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 <linux/atomic.h>
37 #include <linux/jump_label.h>
38 #include "kvm_cache_regs.h"
45 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
47 #define mod_64(x, y) ((x) % (y))
55 #define APIC_BUS_CYCLE_NS 1
57 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
58 #define apic_debug(fmt, arg...)
60 #define APIC_LVT_NUM 6
61 /* 14 is the version for Xeon and Pentium 8.4.8*/
62 #define APIC_VERSION (0x14UL | ((APIC_LVT_NUM - 1) << 16))
63 #define LAPIC_MMIO_LENGTH (1 << 12)
64 /* followed define is not in apicdef.h */
65 #define APIC_SHORT_MASK 0xc0000
66 #define APIC_DEST_NOSHORT 0x0
67 #define APIC_DEST_MASK 0x800
68 #define MAX_APIC_VECTOR 256
69 #define APIC_VECTORS_PER_REG 32
71 #define APIC_BROADCAST 0xFF
72 #define X2APIC_BROADCAST 0xFFFFFFFFul
74 #define VEC_POS(v) ((v) & (32 - 1))
75 #define REG_POS(v) (((v) >> 5) << 4)
77 static inline void apic_set_reg(struct kvm_lapic
*apic
, int reg_off
, u32 val
)
79 *((u32
*) (apic
->regs
+ reg_off
)) = val
;
82 static inline int apic_test_vector(int vec
, void *bitmap
)
84 return test_bit(VEC_POS(vec
), (bitmap
) + REG_POS(vec
));
87 bool kvm_apic_pending_eoi(struct kvm_vcpu
*vcpu
, int vector
)
89 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
91 return apic_test_vector(vector
, apic
->regs
+ APIC_ISR
) ||
92 apic_test_vector(vector
, apic
->regs
+ APIC_IRR
);
95 static inline void apic_set_vector(int vec
, void *bitmap
)
97 set_bit(VEC_POS(vec
), (bitmap
) + REG_POS(vec
));
100 static inline void apic_clear_vector(int vec
, void *bitmap
)
102 clear_bit(VEC_POS(vec
), (bitmap
) + REG_POS(vec
));
105 static inline int __apic_test_and_set_vector(int vec
, void *bitmap
)
107 return __test_and_set_bit(VEC_POS(vec
), (bitmap
) + REG_POS(vec
));
110 static inline int __apic_test_and_clear_vector(int vec
, void *bitmap
)
112 return __test_and_clear_bit(VEC_POS(vec
), (bitmap
) + REG_POS(vec
));
115 struct static_key_deferred apic_hw_disabled __read_mostly
;
116 struct static_key_deferred apic_sw_disabled __read_mostly
;
118 static inline int apic_enabled(struct kvm_lapic
*apic
)
120 return kvm_apic_sw_enabled(apic
) && kvm_apic_hw_enabled(apic
);
124 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
127 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
128 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
130 static inline int kvm_apic_id(struct kvm_lapic
*apic
)
132 return (kvm_apic_get_reg(apic
, APIC_ID
) >> 24) & 0xff;
135 #define KVM_X2APIC_CID_BITS 0
137 static void recalculate_apic_map(struct kvm
*kvm
)
139 struct kvm_apic_map
*new, *old
= NULL
;
140 struct kvm_vcpu
*vcpu
;
143 new = kzalloc(sizeof(struct kvm_apic_map
), GFP_KERNEL
);
145 mutex_lock(&kvm
->arch
.apic_map_lock
);
151 /* flat mode is default */
154 new->lid_mask
= 0xff;
155 new->broadcast
= APIC_BROADCAST
;
157 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
158 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
160 if (!kvm_apic_present(vcpu
))
163 if (apic_x2apic_mode(apic
)) {
166 new->cid_mask
= (1 << KVM_X2APIC_CID_BITS
) - 1;
167 new->lid_mask
= 0xffff;
168 new->broadcast
= X2APIC_BROADCAST
;
169 } else if (kvm_apic_get_reg(apic
, APIC_LDR
)) {
170 if (kvm_apic_get_reg(apic
, APIC_DFR
) ==
178 new->lid_mask
= 0xff;
183 * All APICs have to be configured in the same mode by an OS.
184 * We take advatage of this while building logical id loockup
185 * table. After reset APICs are in software disabled mode, so if
186 * we find apic with different setting we assume this is the mode
187 * OS wants all apics to be in; build lookup table accordingly.
189 if (kvm_apic_sw_enabled(apic
))
193 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
194 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
198 aid
= kvm_apic_id(apic
);
199 ldr
= kvm_apic_get_reg(apic
, APIC_LDR
);
200 cid
= apic_cluster_id(new, ldr
);
201 lid
= apic_logical_id(new, ldr
);
203 if (aid
< ARRAY_SIZE(new->phys_map
))
204 new->phys_map
[aid
] = apic
;
205 if (lid
&& cid
< ARRAY_SIZE(new->logical_map
))
206 new->logical_map
[cid
][ffs(lid
) - 1] = apic
;
209 old
= rcu_dereference_protected(kvm
->arch
.apic_map
,
210 lockdep_is_held(&kvm
->arch
.apic_map_lock
));
211 rcu_assign_pointer(kvm
->arch
.apic_map
, new);
212 mutex_unlock(&kvm
->arch
.apic_map_lock
);
217 kvm_vcpu_request_scan_ioapic(kvm
);
220 static inline void apic_set_spiv(struct kvm_lapic
*apic
, u32 val
)
222 bool enabled
= val
& APIC_SPIV_APIC_ENABLED
;
224 apic_set_reg(apic
, APIC_SPIV
, val
);
226 if (enabled
!= apic
->sw_enabled
) {
227 apic
->sw_enabled
= enabled
;
229 static_key_slow_dec_deferred(&apic_sw_disabled
);
230 recalculate_apic_map(apic
->vcpu
->kvm
);
232 static_key_slow_inc(&apic_sw_disabled
.key
);
236 static inline void kvm_apic_set_id(struct kvm_lapic
*apic
, u8 id
)
238 apic_set_reg(apic
, APIC_ID
, id
<< 24);
239 recalculate_apic_map(apic
->vcpu
->kvm
);
242 static inline void kvm_apic_set_ldr(struct kvm_lapic
*apic
, u32 id
)
244 apic_set_reg(apic
, APIC_LDR
, id
);
245 recalculate_apic_map(apic
->vcpu
->kvm
);
248 static inline int apic_lvt_enabled(struct kvm_lapic
*apic
, int lvt_type
)
250 return !(kvm_apic_get_reg(apic
, lvt_type
) & APIC_LVT_MASKED
);
253 static inline int apic_lvt_vector(struct kvm_lapic
*apic
, int lvt_type
)
255 return kvm_apic_get_reg(apic
, lvt_type
) & APIC_VECTOR_MASK
;
258 static inline int apic_lvtt_oneshot(struct kvm_lapic
*apic
)
260 return apic
->lapic_timer
.timer_mode
== APIC_LVT_TIMER_ONESHOT
;
263 static inline int apic_lvtt_period(struct kvm_lapic
*apic
)
265 return apic
->lapic_timer
.timer_mode
== APIC_LVT_TIMER_PERIODIC
;
268 static inline int apic_lvtt_tscdeadline(struct kvm_lapic
*apic
)
270 return apic
->lapic_timer
.timer_mode
== APIC_LVT_TIMER_TSCDEADLINE
;
273 static inline int apic_lvt_nmi_mode(u32 lvt_val
)
275 return (lvt_val
& (APIC_MODE_MASK
| APIC_LVT_MASKED
)) == APIC_DM_NMI
;
278 void kvm_apic_set_version(struct kvm_vcpu
*vcpu
)
280 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
281 struct kvm_cpuid_entry2
*feat
;
282 u32 v
= APIC_VERSION
;
284 if (!kvm_vcpu_has_lapic(vcpu
))
287 feat
= kvm_find_cpuid_entry(apic
->vcpu
, 0x1, 0);
288 if (feat
&& (feat
->ecx
& (1 << (X86_FEATURE_X2APIC
& 31))))
289 v
|= APIC_LVR_DIRECTED_EOI
;
290 apic_set_reg(apic
, APIC_LVR
, v
);
293 static const unsigned int apic_lvt_mask
[APIC_LVT_NUM
] = {
294 LVT_MASK
, /* part LVTT mask, timer mode mask added at runtime */
295 LVT_MASK
| APIC_MODE_MASK
, /* LVTTHMR */
296 LVT_MASK
| APIC_MODE_MASK
, /* LVTPC */
297 LINT_MASK
, LINT_MASK
, /* LVT0-1 */
298 LVT_MASK
/* LVTERR */
301 static int find_highest_vector(void *bitmap
)
306 for (vec
= MAX_APIC_VECTOR
- APIC_VECTORS_PER_REG
;
307 vec
>= 0; vec
-= APIC_VECTORS_PER_REG
) {
308 reg
= bitmap
+ REG_POS(vec
);
310 return fls(*reg
) - 1 + vec
;
316 static u8
count_vectors(void *bitmap
)
322 for (vec
= 0; vec
< MAX_APIC_VECTOR
; vec
+= APIC_VECTORS_PER_REG
) {
323 reg
= bitmap
+ REG_POS(vec
);
324 count
+= hweight32(*reg
);
330 void kvm_apic_update_irr(struct kvm_vcpu
*vcpu
, u32
*pir
)
333 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
335 for (i
= 0; i
<= 7; i
++) {
336 pir_val
= xchg(&pir
[i
], 0);
338 *((u32
*)(apic
->regs
+ APIC_IRR
+ i
* 0x10)) |= pir_val
;
341 EXPORT_SYMBOL_GPL(kvm_apic_update_irr
);
343 static inline void apic_set_irr(int vec
, struct kvm_lapic
*apic
)
345 apic_set_vector(vec
, apic
->regs
+ APIC_IRR
);
347 * irr_pending must be true if any interrupt is pending; set it after
348 * APIC_IRR to avoid race with apic_clear_irr
350 apic
->irr_pending
= true;
353 static inline int apic_search_irr(struct kvm_lapic
*apic
)
355 return find_highest_vector(apic
->regs
+ APIC_IRR
);
358 static inline int apic_find_highest_irr(struct kvm_lapic
*apic
)
363 * Note that irr_pending is just a hint. It will be always
364 * true with virtual interrupt delivery enabled.
366 if (!apic
->irr_pending
)
369 kvm_x86_ops
->sync_pir_to_irr(apic
->vcpu
);
370 result
= apic_search_irr(apic
);
371 ASSERT(result
== -1 || result
>= 16);
376 static inline void apic_clear_irr(int vec
, struct kvm_lapic
*apic
)
378 struct kvm_vcpu
*vcpu
;
382 if (unlikely(kvm_apic_vid_enabled(vcpu
->kvm
))) {
383 /* try to update RVI */
384 apic_clear_vector(vec
, apic
->regs
+ APIC_IRR
);
385 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
387 apic
->irr_pending
= false;
388 apic_clear_vector(vec
, apic
->regs
+ APIC_IRR
);
389 if (apic_search_irr(apic
) != -1)
390 apic
->irr_pending
= true;
394 static inline void apic_set_isr(int vec
, struct kvm_lapic
*apic
)
396 struct kvm_vcpu
*vcpu
;
398 if (__apic_test_and_set_vector(vec
, apic
->regs
+ APIC_ISR
))
404 * With APIC virtualization enabled, all caching is disabled
405 * because the processor can modify ISR under the hood. Instead
408 if (unlikely(kvm_apic_vid_enabled(vcpu
->kvm
)))
409 kvm_x86_ops
->hwapic_isr_update(vcpu
->kvm
, vec
);
412 BUG_ON(apic
->isr_count
> MAX_APIC_VECTOR
);
414 * ISR (in service register) bit is set when injecting an interrupt.
415 * The highest vector is injected. Thus the latest bit set matches
416 * the highest bit in ISR.
418 apic
->highest_isr_cache
= vec
;
422 static inline int apic_find_highest_isr(struct kvm_lapic
*apic
)
427 * Note that isr_count is always 1, and highest_isr_cache
428 * is always -1, with APIC virtualization enabled.
430 if (!apic
->isr_count
)
432 if (likely(apic
->highest_isr_cache
!= -1))
433 return apic
->highest_isr_cache
;
435 result
= find_highest_vector(apic
->regs
+ APIC_ISR
);
436 ASSERT(result
== -1 || result
>= 16);
441 static inline void apic_clear_isr(int vec
, struct kvm_lapic
*apic
)
443 struct kvm_vcpu
*vcpu
;
444 if (!__apic_test_and_clear_vector(vec
, apic
->regs
+ APIC_ISR
))
450 * We do get here for APIC virtualization enabled if the guest
451 * uses the Hyper-V APIC enlightenment. In this case we may need
452 * to trigger a new interrupt delivery by writing the SVI field;
453 * on the other hand isr_count and highest_isr_cache are unused
454 * and must be left alone.
456 if (unlikely(kvm_apic_vid_enabled(vcpu
->kvm
)))
457 kvm_x86_ops
->hwapic_isr_update(vcpu
->kvm
,
458 apic_find_highest_isr(apic
));
461 BUG_ON(apic
->isr_count
< 0);
462 apic
->highest_isr_cache
= -1;
466 int kvm_lapic_find_highest_irr(struct kvm_vcpu
*vcpu
)
470 /* This may race with setting of irr in __apic_accept_irq() and
471 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
472 * will cause vmexit immediately and the value will be recalculated
473 * on the next vmentry.
475 if (!kvm_vcpu_has_lapic(vcpu
))
477 highest_irr
= apic_find_highest_irr(vcpu
->arch
.apic
);
482 static int __apic_accept_irq(struct kvm_lapic
*apic
, int delivery_mode
,
483 int vector
, int level
, int trig_mode
,
484 unsigned long *dest_map
);
486 int kvm_apic_set_irq(struct kvm_vcpu
*vcpu
, struct kvm_lapic_irq
*irq
,
487 unsigned long *dest_map
)
489 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
491 return __apic_accept_irq(apic
, irq
->delivery_mode
, irq
->vector
,
492 irq
->level
, irq
->trig_mode
, dest_map
);
495 static int pv_eoi_put_user(struct kvm_vcpu
*vcpu
, u8 val
)
498 return kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.pv_eoi
.data
, &val
,
502 static int pv_eoi_get_user(struct kvm_vcpu
*vcpu
, u8
*val
)
505 return kvm_read_guest_cached(vcpu
->kvm
, &vcpu
->arch
.pv_eoi
.data
, val
,
509 static inline bool pv_eoi_enabled(struct kvm_vcpu
*vcpu
)
511 return vcpu
->arch
.pv_eoi
.msr_val
& KVM_MSR_ENABLED
;
514 static bool pv_eoi_get_pending(struct kvm_vcpu
*vcpu
)
517 if (pv_eoi_get_user(vcpu
, &val
) < 0)
518 apic_debug("Can't read EOI MSR value: 0x%llx\n",
519 (unsigned long long)vcpu
->arch
.pv_eoi
.msr_val
);
523 static void pv_eoi_set_pending(struct kvm_vcpu
*vcpu
)
525 if (pv_eoi_put_user(vcpu
, KVM_PV_EOI_ENABLED
) < 0) {
526 apic_debug("Can't set EOI MSR value: 0x%llx\n",
527 (unsigned long long)vcpu
->arch
.pv_eoi
.msr_val
);
530 __set_bit(KVM_APIC_PV_EOI_PENDING
, &vcpu
->arch
.apic_attention
);
533 static void pv_eoi_clr_pending(struct kvm_vcpu
*vcpu
)
535 if (pv_eoi_put_user(vcpu
, KVM_PV_EOI_DISABLED
) < 0) {
536 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
537 (unsigned long long)vcpu
->arch
.pv_eoi
.msr_val
);
540 __clear_bit(KVM_APIC_PV_EOI_PENDING
, &vcpu
->arch
.apic_attention
);
543 void kvm_apic_update_tmr(struct kvm_vcpu
*vcpu
, u32
*tmr
)
545 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
548 for (i
= 0; i
< 8; i
++)
549 apic_set_reg(apic
, APIC_TMR
+ 0x10 * i
, tmr
[i
]);
552 static void apic_update_ppr(struct kvm_lapic
*apic
)
554 u32 tpr
, isrv
, ppr
, old_ppr
;
557 old_ppr
= kvm_apic_get_reg(apic
, APIC_PROCPRI
);
558 tpr
= kvm_apic_get_reg(apic
, APIC_TASKPRI
);
559 isr
= apic_find_highest_isr(apic
);
560 isrv
= (isr
!= -1) ? isr
: 0;
562 if ((tpr
& 0xf0) >= (isrv
& 0xf0))
567 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
568 apic
, ppr
, isr
, isrv
);
570 if (old_ppr
!= ppr
) {
571 apic_set_reg(apic
, APIC_PROCPRI
, ppr
);
573 kvm_make_request(KVM_REQ_EVENT
, apic
->vcpu
);
577 static void apic_set_tpr(struct kvm_lapic
*apic
, u32 tpr
)
579 apic_set_reg(apic
, APIC_TASKPRI
, tpr
);
580 apic_update_ppr(apic
);
583 static int kvm_apic_broadcast(struct kvm_lapic
*apic
, u32 dest
)
585 return dest
== (apic_x2apic_mode(apic
) ?
586 X2APIC_BROADCAST
: APIC_BROADCAST
);
589 int kvm_apic_match_physical_addr(struct kvm_lapic
*apic
, u32 dest
)
591 return kvm_apic_id(apic
) == dest
|| kvm_apic_broadcast(apic
, dest
);
594 int kvm_apic_match_logical_addr(struct kvm_lapic
*apic
, u32 mda
)
599 if (kvm_apic_broadcast(apic
, mda
))
602 if (apic_x2apic_mode(apic
)) {
603 logical_id
= kvm_apic_get_reg(apic
, APIC_LDR
);
604 return logical_id
& mda
;
607 logical_id
= GET_APIC_LOGICAL_ID(kvm_apic_get_reg(apic
, APIC_LDR
));
609 switch (kvm_apic_get_reg(apic
, APIC_DFR
)) {
611 if (logical_id
& mda
)
614 case APIC_DFR_CLUSTER
:
615 if (((logical_id
>> 4) == (mda
>> 0x4))
616 && (logical_id
& mda
& 0xf))
620 apic_debug("Bad DFR vcpu %d: %08x\n",
621 apic
->vcpu
->vcpu_id
, kvm_apic_get_reg(apic
, APIC_DFR
));
628 int kvm_apic_match_dest(struct kvm_vcpu
*vcpu
, struct kvm_lapic
*source
,
629 int short_hand
, unsigned int dest
, int dest_mode
)
632 struct kvm_lapic
*target
= vcpu
->arch
.apic
;
634 apic_debug("target %p, source %p, dest 0x%x, "
635 "dest_mode 0x%x, short_hand 0x%x\n",
636 target
, source
, dest
, dest_mode
, short_hand
);
639 switch (short_hand
) {
640 case APIC_DEST_NOSHORT
:
643 result
= kvm_apic_match_physical_addr(target
, dest
);
646 result
= kvm_apic_match_logical_addr(target
, dest
);
649 result
= (target
== source
);
651 case APIC_DEST_ALLINC
:
654 case APIC_DEST_ALLBUT
:
655 result
= (target
!= source
);
658 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
666 bool kvm_irq_delivery_to_apic_fast(struct kvm
*kvm
, struct kvm_lapic
*src
,
667 struct kvm_lapic_irq
*irq
, int *r
, unsigned long *dest_map
)
669 struct kvm_apic_map
*map
;
670 unsigned long bitmap
= 1;
671 struct kvm_lapic
**dst
;
677 if (irq
->shorthand
== APIC_DEST_SELF
) {
678 *r
= kvm_apic_set_irq(src
->vcpu
, irq
, dest_map
);
686 map
= rcu_dereference(kvm
->arch
.apic_map
);
691 if (irq
->dest_id
== map
->broadcast
)
696 if (irq
->dest_mode
== 0) { /* physical mode */
697 if (irq
->dest_id
>= ARRAY_SIZE(map
->phys_map
))
700 dst
= &map
->phys_map
[irq
->dest_id
];
702 u32 mda
= irq
->dest_id
<< (32 - map
->ldr_bits
);
704 dst
= map
->logical_map
[apic_cluster_id(map
, mda
)];
706 bitmap
= apic_logical_id(map
, mda
);
708 if (irq
->delivery_mode
== APIC_DM_LOWEST
) {
710 for_each_set_bit(i
, &bitmap
, 16) {
715 else if (kvm_apic_compare_prio(dst
[i
]->vcpu
, dst
[l
]->vcpu
) < 0)
719 bitmap
= (l
>= 0) ? 1 << l
: 0;
723 for_each_set_bit(i
, &bitmap
, 16) {
728 *r
+= kvm_apic_set_irq(dst
[i
]->vcpu
, irq
, dest_map
);
736 * Add a pending IRQ into lapic.
737 * Return 1 if successfully added and 0 if discarded.
739 static int __apic_accept_irq(struct kvm_lapic
*apic
, int delivery_mode
,
740 int vector
, int level
, int trig_mode
,
741 unsigned long *dest_map
)
744 struct kvm_vcpu
*vcpu
= apic
->vcpu
;
746 trace_kvm_apic_accept_irq(vcpu
->vcpu_id
, delivery_mode
,
748 switch (delivery_mode
) {
750 vcpu
->arch
.apic_arb_prio
++;
752 /* FIXME add logic for vcpu on reset */
753 if (unlikely(!apic_enabled(apic
)))
759 __set_bit(vcpu
->vcpu_id
, dest_map
);
761 if (kvm_x86_ops
->deliver_posted_interrupt
)
762 kvm_x86_ops
->deliver_posted_interrupt(vcpu
, vector
);
764 apic_set_irr(vector
, apic
);
766 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
773 vcpu
->arch
.pv
.pv_unhalted
= 1;
774 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
779 apic_debug("Ignoring guest SMI\n");
784 kvm_inject_nmi(vcpu
);
789 if (!trig_mode
|| level
) {
791 /* assumes that there are only KVM_APIC_INIT/SIPI */
792 apic
->pending_events
= (1UL << KVM_APIC_INIT
);
793 /* make sure pending_events is visible before sending
796 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
799 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
804 case APIC_DM_STARTUP
:
805 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
806 vcpu
->vcpu_id
, vector
);
808 apic
->sipi_vector
= vector
;
809 /* make sure sipi_vector is visible for the receiver */
811 set_bit(KVM_APIC_SIPI
, &apic
->pending_events
);
812 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
818 * Should only be called by kvm_apic_local_deliver() with LVT0,
819 * before NMI watchdog was enabled. Already handled by
820 * kvm_apic_accept_pic_intr().
825 printk(KERN_ERR
"TODO: unsupported delivery mode %x\n",
832 int kvm_apic_compare_prio(struct kvm_vcpu
*vcpu1
, struct kvm_vcpu
*vcpu2
)
834 return vcpu1
->arch
.apic_arb_prio
- vcpu2
->arch
.apic_arb_prio
;
837 static void kvm_ioapic_send_eoi(struct kvm_lapic
*apic
, int vector
)
839 if (!(kvm_apic_get_reg(apic
, APIC_SPIV
) & APIC_SPIV_DIRECTED_EOI
) &&
840 kvm_ioapic_handles_vector(apic
->vcpu
->kvm
, vector
)) {
842 if (apic_test_vector(vector
, apic
->regs
+ APIC_TMR
))
843 trigger_mode
= IOAPIC_LEVEL_TRIG
;
845 trigger_mode
= IOAPIC_EDGE_TRIG
;
846 kvm_ioapic_update_eoi(apic
->vcpu
, vector
, trigger_mode
);
850 static int apic_set_eoi(struct kvm_lapic
*apic
)
852 int vector
= apic_find_highest_isr(apic
);
854 trace_kvm_eoi(apic
, vector
);
857 * Not every write EOI will has corresponding ISR,
858 * one example is when Kernel check timer on setup_IO_APIC
863 apic_clear_isr(vector
, apic
);
864 apic_update_ppr(apic
);
866 kvm_ioapic_send_eoi(apic
, vector
);
867 kvm_make_request(KVM_REQ_EVENT
, apic
->vcpu
);
872 * this interface assumes a trap-like exit, which has already finished
873 * desired side effect including vISR and vPPR update.
875 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu
*vcpu
, int vector
)
877 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
879 trace_kvm_eoi(apic
, vector
);
881 kvm_ioapic_send_eoi(apic
, vector
);
882 kvm_make_request(KVM_REQ_EVENT
, apic
->vcpu
);
884 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated
);
886 static void apic_send_ipi(struct kvm_lapic
*apic
)
888 u32 icr_low
= kvm_apic_get_reg(apic
, APIC_ICR
);
889 u32 icr_high
= kvm_apic_get_reg(apic
, APIC_ICR2
);
890 struct kvm_lapic_irq irq
;
892 irq
.vector
= icr_low
& APIC_VECTOR_MASK
;
893 irq
.delivery_mode
= icr_low
& APIC_MODE_MASK
;
894 irq
.dest_mode
= icr_low
& APIC_DEST_MASK
;
895 irq
.level
= icr_low
& APIC_INT_ASSERT
;
896 irq
.trig_mode
= icr_low
& APIC_INT_LEVELTRIG
;
897 irq
.shorthand
= icr_low
& APIC_SHORT_MASK
;
898 if (apic_x2apic_mode(apic
))
899 irq
.dest_id
= icr_high
;
901 irq
.dest_id
= GET_APIC_DEST_FIELD(icr_high
);
903 trace_kvm_apic_ipi(icr_low
, irq
.dest_id
);
905 apic_debug("icr_high 0x%x, icr_low 0x%x, "
906 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
907 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
908 icr_high
, icr_low
, irq
.shorthand
, irq
.dest_id
,
909 irq
.trig_mode
, irq
.level
, irq
.dest_mode
, irq
.delivery_mode
,
912 kvm_irq_delivery_to_apic(apic
->vcpu
->kvm
, apic
, &irq
, NULL
);
915 static u32
apic_get_tmcct(struct kvm_lapic
*apic
)
921 ASSERT(apic
!= NULL
);
923 /* if initial count is 0, current count should also be 0 */
924 if (kvm_apic_get_reg(apic
, APIC_TMICT
) == 0 ||
925 apic
->lapic_timer
.period
== 0)
928 remaining
= hrtimer_get_remaining(&apic
->lapic_timer
.timer
);
929 if (ktime_to_ns(remaining
) < 0)
930 remaining
= ktime_set(0, 0);
932 ns
= mod_64(ktime_to_ns(remaining
), apic
->lapic_timer
.period
);
933 tmcct
= div64_u64(ns
,
934 (APIC_BUS_CYCLE_NS
* apic
->divide_count
));
939 static void __report_tpr_access(struct kvm_lapic
*apic
, bool write
)
941 struct kvm_vcpu
*vcpu
= apic
->vcpu
;
942 struct kvm_run
*run
= vcpu
->run
;
944 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS
, vcpu
);
945 run
->tpr_access
.rip
= kvm_rip_read(vcpu
);
946 run
->tpr_access
.is_write
= write
;
949 static inline void report_tpr_access(struct kvm_lapic
*apic
, bool write
)
951 if (apic
->vcpu
->arch
.tpr_access_reporting
)
952 __report_tpr_access(apic
, write
);
955 static u32
__apic_read(struct kvm_lapic
*apic
, unsigned int offset
)
959 if (offset
>= LAPIC_MMIO_LENGTH
)
964 if (apic_x2apic_mode(apic
))
965 val
= kvm_apic_id(apic
);
967 val
= kvm_apic_id(apic
) << 24;
970 apic_debug("Access APIC ARBPRI register which is for P6\n");
973 case APIC_TMCCT
: /* Timer CCR */
974 if (apic_lvtt_tscdeadline(apic
))
977 val
= apic_get_tmcct(apic
);
980 apic_update_ppr(apic
);
981 val
= kvm_apic_get_reg(apic
, offset
);
984 report_tpr_access(apic
, false);
987 val
= kvm_apic_get_reg(apic
, offset
);
994 static inline struct kvm_lapic
*to_lapic(struct kvm_io_device
*dev
)
996 return container_of(dev
, struct kvm_lapic
, dev
);
999 static int apic_reg_read(struct kvm_lapic
*apic
, u32 offset
, int len
,
1002 unsigned char alignment
= offset
& 0xf;
1004 /* this bitmask has a bit cleared for each reserved register */
1005 static const u64 rmask
= 0x43ff01ffffffe70cULL
;
1007 if ((alignment
+ len
) > 4) {
1008 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1013 if (offset
> 0x3f0 || !(rmask
& (1ULL << (offset
>> 4)))) {
1014 apic_debug("KVM_APIC_READ: read reserved register %x\n",
1019 result
= __apic_read(apic
, offset
& ~0xf);
1021 trace_kvm_apic_read(offset
, result
);
1027 memcpy(data
, (char *)&result
+ alignment
, len
);
1030 printk(KERN_ERR
"Local APIC read with len = %x, "
1031 "should be 1,2, or 4 instead\n", len
);
1037 static int apic_mmio_in_range(struct kvm_lapic
*apic
, gpa_t addr
)
1039 return kvm_apic_hw_enabled(apic
) &&
1040 addr
>= apic
->base_address
&&
1041 addr
< apic
->base_address
+ LAPIC_MMIO_LENGTH
;
1044 static int apic_mmio_read(struct kvm_io_device
*this,
1045 gpa_t address
, int len
, void *data
)
1047 struct kvm_lapic
*apic
= to_lapic(this);
1048 u32 offset
= address
- apic
->base_address
;
1050 if (!apic_mmio_in_range(apic
, address
))
1053 apic_reg_read(apic
, offset
, len
, data
);
1058 static void update_divide_count(struct kvm_lapic
*apic
)
1060 u32 tmp1
, tmp2
, tdcr
;
1062 tdcr
= kvm_apic_get_reg(apic
, APIC_TDCR
);
1064 tmp2
= ((tmp1
& 0x3) | ((tmp1
& 0x8) >> 1)) + 1;
1065 apic
->divide_count
= 0x1 << (tmp2
& 0x7);
1067 apic_debug("timer divide count is 0x%x\n",
1068 apic
->divide_count
);
1071 static void apic_timer_expired(struct kvm_lapic
*apic
)
1073 struct kvm_vcpu
*vcpu
= apic
->vcpu
;
1074 wait_queue_head_t
*q
= &vcpu
->wq
;
1077 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1080 if (atomic_read(&apic
->lapic_timer
.pending
))
1083 atomic_inc(&apic
->lapic_timer
.pending
);
1084 /* FIXME: this code should not know anything about vcpus */
1085 kvm_make_request(KVM_REQ_PENDING_TIMER
, vcpu
);
1087 if (waitqueue_active(q
))
1088 wake_up_interruptible(q
);
1091 static void start_apic_timer(struct kvm_lapic
*apic
)
1094 atomic_set(&apic
->lapic_timer
.pending
, 0);
1096 if (apic_lvtt_period(apic
) || apic_lvtt_oneshot(apic
)) {
1097 /* lapic timer in oneshot or periodic mode */
1098 now
= apic
->lapic_timer
.timer
.base
->get_time();
1099 apic
->lapic_timer
.period
= (u64
)kvm_apic_get_reg(apic
, APIC_TMICT
)
1100 * APIC_BUS_CYCLE_NS
* apic
->divide_count
;
1102 if (!apic
->lapic_timer
.period
)
1105 * Do not allow the guest to program periodic timers with small
1106 * interval, since the hrtimers are not throttled by the host
1109 if (apic_lvtt_period(apic
)) {
1110 s64 min_period
= min_timer_period_us
* 1000LL;
1112 if (apic
->lapic_timer
.period
< min_period
) {
1113 pr_info_ratelimited(
1114 "kvm: vcpu %i: requested %lld ns "
1115 "lapic timer period limited to %lld ns\n",
1116 apic
->vcpu
->vcpu_id
,
1117 apic
->lapic_timer
.period
, min_period
);
1118 apic
->lapic_timer
.period
= min_period
;
1122 hrtimer_start(&apic
->lapic_timer
.timer
,
1123 ktime_add_ns(now
, apic
->lapic_timer
.period
),
1126 apic_debug("%s: bus cycle is %" PRId64
"ns, now 0x%016"
1128 "timer initial count 0x%x, period %lldns, "
1129 "expire @ 0x%016" PRIx64
".\n", __func__
,
1130 APIC_BUS_CYCLE_NS
, ktime_to_ns(now
),
1131 kvm_apic_get_reg(apic
, APIC_TMICT
),
1132 apic
->lapic_timer
.period
,
1133 ktime_to_ns(ktime_add_ns(now
,
1134 apic
->lapic_timer
.period
)));
1135 } else if (apic_lvtt_tscdeadline(apic
)) {
1136 /* lapic timer in tsc deadline mode */
1137 u64 guest_tsc
, tscdeadline
= apic
->lapic_timer
.tscdeadline
;
1139 struct kvm_vcpu
*vcpu
= apic
->vcpu
;
1140 unsigned long this_tsc_khz
= vcpu
->arch
.virtual_tsc_khz
;
1141 unsigned long flags
;
1143 if (unlikely(!tscdeadline
|| !this_tsc_khz
))
1146 local_irq_save(flags
);
1148 now
= apic
->lapic_timer
.timer
.base
->get_time();
1149 guest_tsc
= kvm_x86_ops
->read_l1_tsc(vcpu
, native_read_tsc());
1150 if (likely(tscdeadline
> guest_tsc
)) {
1151 ns
= (tscdeadline
- guest_tsc
) * 1000000ULL;
1152 do_div(ns
, this_tsc_khz
);
1153 hrtimer_start(&apic
->lapic_timer
.timer
,
1154 ktime_add_ns(now
, ns
), HRTIMER_MODE_ABS
);
1156 apic_timer_expired(apic
);
1158 local_irq_restore(flags
);
1162 static void apic_manage_nmi_watchdog(struct kvm_lapic
*apic
, u32 lvt0_val
)
1164 int nmi_wd_enabled
= apic_lvt_nmi_mode(kvm_apic_get_reg(apic
, APIC_LVT0
));
1166 if (apic_lvt_nmi_mode(lvt0_val
)) {
1167 if (!nmi_wd_enabled
) {
1168 apic_debug("Receive NMI setting on APIC_LVT0 "
1169 "for cpu %d\n", apic
->vcpu
->vcpu_id
);
1170 apic
->vcpu
->kvm
->arch
.vapics_in_nmi_mode
++;
1172 } else if (nmi_wd_enabled
)
1173 apic
->vcpu
->kvm
->arch
.vapics_in_nmi_mode
--;
1176 static int apic_reg_write(struct kvm_lapic
*apic
, u32 reg
, u32 val
)
1180 trace_kvm_apic_write(reg
, val
);
1183 case APIC_ID
: /* Local APIC ID */
1184 if (!apic_x2apic_mode(apic
))
1185 kvm_apic_set_id(apic
, val
>> 24);
1191 report_tpr_access(apic
, true);
1192 apic_set_tpr(apic
, val
& 0xff);
1200 if (!apic_x2apic_mode(apic
))
1201 kvm_apic_set_ldr(apic
, val
& APIC_LDR_MASK
);
1207 if (!apic_x2apic_mode(apic
)) {
1208 apic_set_reg(apic
, APIC_DFR
, val
| 0x0FFFFFFF);
1209 recalculate_apic_map(apic
->vcpu
->kvm
);
1216 if (kvm_apic_get_reg(apic
, APIC_LVR
) & APIC_LVR_DIRECTED_EOI
)
1217 mask
|= APIC_SPIV_DIRECTED_EOI
;
1218 apic_set_spiv(apic
, val
& mask
);
1219 if (!(val
& APIC_SPIV_APIC_ENABLED
)) {
1223 for (i
= 0; i
< APIC_LVT_NUM
; i
++) {
1224 lvt_val
= kvm_apic_get_reg(apic
,
1225 APIC_LVTT
+ 0x10 * i
);
1226 apic_set_reg(apic
, APIC_LVTT
+ 0x10 * i
,
1227 lvt_val
| APIC_LVT_MASKED
);
1229 atomic_set(&apic
->lapic_timer
.pending
, 0);
1235 /* No delay here, so we always clear the pending bit */
1236 apic_set_reg(apic
, APIC_ICR
, val
& ~(1 << 12));
1237 apic_send_ipi(apic
);
1241 if (!apic_x2apic_mode(apic
))
1243 apic_set_reg(apic
, APIC_ICR2
, val
);
1247 apic_manage_nmi_watchdog(apic
, val
);
1252 /* TODO: Check vector */
1253 if (!kvm_apic_sw_enabled(apic
))
1254 val
|= APIC_LVT_MASKED
;
1256 val
&= apic_lvt_mask
[(reg
- APIC_LVTT
) >> 4];
1257 apic_set_reg(apic
, reg
, val
);
1262 u32 timer_mode
= val
& apic
->lapic_timer
.timer_mode_mask
;
1264 if (apic
->lapic_timer
.timer_mode
!= timer_mode
) {
1265 apic
->lapic_timer
.timer_mode
= timer_mode
;
1266 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1269 if (!kvm_apic_sw_enabled(apic
))
1270 val
|= APIC_LVT_MASKED
;
1271 val
&= (apic_lvt_mask
[0] | apic
->lapic_timer
.timer_mode_mask
);
1272 apic_set_reg(apic
, APIC_LVTT
, val
);
1277 if (apic_lvtt_tscdeadline(apic
))
1280 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1281 apic_set_reg(apic
, APIC_TMICT
, val
);
1282 start_apic_timer(apic
);
1287 apic_debug("KVM_WRITE:TDCR %x\n", val
);
1288 apic_set_reg(apic
, APIC_TDCR
, val
);
1289 update_divide_count(apic
);
1293 if (apic_x2apic_mode(apic
) && val
!= 0) {
1294 apic_debug("KVM_WRITE:ESR not zero %x\n", val
);
1300 if (apic_x2apic_mode(apic
)) {
1301 apic_reg_write(apic
, APIC_ICR
, 0x40000 | (val
& 0xff));
1310 apic_debug("Local APIC Write to read-only register %x\n", reg
);
1314 static int apic_mmio_write(struct kvm_io_device
*this,
1315 gpa_t address
, int len
, const void *data
)
1317 struct kvm_lapic
*apic
= to_lapic(this);
1318 unsigned int offset
= address
- apic
->base_address
;
1321 if (!apic_mmio_in_range(apic
, address
))
1325 * APIC register must be aligned on 128-bits boundary.
1326 * 32/64/128 bits registers must be accessed thru 32 bits.
1329 if (len
!= 4 || (offset
& 0xf)) {
1330 /* Don't shout loud, $infamous_os would cause only noise. */
1331 apic_debug("apic write: bad size=%d %lx\n", len
, (long)address
);
1337 /* too common printing */
1338 if (offset
!= APIC_EOI
)
1339 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1340 "0x%x\n", __func__
, offset
, len
, val
);
1342 apic_reg_write(apic
, offset
& 0xff0, val
);
1347 void kvm_lapic_set_eoi(struct kvm_vcpu
*vcpu
)
1349 if (kvm_vcpu_has_lapic(vcpu
))
1350 apic_reg_write(vcpu
->arch
.apic
, APIC_EOI
, 0);
1352 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi
);
1354 /* emulate APIC access in a trap manner */
1355 void kvm_apic_write_nodecode(struct kvm_vcpu
*vcpu
, u32 offset
)
1359 /* hw has done the conditional check and inst decode */
1362 apic_reg_read(vcpu
->arch
.apic
, offset
, 4, &val
);
1364 /* TODO: optimize to just emulate side effect w/o one more write */
1365 apic_reg_write(vcpu
->arch
.apic
, offset
, val
);
1367 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode
);
1369 void kvm_free_lapic(struct kvm_vcpu
*vcpu
)
1371 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1373 if (!vcpu
->arch
.apic
)
1376 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1378 if (!(vcpu
->arch
.apic_base
& MSR_IA32_APICBASE_ENABLE
))
1379 static_key_slow_dec_deferred(&apic_hw_disabled
);
1381 if (!apic
->sw_enabled
)
1382 static_key_slow_dec_deferred(&apic_sw_disabled
);
1385 free_page((unsigned long)apic
->regs
);
1391 *----------------------------------------------------------------------
1393 *----------------------------------------------------------------------
1396 u64
kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu
*vcpu
)
1398 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1400 if (!kvm_vcpu_has_lapic(vcpu
) || apic_lvtt_oneshot(apic
) ||
1401 apic_lvtt_period(apic
))
1404 return apic
->lapic_timer
.tscdeadline
;
1407 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu
*vcpu
, u64 data
)
1409 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1411 if (!kvm_vcpu_has_lapic(vcpu
) || apic_lvtt_oneshot(apic
) ||
1412 apic_lvtt_period(apic
))
1415 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1416 apic
->lapic_timer
.tscdeadline
= data
;
1417 start_apic_timer(apic
);
1420 void kvm_lapic_set_tpr(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
1422 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1424 if (!kvm_vcpu_has_lapic(vcpu
))
1427 apic_set_tpr(apic
, ((cr8
& 0x0f) << 4)
1428 | (kvm_apic_get_reg(apic
, APIC_TASKPRI
) & 4));
1431 u64
kvm_lapic_get_cr8(struct kvm_vcpu
*vcpu
)
1435 if (!kvm_vcpu_has_lapic(vcpu
))
1438 tpr
= (u64
) kvm_apic_get_reg(vcpu
->arch
.apic
, APIC_TASKPRI
);
1440 return (tpr
& 0xf0) >> 4;
1443 void kvm_lapic_set_base(struct kvm_vcpu
*vcpu
, u64 value
)
1445 u64 old_value
= vcpu
->arch
.apic_base
;
1446 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1449 value
|= MSR_IA32_APICBASE_BSP
;
1450 vcpu
->arch
.apic_base
= value
;
1454 if (!kvm_vcpu_is_bsp(apic
->vcpu
))
1455 value
&= ~MSR_IA32_APICBASE_BSP
;
1456 vcpu
->arch
.apic_base
= value
;
1458 /* update jump label if enable bit changes */
1459 if ((old_value
^ value
) & MSR_IA32_APICBASE_ENABLE
) {
1460 if (value
& MSR_IA32_APICBASE_ENABLE
)
1461 static_key_slow_dec_deferred(&apic_hw_disabled
);
1463 static_key_slow_inc(&apic_hw_disabled
.key
);
1464 recalculate_apic_map(vcpu
->kvm
);
1467 if ((old_value
^ value
) & X2APIC_ENABLE
) {
1468 if (value
& X2APIC_ENABLE
) {
1469 u32 id
= kvm_apic_id(apic
);
1470 u32 ldr
= ((id
>> 4) << 16) | (1 << (id
& 0xf));
1471 kvm_apic_set_ldr(apic
, ldr
);
1472 kvm_x86_ops
->set_virtual_x2apic_mode(vcpu
, true);
1474 kvm_x86_ops
->set_virtual_x2apic_mode(vcpu
, false);
1477 apic
->base_address
= apic
->vcpu
->arch
.apic_base
&
1478 MSR_IA32_APICBASE_BASE
;
1480 if ((value
& MSR_IA32_APICBASE_ENABLE
) &&
1481 apic
->base_address
!= APIC_DEFAULT_PHYS_BASE
)
1482 pr_warn_once("APIC base relocation is unsupported by KVM");
1484 /* with FSB delivery interrupt, we can restart APIC functionality */
1485 apic_debug("apic base msr is 0x%016" PRIx64
", and base address is "
1486 "0x%lx.\n", apic
->vcpu
->arch
.apic_base
, apic
->base_address
);
1490 void kvm_lapic_reset(struct kvm_vcpu
*vcpu
)
1492 struct kvm_lapic
*apic
;
1495 apic_debug("%s\n", __func__
);
1498 apic
= vcpu
->arch
.apic
;
1499 ASSERT(apic
!= NULL
);
1501 /* Stop the timer in case it's a reset to an active apic */
1502 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1504 kvm_apic_set_id(apic
, vcpu
->vcpu_id
);
1505 kvm_apic_set_version(apic
->vcpu
);
1507 for (i
= 0; i
< APIC_LVT_NUM
; i
++)
1508 apic_set_reg(apic
, APIC_LVTT
+ 0x10 * i
, APIC_LVT_MASKED
);
1509 apic
->lapic_timer
.timer_mode
= 0;
1510 apic_set_reg(apic
, APIC_LVT0
,
1511 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT
));
1513 apic_set_reg(apic
, APIC_DFR
, 0xffffffffU
);
1514 apic_set_spiv(apic
, 0xff);
1515 apic_set_reg(apic
, APIC_TASKPRI
, 0);
1516 kvm_apic_set_ldr(apic
, 0);
1517 apic_set_reg(apic
, APIC_ESR
, 0);
1518 apic_set_reg(apic
, APIC_ICR
, 0);
1519 apic_set_reg(apic
, APIC_ICR2
, 0);
1520 apic_set_reg(apic
, APIC_TDCR
, 0);
1521 apic_set_reg(apic
, APIC_TMICT
, 0);
1522 for (i
= 0; i
< 8; i
++) {
1523 apic_set_reg(apic
, APIC_IRR
+ 0x10 * i
, 0);
1524 apic_set_reg(apic
, APIC_ISR
+ 0x10 * i
, 0);
1525 apic_set_reg(apic
, APIC_TMR
+ 0x10 * i
, 0);
1527 apic
->irr_pending
= kvm_apic_vid_enabled(vcpu
->kvm
);
1528 apic
->isr_count
= kvm_apic_vid_enabled(vcpu
->kvm
);
1529 apic
->highest_isr_cache
= -1;
1530 update_divide_count(apic
);
1531 atomic_set(&apic
->lapic_timer
.pending
, 0);
1532 if (kvm_vcpu_is_bsp(vcpu
))
1533 kvm_lapic_set_base(vcpu
,
1534 vcpu
->arch
.apic_base
| MSR_IA32_APICBASE_BSP
);
1535 vcpu
->arch
.pv_eoi
.msr_val
= 0;
1536 apic_update_ppr(apic
);
1538 vcpu
->arch
.apic_arb_prio
= 0;
1539 vcpu
->arch
.apic_attention
= 0;
1541 apic_debug("%s: vcpu=%p, id=%d, base_msr="
1542 "0x%016" PRIx64
", base_address=0x%0lx.\n", __func__
,
1543 vcpu
, kvm_apic_id(apic
),
1544 vcpu
->arch
.apic_base
, apic
->base_address
);
1548 *----------------------------------------------------------------------
1550 *----------------------------------------------------------------------
1553 static bool lapic_is_periodic(struct kvm_lapic
*apic
)
1555 return apic_lvtt_period(apic
);
1558 int apic_has_pending_timer(struct kvm_vcpu
*vcpu
)
1560 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1562 if (kvm_vcpu_has_lapic(vcpu
) && apic_enabled(apic
) &&
1563 apic_lvt_enabled(apic
, APIC_LVTT
))
1564 return atomic_read(&apic
->lapic_timer
.pending
);
1569 int kvm_apic_local_deliver(struct kvm_lapic
*apic
, int lvt_type
)
1571 u32 reg
= kvm_apic_get_reg(apic
, lvt_type
);
1572 int vector
, mode
, trig_mode
;
1574 if (kvm_apic_hw_enabled(apic
) && !(reg
& APIC_LVT_MASKED
)) {
1575 vector
= reg
& APIC_VECTOR_MASK
;
1576 mode
= reg
& APIC_MODE_MASK
;
1577 trig_mode
= reg
& APIC_LVT_LEVEL_TRIGGER
;
1578 return __apic_accept_irq(apic
, mode
, vector
, 1, trig_mode
,
1584 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu
*vcpu
)
1586 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1589 kvm_apic_local_deliver(apic
, APIC_LVT0
);
1592 static const struct kvm_io_device_ops apic_mmio_ops
= {
1593 .read
= apic_mmio_read
,
1594 .write
= apic_mmio_write
,
1597 static enum hrtimer_restart
apic_timer_fn(struct hrtimer
*data
)
1599 struct kvm_timer
*ktimer
= container_of(data
, struct kvm_timer
, timer
);
1600 struct kvm_lapic
*apic
= container_of(ktimer
, struct kvm_lapic
, lapic_timer
);
1602 apic_timer_expired(apic
);
1604 if (lapic_is_periodic(apic
)) {
1605 hrtimer_add_expires_ns(&ktimer
->timer
, ktimer
->period
);
1606 return HRTIMER_RESTART
;
1608 return HRTIMER_NORESTART
;
1611 int kvm_create_lapic(struct kvm_vcpu
*vcpu
)
1613 struct kvm_lapic
*apic
;
1615 ASSERT(vcpu
!= NULL
);
1616 apic_debug("apic_init %d\n", vcpu
->vcpu_id
);
1618 apic
= kzalloc(sizeof(*apic
), GFP_KERNEL
);
1622 vcpu
->arch
.apic
= apic
;
1624 apic
->regs
= (void *)get_zeroed_page(GFP_KERNEL
);
1626 printk(KERN_ERR
"malloc apic regs error for vcpu %x\n",
1628 goto nomem_free_apic
;
1632 hrtimer_init(&apic
->lapic_timer
.timer
, CLOCK_MONOTONIC
,
1634 apic
->lapic_timer
.timer
.function
= apic_timer_fn
;
1637 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1638 * thinking that APIC satet has changed.
1640 vcpu
->arch
.apic_base
= MSR_IA32_APICBASE_ENABLE
;
1641 kvm_lapic_set_base(vcpu
,
1642 APIC_DEFAULT_PHYS_BASE
| MSR_IA32_APICBASE_ENABLE
);
1644 static_key_slow_inc(&apic_sw_disabled
.key
); /* sw disabled at reset */
1645 kvm_lapic_reset(vcpu
);
1646 kvm_iodevice_init(&apic
->dev
, &apic_mmio_ops
);
1655 int kvm_apic_has_interrupt(struct kvm_vcpu
*vcpu
)
1657 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1660 if (!kvm_vcpu_has_lapic(vcpu
) || !apic_enabled(apic
))
1663 apic_update_ppr(apic
);
1664 highest_irr
= apic_find_highest_irr(apic
);
1665 if ((highest_irr
== -1) ||
1666 ((highest_irr
& 0xF0) <= kvm_apic_get_reg(apic
, APIC_PROCPRI
)))
1671 int kvm_apic_accept_pic_intr(struct kvm_vcpu
*vcpu
)
1673 u32 lvt0
= kvm_apic_get_reg(vcpu
->arch
.apic
, APIC_LVT0
);
1676 if (!kvm_apic_hw_enabled(vcpu
->arch
.apic
))
1678 if ((lvt0
& APIC_LVT_MASKED
) == 0 &&
1679 GET_APIC_DELIVERY_MODE(lvt0
) == APIC_MODE_EXTINT
)
1684 void kvm_inject_apic_timer_irqs(struct kvm_vcpu
*vcpu
)
1686 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1688 if (!kvm_vcpu_has_lapic(vcpu
))
1691 if (atomic_read(&apic
->lapic_timer
.pending
) > 0) {
1692 kvm_apic_local_deliver(apic
, APIC_LVTT
);
1693 if (apic_lvtt_tscdeadline(apic
))
1694 apic
->lapic_timer
.tscdeadline
= 0;
1695 atomic_set(&apic
->lapic_timer
.pending
, 0);
1699 int kvm_get_apic_interrupt(struct kvm_vcpu
*vcpu
)
1701 int vector
= kvm_apic_has_interrupt(vcpu
);
1702 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1708 * We get here even with APIC virtualization enabled, if doing
1709 * nested virtualization and L1 runs with the "acknowledge interrupt
1710 * on exit" mode. Then we cannot inject the interrupt via RVI,
1711 * because the process would deliver it through the IDT.
1714 apic_set_isr(vector
, apic
);
1715 apic_update_ppr(apic
);
1716 apic_clear_irr(vector
, apic
);
1720 void kvm_apic_post_state_restore(struct kvm_vcpu
*vcpu
,
1721 struct kvm_lapic_state
*s
)
1723 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1725 kvm_lapic_set_base(vcpu
, vcpu
->arch
.apic_base
);
1726 /* set SPIV separately to get count of SW disabled APICs right */
1727 apic_set_spiv(apic
, *((u32
*)(s
->regs
+ APIC_SPIV
)));
1728 memcpy(vcpu
->arch
.apic
->regs
, s
->regs
, sizeof *s
);
1729 /* call kvm_apic_set_id() to put apic into apic_map */
1730 kvm_apic_set_id(apic
, kvm_apic_id(apic
));
1731 kvm_apic_set_version(vcpu
);
1733 apic_update_ppr(apic
);
1734 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1735 update_divide_count(apic
);
1736 start_apic_timer(apic
);
1737 apic
->irr_pending
= true;
1738 apic
->isr_count
= kvm_apic_vid_enabled(vcpu
->kvm
) ?
1739 1 : count_vectors(apic
->regs
+ APIC_ISR
);
1740 apic
->highest_isr_cache
= -1;
1741 if (kvm_x86_ops
->hwapic_irr_update
)
1742 kvm_x86_ops
->hwapic_irr_update(vcpu
,
1743 apic_find_highest_irr(apic
));
1744 kvm_x86_ops
->hwapic_isr_update(vcpu
->kvm
, apic_find_highest_isr(apic
));
1745 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
1746 kvm_rtc_eoi_tracking_restore_one(vcpu
);
1749 void __kvm_migrate_apic_timer(struct kvm_vcpu
*vcpu
)
1751 struct hrtimer
*timer
;
1753 if (!kvm_vcpu_has_lapic(vcpu
))
1756 timer
= &vcpu
->arch
.apic
->lapic_timer
.timer
;
1757 if (hrtimer_cancel(timer
))
1758 hrtimer_start_expires(timer
, HRTIMER_MODE_ABS
);
1762 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1764 * Detect whether guest triggered PV EOI since the
1765 * last entry. If yes, set EOI on guests's behalf.
1766 * Clear PV EOI in guest memory in any case.
1768 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu
*vcpu
,
1769 struct kvm_lapic
*apic
)
1774 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1775 * and KVM_PV_EOI_ENABLED in guest memory as follows:
1777 * KVM_APIC_PV_EOI_PENDING is unset:
1778 * -> host disabled PV EOI.
1779 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1780 * -> host enabled PV EOI, guest did not execute EOI yet.
1781 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1782 * -> host enabled PV EOI, guest executed EOI.
1784 BUG_ON(!pv_eoi_enabled(vcpu
));
1785 pending
= pv_eoi_get_pending(vcpu
);
1787 * Clear pending bit in any case: it will be set again on vmentry.
1788 * While this might not be ideal from performance point of view,
1789 * this makes sure pv eoi is only enabled when we know it's safe.
1791 pv_eoi_clr_pending(vcpu
);
1794 vector
= apic_set_eoi(apic
);
1795 trace_kvm_pv_eoi(apic
, vector
);
1798 void kvm_lapic_sync_from_vapic(struct kvm_vcpu
*vcpu
)
1802 if (test_bit(KVM_APIC_PV_EOI_PENDING
, &vcpu
->arch
.apic_attention
))
1803 apic_sync_pv_eoi_from_guest(vcpu
, vcpu
->arch
.apic
);
1805 if (!test_bit(KVM_APIC_CHECK_VAPIC
, &vcpu
->arch
.apic_attention
))
1808 kvm_read_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apic
->vapic_cache
, &data
,
1811 apic_set_tpr(vcpu
->arch
.apic
, data
& 0xff);
1815 * apic_sync_pv_eoi_to_guest - called before vmentry
1817 * Detect whether it's safe to enable PV EOI and
1820 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu
*vcpu
,
1821 struct kvm_lapic
*apic
)
1823 if (!pv_eoi_enabled(vcpu
) ||
1824 /* IRR set or many bits in ISR: could be nested. */
1825 apic
->irr_pending
||
1826 /* Cache not set: could be safe but we don't bother. */
1827 apic
->highest_isr_cache
== -1 ||
1828 /* Need EOI to update ioapic. */
1829 kvm_ioapic_handles_vector(vcpu
->kvm
, apic
->highest_isr_cache
)) {
1831 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1832 * so we need not do anything here.
1837 pv_eoi_set_pending(apic
->vcpu
);
1840 void kvm_lapic_sync_to_vapic(struct kvm_vcpu
*vcpu
)
1843 int max_irr
, max_isr
;
1844 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1846 apic_sync_pv_eoi_to_guest(vcpu
, apic
);
1848 if (!test_bit(KVM_APIC_CHECK_VAPIC
, &vcpu
->arch
.apic_attention
))
1851 tpr
= kvm_apic_get_reg(apic
, APIC_TASKPRI
) & 0xff;
1852 max_irr
= apic_find_highest_irr(apic
);
1855 max_isr
= apic_find_highest_isr(apic
);
1858 data
= (tpr
& 0xff) | ((max_isr
& 0xf0) << 8) | (max_irr
<< 24);
1860 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apic
->vapic_cache
, &data
,
1864 int kvm_lapic_set_vapic_addr(struct kvm_vcpu
*vcpu
, gpa_t vapic_addr
)
1867 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
,
1868 &vcpu
->arch
.apic
->vapic_cache
,
1869 vapic_addr
, sizeof(u32
)))
1871 __set_bit(KVM_APIC_CHECK_VAPIC
, &vcpu
->arch
.apic_attention
);
1873 __clear_bit(KVM_APIC_CHECK_VAPIC
, &vcpu
->arch
.apic_attention
);
1876 vcpu
->arch
.apic
->vapic_addr
= vapic_addr
;
1880 int kvm_x2apic_msr_write(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1882 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1883 u32 reg
= (msr
- APIC_BASE_MSR
) << 4;
1885 if (!irqchip_in_kernel(vcpu
->kvm
) || !apic_x2apic_mode(apic
))
1888 if (reg
== APIC_ICR2
)
1891 /* if this is ICR write vector before command */
1892 if (reg
== APIC_ICR
)
1893 apic_reg_write(apic
, APIC_ICR2
, (u32
)(data
>> 32));
1894 return apic_reg_write(apic
, reg
, (u32
)data
);
1897 int kvm_x2apic_msr_read(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*data
)
1899 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1900 u32 reg
= (msr
- APIC_BASE_MSR
) << 4, low
, high
= 0;
1902 if (!irqchip_in_kernel(vcpu
->kvm
) || !apic_x2apic_mode(apic
))
1905 if (reg
== APIC_DFR
|| reg
== APIC_ICR2
) {
1906 apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
1911 if (apic_reg_read(apic
, reg
, 4, &low
))
1913 if (reg
== APIC_ICR
)
1914 apic_reg_read(apic
, APIC_ICR2
, 4, &high
);
1916 *data
= (((u64
)high
) << 32) | low
;
1921 int kvm_hv_vapic_msr_write(struct kvm_vcpu
*vcpu
, u32 reg
, u64 data
)
1923 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1925 if (!kvm_vcpu_has_lapic(vcpu
))
1928 /* if this is ICR write vector before command */
1929 if (reg
== APIC_ICR
)
1930 apic_reg_write(apic
, APIC_ICR2
, (u32
)(data
>> 32));
1931 return apic_reg_write(apic
, reg
, (u32
)data
);
1934 int kvm_hv_vapic_msr_read(struct kvm_vcpu
*vcpu
, u32 reg
, u64
*data
)
1936 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1939 if (!kvm_vcpu_has_lapic(vcpu
))
1942 if (apic_reg_read(apic
, reg
, 4, &low
))
1944 if (reg
== APIC_ICR
)
1945 apic_reg_read(apic
, APIC_ICR2
, 4, &high
);
1947 *data
= (((u64
)high
) << 32) | low
;
1952 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu
*vcpu
, u64 data
)
1954 u64 addr
= data
& ~KVM_MSR_ENABLED
;
1955 if (!IS_ALIGNED(addr
, 4))
1958 vcpu
->arch
.pv_eoi
.msr_val
= data
;
1959 if (!pv_eoi_enabled(vcpu
))
1961 return kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.pv_eoi
.data
,
1965 void kvm_apic_accept_events(struct kvm_vcpu
*vcpu
)
1967 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1971 if (!kvm_vcpu_has_lapic(vcpu
) || !apic
->pending_events
)
1974 pe
= xchg(&apic
->pending_events
, 0);
1976 if (test_bit(KVM_APIC_INIT
, &pe
)) {
1977 kvm_lapic_reset(vcpu
);
1978 kvm_vcpu_reset(vcpu
);
1979 if (kvm_vcpu_is_bsp(apic
->vcpu
))
1980 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
1982 vcpu
->arch
.mp_state
= KVM_MP_STATE_INIT_RECEIVED
;
1984 if (test_bit(KVM_APIC_SIPI
, &pe
) &&
1985 vcpu
->arch
.mp_state
== KVM_MP_STATE_INIT_RECEIVED
) {
1986 /* evaluate pending_events before reading the vector */
1988 sipi_vector
= apic
->sipi_vector
;
1989 apic_debug("vcpu %d received sipi with vector # %x\n",
1990 vcpu
->vcpu_id
, sipi_vector
);
1991 kvm_vcpu_deliver_sipi_vector(vcpu
, sipi_vector
);
1992 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
1996 void kvm_lapic_init(void)
1998 /* do not patch jump label more than once per second */
1999 jump_label_rate_limit(&apic_hw_disabled
, HZ
);
2000 jump_label_rate_limit(&apic_sw_disabled
, HZ
);