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/export.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 static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map
*map
,
119 u32 dest_id
, struct kvm_lapic
***cluster
, u16
*mask
) {
121 case KVM_APIC_MODE_X2APIC
: {
122 u32 offset
= (dest_id
>> 16) * 16;
123 u32 max_apic_id
= map
->max_apic_id
;
125 if (offset
<= max_apic_id
) {
126 u8 cluster_size
= min(max_apic_id
- offset
+ 1, 16U);
128 *cluster
= &map
->phys_map
[offset
];
129 *mask
= dest_id
& (0xffff >> (16 - cluster_size
));
136 case KVM_APIC_MODE_XAPIC_FLAT
:
137 *cluster
= map
->xapic_flat_map
;
138 *mask
= dest_id
& 0xff;
140 case KVM_APIC_MODE_XAPIC_CLUSTER
:
141 *cluster
= map
->xapic_cluster_map
[(dest_id
>> 4) & 0xf];
142 *mask
= dest_id
& 0xf;
150 static void kvm_apic_map_free(struct rcu_head
*rcu
)
152 struct kvm_apic_map
*map
= container_of(rcu
, struct kvm_apic_map
, rcu
);
157 static void recalculate_apic_map(struct kvm
*kvm
)
159 struct kvm_apic_map
*new, *old
= NULL
;
160 struct kvm_vcpu
*vcpu
;
164 mutex_lock(&kvm
->arch
.apic_map_lock
);
166 kvm_for_each_vcpu(i
, vcpu
, kvm
)
167 if (kvm_apic_present(vcpu
))
168 max_id
= max(max_id
, kvm_apic_id(vcpu
->arch
.apic
));
170 new = kvm_kvzalloc(sizeof(struct kvm_apic_map
) +
171 sizeof(struct kvm_lapic
*) * ((u64
)max_id
+ 1));
176 new->max_apic_id
= max_id
;
178 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
179 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
180 struct kvm_lapic
**cluster
;
184 if (!kvm_apic_present(vcpu
))
187 aid
= kvm_apic_id(apic
);
188 ldr
= kvm_lapic_get_reg(apic
, APIC_LDR
);
190 if (aid
<= new->max_apic_id
)
191 new->phys_map
[aid
] = apic
;
193 if (apic_x2apic_mode(apic
)) {
194 new->mode
|= KVM_APIC_MODE_X2APIC
;
196 ldr
= GET_APIC_LOGICAL_ID(ldr
);
197 if (kvm_lapic_get_reg(apic
, APIC_DFR
) == APIC_DFR_FLAT
)
198 new->mode
|= KVM_APIC_MODE_XAPIC_FLAT
;
200 new->mode
|= KVM_APIC_MODE_XAPIC_CLUSTER
;
203 if (!kvm_apic_map_get_logical_dest(new, ldr
, &cluster
, &mask
))
207 cluster
[ffs(mask
) - 1] = apic
;
210 old
= rcu_dereference_protected(kvm
->arch
.apic_map
,
211 lockdep_is_held(&kvm
->arch
.apic_map_lock
));
212 rcu_assign_pointer(kvm
->arch
.apic_map
, new);
213 mutex_unlock(&kvm
->arch
.apic_map_lock
);
216 call_rcu(&old
->rcu
, kvm_apic_map_free
);
218 kvm_make_scan_ioapic_request(kvm
);
221 static inline void apic_set_spiv(struct kvm_lapic
*apic
, u32 val
)
223 bool enabled
= val
& APIC_SPIV_APIC_ENABLED
;
225 kvm_lapic_set_reg(apic
, APIC_SPIV
, val
);
227 if (enabled
!= apic
->sw_enabled
) {
228 apic
->sw_enabled
= enabled
;
230 static_key_slow_dec_deferred(&apic_sw_disabled
);
231 recalculate_apic_map(apic
->vcpu
->kvm
);
233 static_key_slow_inc(&apic_sw_disabled
.key
);
237 static inline void kvm_apic_set_xapic_id(struct kvm_lapic
*apic
, u8 id
)
239 kvm_lapic_set_reg(apic
, APIC_ID
, id
<< 24);
240 recalculate_apic_map(apic
->vcpu
->kvm
);
243 static inline void kvm_apic_set_ldr(struct kvm_lapic
*apic
, u32 id
)
245 kvm_lapic_set_reg(apic
, APIC_LDR
, id
);
246 recalculate_apic_map(apic
->vcpu
->kvm
);
249 static inline void kvm_apic_set_x2apic_id(struct kvm_lapic
*apic
, u32 id
)
251 u32 ldr
= ((id
>> 4) << 16) | (1 << (id
& 0xf));
253 kvm_lapic_set_reg(apic
, APIC_ID
, id
);
254 kvm_lapic_set_reg(apic
, APIC_LDR
, ldr
);
255 recalculate_apic_map(apic
->vcpu
->kvm
);
258 static inline int apic_lvt_enabled(struct kvm_lapic
*apic
, int lvt_type
)
260 return !(kvm_lapic_get_reg(apic
, lvt_type
) & APIC_LVT_MASKED
);
263 static inline int apic_lvt_vector(struct kvm_lapic
*apic
, int lvt_type
)
265 return kvm_lapic_get_reg(apic
, lvt_type
) & APIC_VECTOR_MASK
;
268 static inline int apic_lvtt_oneshot(struct kvm_lapic
*apic
)
270 return apic
->lapic_timer
.timer_mode
== APIC_LVT_TIMER_ONESHOT
;
273 static inline int apic_lvtt_period(struct kvm_lapic
*apic
)
275 return apic
->lapic_timer
.timer_mode
== APIC_LVT_TIMER_PERIODIC
;
278 static inline int apic_lvtt_tscdeadline(struct kvm_lapic
*apic
)
280 return apic
->lapic_timer
.timer_mode
== APIC_LVT_TIMER_TSCDEADLINE
;
283 static inline int apic_lvt_nmi_mode(u32 lvt_val
)
285 return (lvt_val
& (APIC_MODE_MASK
| APIC_LVT_MASKED
)) == APIC_DM_NMI
;
288 void kvm_apic_set_version(struct kvm_vcpu
*vcpu
)
290 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
291 struct kvm_cpuid_entry2
*feat
;
292 u32 v
= APIC_VERSION
;
294 if (!lapic_in_kernel(vcpu
))
297 feat
= kvm_find_cpuid_entry(apic
->vcpu
, 0x1, 0);
298 if (feat
&& (feat
->ecx
& (1 << (X86_FEATURE_X2APIC
& 31))))
299 v
|= APIC_LVR_DIRECTED_EOI
;
300 kvm_lapic_set_reg(apic
, APIC_LVR
, v
);
303 static const unsigned int apic_lvt_mask
[KVM_APIC_LVT_NUM
] = {
304 LVT_MASK
, /* part LVTT mask, timer mode mask added at runtime */
305 LVT_MASK
| APIC_MODE_MASK
, /* LVTTHMR */
306 LVT_MASK
| APIC_MODE_MASK
, /* LVTPC */
307 LINT_MASK
, LINT_MASK
, /* LVT0-1 */
308 LVT_MASK
/* LVTERR */
311 static int find_highest_vector(void *bitmap
)
316 for (vec
= MAX_APIC_VECTOR
- APIC_VECTORS_PER_REG
;
317 vec
>= 0; vec
-= APIC_VECTORS_PER_REG
) {
318 reg
= bitmap
+ REG_POS(vec
);
320 return fls(*reg
) - 1 + vec
;
326 static u8
count_vectors(void *bitmap
)
332 for (vec
= 0; vec
< MAX_APIC_VECTOR
; vec
+= APIC_VECTORS_PER_REG
) {
333 reg
= bitmap
+ REG_POS(vec
);
334 count
+= hweight32(*reg
);
340 void __kvm_apic_update_irr(u32
*pir
, void *regs
)
344 for (i
= 0; i
<= 7; i
++) {
345 pir_val
= READ_ONCE(pir
[i
]);
347 pir_val
= xchg(&pir
[i
], 0);
348 *((u32
*)(regs
+ APIC_IRR
+ i
* 0x10)) |= pir_val
;
352 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr
);
354 void kvm_apic_update_irr(struct kvm_vcpu
*vcpu
, u32
*pir
)
356 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
358 __kvm_apic_update_irr(pir
, apic
->regs
);
360 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
362 EXPORT_SYMBOL_GPL(kvm_apic_update_irr
);
364 static inline int apic_search_irr(struct kvm_lapic
*apic
)
366 return find_highest_vector(apic
->regs
+ APIC_IRR
);
369 static inline int apic_find_highest_irr(struct kvm_lapic
*apic
)
374 * Note that irr_pending is just a hint. It will be always
375 * true with virtual interrupt delivery enabled.
377 if (!apic
->irr_pending
)
380 if (apic
->vcpu
->arch
.apicv_active
)
381 kvm_x86_ops
->sync_pir_to_irr(apic
->vcpu
);
382 result
= apic_search_irr(apic
);
383 ASSERT(result
== -1 || result
>= 16);
388 static inline void apic_clear_irr(int vec
, struct kvm_lapic
*apic
)
390 struct kvm_vcpu
*vcpu
;
394 if (unlikely(vcpu
->arch
.apicv_active
)) {
395 /* try to update RVI */
396 apic_clear_vector(vec
, apic
->regs
+ APIC_IRR
);
397 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
399 apic
->irr_pending
= false;
400 apic_clear_vector(vec
, apic
->regs
+ APIC_IRR
);
401 if (apic_search_irr(apic
) != -1)
402 apic
->irr_pending
= true;
406 static inline void apic_set_isr(int vec
, struct kvm_lapic
*apic
)
408 struct kvm_vcpu
*vcpu
;
410 if (__apic_test_and_set_vector(vec
, apic
->regs
+ APIC_ISR
))
416 * With APIC virtualization enabled, all caching is disabled
417 * because the processor can modify ISR under the hood. Instead
420 if (unlikely(vcpu
->arch
.apicv_active
))
421 kvm_x86_ops
->hwapic_isr_update(vcpu
, vec
);
424 BUG_ON(apic
->isr_count
> MAX_APIC_VECTOR
);
426 * ISR (in service register) bit is set when injecting an interrupt.
427 * The highest vector is injected. Thus the latest bit set matches
428 * the highest bit in ISR.
430 apic
->highest_isr_cache
= vec
;
434 static inline int apic_find_highest_isr(struct kvm_lapic
*apic
)
439 * Note that isr_count is always 1, and highest_isr_cache
440 * is always -1, with APIC virtualization enabled.
442 if (!apic
->isr_count
)
444 if (likely(apic
->highest_isr_cache
!= -1))
445 return apic
->highest_isr_cache
;
447 result
= find_highest_vector(apic
->regs
+ APIC_ISR
);
448 ASSERT(result
== -1 || result
>= 16);
453 static inline void apic_clear_isr(int vec
, struct kvm_lapic
*apic
)
455 struct kvm_vcpu
*vcpu
;
456 if (!__apic_test_and_clear_vector(vec
, apic
->regs
+ APIC_ISR
))
462 * We do get here for APIC virtualization enabled if the guest
463 * uses the Hyper-V APIC enlightenment. In this case we may need
464 * to trigger a new interrupt delivery by writing the SVI field;
465 * on the other hand isr_count and highest_isr_cache are unused
466 * and must be left alone.
468 if (unlikely(vcpu
->arch
.apicv_active
))
469 kvm_x86_ops
->hwapic_isr_update(vcpu
,
470 apic_find_highest_isr(apic
));
473 BUG_ON(apic
->isr_count
< 0);
474 apic
->highest_isr_cache
= -1;
478 int kvm_lapic_find_highest_irr(struct kvm_vcpu
*vcpu
)
480 /* This may race with setting of irr in __apic_accept_irq() and
481 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
482 * will cause vmexit immediately and the value will be recalculated
483 * on the next vmentry.
485 return apic_find_highest_irr(vcpu
->arch
.apic
);
488 static int __apic_accept_irq(struct kvm_lapic
*apic
, int delivery_mode
,
489 int vector
, int level
, int trig_mode
,
490 struct dest_map
*dest_map
);
492 int kvm_apic_set_irq(struct kvm_vcpu
*vcpu
, struct kvm_lapic_irq
*irq
,
493 struct dest_map
*dest_map
)
495 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
497 return __apic_accept_irq(apic
, irq
->delivery_mode
, irq
->vector
,
498 irq
->level
, irq
->trig_mode
, dest_map
);
501 static int pv_eoi_put_user(struct kvm_vcpu
*vcpu
, u8 val
)
504 return kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.pv_eoi
.data
, &val
,
508 static int pv_eoi_get_user(struct kvm_vcpu
*vcpu
, u8
*val
)
511 return kvm_read_guest_cached(vcpu
->kvm
, &vcpu
->arch
.pv_eoi
.data
, val
,
515 static inline bool pv_eoi_enabled(struct kvm_vcpu
*vcpu
)
517 return vcpu
->arch
.pv_eoi
.msr_val
& KVM_MSR_ENABLED
;
520 static bool pv_eoi_get_pending(struct kvm_vcpu
*vcpu
)
523 if (pv_eoi_get_user(vcpu
, &val
) < 0)
524 apic_debug("Can't read EOI MSR value: 0x%llx\n",
525 (unsigned long long)vcpu
->arch
.pv_eoi
.msr_val
);
529 static void pv_eoi_set_pending(struct kvm_vcpu
*vcpu
)
531 if (pv_eoi_put_user(vcpu
, KVM_PV_EOI_ENABLED
) < 0) {
532 apic_debug("Can't set EOI MSR value: 0x%llx\n",
533 (unsigned long long)vcpu
->arch
.pv_eoi
.msr_val
);
536 __set_bit(KVM_APIC_PV_EOI_PENDING
, &vcpu
->arch
.apic_attention
);
539 static void pv_eoi_clr_pending(struct kvm_vcpu
*vcpu
)
541 if (pv_eoi_put_user(vcpu
, KVM_PV_EOI_DISABLED
) < 0) {
542 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
543 (unsigned long long)vcpu
->arch
.pv_eoi
.msr_val
);
546 __clear_bit(KVM_APIC_PV_EOI_PENDING
, &vcpu
->arch
.apic_attention
);
549 static void apic_update_ppr(struct kvm_lapic
*apic
)
551 u32 tpr
, isrv
, ppr
, old_ppr
;
554 old_ppr
= kvm_lapic_get_reg(apic
, APIC_PROCPRI
);
555 tpr
= kvm_lapic_get_reg(apic
, APIC_TASKPRI
);
556 isr
= apic_find_highest_isr(apic
);
557 isrv
= (isr
!= -1) ? isr
: 0;
559 if ((tpr
& 0xf0) >= (isrv
& 0xf0))
564 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
565 apic
, ppr
, isr
, isrv
);
567 if (old_ppr
!= ppr
) {
568 kvm_lapic_set_reg(apic
, APIC_PROCPRI
, ppr
);
570 kvm_make_request(KVM_REQ_EVENT
, apic
->vcpu
);
574 static void apic_set_tpr(struct kvm_lapic
*apic
, u32 tpr
)
576 kvm_lapic_set_reg(apic
, APIC_TASKPRI
, tpr
);
577 apic_update_ppr(apic
);
580 static bool kvm_apic_broadcast(struct kvm_lapic
*apic
, u32 mda
)
582 if (apic_x2apic_mode(apic
))
583 return mda
== X2APIC_BROADCAST
;
585 return GET_APIC_DEST_FIELD(mda
) == APIC_BROADCAST
;
588 static bool kvm_apic_match_physical_addr(struct kvm_lapic
*apic
, u32 mda
)
590 if (kvm_apic_broadcast(apic
, mda
))
593 if (apic_x2apic_mode(apic
))
594 return mda
== kvm_apic_id(apic
);
596 return mda
== SET_APIC_DEST_FIELD(kvm_apic_id(apic
));
599 static bool kvm_apic_match_logical_addr(struct kvm_lapic
*apic
, u32 mda
)
603 if (kvm_apic_broadcast(apic
, mda
))
606 logical_id
= kvm_lapic_get_reg(apic
, APIC_LDR
);
608 if (apic_x2apic_mode(apic
))
609 return ((logical_id
>> 16) == (mda
>> 16))
610 && (logical_id
& mda
& 0xffff) != 0;
612 logical_id
= GET_APIC_LOGICAL_ID(logical_id
);
613 mda
= GET_APIC_DEST_FIELD(mda
);
615 switch (kvm_lapic_get_reg(apic
, APIC_DFR
)) {
617 return (logical_id
& mda
) != 0;
618 case APIC_DFR_CLUSTER
:
619 return ((logical_id
>> 4) == (mda
>> 4))
620 && (logical_id
& mda
& 0xf) != 0;
622 apic_debug("Bad DFR vcpu %d: %08x\n",
623 apic
->vcpu
->vcpu_id
, kvm_lapic_get_reg(apic
, APIC_DFR
));
628 /* The KVM local APIC implementation has two quirks:
630 * - the xAPIC MDA stores the destination at bits 24-31, while this
631 * is not true of struct kvm_lapic_irq's dest_id field. This is
632 * just a quirk in the API and is not problematic.
634 * - in-kernel IOAPIC messages have to be delivered directly to
635 * x2APIC, because the kernel does not support interrupt remapping.
636 * In order to support broadcast without interrupt remapping, x2APIC
637 * rewrites the destination of non-IPI messages from APIC_BROADCAST
638 * to X2APIC_BROADCAST.
640 * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API. This is
641 * important when userspace wants to use x2APIC-format MSIs, because
642 * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
644 static u32
kvm_apic_mda(struct kvm_vcpu
*vcpu
, unsigned int dest_id
,
645 struct kvm_lapic
*source
, struct kvm_lapic
*target
)
647 bool ipi
= source
!= NULL
;
648 bool x2apic_mda
= apic_x2apic_mode(ipi
? source
: target
);
650 if (!vcpu
->kvm
->arch
.x2apic_broadcast_quirk_disabled
&&
651 !ipi
&& dest_id
== APIC_BROADCAST
&& x2apic_mda
)
652 return X2APIC_BROADCAST
;
654 return x2apic_mda
? dest_id
: SET_APIC_DEST_FIELD(dest_id
);
657 bool kvm_apic_match_dest(struct kvm_vcpu
*vcpu
, struct kvm_lapic
*source
,
658 int short_hand
, unsigned int dest
, int dest_mode
)
660 struct kvm_lapic
*target
= vcpu
->arch
.apic
;
661 u32 mda
= kvm_apic_mda(vcpu
, dest
, source
, target
);
663 apic_debug("target %p, source %p, dest 0x%x, "
664 "dest_mode 0x%x, short_hand 0x%x\n",
665 target
, source
, dest
, dest_mode
, short_hand
);
668 switch (short_hand
) {
669 case APIC_DEST_NOSHORT
:
670 if (dest_mode
== APIC_DEST_PHYSICAL
)
671 return kvm_apic_match_physical_addr(target
, mda
);
673 return kvm_apic_match_logical_addr(target
, mda
);
675 return target
== source
;
676 case APIC_DEST_ALLINC
:
678 case APIC_DEST_ALLBUT
:
679 return target
!= source
;
681 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
686 EXPORT_SYMBOL_GPL(kvm_apic_match_dest
);
688 int kvm_vector_to_index(u32 vector
, u32 dest_vcpus
,
689 const unsigned long *bitmap
, u32 bitmap_size
)
694 mod
= vector
% dest_vcpus
;
696 for (i
= 0; i
<= mod
; i
++) {
697 idx
= find_next_bit(bitmap
, bitmap_size
, idx
+ 1);
698 BUG_ON(idx
== bitmap_size
);
704 static void kvm_apic_disabled_lapic_found(struct kvm
*kvm
)
706 if (!kvm
->arch
.disabled_lapic_found
) {
707 kvm
->arch
.disabled_lapic_found
= true;
709 "Disabled LAPIC found during irq injection\n");
713 static bool kvm_apic_is_broadcast_dest(struct kvm
*kvm
, struct kvm_lapic
**src
,
714 struct kvm_lapic_irq
*irq
, struct kvm_apic_map
*map
)
716 if (kvm
->arch
.x2apic_broadcast_quirk_disabled
) {
717 if ((irq
->dest_id
== APIC_BROADCAST
&&
718 map
->mode
!= KVM_APIC_MODE_X2APIC
))
720 if (irq
->dest_id
== X2APIC_BROADCAST
)
723 bool x2apic_ipi
= src
&& *src
&& apic_x2apic_mode(*src
);
724 if (irq
->dest_id
== (x2apic_ipi
?
725 X2APIC_BROADCAST
: APIC_BROADCAST
))
732 /* Return true if the interrupt can be handled by using *bitmap as index mask
733 * for valid destinations in *dst array.
734 * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
735 * Note: we may have zero kvm_lapic destinations when we return true, which
736 * means that the interrupt should be dropped. In this case, *bitmap would be
737 * zero and *dst undefined.
739 static inline bool kvm_apic_map_get_dest_lapic(struct kvm
*kvm
,
740 struct kvm_lapic
**src
, struct kvm_lapic_irq
*irq
,
741 struct kvm_apic_map
*map
, struct kvm_lapic
***dst
,
742 unsigned long *bitmap
)
746 if (irq
->shorthand
== APIC_DEST_SELF
&& src
) {
750 } else if (irq
->shorthand
)
753 if (!map
|| kvm_apic_is_broadcast_dest(kvm
, src
, irq
, map
))
756 if (irq
->dest_mode
== APIC_DEST_PHYSICAL
) {
757 if (irq
->dest_id
> map
->max_apic_id
) {
760 *dst
= &map
->phys_map
[irq
->dest_id
];
767 if (!kvm_apic_map_get_logical_dest(map
, irq
->dest_id
, dst
,
771 if (!kvm_lowest_prio_delivery(irq
))
774 if (!kvm_vector_hashing_enabled()) {
776 for_each_set_bit(i
, bitmap
, 16) {
781 else if (kvm_apic_compare_prio((*dst
)[i
]->vcpu
,
782 (*dst
)[lowest
]->vcpu
) < 0)
789 lowest
= kvm_vector_to_index(irq
->vector
, hweight16(*bitmap
),
792 if (!(*dst
)[lowest
]) {
793 kvm_apic_disabled_lapic_found(kvm
);
799 *bitmap
= (lowest
>= 0) ? 1 << lowest
: 0;
804 bool kvm_irq_delivery_to_apic_fast(struct kvm
*kvm
, struct kvm_lapic
*src
,
805 struct kvm_lapic_irq
*irq
, int *r
, struct dest_map
*dest_map
)
807 struct kvm_apic_map
*map
;
808 unsigned long bitmap
;
809 struct kvm_lapic
**dst
= NULL
;
815 if (irq
->shorthand
== APIC_DEST_SELF
) {
816 *r
= kvm_apic_set_irq(src
->vcpu
, irq
, dest_map
);
821 map
= rcu_dereference(kvm
->arch
.apic_map
);
823 ret
= kvm_apic_map_get_dest_lapic(kvm
, &src
, irq
, map
, &dst
, &bitmap
);
825 for_each_set_bit(i
, &bitmap
, 16) {
830 *r
+= kvm_apic_set_irq(dst
[i
]->vcpu
, irq
, dest_map
);
838 * This routine tries to handler interrupts in posted mode, here is how
839 * it deals with different cases:
840 * - For single-destination interrupts, handle it in posted mode
841 * - Else if vector hashing is enabled and it is a lowest-priority
842 * interrupt, handle it in posted mode and use the following mechanism
843 * to find the destinaiton vCPU.
844 * 1. For lowest-priority interrupts, store all the possible
845 * destination vCPUs in an array.
846 * 2. Use "guest vector % max number of destination vCPUs" to find
847 * the right destination vCPU in the array for the lowest-priority
849 * - Otherwise, use remapped mode to inject the interrupt.
851 bool kvm_intr_is_single_vcpu_fast(struct kvm
*kvm
, struct kvm_lapic_irq
*irq
,
852 struct kvm_vcpu
**dest_vcpu
)
854 struct kvm_apic_map
*map
;
855 unsigned long bitmap
;
856 struct kvm_lapic
**dst
= NULL
;
863 map
= rcu_dereference(kvm
->arch
.apic_map
);
865 if (kvm_apic_map_get_dest_lapic(kvm
, NULL
, irq
, map
, &dst
, &bitmap
) &&
866 hweight16(bitmap
) == 1) {
867 unsigned long i
= find_first_bit(&bitmap
, 16);
870 *dest_vcpu
= dst
[i
]->vcpu
;
880 * Add a pending IRQ into lapic.
881 * Return 1 if successfully added and 0 if discarded.
883 static int __apic_accept_irq(struct kvm_lapic
*apic
, int delivery_mode
,
884 int vector
, int level
, int trig_mode
,
885 struct dest_map
*dest_map
)
888 struct kvm_vcpu
*vcpu
= apic
->vcpu
;
890 trace_kvm_apic_accept_irq(vcpu
->vcpu_id
, delivery_mode
,
892 switch (delivery_mode
) {
894 vcpu
->arch
.apic_arb_prio
++;
896 if (unlikely(trig_mode
&& !level
))
899 /* FIXME add logic for vcpu on reset */
900 if (unlikely(!apic_enabled(apic
)))
906 __set_bit(vcpu
->vcpu_id
, dest_map
->map
);
907 dest_map
->vectors
[vcpu
->vcpu_id
] = vector
;
910 if (apic_test_vector(vector
, apic
->regs
+ APIC_TMR
) != !!trig_mode
) {
912 kvm_lapic_set_vector(vector
, apic
->regs
+ APIC_TMR
);
914 apic_clear_vector(vector
, apic
->regs
+ APIC_TMR
);
917 if (vcpu
->arch
.apicv_active
)
918 kvm_x86_ops
->deliver_posted_interrupt(vcpu
, vector
);
920 kvm_lapic_set_irr(vector
, apic
);
922 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
929 vcpu
->arch
.pv
.pv_unhalted
= 1;
930 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
936 kvm_make_request(KVM_REQ_SMI
, vcpu
);
942 kvm_inject_nmi(vcpu
);
947 if (!trig_mode
|| level
) {
949 /* assumes that there are only KVM_APIC_INIT/SIPI */
950 apic
->pending_events
= (1UL << KVM_APIC_INIT
);
951 /* make sure pending_events is visible before sending
954 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
957 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
962 case APIC_DM_STARTUP
:
963 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
964 vcpu
->vcpu_id
, vector
);
966 apic
->sipi_vector
= vector
;
967 /* make sure sipi_vector is visible for the receiver */
969 set_bit(KVM_APIC_SIPI
, &apic
->pending_events
);
970 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
976 * Should only be called by kvm_apic_local_deliver() with LVT0,
977 * before NMI watchdog was enabled. Already handled by
978 * kvm_apic_accept_pic_intr().
983 printk(KERN_ERR
"TODO: unsupported delivery mode %x\n",
990 int kvm_apic_compare_prio(struct kvm_vcpu
*vcpu1
, struct kvm_vcpu
*vcpu2
)
992 return vcpu1
->arch
.apic_arb_prio
- vcpu2
->arch
.apic_arb_prio
;
995 static bool kvm_ioapic_handles_vector(struct kvm_lapic
*apic
, int vector
)
997 return test_bit(vector
, apic
->vcpu
->arch
.ioapic_handled_vectors
);
1000 static void kvm_ioapic_send_eoi(struct kvm_lapic
*apic
, int vector
)
1004 /* Eoi the ioapic only if the ioapic doesn't own the vector. */
1005 if (!kvm_ioapic_handles_vector(apic
, vector
))
1008 /* Request a KVM exit to inform the userspace IOAPIC. */
1009 if (irqchip_split(apic
->vcpu
->kvm
)) {
1010 apic
->vcpu
->arch
.pending_ioapic_eoi
= vector
;
1011 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT
, apic
->vcpu
);
1015 if (apic_test_vector(vector
, apic
->regs
+ APIC_TMR
))
1016 trigger_mode
= IOAPIC_LEVEL_TRIG
;
1018 trigger_mode
= IOAPIC_EDGE_TRIG
;
1020 kvm_ioapic_update_eoi(apic
->vcpu
, vector
, trigger_mode
);
1023 static int apic_set_eoi(struct kvm_lapic
*apic
)
1025 int vector
= apic_find_highest_isr(apic
);
1027 trace_kvm_eoi(apic
, vector
);
1030 * Not every write EOI will has corresponding ISR,
1031 * one example is when Kernel check timer on setup_IO_APIC
1036 apic_clear_isr(vector
, apic
);
1037 apic_update_ppr(apic
);
1039 if (test_bit(vector
, vcpu_to_synic(apic
->vcpu
)->vec_bitmap
))
1040 kvm_hv_synic_send_eoi(apic
->vcpu
, vector
);
1042 kvm_ioapic_send_eoi(apic
, vector
);
1043 kvm_make_request(KVM_REQ_EVENT
, apic
->vcpu
);
1048 * this interface assumes a trap-like exit, which has already finished
1049 * desired side effect including vISR and vPPR update.
1051 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu
*vcpu
, int vector
)
1053 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1055 trace_kvm_eoi(apic
, vector
);
1057 kvm_ioapic_send_eoi(apic
, vector
);
1058 kvm_make_request(KVM_REQ_EVENT
, apic
->vcpu
);
1060 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated
);
1062 static void apic_send_ipi(struct kvm_lapic
*apic
)
1064 u32 icr_low
= kvm_lapic_get_reg(apic
, APIC_ICR
);
1065 u32 icr_high
= kvm_lapic_get_reg(apic
, APIC_ICR2
);
1066 struct kvm_lapic_irq irq
;
1068 irq
.vector
= icr_low
& APIC_VECTOR_MASK
;
1069 irq
.delivery_mode
= icr_low
& APIC_MODE_MASK
;
1070 irq
.dest_mode
= icr_low
& APIC_DEST_MASK
;
1071 irq
.level
= (icr_low
& APIC_INT_ASSERT
) != 0;
1072 irq
.trig_mode
= icr_low
& APIC_INT_LEVELTRIG
;
1073 irq
.shorthand
= icr_low
& APIC_SHORT_MASK
;
1074 irq
.msi_redir_hint
= false;
1075 if (apic_x2apic_mode(apic
))
1076 irq
.dest_id
= icr_high
;
1078 irq
.dest_id
= GET_APIC_DEST_FIELD(icr_high
);
1080 trace_kvm_apic_ipi(icr_low
, irq
.dest_id
);
1082 apic_debug("icr_high 0x%x, icr_low 0x%x, "
1083 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
1084 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x, "
1085 "msi_redir_hint 0x%x\n",
1086 icr_high
, icr_low
, irq
.shorthand
, irq
.dest_id
,
1087 irq
.trig_mode
, irq
.level
, irq
.dest_mode
, irq
.delivery_mode
,
1088 irq
.vector
, irq
.msi_redir_hint
);
1090 kvm_irq_delivery_to_apic(apic
->vcpu
->kvm
, apic
, &irq
, NULL
);
1093 static u32
apic_get_tmcct(struct kvm_lapic
*apic
)
1095 ktime_t remaining
, now
;
1099 ASSERT(apic
!= NULL
);
1101 /* if initial count is 0, current count should also be 0 */
1102 if (kvm_lapic_get_reg(apic
, APIC_TMICT
) == 0 ||
1103 apic
->lapic_timer
.period
== 0)
1107 remaining
= ktime_sub(apic
->lapic_timer
.target_expiration
, now
);
1108 if (ktime_to_ns(remaining
) < 0)
1111 ns
= mod_64(ktime_to_ns(remaining
), apic
->lapic_timer
.period
);
1112 tmcct
= div64_u64(ns
,
1113 (APIC_BUS_CYCLE_NS
* apic
->divide_count
));
1118 static void __report_tpr_access(struct kvm_lapic
*apic
, bool write
)
1120 struct kvm_vcpu
*vcpu
= apic
->vcpu
;
1121 struct kvm_run
*run
= vcpu
->run
;
1123 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS
, vcpu
);
1124 run
->tpr_access
.rip
= kvm_rip_read(vcpu
);
1125 run
->tpr_access
.is_write
= write
;
1128 static inline void report_tpr_access(struct kvm_lapic
*apic
, bool write
)
1130 if (apic
->vcpu
->arch
.tpr_access_reporting
)
1131 __report_tpr_access(apic
, write
);
1134 static u32
__apic_read(struct kvm_lapic
*apic
, unsigned int offset
)
1138 if (offset
>= LAPIC_MMIO_LENGTH
)
1143 apic_debug("Access APIC ARBPRI register which is for P6\n");
1146 case APIC_TMCCT
: /* Timer CCR */
1147 if (apic_lvtt_tscdeadline(apic
))
1150 val
= apic_get_tmcct(apic
);
1153 apic_update_ppr(apic
);
1154 val
= kvm_lapic_get_reg(apic
, offset
);
1157 report_tpr_access(apic
, false);
1160 val
= kvm_lapic_get_reg(apic
, offset
);
1167 static inline struct kvm_lapic
*to_lapic(struct kvm_io_device
*dev
)
1169 return container_of(dev
, struct kvm_lapic
, dev
);
1172 int kvm_lapic_reg_read(struct kvm_lapic
*apic
, u32 offset
, int len
,
1175 unsigned char alignment
= offset
& 0xf;
1177 /* this bitmask has a bit cleared for each reserved register */
1178 static const u64 rmask
= 0x43ff01ffffffe70cULL
;
1180 if ((alignment
+ len
) > 4) {
1181 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1186 if (offset
> 0x3f0 || !(rmask
& (1ULL << (offset
>> 4)))) {
1187 apic_debug("KVM_APIC_READ: read reserved register %x\n",
1192 result
= __apic_read(apic
, offset
& ~0xf);
1194 trace_kvm_apic_read(offset
, result
);
1200 memcpy(data
, (char *)&result
+ alignment
, len
);
1203 printk(KERN_ERR
"Local APIC read with len = %x, "
1204 "should be 1,2, or 4 instead\n", len
);
1209 EXPORT_SYMBOL_GPL(kvm_lapic_reg_read
);
1211 static int apic_mmio_in_range(struct kvm_lapic
*apic
, gpa_t addr
)
1213 return kvm_apic_hw_enabled(apic
) &&
1214 addr
>= apic
->base_address
&&
1215 addr
< apic
->base_address
+ LAPIC_MMIO_LENGTH
;
1218 static int apic_mmio_read(struct kvm_vcpu
*vcpu
, struct kvm_io_device
*this,
1219 gpa_t address
, int len
, void *data
)
1221 struct kvm_lapic
*apic
= to_lapic(this);
1222 u32 offset
= address
- apic
->base_address
;
1224 if (!apic_mmio_in_range(apic
, address
))
1227 kvm_lapic_reg_read(apic
, offset
, len
, data
);
1232 static void update_divide_count(struct kvm_lapic
*apic
)
1234 u32 tmp1
, tmp2
, tdcr
;
1236 tdcr
= kvm_lapic_get_reg(apic
, APIC_TDCR
);
1238 tmp2
= ((tmp1
& 0x3) | ((tmp1
& 0x8) >> 1)) + 1;
1239 apic
->divide_count
= 0x1 << (tmp2
& 0x7);
1241 apic_debug("timer divide count is 0x%x\n",
1242 apic
->divide_count
);
1245 static void apic_update_lvtt(struct kvm_lapic
*apic
)
1247 u32 timer_mode
= kvm_lapic_get_reg(apic
, APIC_LVTT
) &
1248 apic
->lapic_timer
.timer_mode_mask
;
1250 if (apic
->lapic_timer
.timer_mode
!= timer_mode
) {
1251 apic
->lapic_timer
.timer_mode
= timer_mode
;
1252 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1256 static void apic_timer_expired(struct kvm_lapic
*apic
)
1258 struct kvm_vcpu
*vcpu
= apic
->vcpu
;
1259 struct swait_queue_head
*q
= &vcpu
->wq
;
1260 struct kvm_timer
*ktimer
= &apic
->lapic_timer
;
1262 if (atomic_read(&apic
->lapic_timer
.pending
))
1265 atomic_inc(&apic
->lapic_timer
.pending
);
1266 kvm_set_pending_timer(vcpu
);
1268 if (swait_active(q
))
1271 if (apic_lvtt_tscdeadline(apic
))
1272 ktimer
->expired_tscdeadline
= ktimer
->tscdeadline
;
1276 * On APICv, this test will cause a busy wait
1277 * during a higher-priority task.
1280 static bool lapic_timer_int_injected(struct kvm_vcpu
*vcpu
)
1282 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1283 u32 reg
= kvm_lapic_get_reg(apic
, APIC_LVTT
);
1285 if (kvm_apic_hw_enabled(apic
)) {
1286 int vec
= reg
& APIC_VECTOR_MASK
;
1287 void *bitmap
= apic
->regs
+ APIC_ISR
;
1289 if (vcpu
->arch
.apicv_active
)
1290 bitmap
= apic
->regs
+ APIC_IRR
;
1292 if (apic_test_vector(vec
, bitmap
))
1298 void wait_lapic_expire(struct kvm_vcpu
*vcpu
)
1300 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1301 u64 guest_tsc
, tsc_deadline
;
1303 if (!lapic_in_kernel(vcpu
))
1306 if (apic
->lapic_timer
.expired_tscdeadline
== 0)
1309 if (!lapic_timer_int_injected(vcpu
))
1312 tsc_deadline
= apic
->lapic_timer
.expired_tscdeadline
;
1313 apic
->lapic_timer
.expired_tscdeadline
= 0;
1314 guest_tsc
= kvm_read_l1_tsc(vcpu
, rdtsc());
1315 trace_kvm_wait_lapic_expire(vcpu
->vcpu_id
, guest_tsc
- tsc_deadline
);
1317 /* __delay is delay_tsc whenever the hardware has TSC, thus always. */
1318 if (guest_tsc
< tsc_deadline
)
1319 __delay(min(tsc_deadline
- guest_tsc
,
1320 nsec_to_cycles(vcpu
, lapic_timer_advance_ns
)));
1323 static void start_sw_tscdeadline(struct kvm_lapic
*apic
)
1325 u64 guest_tsc
, tscdeadline
= apic
->lapic_timer
.tscdeadline
;
1328 struct kvm_vcpu
*vcpu
= apic
->vcpu
;
1329 unsigned long this_tsc_khz
= vcpu
->arch
.virtual_tsc_khz
;
1330 unsigned long flags
;
1333 if (unlikely(!tscdeadline
|| !this_tsc_khz
))
1336 local_irq_save(flags
);
1339 guest_tsc
= kvm_read_l1_tsc(vcpu
, rdtsc());
1340 if (likely(tscdeadline
> guest_tsc
)) {
1341 ns
= (tscdeadline
- guest_tsc
) * 1000000ULL;
1342 do_div(ns
, this_tsc_khz
);
1343 expire
= ktime_add_ns(now
, ns
);
1344 expire
= ktime_sub_ns(expire
, lapic_timer_advance_ns
);
1345 hrtimer_start(&apic
->lapic_timer
.timer
,
1346 expire
, HRTIMER_MODE_ABS_PINNED
);
1348 apic_timer_expired(apic
);
1350 local_irq_restore(flags
);
1353 static void start_sw_period(struct kvm_lapic
*apic
)
1355 if (!apic
->lapic_timer
.period
)
1358 if (apic_lvtt_oneshot(apic
) &&
1359 ktime_after(ktime_get(),
1360 apic
->lapic_timer
.target_expiration
)) {
1361 apic_timer_expired(apic
);
1365 hrtimer_start(&apic
->lapic_timer
.timer
,
1366 apic
->lapic_timer
.target_expiration
,
1367 HRTIMER_MODE_ABS_PINNED
);
1370 static bool set_target_expiration(struct kvm_lapic
*apic
)
1376 apic
->lapic_timer
.period
= (u64
)kvm_lapic_get_reg(apic
, APIC_TMICT
)
1377 * APIC_BUS_CYCLE_NS
* apic
->divide_count
;
1379 if (!apic
->lapic_timer
.period
)
1383 * Do not allow the guest to program periodic timers with small
1384 * interval, since the hrtimers are not throttled by the host
1387 if (apic_lvtt_period(apic
)) {
1388 s64 min_period
= min_timer_period_us
* 1000LL;
1390 if (apic
->lapic_timer
.period
< min_period
) {
1391 pr_info_ratelimited(
1392 "kvm: vcpu %i: requested %lld ns "
1393 "lapic timer period limited to %lld ns\n",
1394 apic
->vcpu
->vcpu_id
,
1395 apic
->lapic_timer
.period
, min_period
);
1396 apic
->lapic_timer
.period
= min_period
;
1400 apic_debug("%s: bus cycle is %" PRId64
"ns, now 0x%016"
1402 "timer initial count 0x%x, period %lldns, "
1403 "expire @ 0x%016" PRIx64
".\n", __func__
,
1404 APIC_BUS_CYCLE_NS
, ktime_to_ns(now
),
1405 kvm_lapic_get_reg(apic
, APIC_TMICT
),
1406 apic
->lapic_timer
.period
,
1407 ktime_to_ns(ktime_add_ns(now
,
1408 apic
->lapic_timer
.period
)));
1410 apic
->lapic_timer
.tscdeadline
= kvm_read_l1_tsc(apic
->vcpu
, tscl
) +
1411 nsec_to_cycles(apic
->vcpu
, apic
->lapic_timer
.period
);
1412 apic
->lapic_timer
.target_expiration
= ktime_add_ns(now
, apic
->lapic_timer
.period
);
1417 static void advance_periodic_target_expiration(struct kvm_lapic
*apic
)
1419 apic
->lapic_timer
.tscdeadline
+=
1420 nsec_to_cycles(apic
->vcpu
, apic
->lapic_timer
.period
);
1421 apic
->lapic_timer
.target_expiration
=
1422 ktime_add_ns(apic
->lapic_timer
.target_expiration
,
1423 apic
->lapic_timer
.period
);
1426 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu
*vcpu
)
1428 if (!lapic_in_kernel(vcpu
))
1431 return vcpu
->arch
.apic
->lapic_timer
.hv_timer_in_use
;
1433 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use
);
1435 static void cancel_hv_timer(struct kvm_lapic
*apic
)
1437 kvm_x86_ops
->cancel_hv_timer(apic
->vcpu
);
1438 apic
->lapic_timer
.hv_timer_in_use
= false;
1441 static bool start_hv_timer(struct kvm_lapic
*apic
)
1443 u64 tscdeadline
= apic
->lapic_timer
.tscdeadline
;
1445 if ((atomic_read(&apic
->lapic_timer
.pending
) &&
1446 !apic_lvtt_period(apic
)) ||
1447 kvm_x86_ops
->set_hv_timer(apic
->vcpu
, tscdeadline
)) {
1448 if (apic
->lapic_timer
.hv_timer_in_use
)
1449 cancel_hv_timer(apic
);
1451 apic
->lapic_timer
.hv_timer_in_use
= true;
1452 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1454 /* In case the sw timer triggered in the window */
1455 if (atomic_read(&apic
->lapic_timer
.pending
) &&
1456 !apic_lvtt_period(apic
))
1457 cancel_hv_timer(apic
);
1459 trace_kvm_hv_timer_state(apic
->vcpu
->vcpu_id
,
1460 apic
->lapic_timer
.hv_timer_in_use
);
1461 return apic
->lapic_timer
.hv_timer_in_use
;
1464 void kvm_lapic_expired_hv_timer(struct kvm_vcpu
*vcpu
)
1466 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1468 WARN_ON(!apic
->lapic_timer
.hv_timer_in_use
);
1469 WARN_ON(swait_active(&vcpu
->wq
));
1470 cancel_hv_timer(apic
);
1471 apic_timer_expired(apic
);
1473 if (apic_lvtt_period(apic
) && apic
->lapic_timer
.period
) {
1474 advance_periodic_target_expiration(apic
);
1475 if (!start_hv_timer(apic
))
1476 start_sw_period(apic
);
1479 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer
);
1481 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu
*vcpu
)
1483 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1485 WARN_ON(apic
->lapic_timer
.hv_timer_in_use
);
1487 start_hv_timer(apic
);
1489 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer
);
1491 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu
*vcpu
)
1493 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1495 /* Possibly the TSC deadline timer is not enabled yet */
1496 if (!apic
->lapic_timer
.hv_timer_in_use
)
1499 cancel_hv_timer(apic
);
1501 if (atomic_read(&apic
->lapic_timer
.pending
))
1504 if (apic_lvtt_period(apic
) || apic_lvtt_oneshot(apic
))
1505 start_sw_period(apic
);
1506 else if (apic_lvtt_tscdeadline(apic
))
1507 start_sw_tscdeadline(apic
);
1509 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer
);
1511 static void start_apic_timer(struct kvm_lapic
*apic
)
1513 atomic_set(&apic
->lapic_timer
.pending
, 0);
1515 if (apic_lvtt_period(apic
) || apic_lvtt_oneshot(apic
)) {
1516 if (set_target_expiration(apic
) &&
1517 !(kvm_x86_ops
->set_hv_timer
&& start_hv_timer(apic
)))
1518 start_sw_period(apic
);
1519 } else if (apic_lvtt_tscdeadline(apic
)) {
1520 if (!(kvm_x86_ops
->set_hv_timer
&& start_hv_timer(apic
)))
1521 start_sw_tscdeadline(apic
);
1525 static void apic_manage_nmi_watchdog(struct kvm_lapic
*apic
, u32 lvt0_val
)
1527 bool lvt0_in_nmi_mode
= apic_lvt_nmi_mode(lvt0_val
);
1529 if (apic
->lvt0_in_nmi_mode
!= lvt0_in_nmi_mode
) {
1530 apic
->lvt0_in_nmi_mode
= lvt0_in_nmi_mode
;
1531 if (lvt0_in_nmi_mode
) {
1532 apic_debug("Receive NMI setting on APIC_LVT0 "
1533 "for cpu %d\n", apic
->vcpu
->vcpu_id
);
1534 atomic_inc(&apic
->vcpu
->kvm
->arch
.vapics_in_nmi_mode
);
1536 atomic_dec(&apic
->vcpu
->kvm
->arch
.vapics_in_nmi_mode
);
1540 int kvm_lapic_reg_write(struct kvm_lapic
*apic
, u32 reg
, u32 val
)
1544 trace_kvm_apic_write(reg
, val
);
1547 case APIC_ID
: /* Local APIC ID */
1548 if (!apic_x2apic_mode(apic
))
1549 kvm_apic_set_xapic_id(apic
, val
>> 24);
1555 report_tpr_access(apic
, true);
1556 apic_set_tpr(apic
, val
& 0xff);
1564 if (!apic_x2apic_mode(apic
))
1565 kvm_apic_set_ldr(apic
, val
& APIC_LDR_MASK
);
1571 if (!apic_x2apic_mode(apic
)) {
1572 kvm_lapic_set_reg(apic
, APIC_DFR
, val
| 0x0FFFFFFF);
1573 recalculate_apic_map(apic
->vcpu
->kvm
);
1580 if (kvm_lapic_get_reg(apic
, APIC_LVR
) & APIC_LVR_DIRECTED_EOI
)
1581 mask
|= APIC_SPIV_DIRECTED_EOI
;
1582 apic_set_spiv(apic
, val
& mask
);
1583 if (!(val
& APIC_SPIV_APIC_ENABLED
)) {
1587 for (i
= 0; i
< KVM_APIC_LVT_NUM
; i
++) {
1588 lvt_val
= kvm_lapic_get_reg(apic
,
1589 APIC_LVTT
+ 0x10 * i
);
1590 kvm_lapic_set_reg(apic
, APIC_LVTT
+ 0x10 * i
,
1591 lvt_val
| APIC_LVT_MASKED
);
1593 apic_update_lvtt(apic
);
1594 atomic_set(&apic
->lapic_timer
.pending
, 0);
1600 /* No delay here, so we always clear the pending bit */
1601 kvm_lapic_set_reg(apic
, APIC_ICR
, val
& ~(1 << 12));
1602 apic_send_ipi(apic
);
1606 if (!apic_x2apic_mode(apic
))
1608 kvm_lapic_set_reg(apic
, APIC_ICR2
, val
);
1612 apic_manage_nmi_watchdog(apic
, val
);
1617 /* TODO: Check vector */
1618 if (!kvm_apic_sw_enabled(apic
))
1619 val
|= APIC_LVT_MASKED
;
1621 val
&= apic_lvt_mask
[(reg
- APIC_LVTT
) >> 4];
1622 kvm_lapic_set_reg(apic
, reg
, val
);
1627 if (!kvm_apic_sw_enabled(apic
))
1628 val
|= APIC_LVT_MASKED
;
1629 val
&= (apic_lvt_mask
[0] | apic
->lapic_timer
.timer_mode_mask
);
1630 kvm_lapic_set_reg(apic
, APIC_LVTT
, val
);
1631 apic_update_lvtt(apic
);
1635 if (apic_lvtt_tscdeadline(apic
))
1638 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1639 kvm_lapic_set_reg(apic
, APIC_TMICT
, val
);
1640 start_apic_timer(apic
);
1645 apic_debug("KVM_WRITE:TDCR %x\n", val
);
1646 kvm_lapic_set_reg(apic
, APIC_TDCR
, val
);
1647 update_divide_count(apic
);
1651 if (apic_x2apic_mode(apic
) && val
!= 0) {
1652 apic_debug("KVM_WRITE:ESR not zero %x\n", val
);
1658 if (apic_x2apic_mode(apic
)) {
1659 kvm_lapic_reg_write(apic
, APIC_ICR
, 0x40000 | (val
& 0xff));
1668 apic_debug("Local APIC Write to read-only register %x\n", reg
);
1671 EXPORT_SYMBOL_GPL(kvm_lapic_reg_write
);
1673 static int apic_mmio_write(struct kvm_vcpu
*vcpu
, struct kvm_io_device
*this,
1674 gpa_t address
, int len
, const void *data
)
1676 struct kvm_lapic
*apic
= to_lapic(this);
1677 unsigned int offset
= address
- apic
->base_address
;
1680 if (!apic_mmio_in_range(apic
, address
))
1684 * APIC register must be aligned on 128-bits boundary.
1685 * 32/64/128 bits registers must be accessed thru 32 bits.
1688 if (len
!= 4 || (offset
& 0xf)) {
1689 /* Don't shout loud, $infamous_os would cause only noise. */
1690 apic_debug("apic write: bad size=%d %lx\n", len
, (long)address
);
1696 /* too common printing */
1697 if (offset
!= APIC_EOI
)
1698 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1699 "0x%x\n", __func__
, offset
, len
, val
);
1701 kvm_lapic_reg_write(apic
, offset
& 0xff0, val
);
1706 void kvm_lapic_set_eoi(struct kvm_vcpu
*vcpu
)
1708 kvm_lapic_reg_write(vcpu
->arch
.apic
, APIC_EOI
, 0);
1710 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi
);
1712 /* emulate APIC access in a trap manner */
1713 void kvm_apic_write_nodecode(struct kvm_vcpu
*vcpu
, u32 offset
)
1717 /* hw has done the conditional check and inst decode */
1720 kvm_lapic_reg_read(vcpu
->arch
.apic
, offset
, 4, &val
);
1722 /* TODO: optimize to just emulate side effect w/o one more write */
1723 kvm_lapic_reg_write(vcpu
->arch
.apic
, offset
, val
);
1725 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode
);
1727 void kvm_free_lapic(struct kvm_vcpu
*vcpu
)
1729 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1731 if (!vcpu
->arch
.apic
)
1734 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1736 if (!(vcpu
->arch
.apic_base
& MSR_IA32_APICBASE_ENABLE
))
1737 static_key_slow_dec_deferred(&apic_hw_disabled
);
1739 if (!apic
->sw_enabled
)
1740 static_key_slow_dec_deferred(&apic_sw_disabled
);
1743 free_page((unsigned long)apic
->regs
);
1749 *----------------------------------------------------------------------
1751 *----------------------------------------------------------------------
1753 u64
kvm_get_lapic_target_expiration_tsc(struct kvm_vcpu
*vcpu
)
1755 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1757 if (!lapic_in_kernel(vcpu
))
1760 return apic
->lapic_timer
.tscdeadline
;
1763 u64
kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu
*vcpu
)
1765 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1767 if (!lapic_in_kernel(vcpu
) ||
1768 !apic_lvtt_tscdeadline(apic
))
1771 return apic
->lapic_timer
.tscdeadline
;
1774 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu
*vcpu
, u64 data
)
1776 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1778 if (!lapic_in_kernel(vcpu
) || apic_lvtt_oneshot(apic
) ||
1779 apic_lvtt_period(apic
))
1782 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1783 apic
->lapic_timer
.tscdeadline
= data
;
1784 start_apic_timer(apic
);
1787 void kvm_lapic_set_tpr(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
1789 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1791 apic_set_tpr(apic
, ((cr8
& 0x0f) << 4)
1792 | (kvm_lapic_get_reg(apic
, APIC_TASKPRI
) & 4));
1795 u64
kvm_lapic_get_cr8(struct kvm_vcpu
*vcpu
)
1799 tpr
= (u64
) kvm_lapic_get_reg(vcpu
->arch
.apic
, APIC_TASKPRI
);
1801 return (tpr
& 0xf0) >> 4;
1804 void kvm_lapic_set_base(struct kvm_vcpu
*vcpu
, u64 value
)
1806 u64 old_value
= vcpu
->arch
.apic_base
;
1807 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1810 value
|= MSR_IA32_APICBASE_BSP
;
1812 vcpu
->arch
.apic_base
= value
;
1814 if ((old_value
^ value
) & MSR_IA32_APICBASE_ENABLE
)
1815 kvm_update_cpuid(vcpu
);
1820 /* update jump label if enable bit changes */
1821 if ((old_value
^ value
) & MSR_IA32_APICBASE_ENABLE
) {
1822 if (value
& MSR_IA32_APICBASE_ENABLE
) {
1823 kvm_apic_set_xapic_id(apic
, vcpu
->vcpu_id
);
1824 static_key_slow_dec_deferred(&apic_hw_disabled
);
1826 static_key_slow_inc(&apic_hw_disabled
.key
);
1827 recalculate_apic_map(vcpu
->kvm
);
1831 if ((old_value
^ value
) & X2APIC_ENABLE
) {
1832 if (value
& X2APIC_ENABLE
) {
1833 kvm_apic_set_x2apic_id(apic
, vcpu
->vcpu_id
);
1834 kvm_x86_ops
->set_virtual_x2apic_mode(vcpu
, true);
1836 kvm_x86_ops
->set_virtual_x2apic_mode(vcpu
, false);
1839 apic
->base_address
= apic
->vcpu
->arch
.apic_base
&
1840 MSR_IA32_APICBASE_BASE
;
1842 if ((value
& MSR_IA32_APICBASE_ENABLE
) &&
1843 apic
->base_address
!= APIC_DEFAULT_PHYS_BASE
)
1844 pr_warn_once("APIC base relocation is unsupported by KVM");
1846 /* with FSB delivery interrupt, we can restart APIC functionality */
1847 apic_debug("apic base msr is 0x%016" PRIx64
", and base address is "
1848 "0x%lx.\n", apic
->vcpu
->arch
.apic_base
, apic
->base_address
);
1852 void kvm_lapic_reset(struct kvm_vcpu
*vcpu
, bool init_event
)
1854 struct kvm_lapic
*apic
;
1857 apic_debug("%s\n", __func__
);
1860 apic
= vcpu
->arch
.apic
;
1861 ASSERT(apic
!= NULL
);
1863 /* Stop the timer in case it's a reset to an active apic */
1864 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1867 kvm_lapic_set_base(vcpu
, APIC_DEFAULT_PHYS_BASE
|
1868 MSR_IA32_APICBASE_ENABLE
);
1869 kvm_apic_set_xapic_id(apic
, vcpu
->vcpu_id
);
1871 kvm_apic_set_version(apic
->vcpu
);
1873 for (i
= 0; i
< KVM_APIC_LVT_NUM
; i
++)
1874 kvm_lapic_set_reg(apic
, APIC_LVTT
+ 0x10 * i
, APIC_LVT_MASKED
);
1875 apic_update_lvtt(apic
);
1876 if (kvm_check_has_quirk(vcpu
->kvm
, KVM_X86_QUIRK_LINT0_REENABLED
))
1877 kvm_lapic_set_reg(apic
, APIC_LVT0
,
1878 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT
));
1879 apic_manage_nmi_watchdog(apic
, kvm_lapic_get_reg(apic
, APIC_LVT0
));
1881 kvm_lapic_set_reg(apic
, APIC_DFR
, 0xffffffffU
);
1882 apic_set_spiv(apic
, 0xff);
1883 kvm_lapic_set_reg(apic
, APIC_TASKPRI
, 0);
1884 if (!apic_x2apic_mode(apic
))
1885 kvm_apic_set_ldr(apic
, 0);
1886 kvm_lapic_set_reg(apic
, APIC_ESR
, 0);
1887 kvm_lapic_set_reg(apic
, APIC_ICR
, 0);
1888 kvm_lapic_set_reg(apic
, APIC_ICR2
, 0);
1889 kvm_lapic_set_reg(apic
, APIC_TDCR
, 0);
1890 kvm_lapic_set_reg(apic
, APIC_TMICT
, 0);
1891 for (i
= 0; i
< 8; i
++) {
1892 kvm_lapic_set_reg(apic
, APIC_IRR
+ 0x10 * i
, 0);
1893 kvm_lapic_set_reg(apic
, APIC_ISR
+ 0x10 * i
, 0);
1894 kvm_lapic_set_reg(apic
, APIC_TMR
+ 0x10 * i
, 0);
1896 apic
->irr_pending
= vcpu
->arch
.apicv_active
;
1897 apic
->isr_count
= vcpu
->arch
.apicv_active
? 1 : 0;
1898 apic
->highest_isr_cache
= -1;
1899 update_divide_count(apic
);
1900 atomic_set(&apic
->lapic_timer
.pending
, 0);
1901 if (kvm_vcpu_is_bsp(vcpu
))
1902 kvm_lapic_set_base(vcpu
,
1903 vcpu
->arch
.apic_base
| MSR_IA32_APICBASE_BSP
);
1904 vcpu
->arch
.pv_eoi
.msr_val
= 0;
1905 apic_update_ppr(apic
);
1907 vcpu
->arch
.apic_arb_prio
= 0;
1908 vcpu
->arch
.apic_attention
= 0;
1910 apic_debug("%s: vcpu=%p, id=%d, base_msr="
1911 "0x%016" PRIx64
", base_address=0x%0lx.\n", __func__
,
1912 vcpu
, kvm_apic_id(apic
),
1913 vcpu
->arch
.apic_base
, apic
->base_address
);
1917 *----------------------------------------------------------------------
1919 *----------------------------------------------------------------------
1922 static bool lapic_is_periodic(struct kvm_lapic
*apic
)
1924 return apic_lvtt_period(apic
);
1927 int apic_has_pending_timer(struct kvm_vcpu
*vcpu
)
1929 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1931 if (apic_enabled(apic
) && apic_lvt_enabled(apic
, APIC_LVTT
))
1932 return atomic_read(&apic
->lapic_timer
.pending
);
1937 int kvm_apic_local_deliver(struct kvm_lapic
*apic
, int lvt_type
)
1939 u32 reg
= kvm_lapic_get_reg(apic
, lvt_type
);
1940 int vector
, mode
, trig_mode
;
1942 if (kvm_apic_hw_enabled(apic
) && !(reg
& APIC_LVT_MASKED
)) {
1943 vector
= reg
& APIC_VECTOR_MASK
;
1944 mode
= reg
& APIC_MODE_MASK
;
1945 trig_mode
= reg
& APIC_LVT_LEVEL_TRIGGER
;
1946 return __apic_accept_irq(apic
, mode
, vector
, 1, trig_mode
,
1952 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu
*vcpu
)
1954 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1957 kvm_apic_local_deliver(apic
, APIC_LVT0
);
1960 static const struct kvm_io_device_ops apic_mmio_ops
= {
1961 .read
= apic_mmio_read
,
1962 .write
= apic_mmio_write
,
1965 static enum hrtimer_restart
apic_timer_fn(struct hrtimer
*data
)
1967 struct kvm_timer
*ktimer
= container_of(data
, struct kvm_timer
, timer
);
1968 struct kvm_lapic
*apic
= container_of(ktimer
, struct kvm_lapic
, lapic_timer
);
1970 apic_timer_expired(apic
);
1972 if (lapic_is_periodic(apic
)) {
1973 advance_periodic_target_expiration(apic
);
1974 hrtimer_add_expires_ns(&ktimer
->timer
, ktimer
->period
);
1975 return HRTIMER_RESTART
;
1977 return HRTIMER_NORESTART
;
1980 int kvm_create_lapic(struct kvm_vcpu
*vcpu
)
1982 struct kvm_lapic
*apic
;
1984 ASSERT(vcpu
!= NULL
);
1985 apic_debug("apic_init %d\n", vcpu
->vcpu_id
);
1987 apic
= kzalloc(sizeof(*apic
), GFP_KERNEL
);
1991 vcpu
->arch
.apic
= apic
;
1993 apic
->regs
= (void *)get_zeroed_page(GFP_KERNEL
);
1995 printk(KERN_ERR
"malloc apic regs error for vcpu %x\n",
1997 goto nomem_free_apic
;
2001 hrtimer_init(&apic
->lapic_timer
.timer
, CLOCK_MONOTONIC
,
2002 HRTIMER_MODE_ABS_PINNED
);
2003 apic
->lapic_timer
.timer
.function
= apic_timer_fn
;
2006 * APIC is created enabled. This will prevent kvm_lapic_set_base from
2007 * thinking that APIC satet has changed.
2009 vcpu
->arch
.apic_base
= MSR_IA32_APICBASE_ENABLE
;
2010 static_key_slow_inc(&apic_sw_disabled
.key
); /* sw disabled at reset */
2011 kvm_lapic_reset(vcpu
, false);
2012 kvm_iodevice_init(&apic
->dev
, &apic_mmio_ops
);
2021 int kvm_apic_has_interrupt(struct kvm_vcpu
*vcpu
)
2023 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2026 if (!apic_enabled(apic
))
2029 apic_update_ppr(apic
);
2030 highest_irr
= apic_find_highest_irr(apic
);
2031 if ((highest_irr
== -1) ||
2032 ((highest_irr
& 0xF0) <= kvm_lapic_get_reg(apic
, APIC_PROCPRI
)))
2037 int kvm_apic_accept_pic_intr(struct kvm_vcpu
*vcpu
)
2039 u32 lvt0
= kvm_lapic_get_reg(vcpu
->arch
.apic
, APIC_LVT0
);
2042 if (!kvm_apic_hw_enabled(vcpu
->arch
.apic
))
2044 if ((lvt0
& APIC_LVT_MASKED
) == 0 &&
2045 GET_APIC_DELIVERY_MODE(lvt0
) == APIC_MODE_EXTINT
)
2050 void kvm_inject_apic_timer_irqs(struct kvm_vcpu
*vcpu
)
2052 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2054 if (atomic_read(&apic
->lapic_timer
.pending
) > 0) {
2055 kvm_apic_local_deliver(apic
, APIC_LVTT
);
2056 if (apic_lvtt_tscdeadline(apic
))
2057 apic
->lapic_timer
.tscdeadline
= 0;
2058 if (apic_lvtt_oneshot(apic
)) {
2059 apic
->lapic_timer
.tscdeadline
= 0;
2060 apic
->lapic_timer
.target_expiration
= 0;
2062 atomic_set(&apic
->lapic_timer
.pending
, 0);
2066 int kvm_get_apic_interrupt(struct kvm_vcpu
*vcpu
)
2068 int vector
= kvm_apic_has_interrupt(vcpu
);
2069 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2075 * We get here even with APIC virtualization enabled, if doing
2076 * nested virtualization and L1 runs with the "acknowledge interrupt
2077 * on exit" mode. Then we cannot inject the interrupt via RVI,
2078 * because the process would deliver it through the IDT.
2081 apic_set_isr(vector
, apic
);
2082 apic_update_ppr(apic
);
2083 apic_clear_irr(vector
, apic
);
2085 if (test_bit(vector
, vcpu_to_synic(vcpu
)->auto_eoi_bitmap
)) {
2086 apic_clear_isr(vector
, apic
);
2087 apic_update_ppr(apic
);
2093 static int kvm_apic_state_fixup(struct kvm_vcpu
*vcpu
,
2094 struct kvm_lapic_state
*s
, bool set
)
2096 if (apic_x2apic_mode(vcpu
->arch
.apic
)) {
2097 u32
*id
= (u32
*)(s
->regs
+ APIC_ID
);
2099 if (vcpu
->kvm
->arch
.x2apic_format
) {
2100 if (*id
!= vcpu
->vcpu_id
)
2113 int kvm_apic_get_state(struct kvm_vcpu
*vcpu
, struct kvm_lapic_state
*s
)
2115 memcpy(s
->regs
, vcpu
->arch
.apic
->regs
, sizeof(*s
));
2116 return kvm_apic_state_fixup(vcpu
, s
, false);
2119 int kvm_apic_set_state(struct kvm_vcpu
*vcpu
, struct kvm_lapic_state
*s
)
2121 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2125 kvm_lapic_set_base(vcpu
, vcpu
->arch
.apic_base
);
2126 /* set SPIV separately to get count of SW disabled APICs right */
2127 apic_set_spiv(apic
, *((u32
*)(s
->regs
+ APIC_SPIV
)));
2129 r
= kvm_apic_state_fixup(vcpu
, s
, true);
2132 memcpy(vcpu
->arch
.apic
->regs
, s
->regs
, sizeof *s
);
2134 recalculate_apic_map(vcpu
->kvm
);
2135 kvm_apic_set_version(vcpu
);
2137 apic_update_ppr(apic
);
2138 hrtimer_cancel(&apic
->lapic_timer
.timer
);
2139 apic_update_lvtt(apic
);
2140 apic_manage_nmi_watchdog(apic
, kvm_lapic_get_reg(apic
, APIC_LVT0
));
2141 update_divide_count(apic
);
2142 start_apic_timer(apic
);
2143 apic
->irr_pending
= true;
2144 apic
->isr_count
= vcpu
->arch
.apicv_active
?
2145 1 : count_vectors(apic
->regs
+ APIC_ISR
);
2146 apic
->highest_isr_cache
= -1;
2147 if (vcpu
->arch
.apicv_active
) {
2148 if (kvm_x86_ops
->apicv_post_state_restore
)
2149 kvm_x86_ops
->apicv_post_state_restore(vcpu
);
2150 kvm_x86_ops
->hwapic_irr_update(vcpu
,
2151 apic_find_highest_irr(apic
));
2152 kvm_x86_ops
->hwapic_isr_update(vcpu
,
2153 apic_find_highest_isr(apic
));
2155 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
2156 if (ioapic_in_kernel(vcpu
->kvm
))
2157 kvm_rtc_eoi_tracking_restore_one(vcpu
);
2159 vcpu
->arch
.apic_arb_prio
= 0;
2164 void __kvm_migrate_apic_timer(struct kvm_vcpu
*vcpu
)
2166 struct hrtimer
*timer
;
2168 if (!lapic_in_kernel(vcpu
))
2171 timer
= &vcpu
->arch
.apic
->lapic_timer
.timer
;
2172 if (hrtimer_cancel(timer
))
2173 hrtimer_start_expires(timer
, HRTIMER_MODE_ABS_PINNED
);
2177 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2179 * Detect whether guest triggered PV EOI since the
2180 * last entry. If yes, set EOI on guests's behalf.
2181 * Clear PV EOI in guest memory in any case.
2183 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu
*vcpu
,
2184 struct kvm_lapic
*apic
)
2189 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2190 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2192 * KVM_APIC_PV_EOI_PENDING is unset:
2193 * -> host disabled PV EOI.
2194 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2195 * -> host enabled PV EOI, guest did not execute EOI yet.
2196 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2197 * -> host enabled PV EOI, guest executed EOI.
2199 BUG_ON(!pv_eoi_enabled(vcpu
));
2200 pending
= pv_eoi_get_pending(vcpu
);
2202 * Clear pending bit in any case: it will be set again on vmentry.
2203 * While this might not be ideal from performance point of view,
2204 * this makes sure pv eoi is only enabled when we know it's safe.
2206 pv_eoi_clr_pending(vcpu
);
2209 vector
= apic_set_eoi(apic
);
2210 trace_kvm_pv_eoi(apic
, vector
);
2213 void kvm_lapic_sync_from_vapic(struct kvm_vcpu
*vcpu
)
2217 if (test_bit(KVM_APIC_PV_EOI_PENDING
, &vcpu
->arch
.apic_attention
))
2218 apic_sync_pv_eoi_from_guest(vcpu
, vcpu
->arch
.apic
);
2220 if (!test_bit(KVM_APIC_CHECK_VAPIC
, &vcpu
->arch
.apic_attention
))
2223 if (kvm_read_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apic
->vapic_cache
, &data
,
2227 apic_set_tpr(vcpu
->arch
.apic
, data
& 0xff);
2231 * apic_sync_pv_eoi_to_guest - called before vmentry
2233 * Detect whether it's safe to enable PV EOI and
2236 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu
*vcpu
,
2237 struct kvm_lapic
*apic
)
2239 if (!pv_eoi_enabled(vcpu
) ||
2240 /* IRR set or many bits in ISR: could be nested. */
2241 apic
->irr_pending
||
2242 /* Cache not set: could be safe but we don't bother. */
2243 apic
->highest_isr_cache
== -1 ||
2244 /* Need EOI to update ioapic. */
2245 kvm_ioapic_handles_vector(apic
, apic
->highest_isr_cache
)) {
2247 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2248 * so we need not do anything here.
2253 pv_eoi_set_pending(apic
->vcpu
);
2256 void kvm_lapic_sync_to_vapic(struct kvm_vcpu
*vcpu
)
2259 int max_irr
, max_isr
;
2260 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2262 apic_sync_pv_eoi_to_guest(vcpu
, apic
);
2264 if (!test_bit(KVM_APIC_CHECK_VAPIC
, &vcpu
->arch
.apic_attention
))
2267 tpr
= kvm_lapic_get_reg(apic
, APIC_TASKPRI
) & 0xff;
2268 max_irr
= apic_find_highest_irr(apic
);
2271 max_isr
= apic_find_highest_isr(apic
);
2274 data
= (tpr
& 0xff) | ((max_isr
& 0xf0) << 8) | (max_irr
<< 24);
2276 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apic
->vapic_cache
, &data
,
2280 int kvm_lapic_set_vapic_addr(struct kvm_vcpu
*vcpu
, gpa_t vapic_addr
)
2283 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
,
2284 &vcpu
->arch
.apic
->vapic_cache
,
2285 vapic_addr
, sizeof(u32
)))
2287 __set_bit(KVM_APIC_CHECK_VAPIC
, &vcpu
->arch
.apic_attention
);
2289 __clear_bit(KVM_APIC_CHECK_VAPIC
, &vcpu
->arch
.apic_attention
);
2292 vcpu
->arch
.apic
->vapic_addr
= vapic_addr
;
2296 int kvm_x2apic_msr_write(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
2298 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2299 u32 reg
= (msr
- APIC_BASE_MSR
) << 4;
2301 if (!lapic_in_kernel(vcpu
) || !apic_x2apic_mode(apic
))
2304 if (reg
== APIC_ICR2
)
2307 /* if this is ICR write vector before command */
2308 if (reg
== APIC_ICR
)
2309 kvm_lapic_reg_write(apic
, APIC_ICR2
, (u32
)(data
>> 32));
2310 return kvm_lapic_reg_write(apic
, reg
, (u32
)data
);
2313 int kvm_x2apic_msr_read(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*data
)
2315 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2316 u32 reg
= (msr
- APIC_BASE_MSR
) << 4, low
, high
= 0;
2318 if (!lapic_in_kernel(vcpu
) || !apic_x2apic_mode(apic
))
2321 if (reg
== APIC_DFR
|| reg
== APIC_ICR2
) {
2322 apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
2327 if (kvm_lapic_reg_read(apic
, reg
, 4, &low
))
2329 if (reg
== APIC_ICR
)
2330 kvm_lapic_reg_read(apic
, APIC_ICR2
, 4, &high
);
2332 *data
= (((u64
)high
) << 32) | low
;
2337 int kvm_hv_vapic_msr_write(struct kvm_vcpu
*vcpu
, u32 reg
, u64 data
)
2339 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2341 if (!lapic_in_kernel(vcpu
))
2344 /* if this is ICR write vector before command */
2345 if (reg
== APIC_ICR
)
2346 kvm_lapic_reg_write(apic
, APIC_ICR2
, (u32
)(data
>> 32));
2347 return kvm_lapic_reg_write(apic
, reg
, (u32
)data
);
2350 int kvm_hv_vapic_msr_read(struct kvm_vcpu
*vcpu
, u32 reg
, u64
*data
)
2352 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2355 if (!lapic_in_kernel(vcpu
))
2358 if (kvm_lapic_reg_read(apic
, reg
, 4, &low
))
2360 if (reg
== APIC_ICR
)
2361 kvm_lapic_reg_read(apic
, APIC_ICR2
, 4, &high
);
2363 *data
= (((u64
)high
) << 32) | low
;
2368 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu
*vcpu
, u64 data
)
2370 u64 addr
= data
& ~KVM_MSR_ENABLED
;
2371 if (!IS_ALIGNED(addr
, 4))
2374 vcpu
->arch
.pv_eoi
.msr_val
= data
;
2375 if (!pv_eoi_enabled(vcpu
))
2377 return kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.pv_eoi
.data
,
2381 void kvm_apic_accept_events(struct kvm_vcpu
*vcpu
)
2383 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2387 if (!lapic_in_kernel(vcpu
) || !apic
->pending_events
)
2391 * INITs are latched while in SMM. Because an SMM CPU cannot
2392 * be in KVM_MP_STATE_INIT_RECEIVED state, just eat SIPIs
2393 * and delay processing of INIT until the next RSM.
2396 WARN_ON_ONCE(vcpu
->arch
.mp_state
== KVM_MP_STATE_INIT_RECEIVED
);
2397 if (test_bit(KVM_APIC_SIPI
, &apic
->pending_events
))
2398 clear_bit(KVM_APIC_SIPI
, &apic
->pending_events
);
2402 pe
= xchg(&apic
->pending_events
, 0);
2403 if (test_bit(KVM_APIC_INIT
, &pe
)) {
2404 kvm_lapic_reset(vcpu
, true);
2405 kvm_vcpu_reset(vcpu
, true);
2406 if (kvm_vcpu_is_bsp(apic
->vcpu
))
2407 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
2409 vcpu
->arch
.mp_state
= KVM_MP_STATE_INIT_RECEIVED
;
2411 if (test_bit(KVM_APIC_SIPI
, &pe
) &&
2412 vcpu
->arch
.mp_state
== KVM_MP_STATE_INIT_RECEIVED
) {
2413 /* evaluate pending_events before reading the vector */
2415 sipi_vector
= apic
->sipi_vector
;
2416 apic_debug("vcpu %d received sipi with vector # %x\n",
2417 vcpu
->vcpu_id
, sipi_vector
);
2418 kvm_vcpu_deliver_sipi_vector(vcpu
, sipi_vector
);
2419 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
2423 void kvm_lapic_init(void)
2425 /* do not patch jump label more than once per second */
2426 jump_label_rate_limit(&apic_hw_disabled
, HZ
);
2427 jump_label_rate_limit(&apic_sw_disabled
, HZ
);
2430 void kvm_lapic_exit(void)
2432 static_key_deferred_flush(&apic_hw_disabled
);
2433 static_key_deferred_flush(&apic_sw_disabled
);