (s
->attrib
>> SVM_SELECTOR_DB_SHIFT
) & 1;
1874 * AMD CPUs circa 2014 track the G bit for all segments except CS.
1875 * However, the SVM spec states that the G bit is not observed by the
1876 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
1877 * So let's synthesize a legal G bit for all segments, this helps
1878 * running KVM nested. It also helps cross-vendor migration, because
1879 * Intel's vmentry has a check on the 'G' bit.
1881 var
->g
= s
->limit
> 0xfffff;
1884 * AMD's VMCB does not have an explicit unusable field, so emulate it
1885 * for cross vendor migration purposes by "not present"
1887 var
->unusable
= !var
->present
;
1892 * Work around a bug where the busy flag in the tr selector
1902 * The accessed bit must always be set in the segment
1903 * descriptor cache, although it can be cleared in the
1904 * descriptor, the cached bit always remains at 1. Since
1905 * Intel has a check on this, set it here to support
1906 * cross-vendor migration.
1913 * On AMD CPUs sometimes the DB bit in the segment
1914 * descriptor is left as 1, although the whole segment has
1915 * been made unusable. Clear it here to pass an Intel VMX
1916 * entry check when cross vendor migrating.
1920 /* This is symmetric with svm_set_segment() */
1921 var
->dpl
= to_svm(vcpu
)->vmcb
->save
.cpl
;
1926 static int svm_get_cpl(struct kvm_vcpu
*vcpu
)
1928 struct vmcb_save_area
*save
= &to_svm(vcpu
)->vmcb
->save
;
1933 static void svm_get_idt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1935 struct vcpu_svm
*svm
= to_svm(vcpu
);
1937 dt
->size
= svm
->vmcb
->save
.idtr
.limit
;
1938 dt
->address
= svm
->vmcb
->save
.idtr
.base
;
1941 static void svm_set_idt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1943 struct vcpu_svm
*svm
= to_svm(vcpu
);
1945 svm
->vmcb
->save
.idtr
.limit
= dt
->size
;
1946 svm
->vmcb
->save
.idtr
.base
= dt
->address
;
1947 mark_dirty(svm
->vmcb
, VMCB_DT
);
1950 static void svm_get_gdt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1952 struct vcpu_svm
*svm
= to_svm(vcpu
);
1954 dt
->size
= svm
->vmcb
->save
.gdtr
.limit
;
1955 dt
->address
= svm
->vmcb
->save
.gdtr
.base
;
1958 static void svm_set_gdt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
1960 struct vcpu_svm
*svm
= to_svm(vcpu
);
1962 svm
->vmcb
->save
.gdtr
.limit
= dt
->size
;
1963 svm
->vmcb
->save
.gdtr
.base
= dt
->address
;
1964 mark_dirty(svm
->vmcb
, VMCB_DT
);
1967 static void svm_decache_cr0_guest_bits(struct kvm_vcpu
*vcpu
)
1971 static void svm_decache_cr3(struct kvm_vcpu
*vcpu
)
1975 static void svm_decache_cr4_guest_bits(struct kvm_vcpu
*vcpu
)
1979 static void update_cr0_intercept(struct vcpu_svm
*svm
)
1981 ulong gcr0
= svm
->vcpu
.arch
.cr0
;
1982 u64
*hcr0
= &svm
->vmcb
->save
.cr0
;
1984 *hcr0
= (*hcr0
& ~SVM_CR0_SELECTIVE_MASK
)
1985 | (gcr0
& SVM_CR0_SELECTIVE_MASK
);
1987 mark_dirty(svm
->vmcb
, VMCB_CR
);
1989 if (gcr0
== *hcr0
) {
1990 clr_cr_intercept(svm
, INTERCEPT_CR0_READ
);
1991 clr_cr_intercept(svm
, INTERCEPT_CR0_WRITE
);
1993 set_cr_intercept(svm
, INTERCEPT_CR0_READ
);
1994 set_cr_intercept(svm
, INTERCEPT_CR0_WRITE
);
1998 static void svm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
2000 struct vcpu_svm
*svm
= to_svm(vcpu
);
2002 #ifdef CONFIG_X86_64
2003 if (vcpu
->arch
.efer
& EFER_LME
) {
2004 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
2005 vcpu
->arch
.efer
|= EFER_LMA
;
2006 svm
->vmcb
->save
.efer
|= EFER_LMA
| EFER_LME
;
2009 if (is_paging(vcpu
) && !(cr0
& X86_CR0_PG
)) {
2010 vcpu
->arch
.efer
&= ~EFER_LMA
;
2011 svm
->vmcb
->save
.efer
&= ~(EFER_LMA
| EFER_LME
);
2015 vcpu
->arch
.cr0
= cr0
;
2018 cr0
|= X86_CR0_PG
| X86_CR0_WP
;
2021 * re-enable caching here because the QEMU bios
2022 * does not do it - this results in some delay at
2025 if (kvm_check_has_quirk(vcpu
->kvm
, KVM_X86_QUIRK_CD_NW_CLEARED
))
2026 cr0
&= ~(X86_CR0_CD
| X86_CR0_NW
);
2027 svm
->vmcb
->save
.cr0
= cr0
;
2028 mark_dirty(svm
->vmcb
, VMCB_CR
);
2029 update_cr0_intercept(svm
);
2032 static int svm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
2034 unsigned long host_cr4_mce
= cr4_read_shadow() & X86_CR4_MCE
;
2035 unsigned long old_cr4
= to_svm(vcpu
)->vmcb
->save
.cr4
;
2037 if (cr4
& X86_CR4_VMXE
)
2040 if (npt_enabled
&& ((old_cr4
^ cr4
) & X86_CR4_PGE
))
2041 svm_flush_tlb(vcpu
);
2043 vcpu
->arch
.cr4
= cr4
;
2046 cr4
|= host_cr4_mce
;
2047 to_svm(vcpu
)->vmcb
->save
.cr4
= cr4
;
2048 mark_dirty(to_svm(vcpu
)->vmcb
, VMCB_CR
);
2052 static void svm_set_segment(struct kvm_vcpu
*vcpu
,
2053 struct kvm_segment
*var
, int seg
)
2055 struct vcpu_svm
*svm
= to_svm(vcpu
);
2056 struct vmcb_seg
*s
= svm_seg(vcpu
, seg
);
2058 s
->base
= var
->base
;
2059 s
->limit
= var
->limit
;
2060 s
->selector
= var
->selector
;
2061 s
->attrib
= (var
->type
& SVM_SELECTOR_TYPE_MASK
);
2062 s
->attrib
|= (var
->s
& 1) << SVM_SELECTOR_S_SHIFT
;
2063 s
->attrib
|= (var
->dpl
& 3) << SVM_SELECTOR_DPL_SHIFT
;
2064 s
->attrib
|= ((var
->present
& 1) && !var
->unusable
) << SVM_SELECTOR_P_SHIFT
;
2065 s
->attrib
|= (var
->avl
& 1) << SVM_SELECTOR_AVL_SHIFT
;
2066 s
->attrib
|= (var
->l
& 1) << SVM_SELECTOR_L_SHIFT
;
2067 s
->attrib
|= (var
->db
& 1) << SVM_SELECTOR_DB_SHIFT
;
2068 s
->attrib
|= (var
->g
& 1) << SVM_SELECTOR_G_SHIFT
;
2071 * This is always accurate, except if SYSRET returned to a segment
2072 * with SS.DPL != 3. Intel does not have this quirk, and always
2073 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
2074 * would entail passing the CPL to userspace and back.
2076 if (seg
== VCPU_SREG_SS
)
2077 /* This is symmetric with svm_get_segment() */
2078 svm
->vmcb
->save
.cpl
= (var
->dpl
& 3);
2080 mark_dirty(svm
->vmcb
, VMCB_SEG
);
2083 static void update_bp_intercept(struct kvm_vcpu
*vcpu
)
2085 struct vcpu_svm
*svm
= to_svm(vcpu
);
2087 clr_exception_intercept(svm
, BP_VECTOR
);
2089 if (vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
) {
2090 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_SW_BP
)
2091 set_exception_intercept(svm
, BP_VECTOR
);
2093 vcpu
->guest_debug
= 0;
2096 static void new_asid(struct vcpu_svm
*svm
, struct svm_cpu_data
*sd
)
2098 if (sd
->next_asid
> sd
->max_asid
) {
2099 ++sd
->asid_generation
;
2101 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ALL_ASID
;
2104 svm
->asid_generation
= sd
->asid_generation
;
2105 svm
->vmcb
->control
.asid
= sd
->next_asid
++;
2107 mark_dirty(svm
->vmcb
, VMCB_ASID
);
2110 static u64
svm_get_dr6(struct kvm_vcpu
*vcpu
)
2112 return to_svm(vcpu
)->vmcb
->save
.dr6
;
2115 static void svm_set_dr6(struct kvm_vcpu
*vcpu
, unsigned long value
)
2117 struct vcpu_svm
*svm
= to_svm(vcpu
);
2119 svm
->vmcb
->save
.dr6
= value
;
2120 mark_dirty(svm
->vmcb
, VMCB_DR
);
2123 static void svm_sync_dirty_debug_regs(struct kvm_vcpu
*vcpu
)
2125 struct vcpu_svm
*svm
= to_svm(vcpu
);
2127 get_debugreg(vcpu
->arch
.db
[0], 0);
2128 get_debugreg(vcpu
->arch
.db
[1], 1);
2129 get_debugreg(vcpu
->arch
.db
[2], 2);
2130 get_debugreg(vcpu
->arch
.db
[3], 3);
2131 vcpu
->arch
.dr6
= svm_get_dr6(vcpu
);
2132 vcpu
->arch
.dr7
= svm
->vmcb
->save
.dr7
;
2134 vcpu
->arch
.switch_db_regs
&= ~KVM_DEBUGREG_WONT_EXIT
;
2135 set_dr_intercepts(svm
);
2138 static void svm_set_dr7(struct kvm_vcpu
*vcpu
, unsigned long value
)
2140 struct vcpu_svm
*svm
= to_svm(vcpu
);
2142 svm
->vmcb
->save
.dr7
= value
;
2143 mark_dirty(svm
->vmcb
, VMCB_DR
);
2146 static int pf_interception(struct vcpu_svm
*svm
)
2148 u64 fault_address
= svm
->vmcb
->control
.exit_info_2
;
2149 u64 error_code
= svm
->vmcb
->control
.exit_info_1
;
2151 return kvm_handle_page_fault(&svm
->vcpu
, error_code
, fault_address
,
2152 svm
->vmcb
->control
.insn_bytes
,
2153 svm
->vmcb
->control
.insn_len
, !npt_enabled
);
2156 static int db_interception(struct vcpu_svm
*svm
)
2158 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
2160 if (!(svm
->vcpu
.guest_debug
&
2161 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
)) &&
2162 !svm
->nmi_singlestep
) {
2163 kvm_queue_exception(&svm
->vcpu
, DB_VECTOR
);
2167 if (svm
->nmi_singlestep
) {
2168 disable_nmi_singlestep(svm
);
2171 if (svm
->vcpu
.guest_debug
&
2172 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
)) {
2173 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
2174 kvm_run
->debug
.arch
.pc
=
2175 svm
->vmcb
->save
.cs
.base
+ svm
->vmcb
->save
.rip
;
2176 kvm_run
->debug
.arch
.exception
= DB_VECTOR
;
2183 static int bp_interception(struct vcpu_svm
*svm
)
2185 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
2187 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
2188 kvm_run
->debug
.arch
.pc
= svm
->vmcb
->save
.cs
.base
+ svm
->vmcb
->save
.rip
;
2189 kvm_run
->debug
.arch
.exception
= BP_VECTOR
;
2193 static int ud_interception(struct vcpu_svm
*svm
)
2197 er
= emulate_instruction(&svm
->vcpu
, EMULTYPE_TRAP_UD
);
2198 if (er
!= EMULATE_DONE
)
2199 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
2203 static int ac_interception(struct vcpu_svm
*svm
)
2205 kvm_queue_exception_e(&svm
->vcpu
, AC_VECTOR
, 0);
2209 static bool is_erratum_383(void)
2214 if (!erratum_383_found
)
2217 value
= native_read_msr_safe(MSR_IA32_MC0_STATUS
, &err
);
2221 /* Bit 62 may or may not be set for this mce */
2222 value
&= ~(1ULL << 62);
2224 if (value
!= 0xb600000000010015ULL
)
2227 /* Clear MCi_STATUS registers */
2228 for (i
= 0; i
< 6; ++i
)
2229 native_write_msr_safe(MSR_IA32_MCx_STATUS(i
), 0, 0);
2231 value
= native_read_msr_safe(MSR_IA32_MCG_STATUS
, &err
);
2235 value
&= ~(1ULL << 2);
2236 low
= lower_32_bits(value
);
2237 high
= upper_32_bits(value
);
2239 native_write_msr_safe(MSR_IA32_MCG_STATUS
, low
, high
);
2242 /* Flush tlb to evict multi-match entries */
2248 static void svm_handle_mce(struct vcpu_svm
*svm
)
2250 if (is_erratum_383()) {
2252 * Erratum 383 triggered. Guest state is corrupt so kill the
2255 pr_err("KVM: Guest triggered AMD Erratum 383\n");
2257 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, &svm
->vcpu
);
2263 * On an #MC intercept the MCE handler is not called automatically in
2264 * the host. So do it by hand here.
2268 /* not sure if we ever come back to this point */
2273 static int mc_interception(struct vcpu_svm
*svm
)
2278 static int shutdown_interception(struct vcpu_svm
*svm
)
2280 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
2283 * VMCB is undefined after a SHUTDOWN intercept
2284 * so reinitialize it.
2286 clear_page(svm
->vmcb
);
2289 kvm_run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
2293 static int io_interception(struct vcpu_svm
*svm
)
2295 struct kvm_vcpu
*vcpu
= &svm
->vcpu
;
2296 u32 io_info
= svm
->vmcb
->control
.exit_info_1
; /* address size bug? */
2297 int size
, in
, string
, ret
;
2300 ++svm
->vcpu
.stat
.io_exits
;
2301 string
= (io_info
& SVM_IOIO_STR_MASK
) != 0;
2302 in
= (io_info
& SVM_IOIO_TYPE_MASK
) != 0;
2304 return emulate_instruction(vcpu
, 0) == EMULATE_DONE
;
2306 port
= io_info
>> 16;
2307 size
= (io_info
& SVM_IOIO_SIZE_MASK
) >> SVM_IOIO_SIZE_SHIFT
;
2308 svm
->next_rip
= svm
->vmcb
->control
.exit_info_2
;
2309 ret
= kvm_skip_emulated_instruction(&svm
->vcpu
);
2312 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
2313 * KVM_EXIT_DEBUG here.
2316 return kvm_fast_pio_in(vcpu
, size
, port
) && ret
;
2318 return kvm_fast_pio_out(vcpu
, size
, port
) && ret
;
2321 static int nmi_interception(struct vcpu_svm
*svm
)
2326 static int intr_interception(struct vcpu_svm
*svm
)
2328 ++svm
->vcpu
.stat
.irq_exits
;
2332 static int nop_on_interception(struct vcpu_svm
*svm
)
2337 static int halt_interception(struct vcpu_svm
*svm
)
2339 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 1;
2340 return kvm_emulate_halt(&svm
->vcpu
);
2343 static int vmmcall_interception(struct vcpu_svm
*svm
)
2345 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
2346 return kvm_emulate_hypercall(&svm
->vcpu
);
2349 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu
*vcpu
)
2351 struct vcpu_svm
*svm
= to_svm(vcpu
);
2353 return svm
->nested
.nested_cr3
;
2356 static u64
nested_svm_get_tdp_pdptr(struct kvm_vcpu
*vcpu
, int index
)
2358 struct vcpu_svm
*svm
= to_svm(vcpu
);
2359 u64 cr3
= svm
->nested
.nested_cr3
;
2363 ret
= kvm_vcpu_read_guest_page(vcpu
, gpa_to_gfn(cr3
), &pdpte
,
2364 offset_in_page(cr3
) + index
* 8, 8);
2370 static void nested_svm_set_tdp_cr3(struct kvm_vcpu
*vcpu
,
2373 struct vcpu_svm
*svm
= to_svm(vcpu
);
2375 svm
->vmcb
->control
.nested_cr3
= root
;
2376 mark_dirty(svm
->vmcb
, VMCB_NPT
);
2377 svm_flush_tlb(vcpu
);
2380 static void nested_svm_inject_npf_exit(struct kvm_vcpu
*vcpu
,
2381 struct x86_exception
*fault
)
2383 struct vcpu_svm
*svm
= to_svm(vcpu
);
2385 if (svm
->vmcb
->control
.exit_code
!= SVM_EXIT_NPF
) {
2387 * TODO: track the cause of the nested page fault, and
2388 * correctly fill in the high bits of exit_info_1.
2390 svm
->vmcb
->control
.exit_code
= SVM_EXIT_NPF
;
2391 svm
->vmcb
->control
.exit_code_hi
= 0;
2392 svm
->vmcb
->control
.exit_info_1
= (1ULL << 32);
2393 svm
->vmcb
->control
.exit_info_2
= fault
->address
;
2396 svm
->vmcb
->control
.exit_info_1
&= ~0xffffffffULL
;
2397 svm
->vmcb
->control
.exit_info_1
|= fault
->error_code
;
2400 * The present bit is always zero for page structure faults on real
2403 if (svm
->vmcb
->control
.exit_info_1
& (2ULL << 32))
2404 svm
->vmcb
->control
.exit_info_1
&= ~1;
2406 nested_svm_vmexit(svm
);
2409 static void nested_svm_init_mmu_context(struct kvm_vcpu
*vcpu
)
2411 WARN_ON(mmu_is_nested(vcpu
));
2412 kvm_init_shadow_mmu(vcpu
);
2413 vcpu
->arch
.mmu
.set_cr3
= nested_svm_set_tdp_cr3
;
2414 vcpu
->arch
.mmu
.get_cr3
= nested_svm_get_tdp_cr3
;
2415 vcpu
->arch
.mmu
.get_pdptr
= nested_svm_get_tdp_pdptr
;
2416 vcpu
->arch
.mmu
.inject_page_fault
= nested_svm_inject_npf_exit
;
2417 vcpu
->arch
.mmu
.shadow_root_level
= get_npt_level();
2418 reset_shadow_zero_bits_mask(vcpu
, &vcpu
->arch
.mmu
);
2419 vcpu
->arch
.walk_mmu
= &vcpu
->arch
.nested_mmu
;
2422 static void nested_svm_uninit_mmu_context(struct kvm_vcpu
*vcpu
)
2424 vcpu
->arch
.walk_mmu
= &vcpu
->arch
.mmu
;
2427 static int nested_svm_check_permissions(struct vcpu_svm
*svm
)
2429 if (!(svm
->vcpu
.arch
.efer
& EFER_SVME
) ||
2430 !is_paging(&svm
->vcpu
)) {
2431 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
2435 if (svm
->vmcb
->save
.cpl
) {
2436 kvm_inject_gp(&svm
->vcpu
, 0);
2443 static int nested_svm_check_exception(struct vcpu_svm
*svm
, unsigned nr
,
2444 bool has_error_code
, u32 error_code
)
2448 if (!is_guest_mode(&svm
->vcpu
))
2451 vmexit
= nested_svm_intercept(svm
);
2452 if (vmexit
!= NESTED_EXIT_DONE
)
2455 svm
->vmcb
->control
.exit_code
= SVM_EXIT_EXCP_BASE
+ nr
;
2456 svm
->vmcb
->control
.exit_code_hi
= 0;
2457 svm
->vmcb
->control
.exit_info_1
= error_code
;
2460 * FIXME: we should not write CR2 when L1 intercepts an L2 #PF exception.
2461 * The fix is to add the ancillary datum (CR2 or DR6) to structs
2462 * kvm_queued_exception and kvm_vcpu_events, so that CR2 and DR6 can be
2463 * written only when inject_pending_event runs (DR6 would written here
2464 * too). This should be conditional on a new capability---if the
2465 * capability is disabled, kvm_multiple_exception would write the
2466 * ancillary information to CR2 or DR6, for backwards ABI-compatibility.
2468 if (svm
->vcpu
.arch
.exception
.nested_apf
)
2469 svm
->vmcb
->control
.exit_info_2
= svm
->vcpu
.arch
.apf
.nested_apf_token
;
2471 svm
->vmcb
->control
.exit_info_2
= svm
->vcpu
.arch
.cr2
;
2473 svm
->nested
.exit_required
= true;
2477 /* This function returns true if it is save to enable the irq window */
2478 static inline bool nested_svm_intr(struct vcpu_svm
*svm
)
2480 if (!is_guest_mode(&svm
->vcpu
))
2483 if (!(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
))
2486 if (!(svm
->vcpu
.arch
.hflags
& HF_HIF_MASK
))
2490 * if vmexit was already requested (by intercepted exception
2491 * for instance) do not overwrite it with "external interrupt"
2494 if (svm
->nested
.exit_required
)
2497 svm
->vmcb
->control
.exit_code
= SVM_EXIT_INTR
;
2498 svm
->vmcb
->control
.exit_info_1
= 0;
2499 svm
->vmcb
->control
.exit_info_2
= 0;
2501 if (svm
->nested
.intercept
& 1ULL) {
2503 * The #vmexit can't be emulated here directly because this
2504 * code path runs with irqs and preemption disabled. A
2505 * #vmexit emulation might sleep. Only signal request for
2508 svm
->nested
.exit_required
= true;
2509 trace_kvm_nested_intr_vmexit(svm
->vmcb
->save
.rip
);
2516 /* This function returns true if it is save to enable the nmi window */
2517 static inline bool nested_svm_nmi(struct vcpu_svm
*svm
)
2519 if (!is_guest_mode(&svm
->vcpu
))
2522 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_NMI
)))
2525 svm
->vmcb
->control
.exit_code
= SVM_EXIT_NMI
;
2526 svm
->nested
.exit_required
= true;
2531 static void *nested_svm_map(struct vcpu_svm
*svm
, u64 gpa
, struct page
**_page
)
2537 page
= kvm_vcpu_gfn_to_page(&svm
->vcpu
, gpa
>> PAGE_SHIFT
);
2538 if (is_error_page(page
))
2546 kvm_inject_gp(&svm
->vcpu
, 0);
2551 static void nested_svm_unmap(struct page
*page
)
2554 kvm_release_page_dirty(page
);
2557 static int nested_svm_intercept_ioio(struct vcpu_svm
*svm
)
2559 unsigned port
, size
, iopm_len
;
2564 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_IOIO_PROT
)))
2565 return NESTED_EXIT_HOST
;
2567 port
= svm
->vmcb
->control
.exit_info_1
>> 16;
2568 size
= (svm
->vmcb
->control
.exit_info_1
& SVM_IOIO_SIZE_MASK
) >>
2569 SVM_IOIO_SIZE_SHIFT
;
2570 gpa
= svm
->nested
.vmcb_iopm
+ (port
/ 8);
2571 start_bit
= port
% 8;
2572 iopm_len
= (start_bit
+ size
> 8) ? 2 : 1;
2573 mask
= (0xf >> (4 - size
)) << start_bit
;
2576 if (kvm_vcpu_read_guest(&svm
->vcpu
, gpa
, &val
, iopm_len
))
2577 return NESTED_EXIT_DONE
;
2579 return (val
& mask
) ? NESTED_EXIT_DONE
: NESTED_EXIT_HOST
;
2582 static int nested_svm_exit_handled_msr(struct vcpu_svm
*svm
)
2584 u32 offset
, msr
, value
;
2587 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_MSR_PROT
)))
2588 return NESTED_EXIT_HOST
;
2590 msr
= svm
->vcpu
.arch
.regs
[VCPU_REGS_RCX
];
2591 offset
= svm_msrpm_offset(msr
);
2592 write
= svm
->vmcb
->control
.exit_info_1
& 1;
2593 mask
= 1 << ((2 * (msr
& 0xf)) + write
);
2595 if (offset
== MSR_INVALID
)
2596 return NESTED_EXIT_DONE
;
2598 /* Offset is in 32 bit units but need in 8 bit units */
2601 if (kvm_vcpu_read_guest(&svm
->vcpu
, svm
->nested
.vmcb_msrpm
+ offset
, &value
, 4))
2602 return NESTED_EXIT_DONE
;
2604 return (value
& mask
) ? NESTED_EXIT_DONE
: NESTED_EXIT_HOST
;
2607 /* DB exceptions for our internal use must not cause vmexit */
2608 static int nested_svm_intercept_db(struct vcpu_svm
*svm
)
2612 /* if we're not singlestepping, it's not ours */
2613 if (!svm
->nmi_singlestep
)
2614 return NESTED_EXIT_DONE
;
2616 /* if it's not a singlestep exception, it's not ours */
2617 if (kvm_get_dr(&svm
->vcpu
, 6, &dr6
))
2618 return NESTED_EXIT_DONE
;
2619 if (!(dr6
& DR6_BS
))
2620 return NESTED_EXIT_DONE
;
2622 /* if the guest is singlestepping, it should get the vmexit */
2623 if (svm
->nmi_singlestep_guest_rflags
& X86_EFLAGS_TF
) {
2624 disable_nmi_singlestep(svm
);
2625 return NESTED_EXIT_DONE
;
2628 /* it's ours, the nested hypervisor must not see this one */
2629 return NESTED_EXIT_HOST
;
2632 static int nested_svm_exit_special(struct vcpu_svm
*svm
)
2634 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
2636 switch (exit_code
) {
2639 case SVM_EXIT_EXCP_BASE
+ MC_VECTOR
:
2640 return NESTED_EXIT_HOST
;
2642 /* For now we are always handling NPFs when using them */
2644 return NESTED_EXIT_HOST
;
2646 case SVM_EXIT_EXCP_BASE
+ PF_VECTOR
:
2647 /* When we're shadowing, trap PFs, but not async PF */
2648 if (!npt_enabled
&& svm
->vcpu
.arch
.apf
.host_apf_reason
== 0)
2649 return NESTED_EXIT_HOST
;
2655 return NESTED_EXIT_CONTINUE
;
2659 * If this function returns true, this #vmexit was already handled
2661 static int nested_svm_intercept(struct vcpu_svm
*svm
)
2663 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
2664 int vmexit
= NESTED_EXIT_HOST
;
2666 switch (exit_code
) {
2668 vmexit
= nested_svm_exit_handled_msr(svm
);
2671 vmexit
= nested_svm_intercept_ioio(svm
);
2673 case SVM_EXIT_READ_CR0
... SVM_EXIT_WRITE_CR8
: {
2674 u32 bit
= 1U << (exit_code
- SVM_EXIT_READ_CR0
);
2675 if (svm
->nested
.intercept_cr
& bit
)
2676 vmexit
= NESTED_EXIT_DONE
;
2679 case SVM_EXIT_READ_DR0
... SVM_EXIT_WRITE_DR7
: {
2680 u32 bit
= 1U << (exit_code
- SVM_EXIT_READ_DR0
);
2681 if (svm
->nested
.intercept_dr
& bit
)
2682 vmexit
= NESTED_EXIT_DONE
;
2685 case SVM_EXIT_EXCP_BASE
... SVM_EXIT_EXCP_BASE
+ 0x1f: {
2686 u32 excp_bits
= 1 << (exit_code
- SVM_EXIT_EXCP_BASE
);
2687 if (svm
->nested
.intercept_exceptions
& excp_bits
) {
2688 if (exit_code
== SVM_EXIT_EXCP_BASE
+ DB_VECTOR
)
2689 vmexit
= nested_svm_intercept_db(svm
);
2691 vmexit
= NESTED_EXIT_DONE
;
2693 /* async page fault always cause vmexit */
2694 else if ((exit_code
== SVM_EXIT_EXCP_BASE
+ PF_VECTOR
) &&
2695 svm
->vcpu
.arch
.exception
.nested_apf
!= 0)
2696 vmexit
= NESTED_EXIT_DONE
;
2699 case SVM_EXIT_ERR
: {
2700 vmexit
= NESTED_EXIT_DONE
;
2704 u64 exit_bits
= 1ULL << (exit_code
- SVM_EXIT_INTR
);
2705 if (svm
->nested
.intercept
& exit_bits
)
2706 vmexit
= NESTED_EXIT_DONE
;
2713 static int nested_svm_exit_handled(struct vcpu_svm
*svm
)
2717 vmexit
= nested_svm_intercept(svm
);
2719 if (vmexit
== NESTED_EXIT_DONE
)
2720 nested_svm_vmexit(svm
);
2725 static inline void copy_vmcb_control_area(struct vmcb
*dst_vmcb
, struct vmcb
*from_vmcb
)
2727 struct vmcb_control_area
*dst
= &dst_vmcb
->control
;
2728 struct vmcb_control_area
*from
= &from_vmcb
->control
;
2730 dst
->intercept_cr
= from
->intercept_cr
;
2731 dst
->intercept_dr
= from
->intercept_dr
;
2732 dst
->intercept_exceptions
= from
->intercept_exceptions
;
2733 dst
->intercept
= from
->intercept
;
2734 dst
->iopm_base_pa
= from
->iopm_base_pa
;
2735 dst
->msrpm_base_pa
= from
->msrpm_base_pa
;
2736 dst
->tsc_offset
= from
->tsc_offset
;
2737 dst
->asid
= from
->asid
;
2738 dst
->tlb_ctl
= from
->tlb_ctl
;
2739 dst
->int_ctl
= from
->int_ctl
;
2740 dst
->int_vector
= from
->int_vector
;
2741 dst
->int_state
= from
->int_state
;
2742 dst
->exit_code
= from
->exit_code
;
2743 dst
->exit_code_hi
= from
->exit_code_hi
;
2744 dst
->exit_info_1
= from
->exit_info_1
;
2745 dst
->exit_info_2
= from
->exit_info_2
;
2746 dst
->exit_int_info
= from
->exit_int_info
;
2747 dst
->exit_int_info_err
= from
->exit_int_info_err
;
2748 dst
->nested_ctl
= from
->nested_ctl
;
2749 dst
->event_inj
= from
->event_inj
;
2750 dst
->event_inj_err
= from
->event_inj_err
;
2751 dst
->nested_cr3
= from
->nested_cr3
;
2752 dst
->virt_ext
= from
->virt_ext
;
2755 static int nested_svm_vmexit(struct vcpu_svm
*svm
)
2757 struct vmcb
*nested_vmcb
;
2758 struct vmcb
*hsave
= svm
->nested
.hsave
;
2759 struct vmcb
*vmcb
= svm
->vmcb
;
2762 trace_kvm_nested_vmexit_inject(vmcb
->control
.exit_code
,
2763 vmcb
->control
.exit_info_1
,
2764 vmcb
->control
.exit_info_2
,
2765 vmcb
->control
.exit_int_info
,
2766 vmcb
->control
.exit_int_info_err
,
2769 nested_vmcb
= nested_svm_map(svm
, svm
->nested
.vmcb
, &page
);
2773 /* Exit Guest-Mode */
2774 leave_guest_mode(&svm
->vcpu
);
2775 svm
->nested
.vmcb
= 0;
2777 /* Give the current vmcb to the guest */
2780 nested_vmcb
->save
.es
= vmcb
->save
.es
;
2781 nested_vmcb
->save
.cs
= vmcb
->save
.cs
;
2782 nested_vmcb
->save
.ss
= vmcb
->save
.ss
;
2783 nested_vmcb
->save
.ds
= vmcb
->save
.ds
;
2784 nested_vmcb
->save
.gdtr
= vmcb
->save
.gdtr
;
2785 nested_vmcb
->save
.idtr
= vmcb
->save
.idtr
;
2786 nested_vmcb
->save
.efer
= svm
->vcpu
.arch
.efer
;
2787 nested_vmcb
->save
.cr0
= kvm_read_cr0(&svm
->vcpu
);
2788 nested_vmcb
->save
.cr3
= kvm_read_cr3(&svm
->vcpu
);
2789 nested_vmcb
->save
.cr2
= vmcb
->save
.cr2
;
2790 nested_vmcb
->save
.cr4
= svm
->vcpu
.arch
.cr4
;
2791 nested_vmcb
->save
.rflags
= kvm_get_rflags(&svm
->vcpu
);
2792 nested_vmcb
->save
.rip
= vmcb
->save
.rip
;
2793 nested_vmcb
->save
.rsp
= vmcb
->save
.rsp
;
2794 nested_vmcb
->save
.rax
= vmcb
->save
.rax
;
2795 nested_vmcb
->save
.dr7
= vmcb
->save
.dr7
;
2796 nested_vmcb
->save
.dr6
= vmcb
->save
.dr6
;
2797 nested_vmcb
->save
.cpl
= vmcb
->save
.cpl
;
2799 nested_vmcb
->control
.int_ctl
= vmcb
->control
.int_ctl
;
2800 nested_vmcb
->control
.int_vector
= vmcb
->control
.int_vector
;
2801 nested_vmcb
->control
.int_state
= vmcb
->control
.int_state
;
2802 nested_vmcb
->control
.exit_code
= vmcb
->control
.exit_code
;
2803 nested_vmcb
->control
.exit_code_hi
= vmcb
->control
.exit_code_hi
;
2804 nested_vmcb
->control
.exit_info_1
= vmcb
->control
.exit_info_1
;
2805 nested_vmcb
->control
.exit_info_2
= vmcb
->control
.exit_info_2
;
2806 nested_vmcb
->control
.exit_int_info
= vmcb
->control
.exit_int_info
;
2807 nested_vmcb
->control
.exit_int_info_err
= vmcb
->control
.exit_int_info_err
;
2809 if (svm
->nrips_enabled
)
2810 nested_vmcb
->control
.next_rip
= vmcb
->control
.next_rip
;
2813 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2814 * to make sure that we do not lose injected events. So check event_inj
2815 * here and copy it to exit_int_info if it is valid.
2816 * Exit_int_info and event_inj can't be both valid because the case
2817 * below only happens on a VMRUN instruction intercept which has
2818 * no valid exit_int_info set.
2820 if (vmcb
->control
.event_inj
& SVM_EVTINJ_VALID
) {
2821 struct vmcb_control_area
*nc
= &nested_vmcb
->control
;
2823 nc
->exit_int_info
= vmcb
->control
.event_inj
;
2824 nc
->exit_int_info_err
= vmcb
->control
.event_inj_err
;
2827 nested_vmcb
->control
.tlb_ctl
= 0;
2828 nested_vmcb
->control
.event_inj
= 0;
2829 nested_vmcb
->control
.event_inj_err
= 0;
2831 /* We always set V_INTR_MASKING and remember the old value in hflags */
2832 if (!(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
))
2833 nested_vmcb
->control
.int_ctl
&= ~V_INTR_MASKING_MASK
;
2835 /* Restore the original control entries */
2836 copy_vmcb_control_area(vmcb
, hsave
);
2838 kvm_clear_exception_queue(&svm
->vcpu
);
2839 kvm_clear_interrupt_queue(&svm
->vcpu
);
2841 svm
->nested
.nested_cr3
= 0;
2843 /* Restore selected save entries */
2844 svm
->vmcb
->save
.es
= hsave
->save
.es
;
2845 svm
->vmcb
->save
.cs
= hsave
->save
.cs
;
2846 svm
->vmcb
->save
.ss
= hsave
->save
.ss
;
2847 svm
->vmcb
->save
.ds
= hsave
->save
.ds
;
2848 svm
->vmcb
->save
.gdtr
= hsave
->save
.gdtr
;
2849 svm
->vmcb
->save
.idtr
= hsave
->save
.idtr
;
2850 kvm_set_rflags(&svm
->vcpu
, hsave
->save
.rflags
);
2851 svm_set_efer(&svm
->vcpu
, hsave
->save
.efer
);
2852 svm_set_cr0(&svm
->vcpu
, hsave
->save
.cr0
| X86_CR0_PE
);
2853 svm_set_cr4(&svm
->vcpu
, hsave
->save
.cr4
);
2855 svm
->vmcb
->save
.cr3
= hsave
->save
.cr3
;
2856 svm
->vcpu
.arch
.cr3
= hsave
->save
.cr3
;
2858 (void)kvm_set_cr3(&svm
->vcpu
, hsave
->save
.cr3
);
2860 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
, hsave
->save
.rax
);
2861 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RSP
, hsave
->save
.rsp
);
2862 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RIP
, hsave
->save
.rip
);
2863 svm
->vmcb
->save
.dr7
= 0;
2864 svm
->vmcb
->save
.cpl
= 0;
2865 svm
->vmcb
->control
.exit_int_info
= 0;
2867 mark_all_dirty(svm
->vmcb
);
2869 nested_svm_unmap(page
);
2871 nested_svm_uninit_mmu_context(&svm
->vcpu
);
2872 kvm_mmu_reset_context(&svm
->vcpu
);
2873 kvm_mmu_load(&svm
->vcpu
);
2878 static bool nested_svm_vmrun_msrpm(struct vcpu_svm
*svm
)
2881 * This function merges the msr permission bitmaps of kvm and the
2882 * nested vmcb. It is optimized in that it only merges the parts where
2883 * the kvm msr permission bitmap may contain zero bits
2887 if (!(svm
->nested
.intercept
& (1ULL << INTERCEPT_MSR_PROT
)))
2890 for (i
= 0; i
< MSRPM_OFFSETS
; i
++) {
2894 if (msrpm_offsets
[i
] == 0xffffffff)
2897 p
= msrpm_offsets
[i
];
2898 offset
= svm
->nested
.vmcb_msrpm
+ (p
* 4);
2900 if (kvm_vcpu_read_guest(&svm
->vcpu
, offset
, &value
, 4))
2903 svm
->nested
.msrpm
[p
] = svm
->msrpm
[p
] | value
;
2906 svm
->vmcb
->control
.msrpm_base_pa
= __pa(svm
->nested
.msrpm
);
2911 static bool nested_vmcb_checks(struct vmcb
*vmcb
)
2913 if ((vmcb
->control
.intercept
& (1ULL << INTERCEPT_VMRUN
)) == 0)
2916 if (vmcb
->control
.asid
== 0)
2919 if (vmcb
->control
.nested_ctl
&& !npt_enabled
)
2925 static bool nested_svm_vmrun(struct vcpu_svm
*svm
)
2927 struct vmcb
*nested_vmcb
;
2928 struct vmcb
*hsave
= svm
->nested
.hsave
;
2929 struct vmcb
*vmcb
= svm
->vmcb
;
2933 vmcb_gpa
= svm
->vmcb
->save
.rax
;
2935 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, &page
);
2939 if (!nested_vmcb_checks(nested_vmcb
)) {
2940 nested_vmcb
->control
.exit_code
= SVM_EXIT_ERR
;
2941 nested_vmcb
->control
.exit_code_hi
= 0;
2942 nested_vmcb
->control
.exit_info_1
= 0;
2943 nested_vmcb
->control
.exit_info_2
= 0;
2945 nested_svm_unmap(page
);
2950 trace_kvm_nested_vmrun(svm
->vmcb
->save
.rip
, vmcb_gpa
,
2951 nested_vmcb
->save
.rip
,
2952 nested_vmcb
->control
.int_ctl
,
2953 nested_vmcb
->control
.event_inj
,
2954 nested_vmcb
->control
.nested_ctl
);
2956 trace_kvm_nested_intercepts(nested_vmcb
->control
.intercept_cr
& 0xffff,
2957 nested_vmcb
->control
.intercept_cr
>> 16,
2958 nested_vmcb
->control
.intercept_exceptions
,
2959 nested_vmcb
->control
.intercept
);
2961 /* Clear internal status */
2962 kvm_clear_exception_queue(&svm
->vcpu
);
2963 kvm_clear_interrupt_queue(&svm
->vcpu
);
2966 * Save the old vmcb, so we don't need to pick what we save, but can
2967 * restore everything when a VMEXIT occurs
2969 hsave
->save
.es
= vmcb
->save
.es
;
2970 hsave
->save
.cs
= vmcb
->save
.cs
;
2971 hsave
->save
.ss
= vmcb
->save
.ss
;
2972 hsave
->save
.ds
= vmcb
->save
.ds
;
2973 hsave
->save
.gdtr
= vmcb
->save
.gdtr
;
2974 hsave
->save
.idtr
= vmcb
->save
.idtr
;
2975 hsave
->save
.efer
= svm
->vcpu
.arch
.efer
;
2976 hsave
->save
.cr0
= kvm_read_cr0(&svm
->vcpu
);
2977 hsave
->save
.cr4
= svm
->vcpu
.arch
.cr4
;
2978 hsave
->save
.rflags
= kvm_get_rflags(&svm
->vcpu
);
2979 hsave
->save
.rip
= kvm_rip_read(&svm
->vcpu
);
2980 hsave
->save
.rsp
= vmcb
->save
.rsp
;
2981 hsave
->save
.rax
= vmcb
->save
.rax
;
2983 hsave
->save
.cr3
= vmcb
->save
.cr3
;
2985 hsave
->save
.cr3
= kvm_read_cr3(&svm
->vcpu
);
2987 copy_vmcb_control_area(hsave
, vmcb
);
2989 if (kvm_get_rflags(&svm
->vcpu
) & X86_EFLAGS_IF
)
2990 svm
->vcpu
.arch
.hflags
|= HF_HIF_MASK
;
2992 svm
->vcpu
.arch
.hflags
&= ~HF_HIF_MASK
;
2994 if (nested_vmcb
->control
.nested_ctl
) {
2995 kvm_mmu_unload(&svm
->vcpu
);
2996 svm
->nested
.nested_cr3
= nested_vmcb
->control
.nested_cr3
;
2997 nested_svm_init_mmu_context(&svm
->vcpu
);
3000 /* Load the nested guest state */
3001 svm
->vmcb
->save
.es
= nested_vmcb
->save
.es
;
3002 svm
->vmcb
->save
.cs
= nested_vmcb
->save
.cs
;
3003 svm
->vmcb
->save
.ss
= nested_vmcb
->save
.ss
;
3004 svm
->vmcb
->save
.ds
= nested_vmcb
->save
.ds
;
3005 svm
->vmcb
->save
.gdtr
= nested_vmcb
->save
.gdtr
;
3006 svm
->vmcb
->save
.idtr
= nested_vmcb
->save
.idtr
;
3007 kvm_set_rflags(&svm
->vcpu
, nested_vmcb
->save
.rflags
);
3008 svm_set_efer(&svm
->vcpu
, nested_vmcb
->save
.efer
);
3009 svm_set_cr0(&svm
->vcpu
, nested_vmcb
->save
.cr0
);
3010 svm_set_cr4(&svm
->vcpu
, nested_vmcb
->save
.cr4
);
3012 svm
->vmcb
->save
.cr3
= nested_vmcb
->save
.cr3
;
3013 svm
->vcpu
.arch
.cr3
= nested_vmcb
->save
.cr3
;
3015 (void)kvm_set_cr3(&svm
->vcpu
, nested_vmcb
->save
.cr3
);
3017 /* Guest paging mode is active - reset mmu */
3018 kvm_mmu_reset_context(&svm
->vcpu
);
3020 svm
->vmcb
->save
.cr2
= svm
->vcpu
.arch
.cr2
= nested_vmcb
->save
.cr2
;
3021 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
, nested_vmcb
->save
.rax
);
3022 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RSP
, nested_vmcb
->save
.rsp
);
3023 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RIP
, nested_vmcb
->save
.rip
);
3025 /* In case we don't even reach vcpu_run, the fields are not updated */
3026 svm
->vmcb
->save
.rax
= nested_vmcb
->save
.rax
;
3027 svm
->vmcb
->save
.rsp
= nested_vmcb
->save
.rsp
;
3028 svm
->vmcb
->save
.rip
= nested_vmcb
->save
.rip
;
3029 svm
->vmcb
->save
.dr7
= nested_vmcb
->save
.dr7
;
3030 svm
->vmcb
->save
.dr6
= nested_vmcb
->save
.dr6
;
3031 svm
->vmcb
->save
.cpl
= nested_vmcb
->save
.cpl
;
3033 svm
->nested
.vmcb_msrpm
= nested_vmcb
->control
.msrpm_base_pa
& ~0x0fffULL
;
3034 svm
->nested
.vmcb_iopm
= nested_vmcb
->control
.iopm_base_pa
& ~0x0fffULL
;
3036 /* cache intercepts */
3037 svm
->nested
.intercept_cr
= nested_vmcb
->control
.intercept_cr
;
3038 svm
->nested
.intercept_dr
= nested_vmcb
->control
.intercept_dr
;
3039 svm
->nested
.intercept_exceptions
= nested_vmcb
->control
.intercept_exceptions
;
3040 svm
->nested
.intercept
= nested_vmcb
->control
.intercept
;
3042 svm_flush_tlb(&svm
->vcpu
);
3043 svm
->vmcb
->control
.int_ctl
= nested_vmcb
->control
.int_ctl
| V_INTR_MASKING_MASK
;
3044 if (nested_vmcb
->control
.int_ctl
& V_INTR_MASKING_MASK
)
3045 svm
->vcpu
.arch
.hflags
|= HF_VINTR_MASK
;
3047 svm
->vcpu
.arch
.hflags
&= ~HF_VINTR_MASK
;
3049 if (svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
) {
3050 /* We only want the cr8 intercept bits of the guest */
3051 clr_cr_intercept(svm
, INTERCEPT_CR8_READ
);
3052 clr_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
3055 /* We don't want to see VMMCALLs from a nested guest */
3056 clr_intercept(svm
, INTERCEPT_VMMCALL
);
3058 svm
->vmcb
->control
.virt_ext
= nested_vmcb
->control
.virt_ext
;
3059 svm
->vmcb
->control
.int_vector
= nested_vmcb
->control
.int_vector
;
3060 svm
->vmcb
->control
.int_state
= nested_vmcb
->control
.int_state
;
3061 svm
->vmcb
->control
.tsc_offset
+= nested_vmcb
->control
.tsc_offset
;
3062 svm
->vmcb
->control
.event_inj
= nested_vmcb
->control
.event_inj
;
3063 svm
->vmcb
->control
.event_inj_err
= nested_vmcb
->control
.event_inj_err
;
3065 nested_svm_unmap(page
);
3067 /* Enter Guest-Mode */
3068 enter_guest_mode(&svm
->vcpu
);
3071 * Merge guest and host intercepts - must be called with vcpu in
3072 * guest-mode to take affect here
3074 recalc_intercepts(svm
);
3076 svm
->nested
.vmcb
= vmcb_gpa
;
3080 mark_all_dirty(svm
->vmcb
);
3085 static void nested_svm_vmloadsave(struct vmcb
*from_vmcb
, struct vmcb
*to_vmcb
)
3087 to_vmcb
->save
.fs
= from_vmcb
->save
.fs
;
3088 to_vmcb
->save
.gs
= from_vmcb
->save
.gs
;
3089 to_vmcb
->save
.tr
= from_vmcb
->save
.tr
;
3090 to_vmcb
->save
.ldtr
= from_vmcb
->save
.ldtr
;
3091 to_vmcb
->save
.kernel_gs_base
= from_vmcb
->save
.kernel_gs_base
;
3092 to_vmcb
->save
.star
= from_vmcb
->save
.star
;
3093 to_vmcb
->save
.lstar
= from_vmcb
->save
.lstar
;
3094 to_vmcb
->save
.cstar
= from_vmcb
->save
.cstar
;
3095 to_vmcb
->save
.sfmask
= from_vmcb
->save
.sfmask
;
3096 to_vmcb
->save
.sysenter_cs
= from_vmcb
->save
.sysenter_cs
;
3097 to_vmcb
->save
.sysenter_esp
= from_vmcb
->save
.sysenter_esp
;
3098 to_vmcb
->save
.sysenter_eip
= from_vmcb
->save
.sysenter_eip
;
3101 static int vmload_interception(struct vcpu_svm
*svm
)
3103 struct vmcb
*nested_vmcb
;
3107 if (nested_svm_check_permissions(svm
))
3110 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, &page
);
3114 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
3115 ret
= kvm_skip_emulated_instruction(&svm
->vcpu
);
3117 nested_svm_vmloadsave(nested_vmcb
, svm
->vmcb
);
3118 nested_svm_unmap(page
);
3123 static int vmsave_interception(struct vcpu_svm
*svm
)
3125 struct vmcb
*nested_vmcb
;
3129 if (nested_svm_check_permissions(svm
))
3132 nested_vmcb
= nested_svm_map(svm
, svm
->vmcb
->save
.rax
, &page
);
3136 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
3137 ret
= kvm_skip_emulated_instruction(&svm
->vcpu
);
3139 nested_svm_vmloadsave(svm
->vmcb
, nested_vmcb
);
3140 nested_svm_unmap(page
);
3145 static int vmrun_interception(struct vcpu_svm
*svm
)
3147 if (nested_svm_check_permissions(svm
))
3150 /* Save rip after vmrun instruction */
3151 kvm_rip_write(&svm
->vcpu
, kvm_rip_read(&svm
->vcpu
) + 3);
3153 if (!nested_svm_vmrun(svm
))
3156 if (!nested_svm_vmrun_msrpm(svm
))
3163 svm
->vmcb
->control
.exit_code
= SVM_EXIT_ERR
;
3164 svm
->vmcb
->control
.exit_code_hi
= 0;
3165 svm
->vmcb
->control
.exit_info_1
= 0;
3166 svm
->vmcb
->control
.exit_info_2
= 0;
3168 nested_svm_vmexit(svm
);
3173 static int stgi_interception(struct vcpu_svm
*svm
)
3177 if (nested_svm_check_permissions(svm
))
3180 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
3181 ret
= kvm_skip_emulated_instruction(&svm
->vcpu
);
3182 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
3189 static int clgi_interception(struct vcpu_svm
*svm
)
3193 if (nested_svm_check_permissions(svm
))
3196 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
3197 ret
= kvm_skip_emulated_instruction(&svm
->vcpu
);
3201 /* After a CLGI no interrupts should come */
3202 if (!kvm_vcpu_apicv_active(&svm
->vcpu
)) {
3203 svm_clear_vintr(svm
);
3204 svm
->vmcb
->control
.int_ctl
&= ~V_IRQ_MASK
;
3205 mark_dirty(svm
->vmcb
, VMCB_INTR
);
3211 static int invlpga_interception(struct vcpu_svm
*svm
)
3213 struct kvm_vcpu
*vcpu
= &svm
->vcpu
;
3215 trace_kvm_invlpga(svm
->vmcb
->save
.rip
, kvm_register_read(&svm
->vcpu
, VCPU_REGS_RCX
),
3216 kvm_register_read(&svm
->vcpu
, VCPU_REGS_RAX
));
3218 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
3219 kvm_mmu_invlpg(vcpu
, kvm_register_read(&svm
->vcpu
, VCPU_REGS_RAX
));
3221 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
3222 return kvm_skip_emulated_instruction(&svm
->vcpu
);
3225 static int skinit_interception(struct vcpu_svm
*svm
)
3227 trace_kvm_skinit(svm
->vmcb
->save
.rip
, kvm_register_read(&svm
->vcpu
, VCPU_REGS_RAX
));
3229 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
3233 static int wbinvd_interception(struct vcpu_svm
*svm
)
3235 return kvm_emulate_wbinvd(&svm
->vcpu
);
3238 static int xsetbv_interception(struct vcpu_svm
*svm
)
3240 u64 new_bv
= kvm_read_edx_eax(&svm
->vcpu
);
3241 u32 index
= kvm_register_read(&svm
->vcpu
, VCPU_REGS_RCX
);
3243 if (kvm_set_xcr(&svm
->vcpu
, index
, new_bv
) == 0) {
3244 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 3;
3245 return kvm_skip_emulated_instruction(&svm
->vcpu
);
3251 static int task_switch_interception(struct vcpu_svm
*svm
)
3255 int int_type
= svm
->vmcb
->control
.exit_int_info
&
3256 SVM_EXITINTINFO_TYPE_MASK
;
3257 int int_vec
= svm
->vmcb
->control
.exit_int_info
& SVM_EVTINJ_VEC_MASK
;
3259 svm
->vmcb
->control
.exit_int_info
& SVM_EXITINTINFO_TYPE_MASK
;
3261 svm
->vmcb
->control
.exit_int_info
& SVM_EXITINTINFO_VALID
;
3262 bool has_error_code
= false;
3265 tss_selector
= (u16
)svm
->vmcb
->control
.exit_info_1
;
3267 if (svm
->vmcb
->control
.exit_info_2
&
3268 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET
))
3269 reason
= TASK_SWITCH_IRET
;
3270 else if (svm
->vmcb
->control
.exit_info_2
&
3271 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP
))
3272 reason
= TASK_SWITCH_JMP
;
3274 reason
= TASK_SWITCH_GATE
;
3276 reason
= TASK_SWITCH_CALL
;
3278 if (reason
== TASK_SWITCH_GATE
) {
3280 case SVM_EXITINTINFO_TYPE_NMI
:
3281 svm
->vcpu
.arch
.nmi_injected
= false;
3283 case SVM_EXITINTINFO_TYPE_EXEPT
:
3284 if (svm
->vmcb
->control
.exit_info_2
&
3285 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE
)) {
3286 has_error_code
= true;
3288 (u32
)svm
->vmcb
->control
.exit_info_2
;
3290 kvm_clear_exception_queue(&svm
->vcpu
);
3292 case SVM_EXITINTINFO_TYPE_INTR
:
3293 kvm_clear_interrupt_queue(&svm
->vcpu
);
3300 if (reason
!= TASK_SWITCH_GATE
||
3301 int_type
== SVM_EXITINTINFO_TYPE_SOFT
||
3302 (int_type
== SVM_EXITINTINFO_TYPE_EXEPT
&&
3303 (int_vec
== OF_VECTOR
|| int_vec
== BP_VECTOR
)))
3304 skip_emulated_instruction(&svm
->vcpu
);
3306 if (int_type
!= SVM_EXITINTINFO_TYPE_SOFT
)
3309 if (kvm_task_switch(&svm
->vcpu
, tss_selector
, int_vec
, reason
,
3310 has_error_code
, error_code
) == EMULATE_FAIL
) {
3311 svm
->vcpu
.run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
3312 svm
->vcpu
.run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
3313 svm
->vcpu
.run
->internal
.ndata
= 0;
3319 static int cpuid_interception(struct vcpu_svm
*svm
)
3321 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
3322 return kvm_emulate_cpuid(&svm
->vcpu
);
3325 static int iret_interception(struct vcpu_svm
*svm
)
3327 ++svm
->vcpu
.stat
.nmi_window_exits
;
3328 clr_intercept(svm
, INTERCEPT_IRET
);
3329 svm
->vcpu
.arch
.hflags
|= HF_IRET_MASK
;
3330 svm
->nmi_iret_rip
= kvm_rip_read(&svm
->vcpu
);
3331 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
3335 static int invlpg_interception(struct vcpu_svm
*svm
)
3337 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS
))
3338 return emulate_instruction(&svm
->vcpu
, 0) == EMULATE_DONE
;
3340 kvm_mmu_invlpg(&svm
->vcpu
, svm
->vmcb
->control
.exit_info_1
);
3341 return kvm_skip_emulated_instruction(&svm
->vcpu
);
3344 static int emulate_on_interception(struct vcpu_svm
*svm
)
3346 return emulate_instruction(&svm
->vcpu
, 0) == EMULATE_DONE
;
3349 static int rdpmc_interception(struct vcpu_svm
*svm
)
3353 if (!static_cpu_has(X86_FEATURE_NRIPS
))
3354 return emulate_on_interception(svm
);
3356 err
= kvm_rdpmc(&svm
->vcpu
);
3357 return kvm_complete_insn_gp(&svm
->vcpu
, err
);
3360 static bool check_selective_cr0_intercepted(struct vcpu_svm
*svm
,
3363 unsigned long cr0
= svm
->vcpu
.arch
.cr0
;
3367 intercept
= svm
->nested
.intercept
;
3369 if (!is_guest_mode(&svm
->vcpu
) ||
3370 (!(intercept
& (1ULL << INTERCEPT_SELECTIVE_CR0
))))
3373 cr0
&= ~SVM_CR0_SELECTIVE_MASK
;
3374 val
&= ~SVM_CR0_SELECTIVE_MASK
;
3377 svm
->vmcb
->control
.exit_code
= SVM_EXIT_CR0_SEL_WRITE
;
3378 ret
= (nested_svm_exit_handled(svm
) == NESTED_EXIT_DONE
);
3384 #define CR_VALID (1ULL << 63)
3386 static int cr_interception(struct vcpu_svm
*svm
)
3392 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS
))
3393 return emulate_on_interception(svm
);
3395 if (unlikely((svm
->vmcb
->control
.exit_info_1
& CR_VALID
) == 0))
3396 return emulate_on_interception(svm
);
3398 reg
= svm
->vmcb
->control
.exit_info_1
& SVM_EXITINFO_REG_MASK
;
3399 if (svm
->vmcb
->control
.exit_code
== SVM_EXIT_CR0_SEL_WRITE
)
3400 cr
= SVM_EXIT_WRITE_CR0
- SVM_EXIT_READ_CR0
;
3402 cr
= svm
->vmcb
->control
.exit_code
- SVM_EXIT_READ_CR0
;
3405 if (cr
>= 16) { /* mov to cr */
3407 val
= kvm_register_read(&svm
->vcpu
, reg
);
3410 if (!check_selective_cr0_intercepted(svm
, val
))
3411 err
= kvm_set_cr0(&svm
->vcpu
, val
);
3417 err
= kvm_set_cr3(&svm
->vcpu
, val
);
3420 err
= kvm_set_cr4(&svm
->vcpu
, val
);
3423 err
= kvm_set_cr8(&svm
->vcpu
, val
);
3426 WARN(1, "unhandled write to CR%d", cr
);
3427 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
3430 } else { /* mov from cr */
3433 val
= kvm_read_cr0(&svm
->vcpu
);
3436 val
= svm
->vcpu
.arch
.cr2
;
3439 val
= kvm_read_cr3(&svm
->vcpu
);
3442 val
= kvm_read_cr4(&svm
->vcpu
);
3445 val
= kvm_get_cr8(&svm
->vcpu
);
3448 WARN(1, "unhandled read from CR%d", cr
);
3449 kvm_queue_exception(&svm
->vcpu
, UD_VECTOR
);
3452 kvm_register_write(&svm
->vcpu
, reg
, val
);
3454 return kvm_complete_insn_gp(&svm
->vcpu
, err
);
3457 static int dr_interception(struct vcpu_svm
*svm
)
3462 if (svm
->vcpu
.guest_debug
== 0) {
3464 * No more DR vmexits; force a reload of the debug registers
3465 * and reenter on this instruction. The next vmexit will
3466 * retrieve the full state of the debug registers.
3468 clr_dr_intercepts(svm
);
3469 svm
->vcpu
.arch
.switch_db_regs
|= KVM_DEBUGREG_WONT_EXIT
;
3473 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS
))
3474 return emulate_on_interception(svm
);
3476 reg
= svm
->vmcb
->control
.exit_info_1
& SVM_EXITINFO_REG_MASK
;
3477 dr
= svm
->vmcb
->control
.exit_code
- SVM_EXIT_READ_DR0
;
3479 if (dr
>= 16) { /* mov to DRn */
3480 if (!kvm_require_dr(&svm
->vcpu
, dr
- 16))
3482 val
= kvm_register_read(&svm
->vcpu
, reg
);
3483 kvm_set_dr(&svm
->vcpu
, dr
- 16, val
);
3485 if (!kvm_require_dr(&svm
->vcpu
, dr
))
3487 kvm_get_dr(&svm
->vcpu
, dr
, &val
);
3488 kvm_register_write(&svm
->vcpu
, reg
, val
);
3491 return kvm_skip_emulated_instruction(&svm
->vcpu
);
3494 static int cr8_write_interception(struct vcpu_svm
*svm
)
3496 struct kvm_run
*kvm_run
= svm
->vcpu
.run
;
3499 u8 cr8_prev
= kvm_get_cr8(&svm
->vcpu
);
3500 /* instruction emulation calls kvm_set_cr8() */
3501 r
= cr_interception(svm
);
3502 if (lapic_in_kernel(&svm
->vcpu
))
3504 if (cr8_prev
<= kvm_get_cr8(&svm
->vcpu
))
3506 kvm_run
->exit_reason
= KVM_EXIT_SET_TPR
;
3510 static int svm_get_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
3512 struct vcpu_svm
*svm
= to_svm(vcpu
);
3514 switch (msr_info
->index
) {
3515 case MSR_IA32_TSC
: {
3516 msr_info
->data
= svm
->vmcb
->control
.tsc_offset
+
3517 kvm_scale_tsc(vcpu
, rdtsc());
3522 msr_info
->data
= svm
->vmcb
->save
.star
;
3524 #ifdef CONFIG_X86_64
3526 msr_info
->data
= svm
->vmcb
->save
.lstar
;
3529 msr_info
->data
= svm
->vmcb
->save
.cstar
;
3531 case MSR_KERNEL_GS_BASE
:
3532 msr_info
->data
= svm
->vmcb
->save
.kernel_gs_base
;
3534 case MSR_SYSCALL_MASK
:
3535 msr_info
->data
= svm
->vmcb
->save
.sfmask
;
3538 case MSR_IA32_SYSENTER_CS
:
3539 msr_info
->data
= svm
->vmcb
->save
.sysenter_cs
;
3541 case MSR_IA32_SYSENTER_EIP
:
3542 msr_info
->data
= svm
->sysenter_eip
;
3544 case MSR_IA32_SYSENTER_ESP
:
3545 msr_info
->data
= svm
->sysenter_esp
;
3548 if (!boot_cpu_has(X86_FEATURE_RDTSCP
))
3550 msr_info
->data
= svm
->tsc_aux
;
3553 * Nobody will change the following 5 values in the VMCB so we can
3554 * safely return them on rdmsr. They will always be 0 until LBRV is
3557 case MSR_IA32_DEBUGCTLMSR
:
3558 msr_info
->data
= svm
->vmcb
->save
.dbgctl
;
3560 case MSR_IA32_LASTBRANCHFROMIP
:
3561 msr_info
->data
= svm
->vmcb
->save
.br_from
;
3563 case MSR_IA32_LASTBRANCHTOIP
:
3564 msr_info
->data
= svm
->vmcb
->save
.br_to
;
3566 case MSR_IA32_LASTINTFROMIP
:
3567 msr_info
->data
= svm
->vmcb
->save
.last_excp_from
;
3569 case MSR_IA32_LASTINTTOIP
:
3570 msr_info
->data
= svm
->vmcb
->save
.last_excp_to
;
3572 case MSR_VM_HSAVE_PA
:
3573 msr_info
->data
= svm
->nested
.hsave_msr
;
3576 msr_info
->data
= svm
->nested
.vm_cr_msr
;
3578 case MSR_IA32_SPEC_CTRL
:
3579 msr_info
->data
= svm
->spec_ctrl
;
3581 case MSR_IA32_UCODE_REV
:
3582 msr_info
->data
= 0x01000065;
3584 case MSR_F15H_IC_CFG
: {
3588 family
= guest_cpuid_family(vcpu
);
3589 model
= guest_cpuid_model(vcpu
);
3591 if (family
< 0 || model
< 0)
3592 return kvm_get_msr_common(vcpu
, msr_info
);
3596 if (family
== 0x15 &&
3597 (model
>= 0x2 && model
< 0x20))
3598 msr_info
->data
= 0x1E;
3602 return kvm_get_msr_common(vcpu
, msr_info
);
3607 static int rdmsr_interception(struct vcpu_svm
*svm
)
3609 u32 ecx
= kvm_register_read(&svm
->vcpu
, VCPU_REGS_RCX
);
3610 struct msr_data msr_info
;
3612 msr_info
.index
= ecx
;
3613 msr_info
.host_initiated
= false;
3614 if (svm_get_msr(&svm
->vcpu
, &msr_info
)) {
3615 trace_kvm_msr_read_ex(ecx
);
3616 kvm_inject_gp(&svm
->vcpu
, 0);
3619 trace_kvm_msr_read(ecx
, msr_info
.data
);
3621 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RAX
,
3622 msr_info
.data
& 0xffffffff);
3623 kvm_register_write(&svm
->vcpu
, VCPU_REGS_RDX
,
3624 msr_info
.data
>> 32);
3625 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
3626 return kvm_skip_emulated_instruction(&svm
->vcpu
);
3630 static int svm_set_vm_cr(struct kvm_vcpu
*vcpu
, u64 data
)
3632 struct vcpu_svm
*svm
= to_svm(vcpu
);
3633 int svm_dis
, chg_mask
;
3635 if (data
& ~SVM_VM_CR_VALID_MASK
)
3638 chg_mask
= SVM_VM_CR_VALID_MASK
;
3640 if (svm
->nested
.vm_cr_msr
& SVM_VM_CR_SVM_DIS_MASK
)
3641 chg_mask
&= ~(SVM_VM_CR_SVM_LOCK_MASK
| SVM_VM_CR_SVM_DIS_MASK
);
3643 svm
->nested
.vm_cr_msr
&= ~chg_mask
;
3644 svm
->nested
.vm_cr_msr
|= (data
& chg_mask
);
3646 svm_dis
= svm
->nested
.vm_cr_msr
& SVM_VM_CR_SVM_DIS_MASK
;
3648 /* check for svm_disable while efer.svme is set */
3649 if (svm_dis
&& (vcpu
->arch
.efer
& EFER_SVME
))
3655 static int svm_set_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
3657 struct vcpu_svm
*svm
= to_svm(vcpu
);
3659 u32 ecx
= msr
->index
;
3660 u64 data
= msr
->data
;
3663 kvm_write_tsc(vcpu
, msr
);
3666 svm
->vmcb
->save
.star
= data
;
3668 #ifdef CONFIG_X86_64
3670 svm
->vmcb
->save
.lstar
= data
;
3673 svm
->vmcb
->save
.cstar
= data
;
3675 case MSR_KERNEL_GS_BASE
:
3676 svm
->vmcb
->save
.kernel_gs_base
= data
;
3678 case MSR_SYSCALL_MASK
:
3679 svm
->vmcb
->save
.sfmask
= data
;
3682 case MSR_IA32_SYSENTER_CS
:
3683 svm
->vmcb
->save
.sysenter_cs
= data
;
3685 case MSR_IA32_SYSENTER_EIP
:
3686 svm
->sysenter_eip
= data
;
3687 svm
->vmcb
->save
.sysenter_eip
= data
;
3689 case MSR_IA32_SYSENTER_ESP
:
3690 svm
->sysenter_esp
= data
;
3691 svm
->vmcb
->save
.sysenter_esp
= data
;
3694 if (!boot_cpu_has(X86_FEATURE_RDTSCP
))
3698 * This is rare, so we update the MSR here instead of using
3699 * direct_access_msrs. Doing that would require a rdmsr in
3702 svm
->tsc_aux
= data
;
3703 wrmsrl(MSR_TSC_AUX
, svm
->tsc_aux
);
3705 case MSR_IA32_DEBUGCTLMSR
:
3706 if (!boot_cpu_has(X86_FEATURE_LBRV
)) {
3707 vcpu_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3711 if (data
& DEBUGCTL_RESERVED_BITS
)
3714 svm
->vmcb
->save
.dbgctl
= data
;
3715 mark_dirty(svm
->vmcb
, VMCB_LBR
);
3716 if (data
& (1ULL<<0))
3717 svm_enable_lbrv(svm
);
3719 svm_disable_lbrv(svm
);
3721 case MSR_VM_HSAVE_PA
:
3722 svm
->nested
.hsave_msr
= data
;
3725 return svm_set_vm_cr(vcpu
, data
);
3727 vcpu_unimpl(vcpu
, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx
, data
);
3729 case MSR_IA32_SPEC_CTRL
:
3730 svm
->spec_ctrl
= data
;
3732 case MSR_IA32_APICBASE
:
3733 if (kvm_vcpu_apicv_active(vcpu
))
3734 avic_update_vapic_bar(to_svm(vcpu
), data
);
3735 /* Follow through */
3737 return kvm_set_msr_common(vcpu
, msr
);
3742 static int wrmsr_interception(struct vcpu_svm
*svm
)
3744 struct msr_data msr
;
3745 u32 ecx
= kvm_register_read(&svm
->vcpu
, VCPU_REGS_RCX
);
3746 u64 data
= kvm_read_edx_eax(&svm
->vcpu
);
3750 msr
.host_initiated
= false;
3752 svm
->next_rip
= kvm_rip_read(&svm
->vcpu
) + 2;
3753 if (kvm_set_msr(&svm
->vcpu
, &msr
)) {
3754 trace_kvm_msr_write_ex(ecx
, data
);
3755 kvm_inject_gp(&svm
->vcpu
, 0);
3758 trace_kvm_msr_write(ecx
, data
);
3759 return kvm_skip_emulated_instruction(&svm
->vcpu
);
3763 static int msr_interception(struct vcpu_svm
*svm
)
3765 if (svm
->vmcb
->control
.exit_info_1
)
3766 return wrmsr_interception(svm
);
3768 return rdmsr_interception(svm
);
3771 static int interrupt_window_interception(struct vcpu_svm
*svm
)
3773 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
3774 svm_clear_vintr(svm
);
3775 svm
->vmcb
->control
.int_ctl
&= ~V_IRQ_MASK
;
3776 mark_dirty(svm
->vmcb
, VMCB_INTR
);
3777 ++svm
->vcpu
.stat
.irq_window_exits
;
3781 static int pause_interception(struct vcpu_svm
*svm
)
3783 kvm_vcpu_on_spin(&(svm
->vcpu
));
3787 static int nop_interception(struct vcpu_svm
*svm
)
3789 return kvm_skip_emulated_instruction(&(svm
->vcpu
));
3792 static int monitor_interception(struct vcpu_svm
*svm
)
3794 printk_once(KERN_WARNING
"kvm: MONITOR instruction emulated as NOP!\n");
3795 return nop_interception(svm
);
3798 static int mwait_interception(struct vcpu_svm
*svm
)
3800 printk_once(KERN_WARNING
"kvm: MWAIT instruction emulated as NOP!\n");
3801 return nop_interception(svm
);
3804 enum avic_ipi_failure_cause
{
3805 AVIC_IPI_FAILURE_INVALID_INT_TYPE
,
3806 AVIC_IPI_FAILURE_TARGET_NOT_RUNNING
,
3807 AVIC_IPI_FAILURE_INVALID_TARGET
,
3808 AVIC_IPI_FAILURE_INVALID_BACKING_PAGE
,
3811 static int avic_incomplete_ipi_interception(struct vcpu_svm
*svm
)
3813 u32 icrh
= svm
->vmcb
->control
.exit_info_1
>> 32;
3814 u32 icrl
= svm
->vmcb
->control
.exit_info_1
;
3815 u32 id
= svm
->vmcb
->control
.exit_info_2
>> 32;
3816 u32 index
= svm
->vmcb
->control
.exit_info_2
& 0xFF;
3817 struct kvm_lapic
*apic
= svm
->vcpu
.arch
.apic
;
3819 trace_kvm_avic_incomplete_ipi(svm
->vcpu
.vcpu_id
, icrh
, icrl
, id
, index
);
3822 case AVIC_IPI_FAILURE_INVALID_INT_TYPE
:
3824 * AVIC hardware handles the generation of
3825 * IPIs when the specified Message Type is Fixed
3826 * (also known as fixed delivery mode) and
3827 * the Trigger Mode is edge-triggered. The hardware
3828 * also supports self and broadcast delivery modes
3829 * specified via the Destination Shorthand(DSH)
3830 * field of the ICRL. Logical and physical APIC ID
3831 * formats are supported. All other IPI types cause
3832 * a #VMEXIT, which needs to emulated.
3834 kvm_lapic_reg_write(apic
, APIC_ICR2
, icrh
);
3835 kvm_lapic_reg_write(apic
, APIC_ICR
, icrl
);
3837 case AVIC_IPI_FAILURE_TARGET_NOT_RUNNING
: {
3839 struct kvm_vcpu
*vcpu
;
3840 struct kvm
*kvm
= svm
->vcpu
.kvm
;
3841 struct kvm_lapic
*apic
= svm
->vcpu
.arch
.apic
;
3844 * At this point, we expect that the AVIC HW has already
3845 * set the appropriate IRR bits on the valid target
3846 * vcpus. So, we just need to kick the appropriate vcpu.
3848 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
3849 bool m
= kvm_apic_match_dest(vcpu
, apic
,
3850 icrl
& KVM_APIC_SHORT_MASK
,
3851 GET_APIC_DEST_FIELD(icrh
),
3852 icrl
& KVM_APIC_DEST_MASK
);
3854 if (m
&& !avic_vcpu_is_running(vcpu
))
3855 kvm_vcpu_wake_up(vcpu
);
3859 case AVIC_IPI_FAILURE_INVALID_TARGET
:
3861 case AVIC_IPI_FAILURE_INVALID_BACKING_PAGE
:
3862 WARN_ONCE(1, "Invalid backing page\n");
3865 pr_err("Unknown IPI interception\n");
3871 static u32
*avic_get_logical_id_entry(struct kvm_vcpu
*vcpu
, u32 ldr
, bool flat
)
3873 struct kvm_arch
*vm_data
= &vcpu
->kvm
->arch
;
3875 u32
*logical_apic_id_table
;
3876 int dlid
= GET_APIC_LOGICAL_ID(ldr
);
3881 if (flat
) { /* flat */
3882 index
= ffs(dlid
) - 1;
3885 } else { /* cluster */
3886 int cluster
= (dlid
& 0xf0) >> 4;
3887 int apic
= ffs(dlid
& 0x0f) - 1;
3889 if ((apic
< 0) || (apic
> 7) ||
3892 index
= (cluster
<< 2) + apic
;
3895 logical_apic_id_table
= (u32
*) page_address(vm_data
->avic_logical_id_table_page
);
3897 return &logical_apic_id_table
[index
];
3900 static int avic_ldr_write(struct kvm_vcpu
*vcpu
, u8 g_physical_id
, u32 ldr
,
3904 u32
*entry
, new_entry
;
3906 flat
= kvm_lapic_get_reg(vcpu
->arch
.apic
, APIC_DFR
) == APIC_DFR_FLAT
;
3907 entry
= avic_get_logical_id_entry(vcpu
, ldr
, flat
);
3911 new_entry
= READ_ONCE(*entry
);
3912 new_entry
&= ~AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK
;
3913 new_entry
|= (g_physical_id
& AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK
);
3915 new_entry
|= AVIC_LOGICAL_ID_ENTRY_VALID_MASK
;
3917 new_entry
&= ~AVIC_LOGICAL_ID_ENTRY_VALID_MASK
;
3918 WRITE_ONCE(*entry
, new_entry
);
3923 static int avic_handle_ldr_update(struct kvm_vcpu
*vcpu
)
3926 struct vcpu_svm
*svm
= to_svm(vcpu
);
3927 u32 ldr
= kvm_lapic_get_reg(vcpu
->arch
.apic
, APIC_LDR
);
3932 ret
= avic_ldr_write(vcpu
, vcpu
->vcpu_id
, ldr
, true);
3933 if (ret
&& svm
->ldr_reg
) {
3934 avic_ldr_write(vcpu
, 0, svm
->ldr_reg
, false);
3942 static int avic_handle_apic_id_update(struct kvm_vcpu
*vcpu
)
3945 struct vcpu_svm
*svm
= to_svm(vcpu
);
3946 u32 apic_id_reg
= kvm_lapic_get_reg(vcpu
->arch
.apic
, APIC_ID
);
3947 u32 id
= (apic_id_reg
>> 24) & 0xff;
3949 if (vcpu
->vcpu_id
== id
)
3952 old
= avic_get_physical_id_entry(vcpu
, vcpu
->vcpu_id
);
3953 new = avic_get_physical_id_entry(vcpu
, id
);
3957 /* We need to move physical_id_entry to new offset */
3960 to_svm(vcpu
)->avic_physical_id_cache
= new;
3963 * Also update the guest physical APIC ID in the logical
3964 * APIC ID table entry if already setup the LDR.
3967 avic_handle_ldr_update(vcpu
);
3972 static int avic_handle_dfr_update(struct kvm_vcpu
*vcpu
)
3974 struct vcpu_svm
*svm
= to_svm(vcpu
);
3975 struct kvm_arch
*vm_data
= &vcpu
->kvm
->arch
;
3976 u32 dfr
= kvm_lapic_get_reg(vcpu
->arch
.apic
, APIC_DFR
);
3977 u32 mod
= (dfr
>> 28) & 0xf;
3980 * We assume that all local APICs are using the same type.
3981 * If this changes, we need to flush the AVIC logical
3984 if (vm_data
->ldr_mode
== mod
)
3987 clear_page(page_address(vm_data
->avic_logical_id_table_page
));
3988 vm_data
->ldr_mode
= mod
;
3991 avic_handle_ldr_update(vcpu
);
3995 static int avic_unaccel_trap_write(struct vcpu_svm
*svm
)
3997 struct kvm_lapic
*apic
= svm
->vcpu
.arch
.apic
;
3998 u32 offset
= svm
->vmcb
->control
.exit_info_1
&
3999 AVIC_UNACCEL_ACCESS_OFFSET_MASK
;
4003 if (avic_handle_apic_id_update(&svm
->vcpu
))
4007 if (avic_handle_ldr_update(&svm
->vcpu
))
4011 avic_handle_dfr_update(&svm
->vcpu
);
4017 kvm_lapic_reg_write(apic
, offset
, kvm_lapic_get_reg(apic
, offset
));
4022 static bool is_avic_unaccelerated_access_trap(u32 offset
)
4051 static int avic_unaccelerated_access_interception(struct vcpu_svm
*svm
)
4054 u32 offset
= svm
->vmcb
->control
.exit_info_1
&
4055 AVIC_UNACCEL_ACCESS_OFFSET_MASK
;
4056 u32 vector
= svm
->vmcb
->control
.exit_info_2
&
4057 AVIC_UNACCEL_ACCESS_VECTOR_MASK
;
4058 bool write
= (svm
->vmcb
->control
.exit_info_1
>> 32) &
4059 AVIC_UNACCEL_ACCESS_WRITE_MASK
;
4060 bool trap
= is_avic_unaccelerated_access_trap(offset
);
4062 trace_kvm_avic_unaccelerated_access(svm
->vcpu
.vcpu_id
, offset
,
4063 trap
, write
, vector
);
4066 WARN_ONCE(!write
, "svm: Handling trap read.\n");
4067 ret
= avic_unaccel_trap_write(svm
);
4069 /* Handling Fault */
4070 ret
= (emulate_instruction(&svm
->vcpu
, 0) == EMULATE_DONE
);
4076 static int (*const svm_exit_handlers
[])(struct vcpu_svm
*svm
) = {
4077 [SVM_EXIT_READ_CR0
] = cr_interception
,
4078 [SVM_EXIT_READ_CR3
] = cr_interception
,
4079 [SVM_EXIT_READ_CR4
] = cr_interception
,
4080 [SVM_EXIT_READ_CR8
] = cr_interception
,
4081 [SVM_EXIT_CR0_SEL_WRITE
] = cr_interception
,
4082 [SVM_EXIT_WRITE_CR0
] = cr_interception
,
4083 [SVM_EXIT_WRITE_CR3
] = cr_interception
,
4084 [SVM_EXIT_WRITE_CR4
] = cr_interception
,
4085 [SVM_EXIT_WRITE_CR8
] = cr8_write_interception
,
4086 [SVM_EXIT_READ_DR0
] = dr_interception
,
4087 [SVM_EXIT_READ_DR1
] = dr_interception
,
4088 [SVM_EXIT_READ_DR2
] = dr_interception
,
4089 [SVM_EXIT_READ_DR3
] = dr_interception
,
4090 [SVM_EXIT_READ_DR4
] = dr_interception
,
4091 [SVM_EXIT_READ_DR5
] = dr_interception
,
4092 [SVM_EXIT_READ_DR6
] = dr_interception
,
4093 [SVM_EXIT_READ_DR7
] = dr_interception
,
4094 [SVM_EXIT_WRITE_DR0
] = dr_interception
,
4095 [SVM_EXIT_WRITE_DR1
] = dr_interception
,
4096 [SVM_EXIT_WRITE_DR2
] = dr_interception
,
4097 [SVM_EXIT_WRITE_DR3
] = dr_interception
,
4098 [SVM_EXIT_WRITE_DR4
] = dr_interception
,
4099 [SVM_EXIT_WRITE_DR5
] = dr_interception
,
4100 [SVM_EXIT_WRITE_DR6
] = dr_interception
,
4101 [SVM_EXIT_WRITE_DR7
] = dr_interception
,
4102 [SVM_EXIT_EXCP_BASE
+ DB_VECTOR
] = db_interception
,
4103 [SVM_EXIT_EXCP_BASE
+ BP_VECTOR
] = bp_interception
,
4104 [SVM_EXIT_EXCP_BASE
+ UD_VECTOR
] = ud_interception
,
4105 [SVM_EXIT_EXCP_BASE
+ PF_VECTOR
] = pf_interception
,
4106 [SVM_EXIT_EXCP_BASE
+ MC_VECTOR
] = mc_interception
,
4107 [SVM_EXIT_EXCP_BASE
+ AC_VECTOR
] = ac_interception
,
4108 [SVM_EXIT_INTR
] = intr_interception
,
4109 [SVM_EXIT_NMI
] = nmi_interception
,
4110 [SVM_EXIT_SMI
] = nop_on_interception
,
4111 [SVM_EXIT_INIT
] = nop_on_interception
,
4112 [SVM_EXIT_VINTR
] = interrupt_window_interception
,
4113 [SVM_EXIT_RDPMC
] = rdpmc_interception
,
4114 [SVM_EXIT_CPUID
] = cpuid_interception
,
4115 [SVM_EXIT_IRET
] = iret_interception
,
4116 [SVM_EXIT_INVD
] = emulate_on_interception
,
4117 [SVM_EXIT_PAUSE
] = pause_interception
,
4118 [SVM_EXIT_HLT
] = halt_interception
,
4119 [SVM_EXIT_INVLPG
] = invlpg_interception
,
4120 [SVM_EXIT_INVLPGA
] = invlpga_interception
,
4121 [SVM_EXIT_IOIO
] = io_interception
,
4122 [SVM_EXIT_MSR
] = msr_interception
,
4123 [SVM_EXIT_TASK_SWITCH
] = task_switch_interception
,
4124 [SVM_EXIT_SHUTDOWN
] = shutdown_interception
,
4125 [SVM_EXIT_VMRUN
] = vmrun_interception
,
4126 [SVM_EXIT_VMMCALL
] = vmmcall_interception
,
4127 [SVM_EXIT_VMLOAD
] = vmload_interception
,
4128 [SVM_EXIT_VMSAVE
] = vmsave_interception
,
4129 [SVM_EXIT_STGI
] = stgi_interception
,
4130 [SVM_EXIT_CLGI
] = clgi_interception
,
4131 [SVM_EXIT_SKINIT
] = skinit_interception
,
4132 [SVM_EXIT_WBINVD
] = wbinvd_interception
,
4133 [SVM_EXIT_MONITOR
] = monitor_interception
,
4134 [SVM_EXIT_MWAIT
] = mwait_interception
,
4135 [SVM_EXIT_XSETBV
] = xsetbv_interception
,
4136 [SVM_EXIT_NPF
] = pf_interception
,
4137 [SVM_EXIT_RSM
] = emulate_on_interception
,
4138 [SVM_EXIT_AVIC_INCOMPLETE_IPI
] = avic_incomplete_ipi_interception
,
4139 [SVM_EXIT_AVIC_UNACCELERATED_ACCESS
] = avic_unaccelerated_access_interception
,
4142 static void dump_vmcb(struct kvm_vcpu
*vcpu
)
4144 struct vcpu_svm
*svm
= to_svm(vcpu
);
4145 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
4146 struct vmcb_save_area
*save
= &svm
->vmcb
->save
;
4148 pr_err("VMCB Control Area:\n");
4149 pr_err("%-20s%04x\n", "cr_read:", control
->intercept_cr
& 0xffff);
4150 pr_err("%-20s%04x\n", "cr_write:", control
->intercept_cr
>> 16);
4151 pr_err("%-20s%04x\n", "dr_read:", control
->intercept_dr
& 0xffff);
4152 pr_err("%-20s%04x\n", "dr_write:", control
->intercept_dr
>> 16);
4153 pr_err("%-20s%08x\n", "exceptions:", control
->intercept_exceptions
);
4154 pr_err("%-20s%016llx\n", "intercepts:", control
->intercept
);
4155 pr_err("%-20s%d\n", "pause filter count:", control
->pause_filter_count
);
4156 pr_err("%-20s%016llx\n", "iopm_base_pa:", control
->iopm_base_pa
);
4157 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control
->msrpm_base_pa
);
4158 pr_err("%-20s%016llx\n", "tsc_offset:", control
->tsc_offset
);
4159 pr_err("%-20s%d\n", "asid:", control
->asid
);
4160 pr_err("%-20s%d\n", "tlb_ctl:", control
->tlb_ctl
);
4161 pr_err("%-20s%08x\n", "int_ctl:", control
->int_ctl
);
4162 pr_err("%-20s%08x\n", "int_vector:", control
->int_vector
);
4163 pr_err("%-20s%08x\n", "int_state:", control
->int_state
);
4164 pr_err("%-20s%08x\n", "exit_code:", control
->exit_code
);
4165 pr_err("%-20s%016llx\n", "exit_info1:", control
->exit_info_1
);
4166 pr_err("%-20s%016llx\n", "exit_info2:", control
->exit_info_2
);
4167 pr_err("%-20s%08x\n", "exit_int_info:", control
->exit_int_info
);
4168 pr_err("%-20s%08x\n", "exit_int_info_err:", control
->exit_int_info_err
);
4169 pr_err("%-20s%lld\n", "nested_ctl:", control
->nested_ctl
);
4170 pr_err("%-20s%016llx\n", "nested_cr3:", control
->nested_cr3
);
4171 pr_err("%-20s%016llx\n", "avic_vapic_bar:", control
->avic_vapic_bar
);
4172 pr_err("%-20s%08x\n", "event_inj:", control
->event_inj
);
4173 pr_err("%-20s%08x\n", "event_inj_err:", control
->event_inj_err
);
4174 pr_err("%-20s%lld\n", "virt_ext:", control
->virt_ext
);
4175 pr_err("%-20s%016llx\n", "next_rip:", control
->next_rip
);
4176 pr_err("%-20s%016llx\n", "avic_backing_page:", control
->avic_backing_page
);
4177 pr_err("%-20s%016llx\n", "avic_logical_id:", control
->avic_logical_id
);
4178 pr_err("%-20s%016llx\n", "avic_physical_id:", control
->avic_physical_id
);
4179 pr_err("VMCB State Save Area:\n");
4180 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4182 save
->es
.selector
, save
->es
.attrib
,
4183 save
->es
.limit
, save
->es
.base
);
4184 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4186 save
->cs
.selector
, save
->cs
.attrib
,
4187 save
->cs
.limit
, save
->cs
.base
);
4188 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4190 save
->ss
.selector
, save
->ss
.attrib
,
4191 save
->ss
.limit
, save
->ss
.base
);
4192 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4194 save
->ds
.selector
, save
->ds
.attrib
,
4195 save
->ds
.limit
, save
->ds
.base
);
4196 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4198 save
->fs
.selector
, save
->fs
.attrib
,
4199 save
->fs
.limit
, save
->fs
.base
);
4200 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4202 save
->gs
.selector
, save
->gs
.attrib
,
4203 save
->gs
.limit
, save
->gs
.base
);
4204 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4206 save
->gdtr
.selector
, save
->gdtr
.attrib
,
4207 save
->gdtr
.limit
, save
->gdtr
.base
);
4208 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4210 save
->ldtr
.selector
, save
->ldtr
.attrib
,
4211 save
->ldtr
.limit
, save
->ldtr
.base
);
4212 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4214 save
->idtr
.selector
, save
->idtr
.attrib
,
4215 save
->idtr
.limit
, save
->idtr
.base
);
4216 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4218 save
->tr
.selector
, save
->tr
.attrib
,
4219 save
->tr
.limit
, save
->tr
.base
);
4220 pr_err("cpl: %d efer: %016llx\n",
4221 save
->cpl
, save
->efer
);
4222 pr_err("%-15s %016llx %-13s %016llx\n",
4223 "cr0:", save
->cr0
, "cr2:", save
->cr2
);
4224 pr_err("%-15s %016llx %-13s %016llx\n",
4225 "cr3:", save
->cr3
, "cr4:", save
->cr4
);
4226 pr_err("%-15s %016llx %-13s %016llx\n",
4227 "dr6:", save
->dr6
, "dr7:", save
->dr7
);
4228 pr_err("%-15s %016llx %-13s %016llx\n",
4229 "rip:", save
->rip
, "rflags:", save
->rflags
);
4230 pr_err("%-15s %016llx %-13s %016llx\n",
4231 "rsp:", save
->rsp
, "rax:", save
->rax
);
4232 pr_err("%-15s %016llx %-13s %016llx\n",
4233 "star:", save
->star
, "lstar:", save
->lstar
);
4234 pr_err("%-15s %016llx %-13s %016llx\n",
4235 "cstar:", save
->cstar
, "sfmask:", save
->sfmask
);
4236 pr_err("%-15s %016llx %-13s %016llx\n",
4237 "kernel_gs_base:", save
->kernel_gs_base
,
4238 "sysenter_cs:", save
->sysenter_cs
);
4239 pr_err("%-15s %016llx %-13s %016llx\n",
4240 "sysenter_esp:", save
->sysenter_esp
,
4241 "sysenter_eip:", save
->sysenter_eip
);
4242 pr_err("%-15s %016llx %-13s %016llx\n",
4243 "gpat:", save
->g_pat
, "dbgctl:", save
->dbgctl
);
4244 pr_err("%-15s %016llx %-13s %016llx\n",
4245 "br_from:", save
->br_from
, "br_to:", save
->br_to
);
4246 pr_err("%-15s %016llx %-13s %016llx\n",
4247 "excp_from:", save
->last_excp_from
,
4248 "excp_to:", save
->last_excp_to
);
4251 static void svm_get_exit_info(struct kvm_vcpu
*vcpu
, u64
*info1
, u64
*info2
)
4253 struct vmcb_control_area
*control
= &to_svm(vcpu
)->vmcb
->control
;
4255 *info1
= control
->exit_info_1
;
4256 *info2
= control
->exit_info_2
;
4259 static int handle_exit(struct kvm_vcpu
*vcpu
)
4261 struct vcpu_svm
*svm
= to_svm(vcpu
);
4262 struct kvm_run
*kvm_run
= vcpu
->run
;
4263 u32 exit_code
= svm
->vmcb
->control
.exit_code
;
4265 trace_kvm_exit(exit_code
, vcpu
, KVM_ISA_SVM
);
4267 vcpu
->arch
.gpa_available
= (exit_code
== SVM_EXIT_NPF
);
4269 if (!is_cr_intercept(svm
, INTERCEPT_CR0_WRITE
))
4270 vcpu
->arch
.cr0
= svm
->vmcb
->save
.cr0
;
4272 vcpu
->arch
.cr3
= svm
->vmcb
->save
.cr3
;
4274 if (unlikely(svm
->nested
.exit_required
)) {
4275 nested_svm_vmexit(svm
);
4276 svm
->nested
.exit_required
= false;
4281 if (is_guest_mode(vcpu
)) {
4284 trace_kvm_nested_vmexit(svm
->vmcb
->save
.rip
, exit_code
,
4285 svm
->vmcb
->control
.exit_info_1
,
4286 svm
->vmcb
->control
.exit_info_2
,
4287 svm
->vmcb
->control
.exit_int_info
,
4288 svm
->vmcb
->control
.exit_int_info_err
,
4291 vmexit
= nested_svm_exit_special(svm
);
4293 if (vmexit
== NESTED_EXIT_CONTINUE
)
4294 vmexit
= nested_svm_exit_handled(svm
);
4296 if (vmexit
== NESTED_EXIT_DONE
)
4300 svm_complete_interrupts(svm
);
4302 if (svm
->vmcb
->control
.exit_code
== SVM_EXIT_ERR
) {
4303 kvm_run
->exit_reason
= KVM_EXIT_FAIL_ENTRY
;
4304 kvm_run
->fail_entry
.hardware_entry_failure_reason
4305 = svm
->vmcb
->control
.exit_code
;
4306 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
4311 if (is_external_interrupt(svm
->vmcb
->control
.exit_int_info
) &&
4312 exit_code
!= SVM_EXIT_EXCP_BASE
+ PF_VECTOR
&&
4313 exit_code
!= SVM_EXIT_NPF
&& exit_code
!= SVM_EXIT_TASK_SWITCH
&&
4314 exit_code
!= SVM_EXIT_INTR
&& exit_code
!= SVM_EXIT_NMI
)
4315 printk(KERN_ERR
"%s: unexpected exit_int_info 0x%x "
4317 __func__
, svm
->vmcb
->control
.exit_int_info
,
4320 if (exit_code
>= ARRAY_SIZE(svm_exit_handlers
)
4321 || !svm_exit_handlers
[exit_code
]) {
4322 WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code
);
4323 kvm_queue_exception(vcpu
, UD_VECTOR
);
4327 return svm_exit_handlers
[exit_code
](svm
);
4330 static void reload_tss(struct kvm_vcpu
*vcpu
)
4332 int cpu
= raw_smp_processor_id();
4334 struct svm_cpu_data
*sd
= per_cpu(svm_data
, cpu
);
4335 sd
->tss_desc
->type
= 9; /* available 32/64-bit TSS */
4339 static void pre_svm_run(struct vcpu_svm
*svm
)
4341 int cpu
= raw_smp_processor_id();
4343 struct svm_cpu_data
*sd
= per_cpu(svm_data
, cpu
);
4345 /* FIXME: handle wraparound of asid_generation */
4346 if (svm
->asid_generation
!= sd
->asid_generation
)
4350 static void svm_inject_nmi(struct kvm_vcpu
*vcpu
)
4352 struct vcpu_svm
*svm
= to_svm(vcpu
);
4354 svm
->vmcb
->control
.event_inj
= SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_NMI
;
4355 vcpu
->arch
.hflags
|= HF_NMI_MASK
;
4356 set_intercept(svm
, INTERCEPT_IRET
);
4357 ++vcpu
->stat
.nmi_injections
;
4360 static inline void svm_inject_irq(struct vcpu_svm
*svm
, int irq
)
4362 struct vmcb_control_area
*control
;
4364 /* The following fields are ignored when AVIC is enabled */
4365 control
= &svm
->vmcb
->control
;
4366 control
->int_vector
= irq
;
4367 control
->int_ctl
&= ~V_INTR_PRIO_MASK
;
4368 control
->int_ctl
|= V_IRQ_MASK
|
4369 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT
);
4370 mark_dirty(svm
->vmcb
, VMCB_INTR
);
4373 static void svm_set_irq(struct kvm_vcpu
*vcpu
)
4375 struct vcpu_svm
*svm
= to_svm(vcpu
);
4377 BUG_ON(!(gif_set(svm
)));
4379 trace_kvm_inj_virq(vcpu
->arch
.interrupt
.nr
);
4380 ++vcpu
->stat
.irq_injections
;
4382 svm
->vmcb
->control
.event_inj
= vcpu
->arch
.interrupt
.nr
|
4383 SVM_EVTINJ_VALID
| SVM_EVTINJ_TYPE_INTR
;
4386 static inline bool svm_nested_virtualize_tpr(struct kvm_vcpu
*vcpu
)
4388 return is_guest_mode(vcpu
) && (vcpu
->arch
.hflags
& HF_VINTR_MASK
);
4391 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
, int tpr
, int irr
)
4393 struct vcpu_svm
*svm
= to_svm(vcpu
);
4395 if (svm_nested_virtualize_tpr(vcpu
) ||
4396 kvm_vcpu_apicv_active(vcpu
))
4399 clr_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
4405 set_cr_intercept(svm
, INTERCEPT_CR8_WRITE
);
4408 static void svm_set_virtual_x2apic_mode(struct kvm_vcpu
*vcpu
, bool set
)
4413 static bool svm_get_enable_apicv(void)
4418 static void svm_hwapic_irr_update(struct kvm_vcpu
*vcpu
, int max_irr
)
4422 static void svm_hwapic_isr_update(struct kvm_vcpu
*vcpu
, int max_isr
)
4426 /* Note: Currently only used by Hyper-V. */
4427 static void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu
*vcpu
)
4429 struct vcpu_svm
*svm
= to_svm(vcpu
);
4430 struct vmcb
*vmcb
= svm
->vmcb
;
4435 vmcb
->control
.int_ctl
&= ~AVIC_ENABLE_MASK
;
4436 mark_dirty(vmcb
, VMCB_INTR
);
4439 static void svm_load_eoi_exitmap(struct kvm_vcpu
*vcpu
, u64
*eoi_exit_bitmap
)
4444 static void svm_deliver_avic_intr(struct kvm_vcpu
*vcpu
, int vec
)
4446 kvm_lapic_set_irr(vec
, vcpu
->arch
.apic
);
4447 smp_mb__after_atomic();
4449 if (avic_vcpu_is_running(vcpu
))
4450 wrmsrl(SVM_AVIC_DOORBELL
,
4451 kvm_cpu_get_apicid(vcpu
->cpu
));
4453 kvm_vcpu_wake_up(vcpu
);
4456 static void svm_ir_list_del(struct vcpu_svm
*svm
, struct amd_iommu_pi_data
*pi
)
4458 unsigned long flags
;
4459 struct amd_svm_iommu_ir
*cur
;
4461 spin_lock_irqsave(&svm
->ir_list_lock
, flags
);
4462 list_for_each_entry(cur
, &svm
->ir_list
, node
) {
4463 if (cur
->data
!= pi
->ir_data
)
4465 list_del(&cur
->node
);
4469 spin_unlock_irqrestore(&svm
->ir_list_lock
, flags
);
4472 static int svm_ir_list_add(struct vcpu_svm
*svm
, struct amd_iommu_pi_data
*pi
)
4475 unsigned long flags
;
4476 struct amd_svm_iommu_ir
*ir
;
4479 * In some cases, the existing irte is updaed and re-set,
4480 * so we need to check here if it's already been * added
4483 if (pi
->ir_data
&& (pi
->prev_ga_tag
!= 0)) {
4484 struct kvm
*kvm
= svm
->vcpu
.kvm
;
4485 u32 vcpu_id
= AVIC_GATAG_TO_VCPUID(pi
->prev_ga_tag
);
4486 struct kvm_vcpu
*prev_vcpu
= kvm_get_vcpu_by_id(kvm
, vcpu_id
);
4487 struct vcpu_svm
*prev_svm
;
4494 prev_svm
= to_svm(prev_vcpu
);
4495 svm_ir_list_del(prev_svm
, pi
);
4499 * Allocating new amd_iommu_pi_data, which will get
4500 * add to the per-vcpu ir_list.
4502 ir
= kzalloc(sizeof(struct amd_svm_iommu_ir
), GFP_KERNEL
);
4507 ir
->data
= pi
->ir_data
;
4509 spin_lock_irqsave(&svm
->ir_list_lock
, flags
);
4510 list_add(&ir
->node
, &svm
->ir_list
);
4511 spin_unlock_irqrestore(&svm
->ir_list_lock
, flags
);
4518 * The HW cannot support posting multicast/broadcast
4519 * interrupts to a vCPU. So, we still use legacy interrupt
4520 * remapping for these kind of interrupts.
4522 * For lowest-priority interrupts, we only support
4523 * those with single CPU as the destination, e.g. user
4524 * configures the interrupts via /proc/irq or uses
4525 * irqbalance to make the interrupts single-CPU.
4528 get_pi_vcpu_info(struct kvm
*kvm
, struct kvm_kernel_irq_routing_entry
*e
,
4529 struct vcpu_data
*vcpu_info
, struct vcpu_svm
**svm
)
4531 struct kvm_lapic_irq irq
;
4532 struct kvm_vcpu
*vcpu
= NULL
;
4534 kvm_set_msi_irq(kvm
, e
, &irq
);
4536 if (!kvm_intr_is_single_vcpu(kvm
, &irq
, &vcpu
)) {
4537 pr_debug("SVM: %s: use legacy intr remap mode for irq %u\n",
4538 __func__
, irq
.vector
);
4542 pr_debug("SVM: %s: use GA mode for irq %u\n", __func__
,
4544 *svm
= to_svm(vcpu
);
4545 vcpu_info
->pi_desc_addr
= page_to_phys((*svm
)->avic_backing_page
);
4546 vcpu_info
->vector
= irq
.vector
;
4552 * svm_update_pi_irte - set IRTE for Posted-Interrupts
4555 * @host_irq: host irq of the interrupt
4556 * @guest_irq: gsi of the interrupt
4557 * @set: set or unset PI
4558 * returns 0 on success, < 0 on failure
4560 static int svm_update_pi_irte(struct kvm
*kvm
, unsigned int host_irq
,
4561 uint32_t guest_irq
, bool set
)
4563 struct kvm_kernel_irq_routing_entry
*e
;
4564 struct kvm_irq_routing_table
*irq_rt
;
4565 int idx
, ret
= -EINVAL
;
4567 if (!kvm_arch_has_assigned_device(kvm
) ||
4568 !irq_remapping_cap(IRQ_POSTING_CAP
))
4571 pr_debug("SVM: %s: host_irq=%#x, guest_irq=%#x, set=%#x\n",
4572 __func__
, host_irq
, guest_irq
, set
);
4574 idx
= srcu_read_lock(&kvm
->irq_srcu
);
4575 irq_rt
= srcu_dereference(kvm
->irq_routing
, &kvm
->irq_srcu
);
4576 WARN_ON(guest_irq
>= irq_rt
->nr_rt_entries
);
4578 hlist_for_each_entry(e
, &irq_rt
->map
[guest_irq
], link
) {
4579 struct vcpu_data vcpu_info
;
4580 struct vcpu_svm
*svm
= NULL
;
4582 if (e
->type
!= KVM_IRQ_ROUTING_MSI
)
4586 * Here, we setup with legacy mode in the following cases:
4587 * 1. When cannot target interrupt to a specific vcpu.
4588 * 2. Unsetting posted interrupt.
4589 * 3. APIC virtialization is disabled for the vcpu.
4591 if (!get_pi_vcpu_info(kvm
, e
, &vcpu_info
, &svm
) && set
&&
4592 kvm_vcpu_apicv_active(&svm
->vcpu
)) {
4593 struct amd_iommu_pi_data pi
;
4595 /* Try to enable guest_mode in IRTE */
4596 pi
.base
= page_to_phys(svm
->avic_backing_page
) & AVIC_HPA_MASK
;
4597 pi
.ga_tag
= AVIC_GATAG(kvm
->arch
.avic_vm_id
,
4599 pi
.is_guest_mode
= true;
4600 pi
.vcpu_data
= &vcpu_info
;
4601 ret
= irq_set_vcpu_affinity(host_irq
, &pi
);
4604 * Here, we successfully setting up vcpu affinity in
4605 * IOMMU guest mode. Now, we need to store the posted
4606 * interrupt information in a per-vcpu ir_list so that
4607 * we can reference to them directly when we update vcpu
4608 * scheduling information in IOMMU irte.
4610 if (!ret
&& pi
.is_guest_mode
)
4611 svm_ir_list_add(svm
, &pi
);
4613 /* Use legacy mode in IRTE */
4614 struct amd_iommu_pi_data pi
;
4617 * Here, pi is used to:
4618 * - Tell IOMMU to use legacy mode for this interrupt.
4619 * - Retrieve ga_tag of prior interrupt remapping data.
4621 pi
.is_guest_mode
= false;
4622 ret
= irq_set_vcpu_affinity(host_irq
, &pi
);
4625 * Check if the posted interrupt was previously
4626 * setup with the guest_mode by checking if the ga_tag
4627 * was cached. If so, we need to clean up the per-vcpu
4630 if (!ret
&& pi
.prev_ga_tag
) {
4631 int id
= AVIC_GATAG_TO_VCPUID(pi
.prev_ga_tag
);
4632 struct kvm_vcpu
*vcpu
;
4634 vcpu
= kvm_get_vcpu_by_id(kvm
, id
);
4636 svm_ir_list_del(to_svm(vcpu
), &pi
);
4641 trace_kvm_pi_irte_update(svm
->vcpu
.vcpu_id
,
4644 vcpu_info
.pi_desc_addr
, set
);
4648 pr_err("%s: failed to update PI IRTE\n", __func__
);
4655 srcu_read_unlock(&kvm
->irq_srcu
, idx
);
4659 static int svm_nmi_allowed(struct kvm_vcpu
*vcpu
)
4661 struct vcpu_svm
*svm
= to_svm(vcpu
);
4662 struct vmcb
*vmcb
= svm
->vmcb
;
4664 ret
= !(vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
) &&
4665 !(svm
->vcpu
.arch
.hflags
& HF_NMI_MASK
);
4666 ret
= ret
&& gif_set(svm
) && nested_svm_nmi(svm
);
4671 static bool svm_get_nmi_mask(struct kvm_vcpu
*vcpu
)
4673 struct vcpu_svm
*svm
= to_svm(vcpu
);
4675 return !!(svm
->vcpu
.arch
.hflags
& HF_NMI_MASK
);
4678 static void svm_set_nmi_mask(struct kvm_vcpu
*vcpu
, bool masked
)
4680 struct vcpu_svm
*svm
= to_svm(vcpu
);
4683 svm
->vcpu
.arch
.hflags
|= HF_NMI_MASK
;
4684 set_intercept(svm
, INTERCEPT_IRET
);
4686 svm
->vcpu
.arch
.hflags
&= ~HF_NMI_MASK
;
4687 clr_intercept(svm
, INTERCEPT_IRET
);
4691 static int svm_interrupt_allowed(struct kvm_vcpu
*vcpu
)
4693 struct vcpu_svm
*svm
= to_svm(vcpu
);
4694 struct vmcb
*vmcb
= svm
->vmcb
;
4697 if (!gif_set(svm
) ||
4698 (vmcb
->control
.int_state
& SVM_INTERRUPT_SHADOW_MASK
))
4701 ret
= !!(kvm_get_rflags(vcpu
) & X86_EFLAGS_IF
);
4703 if (is_guest_mode(vcpu
))
4704 return ret
&& !(svm
->vcpu
.arch
.hflags
& HF_VINTR_MASK
);
4709 static void enable_irq_window(struct kvm_vcpu
*vcpu
)
4711 struct vcpu_svm
*svm
= to_svm(vcpu
);
4713 if (kvm_vcpu_apicv_active(vcpu
))
4717 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
4718 * 1, because that's a separate STGI/VMRUN intercept. The next time we
4719 * get that intercept, this function will be called again though and
4720 * we'll get the vintr intercept.
4722 if (gif_set(svm
) && nested_svm_intr(svm
)) {
4724 svm_inject_irq(svm
, 0x0);
4728 static void enable_nmi_window(struct kvm_vcpu
*vcpu
)
4730 struct vcpu_svm
*svm
= to_svm(vcpu
);
4732 if ((svm
->vcpu
.arch
.hflags
& (HF_NMI_MASK
| HF_IRET_MASK
))
4734 return; /* IRET will cause a vm exit */
4736 if ((svm
->vcpu
.arch
.hflags
& HF_GIF_MASK
) == 0)
4737 return; /* STGI will cause a vm exit */
4739 if (svm
->nested
.exit_required
)
4740 return; /* we're not going to run the guest yet */
4743 * Something prevents NMI from been injected. Single step over possible
4744 * problem (IRET or exception injection or interrupt shadow)
4746 svm
->nmi_singlestep_guest_rflags
= svm_get_rflags(vcpu
);
4747 svm
->nmi_singlestep
= true;
4748 svm
->vmcb
->save
.rflags
|= (X86_EFLAGS_TF
| X86_EFLAGS_RF
);
4751 static int svm_set_tss_addr(struct kvm
*kvm
, unsigned int addr
)
4756 static void svm_flush_tlb(struct kvm_vcpu
*vcpu
)
4758 struct vcpu_svm
*svm
= to_svm(vcpu
);
4760 if (static_cpu_has(X86_FEATURE_FLUSHBYASID
))
4761 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_FLUSH_ASID
;
4763 svm
->asid_generation
--;
4766 static void svm_prepare_guest_switch(struct kvm_vcpu
*vcpu
)
4770 static inline void sync_cr8_to_lapic(struct kvm_vcpu
*vcpu
)
4772 struct vcpu_svm
*svm
= to_svm(vcpu
);
4774 if (svm_nested_virtualize_tpr(vcpu
))
4777 if (!is_cr_intercept(svm
, INTERCEPT_CR8_WRITE
)) {
4778 int cr8
= svm
->vmcb
->control
.int_ctl
& V_TPR_MASK
;
4779 kvm_set_cr8(vcpu
, cr8
);
4783 static inline void sync_lapic_to_cr8(struct kvm_vcpu
*vcpu
)
4785 struct vcpu_svm
*svm
= to_svm(vcpu
);
4788 if (svm_nested_virtualize_tpr(vcpu
) ||
4789 kvm_vcpu_apicv_active(vcpu
))
4792 cr8
= kvm_get_cr8(vcpu
);
4793 svm
->vmcb
->control
.int_ctl
&= ~V_TPR_MASK
;
4794 svm
->vmcb
->control
.int_ctl
|= cr8
& V_TPR_MASK
;
4797 static void svm_complete_interrupts(struct vcpu_svm
*svm
)
4801 u32 exitintinfo
= svm
->vmcb
->control
.exit_int_info
;
4802 unsigned int3_injected
= svm
->int3_injected
;
4804 svm
->int3_injected
= 0;
4807 * If we've made progress since setting HF_IRET_MASK, we've
4808 * executed an IRET and can allow NMI injection.
4810 if ((svm
->vcpu
.arch
.hflags
& HF_IRET_MASK
)
4811 && kvm_rip_read(&svm
->vcpu
) != svm
->nmi_iret_rip
) {
4812 svm
->vcpu
.arch
.hflags
&= ~(HF_NMI_MASK
| HF_IRET_MASK
);
4813 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
4816 svm
->vcpu
.arch
.nmi_injected
= false;
4817 kvm_clear_exception_queue(&svm
->vcpu
);
4818 kvm_clear_interrupt_queue(&svm
->vcpu
);
4820 if (!(exitintinfo
& SVM_EXITINTINFO_VALID
))
4823 kvm_make_request(KVM_REQ_EVENT
, &svm
->vcpu
);
4825 vector
= exitintinfo
& SVM_EXITINTINFO_VEC_MASK
;
4826 type
= exitintinfo
& SVM_EXITINTINFO_TYPE_MASK
;
4829 case SVM_EXITINTINFO_TYPE_NMI
:
4830 svm
->vcpu
.arch
.nmi_injected
= true;
4832 case SVM_EXITINTINFO_TYPE_EXEPT
:
4834 * In case of software exceptions, do not reinject the vector,
4835 * but re-execute the instruction instead. Rewind RIP first
4836 * if we emulated INT3 before.
4838 if (kvm_exception_is_soft(vector
)) {
4839 if (vector
== BP_VECTOR
&& int3_injected
&&
4840 kvm_is_linear_rip(&svm
->vcpu
, svm
->int3_rip
))
4841 kvm_rip_write(&svm
->vcpu
,
4842 kvm_rip_read(&svm
->vcpu
) -
4846 if (exitintinfo
& SVM_EXITINTINFO_VALID_ERR
) {
4847 u32 err
= svm
->vmcb
->control
.exit_int_info_err
;
4848 kvm_requeue_exception_e(&svm
->vcpu
, vector
, err
);
4851 kvm_requeue_exception(&svm
->vcpu
, vector
);
4853 case SVM_EXITINTINFO_TYPE_INTR
:
4854 kvm_queue_interrupt(&svm
->vcpu
, vector
, false);
4861 static void svm_cancel_injection(struct kvm_vcpu
*vcpu
)
4863 struct vcpu_svm
*svm
= to_svm(vcpu
);
4864 struct vmcb_control_area
*control
= &svm
->vmcb
->control
;
4866 control
->exit_int_info
= control
->event_inj
;
4867 control
->exit_int_info_err
= control
->event_inj_err
;
4868 control
->event_inj
= 0;
4869 svm_complete_interrupts(svm
);
4872 static void svm_vcpu_run(struct kvm_vcpu
*vcpu
)
4874 struct vcpu_svm
*svm
= to_svm(vcpu
);
4876 svm
->vmcb
->save
.rax
= vcpu
->arch
.regs
[VCPU_REGS_RAX
];
4877 svm
->vmcb
->save
.rsp
= vcpu
->arch
.regs
[VCPU_REGS_RSP
];
4878 svm
->vmcb
->save
.rip
= vcpu
->arch
.regs
[VCPU_REGS_RIP
];
4881 * A vmexit emulation is required before the vcpu can be executed
4884 if (unlikely(svm
->nested
.exit_required
))
4888 * Disable singlestep if we're injecting an interrupt/exception.
4889 * We don't want our modified rflags to be pushed on the stack where
4890 * we might not be able to easily reset them if we disabled NMI
4893 if (svm
->nmi_singlestep
&& svm
->vmcb
->control
.event_inj
) {
4895 * Event injection happens before external interrupts cause a
4896 * vmexit and interrupts are disabled here, so smp_send_reschedule
4897 * is enough to force an immediate vmexit.
4899 disable_nmi_singlestep(svm
);
4900 smp_send_reschedule(vcpu
->cpu
);
4905 sync_lapic_to_cr8(vcpu
);
4907 svm
->vmcb
->save
.cr2
= vcpu
->arch
.cr2
;
4913 /* SMB: Don't care about ibrs_inuse but rely on guest value */
4914 x86_spec_ctrl_set_guest(svm
->spec_ctrl
, svm
->virt_spec_ctrl
);
4917 "push %%" _ASM_BP
"; \n\t"
4918 "mov %c[rbx](%[svm]), %%" _ASM_BX
" \n\t"
4919 "mov %c[rcx](%[svm]), %%" _ASM_CX
" \n\t"
4920 "mov %c[rdx](%[svm]), %%" _ASM_DX
" \n\t"
4921 "mov %c[rsi](%[svm]), %%" _ASM_SI
" \n\t"
4922 "mov %c[rdi](%[svm]), %%" _ASM_DI
" \n\t"
4923 "mov %c[rbp](%[svm]), %%" _ASM_BP
" \n\t"
4924 #ifdef CONFIG_X86_64
4925 "mov %c[r8](%[svm]), %%r8 \n\t"
4926 "mov %c[r9](%[svm]), %%r9 \n\t"
4927 "mov %c[r10](%[svm]), %%r10 \n\t"
4928 "mov %c[r11](%[svm]), %%r11 \n\t"
4929 "mov %c[r12](%[svm]), %%r12 \n\t"
4930 "mov %c[r13](%[svm]), %%r13 \n\t"
4931 "mov %c[r14](%[svm]), %%r14 \n\t"
4932 "mov %c[r15](%[svm]), %%r15 \n\t"
4935 /* Enter guest mode */
4936 "push %%" _ASM_AX
" \n\t"
4937 "mov %c[vmcb](%[svm]), %%" _ASM_AX
" \n\t"
4938 __ex(SVM_VMLOAD
) "\n\t"
4939 __ex(SVM_VMRUN
) "\n\t"
4940 __ex(SVM_VMSAVE
) "\n\t"
4941 "pop %%" _ASM_AX
" \n\t"
4943 /* Save guest registers, load host registers */
4944 "mov %%" _ASM_BX
", %c[rbx](%[svm]) \n\t"
4945 "mov %%" _ASM_CX
", %c[rcx](%[svm]) \n\t"
4946 "mov %%" _ASM_DX
", %c[rdx](%[svm]) \n\t"
4947 "mov %%" _ASM_SI
", %c[rsi](%[svm]) \n\t"
4948 "mov %%" _ASM_DI
", %c[rdi](%[svm]) \n\t"
4949 "mov %%" _ASM_BP
", %c[rbp](%[svm]) \n\t"
4950 #ifdef CONFIG_X86_64
4951 "mov %%r8, %c[r8](%[svm]) \n\t"
4952 "mov %%r9, %c[r9](%[svm]) \n\t"
4953 "mov %%r10, %c[r10](%[svm]) \n\t"
4954 "mov %%r11, %c[r11](%[svm]) \n\t"
4955 "mov %%r12, %c[r12](%[svm]) \n\t"
4956 "mov %%r13, %c[r13](%[svm]) \n\t"
4957 "mov %%r14, %c[r14](%[svm]) \n\t"
4958 "mov %%r15, %c[r15](%[svm]) \n\t"
4961 * Clear host registers marked as clobbered to prevent
4964 "xor %%" _ASM_BX
", %%" _ASM_BX
" \n\t"
4965 "xor %%" _ASM_CX
", %%" _ASM_CX
" \n\t"
4966 "xor %%" _ASM_DX
", %%" _ASM_DX
" \n\t"
4967 "xor %%" _ASM_SI
", %%" _ASM_SI
" \n\t"
4968 "xor %%" _ASM_DI
", %%" _ASM_DI
" \n\t"
4969 #ifdef CONFIG_X86_64
4970 "xor %%r8, %%r8 \n\t"
4971 "xor %%r9, %%r9 \n\t"
4972 "xor %%r10, %%r10 \n\t"
4973 "xor %%r11, %%r11 \n\t"
4974 "xor %%r12, %%r12 \n\t"
4975 "xor %%r13, %%r13 \n\t"
4976 "xor %%r14, %%r14 \n\t"
4977 "xor %%r15, %%r15 \n\t"
4982 [vmcb
]"i"(offsetof(struct vcpu_svm
, vmcb_pa
)),
4983 [rbx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RBX
])),
4984 [rcx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RCX
])),
4985 [rdx
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RDX
])),
4986 [rsi
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RSI
])),
4987 [rdi
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RDI
])),
4988 [rbp
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_RBP
]))
4989 #ifdef CONFIG_X86_64
4990 , [r8
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R8
])),
4991 [r9
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R9
])),
4992 [r10
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R10
])),
4993 [r11
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R11
])),
4994 [r12
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R12
])),
4995 [r13
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R13
])),
4996 [r14
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R14
])),
4997 [r15
]"i"(offsetof(struct vcpu_svm
, vcpu
.arch
.regs
[VCPU_REGS_R15
]))
5000 #ifdef CONFIG_X86_64
5001 , "rbx", "rcx", "rdx", "rsi", "rdi"
5002 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
5004 , "ebx", "ecx", "edx", "esi", "edi"
5008 /* Eliminate branch target predictions from guest mode */
5011 #ifdef CONFIG_X86_64
5012 wrmsrl(MSR_GS_BASE
, svm
->host
.gs_base
);
5014 loadsegment(fs
, svm
->host
.fs
);
5015 #ifndef CONFIG_X86_32_LAZY_GS
5016 loadsegment(gs
, svm
->host
.gs
);
5020 x86_spec_ctrl_restore_host(svm
->spec_ctrl
, svm
->virt_spec_ctrl
);
5024 local_irq_disable();
5026 vcpu
->arch
.cr2
= svm
->vmcb
->save
.cr2
;
5027 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = svm
->vmcb
->save
.rax
;
5028 vcpu
->arch
.regs
[VCPU_REGS_RSP
] = svm
->vmcb
->save
.rsp
;
5029 vcpu
->arch
.regs
[VCPU_REGS_RIP
] = svm
->vmcb
->save
.rip
;
5031 if (unlikely(svm
->vmcb
->control
.exit_code
== SVM_EXIT_NMI
))
5032 kvm_before_handle_nmi(&svm
->vcpu
);
5036 /* Any pending NMI will happen here */
5038 if (unlikely(svm
->vmcb
->control
.exit_code
== SVM_EXIT_NMI
))
5039 kvm_after_handle_nmi(&svm
->vcpu
);
5041 sync_cr8_to_lapic(vcpu
);
5045 svm
->vmcb
->control
.tlb_ctl
= TLB_CONTROL_DO_NOTHING
;
5047 /* if exit due to PF check for async PF */
5048 if (svm
->vmcb
->control
.exit_code
== SVM_EXIT_EXCP_BASE
+ PF_VECTOR
)
5049 svm
->vcpu
.arch
.apf
.host_apf_reason
= kvm_read_and_reset_pf_reason();
5052 vcpu
->arch
.regs_avail
&= ~(1 << VCPU_EXREG_PDPTR
);
5053 vcpu
->arch
.regs_dirty
&= ~(1 << VCPU_EXREG_PDPTR
);
5057 * We need to handle MC intercepts here before the vcpu has a chance to
5058 * change the physical cpu
5060 if (unlikely(svm
->vmcb
->control
.exit_code
==
5061 SVM_EXIT_EXCP_BASE
+ MC_VECTOR
))
5062 svm_handle_mce(svm
);
5064 mark_all_clean(svm
->vmcb
);
5066 STACK_FRAME_NON_STANDARD(svm_vcpu_run
);
5068 static void svm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
5070 struct vcpu_svm
*svm
= to_svm(vcpu
);
5072 svm
->vmcb
->save
.cr3
= root
;
5073 mark_dirty(svm
->vmcb
, VMCB_CR
);
5074 svm_flush_tlb(vcpu
);
5077 static void set_tdp_cr3(struct kvm_vcpu
*vcpu
, unsigned long root
)
5079 struct vcpu_svm
*svm
= to_svm(vcpu
);
5081 svm
->vmcb
->control
.nested_cr3
= root
;
5082 mark_dirty(svm
->vmcb
, VMCB_NPT
);
5084 /* Also sync guest cr3 here in case we live migrate */
5085 svm
->vmcb
->save
.cr3
= kvm_read_cr3(vcpu
);
5086 mark_dirty(svm
->vmcb
, VMCB_CR
);
5088 svm_flush_tlb(vcpu
);
5091 static int is_disabled(void)
5095 rdmsrl(MSR_VM_CR
, vm_cr
);
5096 if (vm_cr
& (1 << SVM_VM_CR_SVM_DISABLE
))
5103 svm_patch_hypercall(struct kvm_vcpu
*vcpu
, unsigned char *hypercall
)
5106 * Patch in the VMMCALL instruction:
5108 hypercall
[0] = 0x0f;
5109 hypercall
[1] = 0x01;
5110 hypercall
[2] = 0xd9;
5113 static void svm_check_processor_compat(void *rtn
)
5118 static bool svm_cpu_has_accelerated_tpr(void)
5123 static bool svm_has_high_real_mode_segbase(void)
5128 static u64
svm_get_mt_mask(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool is_mmio
)
5133 static void svm_cpuid_update(struct kvm_vcpu
*vcpu
)
5135 struct vcpu_svm
*svm
= to_svm(vcpu
);
5136 struct kvm_cpuid_entry2
*entry
;
5138 /* Update nrips enabled cache */
5139 svm
->nrips_enabled
= !!guest_cpuid_has_nrips(&svm
->vcpu
);
5141 if (!kvm_vcpu_apicv_active(vcpu
))
5144 entry
= kvm_find_cpuid_entry(vcpu
, 1, 0);
5146 entry
->ecx
&= ~bit(X86_FEATURE_X2APIC
);
5149 static void svm_set_supported_cpuid(u32 func
, struct kvm_cpuid_entry2
*entry
)
5154 entry
->ecx
&= ~bit(X86_FEATURE_X2APIC
);
5158 entry
->ecx
|= (1 << 2); /* Set SVM bit */
5161 entry
->eax
= 1; /* SVM revision 1 */
5162 entry
->ebx
= 8; /* Lets support 8 ASIDs in case we add proper
5163 ASID emulation to nested SVM */
5164 entry
->ecx
= 0; /* Reserved */
5165 entry
->edx
= 0; /* Per default do not support any
5166 additional features */
5168 /* Support next_rip if host supports it */
5169 if (boot_cpu_has(X86_FEATURE_NRIPS
))
5170 entry
->edx
|= SVM_FEATURE_NRIP
;
5172 /* Support NPT for the guest if enabled */
5174 entry
->edx
|= SVM_FEATURE_NPT
;
5180 static int svm_get_lpage_level(void)
5182 return PT_PDPE_LEVEL
;
5185 static bool svm_rdtscp_supported(void)
5187 return boot_cpu_has(X86_FEATURE_RDTSCP
);
5190 static bool svm_invpcid_supported(void)
5195 static bool svm_mpx_supported(void)
5200 static bool svm_xsaves_supported(void)
5205 static bool svm_has_wbinvd_exit(void)
5210 #define PRE_EX(exit) { .exit_code = (exit), \
5211 .stage = X86_ICPT_PRE_EXCEPT, }
5212 #define POST_EX(exit) { .exit_code = (exit), \
5213 .stage = X86_ICPT_POST_EXCEPT, }
5214 #define POST_MEM(exit) { .exit_code = (exit), \
5215 .stage = X86_ICPT_POST_MEMACCESS, }
5217 static const struct __x86_intercept
{
5219 enum x86_intercept_stage stage
;
5220 } x86_intercept_map
[] = {
5221 [x86_intercept_cr_read
] = POST_EX(SVM_EXIT_READ_CR0
),
5222 [x86_intercept_cr_write
] = POST_EX(SVM_EXIT_WRITE_CR0
),
5223 [x86_intercept_clts
] = POST_EX(SVM_EXIT_WRITE_CR0
),
5224 [x86_intercept_lmsw
] = POST_EX(SVM_EXIT_WRITE_CR0
),
5225 [x86_intercept_smsw
] = POST_EX(SVM_EXIT_READ_CR0
),
5226 [x86_intercept_dr_read
] = POST_EX(SVM_EXIT_READ_DR0
),
5227 [x86_intercept_dr_write
] = POST_EX(SVM_EXIT_WRITE_DR0
),
5228 [x86_intercept_sldt
] = POST_EX(SVM_EXIT_LDTR_READ
),
5229 [x86_intercept_str
] = POST_EX(SVM_EXIT_TR_READ
),
5230 [x86_intercept_lldt
] = POST_EX(SVM_EXIT_LDTR_WRITE
),
5231 [x86_intercept_ltr
] = POST_EX(SVM_EXIT_TR_WRITE
),
5232 [x86_intercept_sgdt
] = POST_EX(SVM_EXIT_GDTR_READ
),
5233 [x86_intercept_sidt
] = POST_EX(SVM_EXIT_IDTR_READ
),
5234 [x86_intercept_lgdt
] = POST_EX(SVM_EXIT_GDTR_WRITE
),
5235 [x86_intercept_lidt
] = POST_EX(SVM_EXIT_IDTR_WRITE
),
5236 [x86_intercept_vmrun
] = POST_EX(SVM_EXIT_VMRUN
),
5237 [x86_intercept_vmmcall
] = POST_EX(SVM_EXIT_VMMCALL
),
5238 [x86_intercept_vmload
] = POST_EX(SVM_EXIT_VMLOAD
),
5239 [x86_intercept_vmsave
] = POST_EX(SVM_EXIT_VMSAVE
),
5240 [x86_intercept_stgi
] = POST_EX(SVM_EXIT_STGI
),
5241 [x86_intercept_clgi
] = POST_EX(SVM_EXIT_CLGI
),
5242 [x86_intercept_skinit
] = POST_EX(SVM_EXIT_SKINIT
),
5243 [x86_intercept_invlpga
] = POST_EX(SVM_EXIT_INVLPGA
),
5244 [x86_intercept_rdtscp
] = POST_EX(SVM_EXIT_RDTSCP
),
5245 [x86_intercept_monitor
] = POST_MEM(SVM_EXIT_MONITOR
),
5246 [x86_intercept_mwait
] = POST_EX(SVM_EXIT_MWAIT
),
5247 [x86_intercept_invlpg
] = POST_EX(SVM_EXIT_INVLPG
),
5248 [x86_intercept_invd
] = POST_EX(SVM_EXIT_INVD
),
5249 [x86_intercept_wbinvd
] = POST_EX(SVM_EXIT_WBINVD
),
5250 [x86_intercept_wrmsr
] = POST_EX(SVM_EXIT_MSR
),
5251 [x86_intercept_rdtsc
] = POST_EX(SVM_EXIT_RDTSC
),
5252 [x86_intercept_rdmsr
] = POST_EX(SVM_EXIT_MSR
),
5253 [x86_intercept_rdpmc
] = POST_EX(SVM_EXIT_RDPMC
),
5254 [x86_intercept_cpuid
] = PRE_EX(SVM_EXIT_CPUID
),
5255 [x86_intercept_rsm
] = PRE_EX(SVM_EXIT_RSM
),
5256 [x86_intercept_pause
] = PRE_EX(SVM_EXIT_PAUSE
),
5257 [x86_intercept_pushf
] = PRE_EX(SVM_EXIT_PUSHF
),
5258 [x86_intercept_popf
] = PRE_EX(SVM_EXIT_POPF
),
5259 [x86_intercept_intn
] = PRE_EX(SVM_EXIT_SWINT
),
5260 [x86_intercept_iret
] = PRE_EX(SVM_EXIT_IRET
),
5261 [x86_intercept_icebp
] = PRE_EX(SVM_EXIT_ICEBP
),
5262 [x86_intercept_hlt
] = POST_EX(SVM_EXIT_HLT
),
5263 [x86_intercept_in
] = POST_EX(SVM_EXIT_IOIO
),
5264 [x86_intercept_ins
] = POST_EX(SVM_EXIT_IOIO
),
5265 [x86_intercept_out
] = POST_EX(SVM_EXIT_IOIO
),
5266 [x86_intercept_outs
] = POST_EX(SVM_EXIT_IOIO
),
5273 static int svm_check_intercept(struct kvm_vcpu
*vcpu
,
5274 struct x86_instruction_info
*info
,
5275 enum x86_intercept_stage stage
)
5277 struct vcpu_svm
*svm
= to_svm(vcpu
);
5278 int vmexit
, ret
= X86EMUL_CONTINUE
;
5279 struct __x86_intercept icpt_info
;
5280 struct vmcb
*vmcb
= svm
->vmcb
;
5282 if (info
->intercept
>= ARRAY_SIZE(x86_intercept_map
))
5285 icpt_info
= x86_intercept_map
[info
->intercept
];
5287 if (stage
!= icpt_info
.stage
)
5290 switch (icpt_info
.exit_code
) {
5291 case SVM_EXIT_READ_CR0
:
5292 if (info
->intercept
== x86_intercept_cr_read
)
5293 icpt_info
.exit_code
+= info
->modrm_reg
;
5295 case SVM_EXIT_WRITE_CR0
: {
5296 unsigned long cr0
, val
;
5299 if (info
->intercept
== x86_intercept_cr_write
)
5300 icpt_info
.exit_code
+= info
->modrm_reg
;
5302 if (icpt_info
.exit_code
!= SVM_EXIT_WRITE_CR0
||
5303 info
->intercept
== x86_intercept_clts
)
5306 intercept
= svm
->nested
.intercept
;
5308 if (!(intercept
& (1ULL << INTERCEPT_SELECTIVE_CR0
)))
5311 cr0
= vcpu
->arch
.cr0
& ~SVM_CR0_SELECTIVE_MASK
;
5312 val
= info
->src_val
& ~SVM_CR0_SELECTIVE_MASK
;
5314 if (info
->intercept
== x86_intercept_lmsw
) {
5317 /* lmsw can't clear PE - catch this here */
5318 if (cr0
& X86_CR0_PE
)
5323 icpt_info
.exit_code
= SVM_EXIT_CR0_SEL_WRITE
;
5327 case SVM_EXIT_READ_DR0
:
5328 case SVM_EXIT_WRITE_DR0
:
5329 icpt_info
.exit_code
+= info
->modrm_reg
;
5332 if (info
->intercept
== x86_intercept_wrmsr
)
5333 vmcb
->control
.exit_info_1
= 1;
5335 vmcb
->control
.exit_info_1
= 0;
5337 case SVM_EXIT_PAUSE
:
5339 * We get this for NOP only, but pause
5340 * is rep not, check this here
5342 if (info
->rep_prefix
!= REPE_PREFIX
)
5344 case SVM_EXIT_IOIO
: {
5348 if (info
->intercept
== x86_intercept_in
||
5349 info
->intercept
== x86_intercept_ins
) {
5350 exit_info
= ((info
->src_val
& 0xffff) << 16) |
5352 bytes
= info
->dst_bytes
;
5354 exit_info
= (info
->dst_val
& 0xffff) << 16;
5355 bytes
= info
->src_bytes
;
5358 if (info
->intercept
== x86_intercept_outs
||
5359 info
->intercept
== x86_intercept_ins
)
5360 exit_info
|= SVM_IOIO_STR_MASK
;
5362 if (info
->rep_prefix
)
5363 exit_info
|= SVM_IOIO_REP_MASK
;
5365 bytes
= min(bytes
, 4u);
5367 exit_info
|= bytes
<< SVM_IOIO_SIZE_SHIFT
;
5369 exit_info
|= (u32
)info
->ad_bytes
<< (SVM_IOIO_ASIZE_SHIFT
- 1);
5371 vmcb
->control
.exit_info_1
= exit_info
;
5372 vmcb
->control
.exit_info_2
= info
->next_rip
;
5380 /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
5381 if (static_cpu_has(X86_FEATURE_NRIPS
))
5382 vmcb
->control
.next_rip
= info
->next_rip
;
5383 vmcb
->control
.exit_code
= icpt_info
.exit_code
;
5384 vmexit
= nested_svm_exit_handled(svm
);
5386 ret
= (vmexit
== NESTED_EXIT_DONE
) ? X86EMUL_INTERCEPTED
5393 static void svm_handle_external_intr(struct kvm_vcpu
*vcpu
)
5397 * We must have an instruction with interrupts enabled, so
5398 * the timer interrupt isn't delayed by the interrupt shadow.
5401 local_irq_disable();
5404 static void svm_sched_in(struct kvm_vcpu
*vcpu
, int cpu
)
5408 static inline void avic_post_state_restore(struct kvm_vcpu
*vcpu
)
5410 if (avic_handle_apic_id_update(vcpu
) != 0)
5412 if (avic_handle_dfr_update(vcpu
) != 0)
5414 avic_handle_ldr_update(vcpu
);
5417 static void svm_setup_mce(struct kvm_vcpu
*vcpu
)
5419 /* [63:9] are reserved. */
5420 vcpu
->arch
.mcg_cap
&= 0x1ff;
5423 static struct kvm_x86_ops svm_x86_ops __ro_after_init
= {
5424 .cpu_has_kvm_support
= has_svm
,
5425 .disabled_by_bios
= is_disabled
,
5426 .hardware_setup
= svm_hardware_setup
,
5427 .hardware_unsetup
= svm_hardware_unsetup
,
5428 .check_processor_compatibility
= svm_check_processor_compat
,
5429 .hardware_enable
= svm_hardware_enable
,
5430 .hardware_disable
= svm_hardware_disable
,
5431 .cpu_has_accelerated_tpr
= svm_cpu_has_accelerated_tpr
,
5432 .cpu_has_high_real_mode_segbase
= svm_has_high_real_mode_segbase
,
5434 .vcpu_create
= svm_create_vcpu
,
5435 .vcpu_free
= svm_free_vcpu
,
5436 .vcpu_reset
= svm_vcpu_reset
,
5438 .vm_init
= avic_vm_init
,
5439 .vm_destroy
= avic_vm_destroy
,
5441 .prepare_guest_switch
= svm_prepare_guest_switch
,
5442 .vcpu_load
= svm_vcpu_load
,
5443 .vcpu_put
= svm_vcpu_put
,
5444 .vcpu_blocking
= svm_vcpu_blocking
,
5445 .vcpu_unblocking
= svm_vcpu_unblocking
,
5447 .update_bp_intercept
= update_bp_intercept
,
5448 .get_msr
= svm_get_msr
,
5449 .set_msr
= svm_set_msr
,
5450 .get_segment_base
= svm_get_segment_base
,
5451 .get_segment
= svm_get_segment
,
5452 .set_segment
= svm_set_segment
,
5453 .get_cpl
= svm_get_cpl
,
5454 .get_cs_db_l_bits
= kvm_get_cs_db_l_bits
,
5455 .decache_cr0_guest_bits
= svm_decache_cr0_guest_bits
,
5456 .decache_cr3
= svm_decache_cr3
,
5457 .decache_cr4_guest_bits
= svm_decache_cr4_guest_bits
,
5458 .set_cr0
= svm_set_cr0
,
5459 .set_cr3
= svm_set_cr3
,
5460 .set_cr4
= svm_set_cr4
,
5461 .set_efer
= svm_set_efer
,
5462 .get_idt
= svm_get_idt
,
5463 .set_idt
= svm_set_idt
,
5464 .get_gdt
= svm_get_gdt
,
5465 .set_gdt
= svm_set_gdt
,
5466 .get_dr6
= svm_get_dr6
,
5467 .set_dr6
= svm_set_dr6
,
5468 .set_dr7
= svm_set_dr7
,
5469 .sync_dirty_debug_regs
= svm_sync_dirty_debug_regs
,
5470 .cache_reg
= svm_cache_reg
,
5471 .get_rflags
= svm_get_rflags
,
5472 .set_rflags
= svm_set_rflags
,
5474 .tlb_flush
= svm_flush_tlb
,
5476 .run
= svm_vcpu_run
,
5477 .handle_exit
= handle_exit
,
5478 .skip_emulated_instruction
= skip_emulated_instruction
,
5479 .set_interrupt_shadow
= svm_set_interrupt_shadow
,
5480 .get_interrupt_shadow
= svm_get_interrupt_shadow
,
5481 .patch_hypercall
= svm_patch_hypercall
,
5482 .set_irq
= svm_set_irq
,
5483 .set_nmi
= svm_inject_nmi
,
5484 .queue_exception
= svm_queue_exception
,
5485 .cancel_injection
= svm_cancel_injection
,
5486 .interrupt_allowed
= svm_interrupt_allowed
,
5487 .nmi_allowed
= svm_nmi_allowed
,
5488 .get_nmi_mask
= svm_get_nmi_mask
,
5489 .set_nmi_mask
= svm_set_nmi_mask
,
5490 .enable_nmi_window
= enable_nmi_window
,
5491 .enable_irq_window
= enable_irq_window
,
5492 .update_cr8_intercept
= update_cr8_intercept
,
5493 .set_virtual_x2apic_mode
= svm_set_virtual_x2apic_mode
,
5494 .get_enable_apicv
= svm_get_enable_apicv
,
5495 .refresh_apicv_exec_ctrl
= svm_refresh_apicv_exec_ctrl
,
5496 .load_eoi_exitmap
= svm_load_eoi_exitmap
,
5497 .hwapic_irr_update
= svm_hwapic_irr_update
,
5498 .hwapic_isr_update
= svm_hwapic_isr_update
,
5499 .apicv_post_state_restore
= avic_post_state_restore
,
5501 .set_tss_addr
= svm_set_tss_addr
,
5502 .get_tdp_level
= get_npt_level
,
5503 .get_mt_mask
= svm_get_mt_mask
,
5505 .get_exit_info
= svm_get_exit_info
,
5507 .get_lpage_level
= svm_get_lpage_level
,
5509 .cpuid_update
= svm_cpuid_update
,
5511 .rdtscp_supported
= svm_rdtscp_supported
,
5512 .invpcid_supported
= svm_invpcid_supported
,
5513 .mpx_supported
= svm_mpx_supported
,
5514 .xsaves_supported
= svm_xsaves_supported
,
5516 .set_supported_cpuid
= svm_set_supported_cpuid
,
5518 .has_wbinvd_exit
= svm_has_wbinvd_exit
,
5520 .write_tsc_offset
= svm_write_tsc_offset
,
5522 .set_tdp_cr3
= set_tdp_cr3
,
5524 .check_intercept
= svm_check_intercept
,
5525 .handle_external_intr
= svm_handle_external_intr
,
5527 .sched_in
= svm_sched_in
,
5529 .pmu_ops
= &amd_pmu_ops
,
5530 .deliver_posted_interrupt
= svm_deliver_avic_intr
,
5531 .update_pi_irte
= svm_update_pi_irte
,
5532 .setup_mce
= svm_setup_mce
,
5535 static int __init
svm_init(void)
5537 return kvm_init(&svm_x86_ops
, sizeof(struct vcpu_svm
),
5538 __alignof__(struct vcpu_svm
), THIS_MODULE
);
5541 static void __exit
svm_exit(void)
5546 module_init(svm_init
)
5547 module_exit(svm_exit
)