3 * Local APIC virtualization
5 * Copyright (C) 2006 Qumranet, Inc.
6 * Copyright (C) 2007 Novell
7 * Copyright (C) 2007 Intel
8 * Copyright 2009 Red Hat, Inc. and/or its affiliates.
11 * Dor Laor <dor.laor@qumranet.com>
12 * Gregory Haskins <ghaskins@novell.com>
13 * Yaozu (Eddie) Dong <eddie.dong@intel.com>
15 * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
24 #include <linux/highmem.h>
25 #include <linux/smp.h>
26 #include <linux/hrtimer.h>
28 #include <linux/module.h>
29 #include <linux/math64.h>
30 #include <linux/slab.h>
31 #include <asm/processor.h>
34 #include <asm/current.h>
35 #include <asm/apicdef.h>
36 #include <asm/delay.h>
37 #include <linux/atomic.h>
38 #include <linux/jump_label.h>
39 #include "kvm_cache_regs.h"
47 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
49 #define mod_64(x, y) ((x) % (y))
57 #define APIC_BUS_CYCLE_NS 1
59 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
60 #define apic_debug(fmt, arg...)
62 /* 14 is the version for Xeon and Pentium 8.4.8*/
63 #define APIC_VERSION (0x14UL | ((KVM_APIC_LVT_NUM - 1) << 16))
64 #define LAPIC_MMIO_LENGTH (1 << 12)
65 /* followed define is not in apicdef.h */
66 #define APIC_SHORT_MASK 0xc0000
67 #define APIC_DEST_NOSHORT 0x0
68 #define APIC_DEST_MASK 0x800
69 #define MAX_APIC_VECTOR 256
70 #define APIC_VECTORS_PER_REG 32
72 #define APIC_BROADCAST 0xFF
73 #define X2APIC_BROADCAST 0xFFFFFFFFul
75 static inline int apic_test_vector(int vec
, void *bitmap
)
77 return test_bit(VEC_POS(vec
), (bitmap
) + REG_POS(vec
));
80 bool kvm_apic_pending_eoi(struct kvm_vcpu
*vcpu
, int vector
)
82 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
84 return apic_test_vector(vector
, apic
->regs
+ APIC_ISR
) ||
85 apic_test_vector(vector
, apic
->regs
+ APIC_IRR
);
88 static inline void apic_clear_vector(int vec
, void *bitmap
)
90 clear_bit(VEC_POS(vec
), (bitmap
) + REG_POS(vec
));
93 static inline int __apic_test_and_set_vector(int vec
, void *bitmap
)
95 return __test_and_set_bit(VEC_POS(vec
), (bitmap
) + REG_POS(vec
));
98 static inline int __apic_test_and_clear_vector(int vec
, void *bitmap
)
100 return __test_and_clear_bit(VEC_POS(vec
), (bitmap
) + REG_POS(vec
));
103 struct static_key_deferred apic_hw_disabled __read_mostly
;
104 struct static_key_deferred apic_sw_disabled __read_mostly
;
106 static inline int apic_enabled(struct kvm_lapic
*apic
)
108 return kvm_apic_sw_enabled(apic
) && kvm_apic_hw_enabled(apic
);
112 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
115 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
116 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
118 /* The logical map is definitely wrong if we have multiple
119 * modes at the same time. (Physical map is always right.)
121 static inline bool kvm_apic_logical_map_valid(struct kvm_apic_map
*map
)
123 return !(map
->mode
& (map
->mode
- 1));
127 apic_logical_id(struct kvm_apic_map
*map
, u32 dest_id
, u16
*cid
, u16
*lid
)
131 BUILD_BUG_ON(KVM_APIC_MODE_XAPIC_CLUSTER
!= 4);
132 BUILD_BUG_ON(KVM_APIC_MODE_XAPIC_FLAT
!= 8);
133 BUILD_BUG_ON(KVM_APIC_MODE_X2APIC
!= 16);
134 lid_bits
= map
->mode
;
136 *cid
= dest_id
>> lid_bits
;
137 *lid
= dest_id
& ((1 << lid_bits
) - 1);
140 static void recalculate_apic_map(struct kvm
*kvm
)
142 struct kvm_apic_map
*new, *old
= NULL
;
143 struct kvm_vcpu
*vcpu
;
146 new = kzalloc(sizeof(struct kvm_apic_map
), GFP_KERNEL
);
148 mutex_lock(&kvm
->arch
.apic_map_lock
);
153 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
154 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
158 if (!kvm_apic_present(vcpu
))
161 aid
= kvm_apic_id(apic
);
162 ldr
= kvm_lapic_get_reg(apic
, APIC_LDR
);
164 if (aid
< ARRAY_SIZE(new->phys_map
))
165 new->phys_map
[aid
] = apic
;
167 if (apic_x2apic_mode(apic
)) {
168 new->mode
|= KVM_APIC_MODE_X2APIC
;
170 ldr
= GET_APIC_LOGICAL_ID(ldr
);
171 if (kvm_lapic_get_reg(apic
, APIC_DFR
) == APIC_DFR_FLAT
)
172 new->mode
|= KVM_APIC_MODE_XAPIC_FLAT
;
174 new->mode
|= KVM_APIC_MODE_XAPIC_CLUSTER
;
177 if (!kvm_apic_logical_map_valid(new))
180 apic_logical_id(new, ldr
, &cid
, &lid
);
182 if (lid
&& cid
< ARRAY_SIZE(new->logical_map
))
183 new->logical_map
[cid
][ffs(lid
) - 1] = apic
;
186 old
= rcu_dereference_protected(kvm
->arch
.apic_map
,
187 lockdep_is_held(&kvm
->arch
.apic_map_lock
));
188 rcu_assign_pointer(kvm
->arch
.apic_map
, new);
189 mutex_unlock(&kvm
->arch
.apic_map_lock
);
194 kvm_make_scan_ioapic_request(kvm
);
197 static inline void apic_set_spiv(struct kvm_lapic
*apic
, u32 val
)
199 bool enabled
= val
& APIC_SPIV_APIC_ENABLED
;
201 kvm_lapic_set_reg(apic
, APIC_SPIV
, val
);
203 if (enabled
!= apic
->sw_enabled
) {
204 apic
->sw_enabled
= enabled
;
206 static_key_slow_dec_deferred(&apic_sw_disabled
);
207 recalculate_apic_map(apic
->vcpu
->kvm
);
209 static_key_slow_inc(&apic_sw_disabled
.key
);
213 static inline void kvm_apic_set_id(struct kvm_lapic
*apic
, u8 id
)
215 kvm_lapic_set_reg(apic
, APIC_ID
, id
<< 24);
216 recalculate_apic_map(apic
->vcpu
->kvm
);
219 static inline void kvm_apic_set_ldr(struct kvm_lapic
*apic
, u32 id
)
221 kvm_lapic_set_reg(apic
, APIC_LDR
, id
);
222 recalculate_apic_map(apic
->vcpu
->kvm
);
225 static inline void kvm_apic_set_x2apic_id(struct kvm_lapic
*apic
, u8 id
)
227 u32 ldr
= ((id
>> 4) << 16) | (1 << (id
& 0xf));
229 kvm_lapic_set_reg(apic
, APIC_ID
, id
<< 24);
230 kvm_lapic_set_reg(apic
, APIC_LDR
, ldr
);
231 recalculate_apic_map(apic
->vcpu
->kvm
);
234 static inline int apic_lvt_enabled(struct kvm_lapic
*apic
, int lvt_type
)
236 return !(kvm_lapic_get_reg(apic
, lvt_type
) & APIC_LVT_MASKED
);
239 static inline int apic_lvt_vector(struct kvm_lapic
*apic
, int lvt_type
)
241 return kvm_lapic_get_reg(apic
, lvt_type
) & APIC_VECTOR_MASK
;
244 static inline int apic_lvtt_oneshot(struct kvm_lapic
*apic
)
246 return apic
->lapic_timer
.timer_mode
== APIC_LVT_TIMER_ONESHOT
;
249 static inline int apic_lvtt_period(struct kvm_lapic
*apic
)
251 return apic
->lapic_timer
.timer_mode
== APIC_LVT_TIMER_PERIODIC
;
254 static inline int apic_lvtt_tscdeadline(struct kvm_lapic
*apic
)
256 return apic
->lapic_timer
.timer_mode
== APIC_LVT_TIMER_TSCDEADLINE
;
259 static inline int apic_lvt_nmi_mode(u32 lvt_val
)
261 return (lvt_val
& (APIC_MODE_MASK
| APIC_LVT_MASKED
)) == APIC_DM_NMI
;
264 void kvm_apic_set_version(struct kvm_vcpu
*vcpu
)
266 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
267 struct kvm_cpuid_entry2
*feat
;
268 u32 v
= APIC_VERSION
;
270 if (!lapic_in_kernel(vcpu
))
273 feat
= kvm_find_cpuid_entry(apic
->vcpu
, 0x1, 0);
274 if (feat
&& (feat
->ecx
& (1 << (X86_FEATURE_X2APIC
& 31))))
275 v
|= APIC_LVR_DIRECTED_EOI
;
276 kvm_lapic_set_reg(apic
, APIC_LVR
, v
);
279 static const unsigned int apic_lvt_mask
[KVM_APIC_LVT_NUM
] = {
280 LVT_MASK
, /* part LVTT mask, timer mode mask added at runtime */
281 LVT_MASK
| APIC_MODE_MASK
, /* LVTTHMR */
282 LVT_MASK
| APIC_MODE_MASK
, /* LVTPC */
283 LINT_MASK
, LINT_MASK
, /* LVT0-1 */
284 LVT_MASK
/* LVTERR */
287 static int find_highest_vector(void *bitmap
)
292 for (vec
= MAX_APIC_VECTOR
- APIC_VECTORS_PER_REG
;
293 vec
>= 0; vec
-= APIC_VECTORS_PER_REG
) {
294 reg
= bitmap
+ REG_POS(vec
);
296 return fls(*reg
) - 1 + vec
;
302 static u8
count_vectors(void *bitmap
)
308 for (vec
= 0; vec
< MAX_APIC_VECTOR
; vec
+= APIC_VECTORS_PER_REG
) {
309 reg
= bitmap
+ REG_POS(vec
);
310 count
+= hweight32(*reg
);
316 void __kvm_apic_update_irr(u32
*pir
, void *regs
)
320 for (i
= 0; i
<= 7; i
++) {
321 pir_val
= xchg(&pir
[i
], 0);
323 *((u32
*)(regs
+ APIC_IRR
+ i
* 0x10)) |= pir_val
;
326 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr
);
328 void kvm_apic_update_irr(struct kvm_vcpu
*vcpu
, u32
*pir
)
330 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
332 __kvm_apic_update_irr(pir
, apic
->regs
);
334 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
336 EXPORT_SYMBOL_GPL(kvm_apic_update_irr
);
338 static inline int apic_search_irr(struct kvm_lapic
*apic
)
340 return find_highest_vector(apic
->regs
+ APIC_IRR
);
343 static inline int apic_find_highest_irr(struct kvm_lapic
*apic
)
348 * Note that irr_pending is just a hint. It will be always
349 * true with virtual interrupt delivery enabled.
351 if (!apic
->irr_pending
)
354 if (apic
->vcpu
->arch
.apicv_active
)
355 kvm_x86_ops
->sync_pir_to_irr(apic
->vcpu
);
356 result
= apic_search_irr(apic
);
357 ASSERT(result
== -1 || result
>= 16);
362 static inline void apic_clear_irr(int vec
, struct kvm_lapic
*apic
)
364 struct kvm_vcpu
*vcpu
;
368 if (unlikely(vcpu
->arch
.apicv_active
)) {
369 /* try to update RVI */
370 apic_clear_vector(vec
, apic
->regs
+ APIC_IRR
);
371 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
373 apic
->irr_pending
= false;
374 apic_clear_vector(vec
, apic
->regs
+ APIC_IRR
);
375 if (apic_search_irr(apic
) != -1)
376 apic
->irr_pending
= true;
380 static inline void apic_set_isr(int vec
, struct kvm_lapic
*apic
)
382 struct kvm_vcpu
*vcpu
;
384 if (__apic_test_and_set_vector(vec
, apic
->regs
+ APIC_ISR
))
390 * With APIC virtualization enabled, all caching is disabled
391 * because the processor can modify ISR under the hood. Instead
394 if (unlikely(vcpu
->arch
.apicv_active
))
395 kvm_x86_ops
->hwapic_isr_update(vcpu
, vec
);
398 BUG_ON(apic
->isr_count
> MAX_APIC_VECTOR
);
400 * ISR (in service register) bit is set when injecting an interrupt.
401 * The highest vector is injected. Thus the latest bit set matches
402 * the highest bit in ISR.
404 apic
->highest_isr_cache
= vec
;
408 static inline int apic_find_highest_isr(struct kvm_lapic
*apic
)
413 * Note that isr_count is always 1, and highest_isr_cache
414 * is always -1, with APIC virtualization enabled.
416 if (!apic
->isr_count
)
418 if (likely(apic
->highest_isr_cache
!= -1))
419 return apic
->highest_isr_cache
;
421 result
= find_highest_vector(apic
->regs
+ APIC_ISR
);
422 ASSERT(result
== -1 || result
>= 16);
427 static inline void apic_clear_isr(int vec
, struct kvm_lapic
*apic
)
429 struct kvm_vcpu
*vcpu
;
430 if (!__apic_test_and_clear_vector(vec
, apic
->regs
+ APIC_ISR
))
436 * We do get here for APIC virtualization enabled if the guest
437 * uses the Hyper-V APIC enlightenment. In this case we may need
438 * to trigger a new interrupt delivery by writing the SVI field;
439 * on the other hand isr_count and highest_isr_cache are unused
440 * and must be left alone.
442 if (unlikely(vcpu
->arch
.apicv_active
))
443 kvm_x86_ops
->hwapic_isr_update(vcpu
,
444 apic_find_highest_isr(apic
));
447 BUG_ON(apic
->isr_count
< 0);
448 apic
->highest_isr_cache
= -1;
452 int kvm_lapic_find_highest_irr(struct kvm_vcpu
*vcpu
)
454 /* This may race with setting of irr in __apic_accept_irq() and
455 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
456 * will cause vmexit immediately and the value will be recalculated
457 * on the next vmentry.
459 return apic_find_highest_irr(vcpu
->arch
.apic
);
462 static int __apic_accept_irq(struct kvm_lapic
*apic
, int delivery_mode
,
463 int vector
, int level
, int trig_mode
,
464 struct dest_map
*dest_map
);
466 int kvm_apic_set_irq(struct kvm_vcpu
*vcpu
, struct kvm_lapic_irq
*irq
,
467 struct dest_map
*dest_map
)
469 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
471 return __apic_accept_irq(apic
, irq
->delivery_mode
, irq
->vector
,
472 irq
->level
, irq
->trig_mode
, dest_map
);
475 static int pv_eoi_put_user(struct kvm_vcpu
*vcpu
, u8 val
)
478 return kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.pv_eoi
.data
, &val
,
482 static int pv_eoi_get_user(struct kvm_vcpu
*vcpu
, u8
*val
)
485 return kvm_read_guest_cached(vcpu
->kvm
, &vcpu
->arch
.pv_eoi
.data
, val
,
489 static inline bool pv_eoi_enabled(struct kvm_vcpu
*vcpu
)
491 return vcpu
->arch
.pv_eoi
.msr_val
& KVM_MSR_ENABLED
;
494 static bool pv_eoi_get_pending(struct kvm_vcpu
*vcpu
)
497 if (pv_eoi_get_user(vcpu
, &val
) < 0)
498 apic_debug("Can't read EOI MSR value: 0x%llx\n",
499 (unsigned long long)vcpu
->arch
.pv_eoi
.msr_val
);
503 static void pv_eoi_set_pending(struct kvm_vcpu
*vcpu
)
505 if (pv_eoi_put_user(vcpu
, KVM_PV_EOI_ENABLED
) < 0) {
506 apic_debug("Can't set EOI MSR value: 0x%llx\n",
507 (unsigned long long)vcpu
->arch
.pv_eoi
.msr_val
);
510 __set_bit(KVM_APIC_PV_EOI_PENDING
, &vcpu
->arch
.apic_attention
);
513 static void pv_eoi_clr_pending(struct kvm_vcpu
*vcpu
)
515 if (pv_eoi_put_user(vcpu
, KVM_PV_EOI_DISABLED
) < 0) {
516 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
517 (unsigned long long)vcpu
->arch
.pv_eoi
.msr_val
);
520 __clear_bit(KVM_APIC_PV_EOI_PENDING
, &vcpu
->arch
.apic_attention
);
523 static void apic_update_ppr(struct kvm_lapic
*apic
)
525 u32 tpr
, isrv
, ppr
, old_ppr
;
528 old_ppr
= kvm_lapic_get_reg(apic
, APIC_PROCPRI
);
529 tpr
= kvm_lapic_get_reg(apic
, APIC_TASKPRI
);
530 isr
= apic_find_highest_isr(apic
);
531 isrv
= (isr
!= -1) ? isr
: 0;
533 if ((tpr
& 0xf0) >= (isrv
& 0xf0))
538 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
539 apic
, ppr
, isr
, isrv
);
541 if (old_ppr
!= ppr
) {
542 kvm_lapic_set_reg(apic
, APIC_PROCPRI
, ppr
);
544 kvm_make_request(KVM_REQ_EVENT
, apic
->vcpu
);
548 static void apic_set_tpr(struct kvm_lapic
*apic
, u32 tpr
)
550 kvm_lapic_set_reg(apic
, APIC_TASKPRI
, tpr
);
551 apic_update_ppr(apic
);
554 static bool kvm_apic_broadcast(struct kvm_lapic
*apic
, u32 mda
)
556 if (apic_x2apic_mode(apic
))
557 return mda
== X2APIC_BROADCAST
;
559 return GET_APIC_DEST_FIELD(mda
) == APIC_BROADCAST
;
562 static bool kvm_apic_match_physical_addr(struct kvm_lapic
*apic
, u32 mda
)
564 if (kvm_apic_broadcast(apic
, mda
))
567 if (apic_x2apic_mode(apic
))
568 return mda
== kvm_apic_id(apic
);
570 return mda
== SET_APIC_DEST_FIELD(kvm_apic_id(apic
));
573 static bool kvm_apic_match_logical_addr(struct kvm_lapic
*apic
, u32 mda
)
577 if (kvm_apic_broadcast(apic
, mda
))
580 logical_id
= kvm_lapic_get_reg(apic
, APIC_LDR
);
582 if (apic_x2apic_mode(apic
))
583 return ((logical_id
>> 16) == (mda
>> 16))
584 && (logical_id
& mda
& 0xffff) != 0;
586 logical_id
= GET_APIC_LOGICAL_ID(logical_id
);
587 mda
= GET_APIC_DEST_FIELD(mda
);
589 switch (kvm_lapic_get_reg(apic
, APIC_DFR
)) {
591 return (logical_id
& mda
) != 0;
592 case APIC_DFR_CLUSTER
:
593 return ((logical_id
>> 4) == (mda
>> 4))
594 && (logical_id
& mda
& 0xf) != 0;
596 apic_debug("Bad DFR vcpu %d: %08x\n",
597 apic
->vcpu
->vcpu_id
, kvm_lapic_get_reg(apic
, APIC_DFR
));
602 /* KVM APIC implementation has two quirks
603 * - dest always begins at 0 while xAPIC MDA has offset 24,
604 * - IOxAPIC messages have to be delivered (directly) to x2APIC.
606 static u32
kvm_apic_mda(unsigned int dest_id
, struct kvm_lapic
*source
,
607 struct kvm_lapic
*target
)
609 bool ipi
= source
!= NULL
;
610 bool x2apic_mda
= apic_x2apic_mode(ipi
? source
: target
);
612 if (!ipi
&& dest_id
== APIC_BROADCAST
&& x2apic_mda
)
613 return X2APIC_BROADCAST
;
615 return x2apic_mda
? dest_id
: SET_APIC_DEST_FIELD(dest_id
);
618 bool kvm_apic_match_dest(struct kvm_vcpu
*vcpu
, struct kvm_lapic
*source
,
619 int short_hand
, unsigned int dest
, int dest_mode
)
621 struct kvm_lapic
*target
= vcpu
->arch
.apic
;
622 u32 mda
= kvm_apic_mda(dest
, source
, target
);
624 apic_debug("target %p, source %p, dest 0x%x, "
625 "dest_mode 0x%x, short_hand 0x%x\n",
626 target
, source
, dest
, dest_mode
, short_hand
);
629 switch (short_hand
) {
630 case APIC_DEST_NOSHORT
:
631 if (dest_mode
== APIC_DEST_PHYSICAL
)
632 return kvm_apic_match_physical_addr(target
, mda
);
634 return kvm_apic_match_logical_addr(target
, mda
);
636 return target
== source
;
637 case APIC_DEST_ALLINC
:
639 case APIC_DEST_ALLBUT
:
640 return target
!= source
;
642 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
647 EXPORT_SYMBOL_GPL(kvm_apic_match_dest
);
649 int kvm_vector_to_index(u32 vector
, u32 dest_vcpus
,
650 const unsigned long *bitmap
, u32 bitmap_size
)
655 mod
= vector
% dest_vcpus
;
657 for (i
= 0; i
<= mod
; i
++) {
658 idx
= find_next_bit(bitmap
, bitmap_size
, idx
+ 1);
659 BUG_ON(idx
== bitmap_size
);
665 static void kvm_apic_disabled_lapic_found(struct kvm
*kvm
)
667 if (!kvm
->arch
.disabled_lapic_found
) {
668 kvm
->arch
.disabled_lapic_found
= true;
670 "Disabled LAPIC found during irq injection\n");
674 bool kvm_irq_delivery_to_apic_fast(struct kvm
*kvm
, struct kvm_lapic
*src
,
675 struct kvm_lapic_irq
*irq
, int *r
, struct dest_map
*dest_map
)
677 struct kvm_apic_map
*map
;
678 unsigned long bitmap
= 1;
679 struct kvm_lapic
**dst
;
681 bool ret
, x2apic_ipi
;
685 if (irq
->shorthand
== APIC_DEST_SELF
) {
686 *r
= kvm_apic_set_irq(src
->vcpu
, irq
, dest_map
);
693 x2apic_ipi
= src
&& apic_x2apic_mode(src
);
694 if (irq
->dest_id
== (x2apic_ipi
? X2APIC_BROADCAST
: APIC_BROADCAST
))
699 map
= rcu_dereference(kvm
->arch
.apic_map
);
706 if (irq
->dest_mode
== APIC_DEST_PHYSICAL
) {
707 if (irq
->dest_id
>= ARRAY_SIZE(map
->phys_map
))
710 dst
= &map
->phys_map
[irq
->dest_id
];
714 if (!kvm_apic_logical_map_valid(map
)) {
719 apic_logical_id(map
, irq
->dest_id
, &cid
, (u16
*)&bitmap
);
721 if (cid
>= ARRAY_SIZE(map
->logical_map
))
724 dst
= map
->logical_map
[cid
];
726 if (!kvm_lowest_prio_delivery(irq
))
729 if (!kvm_vector_hashing_enabled()) {
731 for_each_set_bit(i
, &bitmap
, 16) {
736 else if (kvm_apic_compare_prio(dst
[i
]->vcpu
,
740 bitmap
= (l
>= 0) ? 1 << l
: 0;
743 unsigned int dest_vcpus
;
745 dest_vcpus
= hweight16(bitmap
);
749 idx
= kvm_vector_to_index(irq
->vector
,
750 dest_vcpus
, &bitmap
, 16);
753 kvm_apic_disabled_lapic_found(kvm
);
757 bitmap
= (idx
>= 0) ? 1 << idx
: 0;
762 for_each_set_bit(i
, &bitmap
, 16) {
767 *r
+= kvm_apic_set_irq(dst
[i
]->vcpu
, irq
, dest_map
);
775 * This routine tries to handler interrupts in posted mode, here is how
776 * it deals with different cases:
777 * - For single-destination interrupts, handle it in posted mode
778 * - Else if vector hashing is enabled and it is a lowest-priority
779 * interrupt, handle it in posted mode and use the following mechanism
780 * to find the destinaiton vCPU.
781 * 1. For lowest-priority interrupts, store all the possible
782 * destination vCPUs in an array.
783 * 2. Use "guest vector % max number of destination vCPUs" to find
784 * the right destination vCPU in the array for the lowest-priority
786 * - Otherwise, use remapped mode to inject the interrupt.
788 bool kvm_intr_is_single_vcpu_fast(struct kvm
*kvm
, struct kvm_lapic_irq
*irq
,
789 struct kvm_vcpu
**dest_vcpu
)
791 struct kvm_apic_map
*map
;
793 struct kvm_lapic
*dst
= NULL
;
799 map
= rcu_dereference(kvm
->arch
.apic_map
);
804 if (irq
->dest_mode
== APIC_DEST_PHYSICAL
) {
805 if (irq
->dest_id
== 0xFF)
808 if (irq
->dest_id
>= ARRAY_SIZE(map
->phys_map
))
811 dst
= map
->phys_map
[irq
->dest_id
];
812 if (dst
&& kvm_apic_present(dst
->vcpu
))
813 *dest_vcpu
= dst
->vcpu
;
818 unsigned long bitmap
= 1;
821 if (!kvm_apic_logical_map_valid(map
))
824 apic_logical_id(map
, irq
->dest_id
, &cid
, (u16
*)&bitmap
);
826 if (cid
>= ARRAY_SIZE(map
->logical_map
))
829 if (kvm_vector_hashing_enabled() &&
830 kvm_lowest_prio_delivery(irq
)) {
832 unsigned int dest_vcpus
;
834 dest_vcpus
= hweight16(bitmap
);
838 idx
= kvm_vector_to_index(irq
->vector
, dest_vcpus
,
841 dst
= map
->logical_map
[cid
][idx
];
843 kvm_apic_disabled_lapic_found(kvm
);
847 *dest_vcpu
= dst
->vcpu
;
849 for_each_set_bit(i
, &bitmap
, 16) {
850 dst
= map
->logical_map
[cid
][i
];
855 if (dst
&& kvm_apic_present(dst
->vcpu
))
856 *dest_vcpu
= dst
->vcpu
;
869 * Add a pending IRQ into lapic.
870 * Return 1 if successfully added and 0 if discarded.
872 static int __apic_accept_irq(struct kvm_lapic
*apic
, int delivery_mode
,
873 int vector
, int level
, int trig_mode
,
874 struct dest_map
*dest_map
)
877 struct kvm_vcpu
*vcpu
= apic
->vcpu
;
879 trace_kvm_apic_accept_irq(vcpu
->vcpu_id
, delivery_mode
,
881 switch (delivery_mode
) {
883 vcpu
->arch
.apic_arb_prio
++;
885 if (unlikely(trig_mode
&& !level
))
888 /* FIXME add logic for vcpu on reset */
889 if (unlikely(!apic_enabled(apic
)))
895 __set_bit(vcpu
->vcpu_id
, dest_map
->map
);
896 dest_map
->vectors
[vcpu
->vcpu_id
] = vector
;
899 if (apic_test_vector(vector
, apic
->regs
+ APIC_TMR
) != !!trig_mode
) {
901 kvm_lapic_set_vector(vector
, apic
->regs
+ APIC_TMR
);
903 apic_clear_vector(vector
, apic
->regs
+ APIC_TMR
);
906 if (vcpu
->arch
.apicv_active
)
907 kvm_x86_ops
->deliver_posted_interrupt(vcpu
, vector
);
909 kvm_lapic_set_irr(vector
, apic
);
911 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
918 vcpu
->arch
.pv
.pv_unhalted
= 1;
919 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
925 kvm_make_request(KVM_REQ_SMI
, vcpu
);
931 kvm_inject_nmi(vcpu
);
936 if (!trig_mode
|| level
) {
938 /* assumes that there are only KVM_APIC_INIT/SIPI */
939 apic
->pending_events
= (1UL << KVM_APIC_INIT
);
940 /* make sure pending_events is visible before sending
943 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
946 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
951 case APIC_DM_STARTUP
:
952 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
953 vcpu
->vcpu_id
, vector
);
955 apic
->sipi_vector
= vector
;
956 /* make sure sipi_vector is visible for the receiver */
958 set_bit(KVM_APIC_SIPI
, &apic
->pending_events
);
959 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
965 * Should only be called by kvm_apic_local_deliver() with LVT0,
966 * before NMI watchdog was enabled. Already handled by
967 * kvm_apic_accept_pic_intr().
972 printk(KERN_ERR
"TODO: unsupported delivery mode %x\n",
979 int kvm_apic_compare_prio(struct kvm_vcpu
*vcpu1
, struct kvm_vcpu
*vcpu2
)
981 return vcpu1
->arch
.apic_arb_prio
- vcpu2
->arch
.apic_arb_prio
;
984 static bool kvm_ioapic_handles_vector(struct kvm_lapic
*apic
, int vector
)
986 return test_bit(vector
, apic
->vcpu
->arch
.ioapic_handled_vectors
);
989 static void kvm_ioapic_send_eoi(struct kvm_lapic
*apic
, int vector
)
993 /* Eoi the ioapic only if the ioapic doesn't own the vector. */
994 if (!kvm_ioapic_handles_vector(apic
, vector
))
997 /* Request a KVM exit to inform the userspace IOAPIC. */
998 if (irqchip_split(apic
->vcpu
->kvm
)) {
999 apic
->vcpu
->arch
.pending_ioapic_eoi
= vector
;
1000 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT
, apic
->vcpu
);
1004 if (apic_test_vector(vector
, apic
->regs
+ APIC_TMR
))
1005 trigger_mode
= IOAPIC_LEVEL_TRIG
;
1007 trigger_mode
= IOAPIC_EDGE_TRIG
;
1009 kvm_ioapic_update_eoi(apic
->vcpu
, vector
, trigger_mode
);
1012 static int apic_set_eoi(struct kvm_lapic
*apic
)
1014 int vector
= apic_find_highest_isr(apic
);
1016 trace_kvm_eoi(apic
, vector
);
1019 * Not every write EOI will has corresponding ISR,
1020 * one example is when Kernel check timer on setup_IO_APIC
1025 apic_clear_isr(vector
, apic
);
1026 apic_update_ppr(apic
);
1028 if (test_bit(vector
, vcpu_to_synic(apic
->vcpu
)->vec_bitmap
))
1029 kvm_hv_synic_send_eoi(apic
->vcpu
, vector
);
1031 kvm_ioapic_send_eoi(apic
, vector
);
1032 kvm_make_request(KVM_REQ_EVENT
, apic
->vcpu
);
1037 * this interface assumes a trap-like exit, which has already finished
1038 * desired side effect including vISR and vPPR update.
1040 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu
*vcpu
, int vector
)
1042 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1044 trace_kvm_eoi(apic
, vector
);
1046 kvm_ioapic_send_eoi(apic
, vector
);
1047 kvm_make_request(KVM_REQ_EVENT
, apic
->vcpu
);
1049 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated
);
1051 static void apic_send_ipi(struct kvm_lapic
*apic
)
1053 u32 icr_low
= kvm_lapic_get_reg(apic
, APIC_ICR
);
1054 u32 icr_high
= kvm_lapic_get_reg(apic
, APIC_ICR2
);
1055 struct kvm_lapic_irq irq
;
1057 irq
.vector
= icr_low
& APIC_VECTOR_MASK
;
1058 irq
.delivery_mode
= icr_low
& APIC_MODE_MASK
;
1059 irq
.dest_mode
= icr_low
& APIC_DEST_MASK
;
1060 irq
.level
= (icr_low
& APIC_INT_ASSERT
) != 0;
1061 irq
.trig_mode
= icr_low
& APIC_INT_LEVELTRIG
;
1062 irq
.shorthand
= icr_low
& APIC_SHORT_MASK
;
1063 irq
.msi_redir_hint
= false;
1064 if (apic_x2apic_mode(apic
))
1065 irq
.dest_id
= icr_high
;
1067 irq
.dest_id
= GET_APIC_DEST_FIELD(icr_high
);
1069 trace_kvm_apic_ipi(icr_low
, irq
.dest_id
);
1071 apic_debug("icr_high 0x%x, icr_low 0x%x, "
1072 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
1073 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x, "
1074 "msi_redir_hint 0x%x\n",
1075 icr_high
, icr_low
, irq
.shorthand
, irq
.dest_id
,
1076 irq
.trig_mode
, irq
.level
, irq
.dest_mode
, irq
.delivery_mode
,
1077 irq
.vector
, irq
.msi_redir_hint
);
1079 kvm_irq_delivery_to_apic(apic
->vcpu
->kvm
, apic
, &irq
, NULL
);
1082 static u32
apic_get_tmcct(struct kvm_lapic
*apic
)
1088 ASSERT(apic
!= NULL
);
1090 /* if initial count is 0, current count should also be 0 */
1091 if (kvm_lapic_get_reg(apic
, APIC_TMICT
) == 0 ||
1092 apic
->lapic_timer
.period
== 0)
1095 remaining
= hrtimer_get_remaining(&apic
->lapic_timer
.timer
);
1096 if (ktime_to_ns(remaining
) < 0)
1097 remaining
= ktime_set(0, 0);
1099 ns
= mod_64(ktime_to_ns(remaining
), apic
->lapic_timer
.period
);
1100 tmcct
= div64_u64(ns
,
1101 (APIC_BUS_CYCLE_NS
* apic
->divide_count
));
1106 static void __report_tpr_access(struct kvm_lapic
*apic
, bool write
)
1108 struct kvm_vcpu
*vcpu
= apic
->vcpu
;
1109 struct kvm_run
*run
= vcpu
->run
;
1111 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS
, vcpu
);
1112 run
->tpr_access
.rip
= kvm_rip_read(vcpu
);
1113 run
->tpr_access
.is_write
= write
;
1116 static inline void report_tpr_access(struct kvm_lapic
*apic
, bool write
)
1118 if (apic
->vcpu
->arch
.tpr_access_reporting
)
1119 __report_tpr_access(apic
, write
);
1122 static u32
__apic_read(struct kvm_lapic
*apic
, unsigned int offset
)
1126 if (offset
>= LAPIC_MMIO_LENGTH
)
1131 if (apic_x2apic_mode(apic
))
1132 val
= kvm_apic_id(apic
);
1134 val
= kvm_apic_id(apic
) << 24;
1137 apic_debug("Access APIC ARBPRI register which is for P6\n");
1140 case APIC_TMCCT
: /* Timer CCR */
1141 if (apic_lvtt_tscdeadline(apic
))
1144 val
= apic_get_tmcct(apic
);
1147 apic_update_ppr(apic
);
1148 val
= kvm_lapic_get_reg(apic
, offset
);
1151 report_tpr_access(apic
, false);
1154 val
= kvm_lapic_get_reg(apic
, offset
);
1161 static inline struct kvm_lapic
*to_lapic(struct kvm_io_device
*dev
)
1163 return container_of(dev
, struct kvm_lapic
, dev
);
1166 int kvm_lapic_reg_read(struct kvm_lapic
*apic
, u32 offset
, int len
,
1169 unsigned char alignment
= offset
& 0xf;
1171 /* this bitmask has a bit cleared for each reserved register */
1172 static const u64 rmask
= 0x43ff01ffffffe70cULL
;
1174 if ((alignment
+ len
) > 4) {
1175 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1180 if (offset
> 0x3f0 || !(rmask
& (1ULL << (offset
>> 4)))) {
1181 apic_debug("KVM_APIC_READ: read reserved register %x\n",
1186 result
= __apic_read(apic
, offset
& ~0xf);
1188 trace_kvm_apic_read(offset
, result
);
1194 memcpy(data
, (char *)&result
+ alignment
, len
);
1197 printk(KERN_ERR
"Local APIC read with len = %x, "
1198 "should be 1,2, or 4 instead\n", len
);
1203 EXPORT_SYMBOL_GPL(kvm_lapic_reg_read
);
1205 static int apic_mmio_in_range(struct kvm_lapic
*apic
, gpa_t addr
)
1207 return kvm_apic_hw_enabled(apic
) &&
1208 addr
>= apic
->base_address
&&
1209 addr
< apic
->base_address
+ LAPIC_MMIO_LENGTH
;
1212 static int apic_mmio_read(struct kvm_vcpu
*vcpu
, struct kvm_io_device
*this,
1213 gpa_t address
, int len
, void *data
)
1215 struct kvm_lapic
*apic
= to_lapic(this);
1216 u32 offset
= address
- apic
->base_address
;
1218 if (!apic_mmio_in_range(apic
, address
))
1221 kvm_lapic_reg_read(apic
, offset
, len
, data
);
1226 static void update_divide_count(struct kvm_lapic
*apic
)
1228 u32 tmp1
, tmp2
, tdcr
;
1230 tdcr
= kvm_lapic_get_reg(apic
, APIC_TDCR
);
1232 tmp2
= ((tmp1
& 0x3) | ((tmp1
& 0x8) >> 1)) + 1;
1233 apic
->divide_count
= 0x1 << (tmp2
& 0x7);
1235 apic_debug("timer divide count is 0x%x\n",
1236 apic
->divide_count
);
1239 static void apic_update_lvtt(struct kvm_lapic
*apic
)
1241 u32 timer_mode
= kvm_lapic_get_reg(apic
, APIC_LVTT
) &
1242 apic
->lapic_timer
.timer_mode_mask
;
1244 if (apic
->lapic_timer
.timer_mode
!= timer_mode
) {
1245 apic
->lapic_timer
.timer_mode
= timer_mode
;
1246 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1250 static void apic_timer_expired(struct kvm_lapic
*apic
)
1252 struct kvm_vcpu
*vcpu
= apic
->vcpu
;
1253 struct swait_queue_head
*q
= &vcpu
->wq
;
1254 struct kvm_timer
*ktimer
= &apic
->lapic_timer
;
1256 if (atomic_read(&apic
->lapic_timer
.pending
))
1259 atomic_inc(&apic
->lapic_timer
.pending
);
1260 kvm_set_pending_timer(vcpu
);
1262 if (swait_active(q
))
1265 if (apic_lvtt_tscdeadline(apic
))
1266 ktimer
->expired_tscdeadline
= ktimer
->tscdeadline
;
1270 * On APICv, this test will cause a busy wait
1271 * during a higher-priority task.
1274 static bool lapic_timer_int_injected(struct kvm_vcpu
*vcpu
)
1276 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1277 u32 reg
= kvm_lapic_get_reg(apic
, APIC_LVTT
);
1279 if (kvm_apic_hw_enabled(apic
)) {
1280 int vec
= reg
& APIC_VECTOR_MASK
;
1281 void *bitmap
= apic
->regs
+ APIC_ISR
;
1283 if (vcpu
->arch
.apicv_active
)
1284 bitmap
= apic
->regs
+ APIC_IRR
;
1286 if (apic_test_vector(vec
, bitmap
))
1292 void wait_lapic_expire(struct kvm_vcpu
*vcpu
)
1294 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1295 u64 guest_tsc
, tsc_deadline
;
1297 if (!lapic_in_kernel(vcpu
))
1300 if (apic
->lapic_timer
.expired_tscdeadline
== 0)
1303 if (!lapic_timer_int_injected(vcpu
))
1306 tsc_deadline
= apic
->lapic_timer
.expired_tscdeadline
;
1307 apic
->lapic_timer
.expired_tscdeadline
= 0;
1308 guest_tsc
= kvm_read_l1_tsc(vcpu
, rdtsc());
1309 trace_kvm_wait_lapic_expire(vcpu
->vcpu_id
, guest_tsc
- tsc_deadline
);
1311 /* __delay is delay_tsc whenever the hardware has TSC, thus always. */
1312 if (guest_tsc
< tsc_deadline
)
1313 __delay(tsc_deadline
- guest_tsc
);
1316 static void start_sw_tscdeadline(struct kvm_lapic
*apic
)
1318 u64 guest_tsc
, tscdeadline
= apic
->lapic_timer
.tscdeadline
;
1321 struct kvm_vcpu
*vcpu
= apic
->vcpu
;
1322 unsigned long this_tsc_khz
= vcpu
->arch
.virtual_tsc_khz
;
1323 unsigned long flags
;
1326 if (unlikely(!tscdeadline
|| !this_tsc_khz
))
1329 local_irq_save(flags
);
1331 now
= apic
->lapic_timer
.timer
.base
->get_time();
1332 guest_tsc
= kvm_read_l1_tsc(vcpu
, rdtsc());
1333 if (likely(tscdeadline
> guest_tsc
)) {
1334 ns
= (tscdeadline
- guest_tsc
) * 1000000ULL;
1335 do_div(ns
, this_tsc_khz
);
1336 expire
= ktime_add_ns(now
, ns
);
1337 expire
= ktime_sub_ns(expire
, lapic_timer_advance_ns
);
1338 hrtimer_start(&apic
->lapic_timer
.timer
,
1339 expire
, HRTIMER_MODE_ABS_PINNED
);
1341 apic_timer_expired(apic
);
1343 local_irq_restore(flags
);
1346 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu
*vcpu
)
1348 return vcpu
->arch
.apic
->lapic_timer
.hv_timer_in_use
;
1350 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use
);
1352 void kvm_lapic_expired_hv_timer(struct kvm_vcpu
*vcpu
)
1354 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1356 WARN_ON(!apic
->lapic_timer
.hv_timer_in_use
);
1357 WARN_ON(swait_active(&vcpu
->wq
));
1358 kvm_x86_ops
->cancel_hv_timer(vcpu
);
1359 apic
->lapic_timer
.hv_timer_in_use
= false;
1360 apic_timer_expired(apic
);
1362 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer
);
1364 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu
*vcpu
)
1366 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1368 WARN_ON(apic
->lapic_timer
.hv_timer_in_use
);
1370 if (apic_lvtt_tscdeadline(apic
) &&
1371 !atomic_read(&apic
->lapic_timer
.pending
)) {
1372 u64 tscdeadline
= apic
->lapic_timer
.tscdeadline
;
1374 if (!kvm_x86_ops
->set_hv_timer(vcpu
, tscdeadline
)) {
1375 apic
->lapic_timer
.hv_timer_in_use
= true;
1376 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1378 /* In case the sw timer triggered in the window */
1379 if (atomic_read(&apic
->lapic_timer
.pending
)) {
1380 apic
->lapic_timer
.hv_timer_in_use
= false;
1381 kvm_x86_ops
->cancel_hv_timer(apic
->vcpu
);
1384 trace_kvm_hv_timer_state(vcpu
->vcpu_id
,
1385 apic
->lapic_timer
.hv_timer_in_use
);
1388 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer
);
1390 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu
*vcpu
)
1392 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1394 /* Possibly the TSC deadline timer is not enabled yet */
1395 if (!apic
->lapic_timer
.hv_timer_in_use
)
1398 kvm_x86_ops
->cancel_hv_timer(vcpu
);
1399 apic
->lapic_timer
.hv_timer_in_use
= false;
1401 if (atomic_read(&apic
->lapic_timer
.pending
))
1404 start_sw_tscdeadline(apic
);
1406 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer
);
1408 static void start_apic_timer(struct kvm_lapic
*apic
)
1412 atomic_set(&apic
->lapic_timer
.pending
, 0);
1414 if (apic_lvtt_period(apic
) || apic_lvtt_oneshot(apic
)) {
1415 /* lapic timer in oneshot or periodic mode */
1416 now
= apic
->lapic_timer
.timer
.base
->get_time();
1417 apic
->lapic_timer
.period
= (u64
)kvm_lapic_get_reg(apic
, APIC_TMICT
)
1418 * APIC_BUS_CYCLE_NS
* apic
->divide_count
;
1420 if (!apic
->lapic_timer
.period
)
1423 * Do not allow the guest to program periodic timers with small
1424 * interval, since the hrtimers are not throttled by the host
1427 if (apic_lvtt_period(apic
)) {
1428 s64 min_period
= min_timer_period_us
* 1000LL;
1430 if (apic
->lapic_timer
.period
< min_period
) {
1431 pr_info_ratelimited(
1432 "kvm: vcpu %i: requested %lld ns "
1433 "lapic timer period limited to %lld ns\n",
1434 apic
->vcpu
->vcpu_id
,
1435 apic
->lapic_timer
.period
, min_period
);
1436 apic
->lapic_timer
.period
= min_period
;
1440 hrtimer_start(&apic
->lapic_timer
.timer
,
1441 ktime_add_ns(now
, apic
->lapic_timer
.period
),
1442 HRTIMER_MODE_ABS_PINNED
);
1444 apic_debug("%s: bus cycle is %" PRId64
"ns, now 0x%016"
1446 "timer initial count 0x%x, period %lldns, "
1447 "expire @ 0x%016" PRIx64
".\n", __func__
,
1448 APIC_BUS_CYCLE_NS
, ktime_to_ns(now
),
1449 kvm_lapic_get_reg(apic
, APIC_TMICT
),
1450 apic
->lapic_timer
.period
,
1451 ktime_to_ns(ktime_add_ns(now
,
1452 apic
->lapic_timer
.period
)));
1453 } else if (apic_lvtt_tscdeadline(apic
)) {
1454 /* lapic timer in tsc deadline mode */
1455 u64 tscdeadline
= apic
->lapic_timer
.tscdeadline
;
1457 if (kvm_x86_ops
->set_hv_timer
&&
1458 !kvm_x86_ops
->set_hv_timer(apic
->vcpu
, tscdeadline
)) {
1459 apic
->lapic_timer
.hv_timer_in_use
= true;
1460 trace_kvm_hv_timer_state(apic
->vcpu
->vcpu_id
,
1461 apic
->lapic_timer
.hv_timer_in_use
);
1463 start_sw_tscdeadline(apic
);
1467 static void apic_manage_nmi_watchdog(struct kvm_lapic
*apic
, u32 lvt0_val
)
1469 bool lvt0_in_nmi_mode
= apic_lvt_nmi_mode(lvt0_val
);
1471 if (apic
->lvt0_in_nmi_mode
!= lvt0_in_nmi_mode
) {
1472 apic
->lvt0_in_nmi_mode
= lvt0_in_nmi_mode
;
1473 if (lvt0_in_nmi_mode
) {
1474 apic_debug("Receive NMI setting on APIC_LVT0 "
1475 "for cpu %d\n", apic
->vcpu
->vcpu_id
);
1476 atomic_inc(&apic
->vcpu
->kvm
->arch
.vapics_in_nmi_mode
);
1478 atomic_dec(&apic
->vcpu
->kvm
->arch
.vapics_in_nmi_mode
);
1482 int kvm_lapic_reg_write(struct kvm_lapic
*apic
, u32 reg
, u32 val
)
1486 trace_kvm_apic_write(reg
, val
);
1489 case APIC_ID
: /* Local APIC ID */
1490 if (!apic_x2apic_mode(apic
))
1491 kvm_apic_set_id(apic
, val
>> 24);
1497 report_tpr_access(apic
, true);
1498 apic_set_tpr(apic
, val
& 0xff);
1506 if (!apic_x2apic_mode(apic
))
1507 kvm_apic_set_ldr(apic
, val
& APIC_LDR_MASK
);
1513 if (!apic_x2apic_mode(apic
)) {
1514 kvm_lapic_set_reg(apic
, APIC_DFR
, val
| 0x0FFFFFFF);
1515 recalculate_apic_map(apic
->vcpu
->kvm
);
1522 if (kvm_lapic_get_reg(apic
, APIC_LVR
) & APIC_LVR_DIRECTED_EOI
)
1523 mask
|= APIC_SPIV_DIRECTED_EOI
;
1524 apic_set_spiv(apic
, val
& mask
);
1525 if (!(val
& APIC_SPIV_APIC_ENABLED
)) {
1529 for (i
= 0; i
< KVM_APIC_LVT_NUM
; i
++) {
1530 lvt_val
= kvm_lapic_get_reg(apic
,
1531 APIC_LVTT
+ 0x10 * i
);
1532 kvm_lapic_set_reg(apic
, APIC_LVTT
+ 0x10 * i
,
1533 lvt_val
| APIC_LVT_MASKED
);
1535 apic_update_lvtt(apic
);
1536 atomic_set(&apic
->lapic_timer
.pending
, 0);
1542 /* No delay here, so we always clear the pending bit */
1543 kvm_lapic_set_reg(apic
, APIC_ICR
, val
& ~(1 << 12));
1544 apic_send_ipi(apic
);
1548 if (!apic_x2apic_mode(apic
))
1550 kvm_lapic_set_reg(apic
, APIC_ICR2
, val
);
1554 apic_manage_nmi_watchdog(apic
, val
);
1559 /* TODO: Check vector */
1560 if (!kvm_apic_sw_enabled(apic
))
1561 val
|= APIC_LVT_MASKED
;
1563 val
&= apic_lvt_mask
[(reg
- APIC_LVTT
) >> 4];
1564 kvm_lapic_set_reg(apic
, reg
, val
);
1569 if (!kvm_apic_sw_enabled(apic
))
1570 val
|= APIC_LVT_MASKED
;
1571 val
&= (apic_lvt_mask
[0] | apic
->lapic_timer
.timer_mode_mask
);
1572 kvm_lapic_set_reg(apic
, APIC_LVTT
, val
);
1573 apic_update_lvtt(apic
);
1577 if (apic_lvtt_tscdeadline(apic
))
1580 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1581 kvm_lapic_set_reg(apic
, APIC_TMICT
, val
);
1582 start_apic_timer(apic
);
1587 apic_debug("KVM_WRITE:TDCR %x\n", val
);
1588 kvm_lapic_set_reg(apic
, APIC_TDCR
, val
);
1589 update_divide_count(apic
);
1593 if (apic_x2apic_mode(apic
) && val
!= 0) {
1594 apic_debug("KVM_WRITE:ESR not zero %x\n", val
);
1600 if (apic_x2apic_mode(apic
)) {
1601 kvm_lapic_reg_write(apic
, APIC_ICR
, 0x40000 | (val
& 0xff));
1610 apic_debug("Local APIC Write to read-only register %x\n", reg
);
1613 EXPORT_SYMBOL_GPL(kvm_lapic_reg_write
);
1615 static int apic_mmio_write(struct kvm_vcpu
*vcpu
, struct kvm_io_device
*this,
1616 gpa_t address
, int len
, const void *data
)
1618 struct kvm_lapic
*apic
= to_lapic(this);
1619 unsigned int offset
= address
- apic
->base_address
;
1622 if (!apic_mmio_in_range(apic
, address
))
1626 * APIC register must be aligned on 128-bits boundary.
1627 * 32/64/128 bits registers must be accessed thru 32 bits.
1630 if (len
!= 4 || (offset
& 0xf)) {
1631 /* Don't shout loud, $infamous_os would cause only noise. */
1632 apic_debug("apic write: bad size=%d %lx\n", len
, (long)address
);
1638 /* too common printing */
1639 if (offset
!= APIC_EOI
)
1640 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1641 "0x%x\n", __func__
, offset
, len
, val
);
1643 kvm_lapic_reg_write(apic
, offset
& 0xff0, val
);
1648 void kvm_lapic_set_eoi(struct kvm_vcpu
*vcpu
)
1650 kvm_lapic_reg_write(vcpu
->arch
.apic
, APIC_EOI
, 0);
1652 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi
);
1654 /* emulate APIC access in a trap manner */
1655 void kvm_apic_write_nodecode(struct kvm_vcpu
*vcpu
, u32 offset
)
1659 /* hw has done the conditional check and inst decode */
1662 kvm_lapic_reg_read(vcpu
->arch
.apic
, offset
, 4, &val
);
1664 /* TODO: optimize to just emulate side effect w/o one more write */
1665 kvm_lapic_reg_write(vcpu
->arch
.apic
, offset
, val
);
1667 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode
);
1669 void kvm_free_lapic(struct kvm_vcpu
*vcpu
)
1671 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1673 if (!vcpu
->arch
.apic
)
1676 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1678 if (!(vcpu
->arch
.apic_base
& MSR_IA32_APICBASE_ENABLE
))
1679 static_key_slow_dec_deferred(&apic_hw_disabled
);
1681 if (!apic
->sw_enabled
)
1682 static_key_slow_dec_deferred(&apic_sw_disabled
);
1685 free_page((unsigned long)apic
->regs
);
1691 *----------------------------------------------------------------------
1693 *----------------------------------------------------------------------
1696 u64
kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu
*vcpu
)
1698 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1700 if (!lapic_in_kernel(vcpu
) || apic_lvtt_oneshot(apic
) ||
1701 apic_lvtt_period(apic
))
1704 return apic
->lapic_timer
.tscdeadline
;
1707 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu
*vcpu
, u64 data
)
1709 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1711 if (!lapic_in_kernel(vcpu
) || apic_lvtt_oneshot(apic
) ||
1712 apic_lvtt_period(apic
))
1715 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1716 apic
->lapic_timer
.tscdeadline
= data
;
1717 start_apic_timer(apic
);
1720 void kvm_lapic_set_tpr(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
1722 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1724 apic_set_tpr(apic
, ((cr8
& 0x0f) << 4)
1725 | (kvm_lapic_get_reg(apic
, APIC_TASKPRI
) & 4));
1728 u64
kvm_lapic_get_cr8(struct kvm_vcpu
*vcpu
)
1732 tpr
= (u64
) kvm_lapic_get_reg(vcpu
->arch
.apic
, APIC_TASKPRI
);
1734 return (tpr
& 0xf0) >> 4;
1737 void kvm_lapic_set_base(struct kvm_vcpu
*vcpu
, u64 value
)
1739 u64 old_value
= vcpu
->arch
.apic_base
;
1740 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1743 value
|= MSR_IA32_APICBASE_BSP
;
1744 vcpu
->arch
.apic_base
= value
;
1748 vcpu
->arch
.apic_base
= value
;
1750 /* update jump label if enable bit changes */
1751 if ((old_value
^ value
) & MSR_IA32_APICBASE_ENABLE
) {
1752 if (value
& MSR_IA32_APICBASE_ENABLE
)
1753 static_key_slow_dec_deferred(&apic_hw_disabled
);
1755 static_key_slow_inc(&apic_hw_disabled
.key
);
1756 recalculate_apic_map(vcpu
->kvm
);
1759 if ((old_value
^ value
) & X2APIC_ENABLE
) {
1760 if (value
& X2APIC_ENABLE
) {
1761 kvm_apic_set_x2apic_id(apic
, vcpu
->vcpu_id
);
1762 kvm_x86_ops
->set_virtual_x2apic_mode(vcpu
, true);
1764 kvm_x86_ops
->set_virtual_x2apic_mode(vcpu
, false);
1767 apic
->base_address
= apic
->vcpu
->arch
.apic_base
&
1768 MSR_IA32_APICBASE_BASE
;
1770 if ((value
& MSR_IA32_APICBASE_ENABLE
) &&
1771 apic
->base_address
!= APIC_DEFAULT_PHYS_BASE
)
1772 pr_warn_once("APIC base relocation is unsupported by KVM");
1774 /* with FSB delivery interrupt, we can restart APIC functionality */
1775 apic_debug("apic base msr is 0x%016" PRIx64
", and base address is "
1776 "0x%lx.\n", apic
->vcpu
->arch
.apic_base
, apic
->base_address
);
1780 void kvm_lapic_reset(struct kvm_vcpu
*vcpu
, bool init_event
)
1782 struct kvm_lapic
*apic
;
1785 apic_debug("%s\n", __func__
);
1788 apic
= vcpu
->arch
.apic
;
1789 ASSERT(apic
!= NULL
);
1791 /* Stop the timer in case it's a reset to an active apic */
1792 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1795 kvm_apic_set_id(apic
, vcpu
->vcpu_id
);
1796 kvm_apic_set_version(apic
->vcpu
);
1798 for (i
= 0; i
< KVM_APIC_LVT_NUM
; i
++)
1799 kvm_lapic_set_reg(apic
, APIC_LVTT
+ 0x10 * i
, APIC_LVT_MASKED
);
1800 apic_update_lvtt(apic
);
1801 if (kvm_check_has_quirk(vcpu
->kvm
, KVM_X86_QUIRK_LINT0_REENABLED
))
1802 kvm_lapic_set_reg(apic
, APIC_LVT0
,
1803 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT
));
1804 apic_manage_nmi_watchdog(apic
, kvm_lapic_get_reg(apic
, APIC_LVT0
));
1806 kvm_lapic_set_reg(apic
, APIC_DFR
, 0xffffffffU
);
1807 apic_set_spiv(apic
, 0xff);
1808 kvm_lapic_set_reg(apic
, APIC_TASKPRI
, 0);
1809 if (!apic_x2apic_mode(apic
))
1810 kvm_apic_set_ldr(apic
, 0);
1811 kvm_lapic_set_reg(apic
, APIC_ESR
, 0);
1812 kvm_lapic_set_reg(apic
, APIC_ICR
, 0);
1813 kvm_lapic_set_reg(apic
, APIC_ICR2
, 0);
1814 kvm_lapic_set_reg(apic
, APIC_TDCR
, 0);
1815 kvm_lapic_set_reg(apic
, APIC_TMICT
, 0);
1816 for (i
= 0; i
< 8; i
++) {
1817 kvm_lapic_set_reg(apic
, APIC_IRR
+ 0x10 * i
, 0);
1818 kvm_lapic_set_reg(apic
, APIC_ISR
+ 0x10 * i
, 0);
1819 kvm_lapic_set_reg(apic
, APIC_TMR
+ 0x10 * i
, 0);
1821 apic
->irr_pending
= vcpu
->arch
.apicv_active
;
1822 apic
->isr_count
= vcpu
->arch
.apicv_active
? 1 : 0;
1823 apic
->highest_isr_cache
= -1;
1824 update_divide_count(apic
);
1825 atomic_set(&apic
->lapic_timer
.pending
, 0);
1826 if (kvm_vcpu_is_bsp(vcpu
))
1827 kvm_lapic_set_base(vcpu
,
1828 vcpu
->arch
.apic_base
| MSR_IA32_APICBASE_BSP
);
1829 vcpu
->arch
.pv_eoi
.msr_val
= 0;
1830 apic_update_ppr(apic
);
1832 vcpu
->arch
.apic_arb_prio
= 0;
1833 vcpu
->arch
.apic_attention
= 0;
1835 apic_debug("%s: vcpu=%p, id=%d, base_msr="
1836 "0x%016" PRIx64
", base_address=0x%0lx.\n", __func__
,
1837 vcpu
, kvm_apic_id(apic
),
1838 vcpu
->arch
.apic_base
, apic
->base_address
);
1842 *----------------------------------------------------------------------
1844 *----------------------------------------------------------------------
1847 static bool lapic_is_periodic(struct kvm_lapic
*apic
)
1849 return apic_lvtt_period(apic
);
1852 int apic_has_pending_timer(struct kvm_vcpu
*vcpu
)
1854 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1856 if (apic_enabled(apic
) && apic_lvt_enabled(apic
, APIC_LVTT
))
1857 return atomic_read(&apic
->lapic_timer
.pending
);
1862 int kvm_apic_local_deliver(struct kvm_lapic
*apic
, int lvt_type
)
1864 u32 reg
= kvm_lapic_get_reg(apic
, lvt_type
);
1865 int vector
, mode
, trig_mode
;
1867 if (kvm_apic_hw_enabled(apic
) && !(reg
& APIC_LVT_MASKED
)) {
1868 vector
= reg
& APIC_VECTOR_MASK
;
1869 mode
= reg
& APIC_MODE_MASK
;
1870 trig_mode
= reg
& APIC_LVT_LEVEL_TRIGGER
;
1871 return __apic_accept_irq(apic
, mode
, vector
, 1, trig_mode
,
1877 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu
*vcpu
)
1879 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1882 kvm_apic_local_deliver(apic
, APIC_LVT0
);
1885 static const struct kvm_io_device_ops apic_mmio_ops
= {
1886 .read
= apic_mmio_read
,
1887 .write
= apic_mmio_write
,
1890 static enum hrtimer_restart
apic_timer_fn(struct hrtimer
*data
)
1892 struct kvm_timer
*ktimer
= container_of(data
, struct kvm_timer
, timer
);
1893 struct kvm_lapic
*apic
= container_of(ktimer
, struct kvm_lapic
, lapic_timer
);
1895 apic_timer_expired(apic
);
1897 if (lapic_is_periodic(apic
)) {
1898 hrtimer_add_expires_ns(&ktimer
->timer
, ktimer
->period
);
1899 return HRTIMER_RESTART
;
1901 return HRTIMER_NORESTART
;
1904 int kvm_create_lapic(struct kvm_vcpu
*vcpu
)
1906 struct kvm_lapic
*apic
;
1908 ASSERT(vcpu
!= NULL
);
1909 apic_debug("apic_init %d\n", vcpu
->vcpu_id
);
1911 apic
= kzalloc(sizeof(*apic
), GFP_KERNEL
);
1915 vcpu
->arch
.apic
= apic
;
1917 apic
->regs
= (void *)get_zeroed_page(GFP_KERNEL
);
1919 printk(KERN_ERR
"malloc apic regs error for vcpu %x\n",
1921 goto nomem_free_apic
;
1925 hrtimer_init(&apic
->lapic_timer
.timer
, CLOCK_MONOTONIC
,
1926 HRTIMER_MODE_ABS_PINNED
);
1927 apic
->lapic_timer
.timer
.function
= apic_timer_fn
;
1930 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1931 * thinking that APIC satet has changed.
1933 vcpu
->arch
.apic_base
= MSR_IA32_APICBASE_ENABLE
;
1934 kvm_lapic_set_base(vcpu
,
1935 APIC_DEFAULT_PHYS_BASE
| MSR_IA32_APICBASE_ENABLE
);
1937 static_key_slow_inc(&apic_sw_disabled
.key
); /* sw disabled at reset */
1938 kvm_lapic_reset(vcpu
, false);
1939 kvm_iodevice_init(&apic
->dev
, &apic_mmio_ops
);
1948 int kvm_apic_has_interrupt(struct kvm_vcpu
*vcpu
)
1950 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1953 if (!apic_enabled(apic
))
1956 apic_update_ppr(apic
);
1957 highest_irr
= apic_find_highest_irr(apic
);
1958 if ((highest_irr
== -1) ||
1959 ((highest_irr
& 0xF0) <= kvm_lapic_get_reg(apic
, APIC_PROCPRI
)))
1964 int kvm_apic_accept_pic_intr(struct kvm_vcpu
*vcpu
)
1966 u32 lvt0
= kvm_lapic_get_reg(vcpu
->arch
.apic
, APIC_LVT0
);
1969 if (!kvm_apic_hw_enabled(vcpu
->arch
.apic
))
1971 if ((lvt0
& APIC_LVT_MASKED
) == 0 &&
1972 GET_APIC_DELIVERY_MODE(lvt0
) == APIC_MODE_EXTINT
)
1977 void kvm_inject_apic_timer_irqs(struct kvm_vcpu
*vcpu
)
1979 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1981 if (atomic_read(&apic
->lapic_timer
.pending
) > 0) {
1982 kvm_apic_local_deliver(apic
, APIC_LVTT
);
1983 if (apic_lvtt_tscdeadline(apic
))
1984 apic
->lapic_timer
.tscdeadline
= 0;
1985 atomic_set(&apic
->lapic_timer
.pending
, 0);
1989 int kvm_get_apic_interrupt(struct kvm_vcpu
*vcpu
)
1991 int vector
= kvm_apic_has_interrupt(vcpu
);
1992 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1998 * We get here even with APIC virtualization enabled, if doing
1999 * nested virtualization and L1 runs with the "acknowledge interrupt
2000 * on exit" mode. Then we cannot inject the interrupt via RVI,
2001 * because the process would deliver it through the IDT.
2004 apic_set_isr(vector
, apic
);
2005 apic_update_ppr(apic
);
2006 apic_clear_irr(vector
, apic
);
2008 if (test_bit(vector
, vcpu_to_synic(vcpu
)->auto_eoi_bitmap
)) {
2009 apic_clear_isr(vector
, apic
);
2010 apic_update_ppr(apic
);
2016 void kvm_apic_post_state_restore(struct kvm_vcpu
*vcpu
,
2017 struct kvm_lapic_state
*s
)
2019 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2021 kvm_lapic_set_base(vcpu
, vcpu
->arch
.apic_base
);
2022 /* set SPIV separately to get count of SW disabled APICs right */
2023 apic_set_spiv(apic
, *((u32
*)(s
->regs
+ APIC_SPIV
)));
2024 memcpy(vcpu
->arch
.apic
->regs
, s
->regs
, sizeof *s
);
2025 /* call kvm_apic_set_id() to put apic into apic_map */
2026 kvm_apic_set_id(apic
, kvm_apic_id(apic
));
2027 kvm_apic_set_version(vcpu
);
2029 apic_update_ppr(apic
);
2030 hrtimer_cancel(&apic
->lapic_timer
.timer
);
2031 apic_update_lvtt(apic
);
2032 apic_manage_nmi_watchdog(apic
, kvm_lapic_get_reg(apic
, APIC_LVT0
));
2033 update_divide_count(apic
);
2034 start_apic_timer(apic
);
2035 apic
->irr_pending
= true;
2036 apic
->isr_count
= vcpu
->arch
.apicv_active
?
2037 1 : count_vectors(apic
->regs
+ APIC_ISR
);
2038 apic
->highest_isr_cache
= -1;
2039 if (vcpu
->arch
.apicv_active
) {
2040 if (kvm_x86_ops
->apicv_post_state_restore
)
2041 kvm_x86_ops
->apicv_post_state_restore(vcpu
);
2042 kvm_x86_ops
->hwapic_irr_update(vcpu
,
2043 apic_find_highest_irr(apic
));
2044 kvm_x86_ops
->hwapic_isr_update(vcpu
,
2045 apic_find_highest_isr(apic
));
2047 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
2048 if (ioapic_in_kernel(vcpu
->kvm
))
2049 kvm_rtc_eoi_tracking_restore_one(vcpu
);
2051 vcpu
->arch
.apic_arb_prio
= 0;
2054 void __kvm_migrate_apic_timer(struct kvm_vcpu
*vcpu
)
2056 struct hrtimer
*timer
;
2058 if (!lapic_in_kernel(vcpu
))
2061 timer
= &vcpu
->arch
.apic
->lapic_timer
.timer
;
2062 if (hrtimer_cancel(timer
))
2063 hrtimer_start_expires(timer
, HRTIMER_MODE_ABS_PINNED
);
2067 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2069 * Detect whether guest triggered PV EOI since the
2070 * last entry. If yes, set EOI on guests's behalf.
2071 * Clear PV EOI in guest memory in any case.
2073 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu
*vcpu
,
2074 struct kvm_lapic
*apic
)
2079 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2080 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2082 * KVM_APIC_PV_EOI_PENDING is unset:
2083 * -> host disabled PV EOI.
2084 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2085 * -> host enabled PV EOI, guest did not execute EOI yet.
2086 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2087 * -> host enabled PV EOI, guest executed EOI.
2089 BUG_ON(!pv_eoi_enabled(vcpu
));
2090 pending
= pv_eoi_get_pending(vcpu
);
2092 * Clear pending bit in any case: it will be set again on vmentry.
2093 * While this might not be ideal from performance point of view,
2094 * this makes sure pv eoi is only enabled when we know it's safe.
2096 pv_eoi_clr_pending(vcpu
);
2099 vector
= apic_set_eoi(apic
);
2100 trace_kvm_pv_eoi(apic
, vector
);
2103 void kvm_lapic_sync_from_vapic(struct kvm_vcpu
*vcpu
)
2107 if (test_bit(KVM_APIC_PV_EOI_PENDING
, &vcpu
->arch
.apic_attention
))
2108 apic_sync_pv_eoi_from_guest(vcpu
, vcpu
->arch
.apic
);
2110 if (!test_bit(KVM_APIC_CHECK_VAPIC
, &vcpu
->arch
.apic_attention
))
2113 if (kvm_read_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apic
->vapic_cache
, &data
,
2117 apic_set_tpr(vcpu
->arch
.apic
, data
& 0xff);
2121 * apic_sync_pv_eoi_to_guest - called before vmentry
2123 * Detect whether it's safe to enable PV EOI and
2126 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu
*vcpu
,
2127 struct kvm_lapic
*apic
)
2129 if (!pv_eoi_enabled(vcpu
) ||
2130 /* IRR set or many bits in ISR: could be nested. */
2131 apic
->irr_pending
||
2132 /* Cache not set: could be safe but we don't bother. */
2133 apic
->highest_isr_cache
== -1 ||
2134 /* Need EOI to update ioapic. */
2135 kvm_ioapic_handles_vector(apic
, apic
->highest_isr_cache
)) {
2137 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2138 * so we need not do anything here.
2143 pv_eoi_set_pending(apic
->vcpu
);
2146 void kvm_lapic_sync_to_vapic(struct kvm_vcpu
*vcpu
)
2149 int max_irr
, max_isr
;
2150 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2152 apic_sync_pv_eoi_to_guest(vcpu
, apic
);
2154 if (!test_bit(KVM_APIC_CHECK_VAPIC
, &vcpu
->arch
.apic_attention
))
2157 tpr
= kvm_lapic_get_reg(apic
, APIC_TASKPRI
) & 0xff;
2158 max_irr
= apic_find_highest_irr(apic
);
2161 max_isr
= apic_find_highest_isr(apic
);
2164 data
= (tpr
& 0xff) | ((max_isr
& 0xf0) << 8) | (max_irr
<< 24);
2166 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apic
->vapic_cache
, &data
,
2170 int kvm_lapic_set_vapic_addr(struct kvm_vcpu
*vcpu
, gpa_t vapic_addr
)
2173 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
,
2174 &vcpu
->arch
.apic
->vapic_cache
,
2175 vapic_addr
, sizeof(u32
)))
2177 __set_bit(KVM_APIC_CHECK_VAPIC
, &vcpu
->arch
.apic_attention
);
2179 __clear_bit(KVM_APIC_CHECK_VAPIC
, &vcpu
->arch
.apic_attention
);
2182 vcpu
->arch
.apic
->vapic_addr
= vapic_addr
;
2186 int kvm_x2apic_msr_write(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
2188 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2189 u32 reg
= (msr
- APIC_BASE_MSR
) << 4;
2191 if (!lapic_in_kernel(vcpu
) || !apic_x2apic_mode(apic
))
2194 if (reg
== APIC_ICR2
)
2197 /* if this is ICR write vector before command */
2198 if (reg
== APIC_ICR
)
2199 kvm_lapic_reg_write(apic
, APIC_ICR2
, (u32
)(data
>> 32));
2200 return kvm_lapic_reg_write(apic
, reg
, (u32
)data
);
2203 int kvm_x2apic_msr_read(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*data
)
2205 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2206 u32 reg
= (msr
- APIC_BASE_MSR
) << 4, low
, high
= 0;
2208 if (!lapic_in_kernel(vcpu
) || !apic_x2apic_mode(apic
))
2211 if (reg
== APIC_DFR
|| reg
== APIC_ICR2
) {
2212 apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
2217 if (kvm_lapic_reg_read(apic
, reg
, 4, &low
))
2219 if (reg
== APIC_ICR
)
2220 kvm_lapic_reg_read(apic
, APIC_ICR2
, 4, &high
);
2222 *data
= (((u64
)high
) << 32) | low
;
2227 int kvm_hv_vapic_msr_write(struct kvm_vcpu
*vcpu
, u32 reg
, u64 data
)
2229 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2231 if (!lapic_in_kernel(vcpu
))
2234 /* if this is ICR write vector before command */
2235 if (reg
== APIC_ICR
)
2236 kvm_lapic_reg_write(apic
, APIC_ICR2
, (u32
)(data
>> 32));
2237 return kvm_lapic_reg_write(apic
, reg
, (u32
)data
);
2240 int kvm_hv_vapic_msr_read(struct kvm_vcpu
*vcpu
, u32 reg
, u64
*data
)
2242 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2245 if (!lapic_in_kernel(vcpu
))
2248 if (kvm_lapic_reg_read(apic
, reg
, 4, &low
))
2250 if (reg
== APIC_ICR
)
2251 kvm_lapic_reg_read(apic
, APIC_ICR2
, 4, &high
);
2253 *data
= (((u64
)high
) << 32) | low
;
2258 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu
*vcpu
, u64 data
)
2260 u64 addr
= data
& ~KVM_MSR_ENABLED
;
2261 if (!IS_ALIGNED(addr
, 4))
2264 vcpu
->arch
.pv_eoi
.msr_val
= data
;
2265 if (!pv_eoi_enabled(vcpu
))
2267 return kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.pv_eoi
.data
,
2271 void kvm_apic_accept_events(struct kvm_vcpu
*vcpu
)
2273 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2277 if (!lapic_in_kernel(vcpu
) || !apic
->pending_events
)
2281 * INITs are latched while in SMM. Because an SMM CPU cannot
2282 * be in KVM_MP_STATE_INIT_RECEIVED state, just eat SIPIs
2283 * and delay processing of INIT until the next RSM.
2286 WARN_ON_ONCE(vcpu
->arch
.mp_state
== KVM_MP_STATE_INIT_RECEIVED
);
2287 if (test_bit(KVM_APIC_SIPI
, &apic
->pending_events
))
2288 clear_bit(KVM_APIC_SIPI
, &apic
->pending_events
);
2292 pe
= xchg(&apic
->pending_events
, 0);
2293 if (test_bit(KVM_APIC_INIT
, &pe
)) {
2294 kvm_lapic_reset(vcpu
, true);
2295 kvm_vcpu_reset(vcpu
, true);
2296 if (kvm_vcpu_is_bsp(apic
->vcpu
))
2297 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
2299 vcpu
->arch
.mp_state
= KVM_MP_STATE_INIT_RECEIVED
;
2301 if (test_bit(KVM_APIC_SIPI
, &pe
) &&
2302 vcpu
->arch
.mp_state
== KVM_MP_STATE_INIT_RECEIVED
) {
2303 /* evaluate pending_events before reading the vector */
2305 sipi_vector
= apic
->sipi_vector
;
2306 apic_debug("vcpu %d received sipi with vector # %x\n",
2307 vcpu
->vcpu_id
, sipi_vector
);
2308 kvm_vcpu_deliver_sipi_vector(vcpu
, sipi_vector
);
2309 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
2313 void kvm_lapic_init(void)
2315 /* do not patch jump label more than once per second */
2316 jump_label_rate_limit(&apic_hw_disabled
, HZ
);
2317 jump_label_rate_limit(&apic_sw_disabled
, HZ
);