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 u8
kvm_xapic_id(struct kvm_lapic
*apic
)
120 return kvm_lapic_get_reg(apic
, APIC_ID
) >> 24;
123 static inline u32
kvm_x2apic_id(struct kvm_lapic
*apic
)
125 return apic
->vcpu
->vcpu_id
;
128 static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map
*map
,
129 u32 dest_id
, struct kvm_lapic
***cluster
, u16
*mask
) {
131 case KVM_APIC_MODE_X2APIC
: {
132 u32 offset
= (dest_id
>> 16) * 16;
133 u32 max_apic_id
= map
->max_apic_id
;
135 if (offset
<= max_apic_id
) {
136 u8 cluster_size
= min(max_apic_id
- offset
+ 1, 16U);
138 *cluster
= &map
->phys_map
[offset
];
139 *mask
= dest_id
& (0xffff >> (16 - cluster_size
));
146 case KVM_APIC_MODE_XAPIC_FLAT
:
147 *cluster
= map
->xapic_flat_map
;
148 *mask
= dest_id
& 0xff;
150 case KVM_APIC_MODE_XAPIC_CLUSTER
:
151 *cluster
= map
->xapic_cluster_map
[(dest_id
>> 4) & 0xf];
152 *mask
= dest_id
& 0xf;
160 static void kvm_apic_map_free(struct rcu_head
*rcu
)
162 struct kvm_apic_map
*map
= container_of(rcu
, struct kvm_apic_map
, rcu
);
167 static void recalculate_apic_map(struct kvm
*kvm
)
169 struct kvm_apic_map
*new, *old
= NULL
;
170 struct kvm_vcpu
*vcpu
;
172 u32 max_id
= 255; /* enough space for any xAPIC ID */
174 mutex_lock(&kvm
->arch
.apic_map_lock
);
176 kvm_for_each_vcpu(i
, vcpu
, kvm
)
177 if (kvm_apic_present(vcpu
))
178 max_id
= max(max_id
, kvm_x2apic_id(vcpu
->arch
.apic
));
180 new = kvm_kvzalloc(sizeof(struct kvm_apic_map
) +
181 sizeof(struct kvm_lapic
*) * ((u64
)max_id
+ 1));
186 new->max_apic_id
= max_id
;
188 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
189 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
190 struct kvm_lapic
**cluster
;
194 if (!kvm_apic_present(vcpu
))
197 aid
= apic_x2apic_mode(apic
) ? kvm_x2apic_id(apic
)
198 : kvm_xapic_id(apic
);
199 if (aid
<= new->max_apic_id
)
200 new->phys_map
[aid
] = apic
;
202 ldr
= kvm_lapic_get_reg(apic
, APIC_LDR
);
204 if (apic_x2apic_mode(apic
)) {
205 new->mode
|= KVM_APIC_MODE_X2APIC
;
207 ldr
= GET_APIC_LOGICAL_ID(ldr
);
208 if (kvm_lapic_get_reg(apic
, APIC_DFR
) == APIC_DFR_FLAT
)
209 new->mode
|= KVM_APIC_MODE_XAPIC_FLAT
;
211 new->mode
|= KVM_APIC_MODE_XAPIC_CLUSTER
;
214 if (!kvm_apic_map_get_logical_dest(new, ldr
, &cluster
, &mask
))
218 cluster
[ffs(mask
) - 1] = apic
;
221 old
= rcu_dereference_protected(kvm
->arch
.apic_map
,
222 lockdep_is_held(&kvm
->arch
.apic_map_lock
));
223 rcu_assign_pointer(kvm
->arch
.apic_map
, new);
224 mutex_unlock(&kvm
->arch
.apic_map_lock
);
227 call_rcu(&old
->rcu
, kvm_apic_map_free
);
229 kvm_make_scan_ioapic_request(kvm
);
232 static inline void apic_set_spiv(struct kvm_lapic
*apic
, u32 val
)
234 bool enabled
= val
& APIC_SPIV_APIC_ENABLED
;
236 kvm_lapic_set_reg(apic
, APIC_SPIV
, val
);
238 if (enabled
!= apic
->sw_enabled
) {
239 apic
->sw_enabled
= enabled
;
241 static_key_slow_dec_deferred(&apic_sw_disabled
);
242 recalculate_apic_map(apic
->vcpu
->kvm
);
244 static_key_slow_inc(&apic_sw_disabled
.key
);
248 static inline void kvm_apic_set_xapic_id(struct kvm_lapic
*apic
, u8 id
)
250 kvm_lapic_set_reg(apic
, APIC_ID
, id
<< 24);
251 recalculate_apic_map(apic
->vcpu
->kvm
);
254 static inline void kvm_apic_set_ldr(struct kvm_lapic
*apic
, u32 id
)
256 kvm_lapic_set_reg(apic
, APIC_LDR
, id
);
257 recalculate_apic_map(apic
->vcpu
->kvm
);
260 static inline void kvm_apic_set_x2apic_id(struct kvm_lapic
*apic
, u32 id
)
262 u32 ldr
= ((id
>> 4) << 16) | (1 << (id
& 0xf));
264 WARN_ON_ONCE(id
!= apic
->vcpu
->vcpu_id
);
266 kvm_lapic_set_reg(apic
, APIC_ID
, id
);
267 kvm_lapic_set_reg(apic
, APIC_LDR
, ldr
);
268 recalculate_apic_map(apic
->vcpu
->kvm
);
271 static inline int apic_lvt_enabled(struct kvm_lapic
*apic
, int lvt_type
)
273 return !(kvm_lapic_get_reg(apic
, lvt_type
) & APIC_LVT_MASKED
);
276 static inline int apic_lvt_vector(struct kvm_lapic
*apic
, int lvt_type
)
278 return kvm_lapic_get_reg(apic
, lvt_type
) & APIC_VECTOR_MASK
;
281 static inline int apic_lvtt_oneshot(struct kvm_lapic
*apic
)
283 return apic
->lapic_timer
.timer_mode
== APIC_LVT_TIMER_ONESHOT
;
286 static inline int apic_lvtt_period(struct kvm_lapic
*apic
)
288 return apic
->lapic_timer
.timer_mode
== APIC_LVT_TIMER_PERIODIC
;
291 static inline int apic_lvtt_tscdeadline(struct kvm_lapic
*apic
)
293 return apic
->lapic_timer
.timer_mode
== APIC_LVT_TIMER_TSCDEADLINE
;
296 static inline int apic_lvt_nmi_mode(u32 lvt_val
)
298 return (lvt_val
& (APIC_MODE_MASK
| APIC_LVT_MASKED
)) == APIC_DM_NMI
;
301 void kvm_apic_set_version(struct kvm_vcpu
*vcpu
)
303 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
304 struct kvm_cpuid_entry2
*feat
;
305 u32 v
= APIC_VERSION
;
307 if (!lapic_in_kernel(vcpu
))
310 feat
= kvm_find_cpuid_entry(apic
->vcpu
, 0x1, 0);
311 if (feat
&& (feat
->ecx
& (1 << (X86_FEATURE_X2APIC
& 31))))
312 v
|= APIC_LVR_DIRECTED_EOI
;
313 kvm_lapic_set_reg(apic
, APIC_LVR
, v
);
316 static const unsigned int apic_lvt_mask
[KVM_APIC_LVT_NUM
] = {
317 LVT_MASK
, /* part LVTT mask, timer mode mask added at runtime */
318 LVT_MASK
| APIC_MODE_MASK
, /* LVTTHMR */
319 LVT_MASK
| APIC_MODE_MASK
, /* LVTPC */
320 LINT_MASK
, LINT_MASK
, /* LVT0-1 */
321 LVT_MASK
/* LVTERR */
324 static int find_highest_vector(void *bitmap
)
329 for (vec
= MAX_APIC_VECTOR
- APIC_VECTORS_PER_REG
;
330 vec
>= 0; vec
-= APIC_VECTORS_PER_REG
) {
331 reg
= bitmap
+ REG_POS(vec
);
333 return fls(*reg
) - 1 + vec
;
339 static u8
count_vectors(void *bitmap
)
345 for (vec
= 0; vec
< MAX_APIC_VECTOR
; vec
+= APIC_VECTORS_PER_REG
) {
346 reg
= bitmap
+ REG_POS(vec
);
347 count
+= hweight32(*reg
);
353 void __kvm_apic_update_irr(u32
*pir
, void *regs
)
357 for (i
= 0; i
<= 7; i
++) {
358 pir_val
= READ_ONCE(pir
[i
]);
360 pir_val
= xchg(&pir
[i
], 0);
361 *((u32
*)(regs
+ APIC_IRR
+ i
* 0x10)) |= pir_val
;
365 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr
);
367 void kvm_apic_update_irr(struct kvm_vcpu
*vcpu
, u32
*pir
)
369 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
371 __kvm_apic_update_irr(pir
, apic
->regs
);
373 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
375 EXPORT_SYMBOL_GPL(kvm_apic_update_irr
);
377 static inline int apic_search_irr(struct kvm_lapic
*apic
)
379 return find_highest_vector(apic
->regs
+ APIC_IRR
);
382 static inline int apic_find_highest_irr(struct kvm_lapic
*apic
)
387 * Note that irr_pending is just a hint. It will be always
388 * true with virtual interrupt delivery enabled.
390 if (!apic
->irr_pending
)
393 if (apic
->vcpu
->arch
.apicv_active
)
394 kvm_x86_ops
->sync_pir_to_irr(apic
->vcpu
);
395 result
= apic_search_irr(apic
);
396 ASSERT(result
== -1 || result
>= 16);
401 static inline void apic_clear_irr(int vec
, struct kvm_lapic
*apic
)
403 struct kvm_vcpu
*vcpu
;
407 if (unlikely(vcpu
->arch
.apicv_active
)) {
408 /* try to update RVI */
409 apic_clear_vector(vec
, apic
->regs
+ APIC_IRR
);
410 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
412 apic
->irr_pending
= false;
413 apic_clear_vector(vec
, apic
->regs
+ APIC_IRR
);
414 if (apic_search_irr(apic
) != -1)
415 apic
->irr_pending
= true;
419 static inline void apic_set_isr(int vec
, struct kvm_lapic
*apic
)
421 struct kvm_vcpu
*vcpu
;
423 if (__apic_test_and_set_vector(vec
, apic
->regs
+ APIC_ISR
))
429 * With APIC virtualization enabled, all caching is disabled
430 * because the processor can modify ISR under the hood. Instead
433 if (unlikely(vcpu
->arch
.apicv_active
))
434 kvm_x86_ops
->hwapic_isr_update(vcpu
, vec
);
437 BUG_ON(apic
->isr_count
> MAX_APIC_VECTOR
);
439 * ISR (in service register) bit is set when injecting an interrupt.
440 * The highest vector is injected. Thus the latest bit set matches
441 * the highest bit in ISR.
443 apic
->highest_isr_cache
= vec
;
447 static inline int apic_find_highest_isr(struct kvm_lapic
*apic
)
452 * Note that isr_count is always 1, and highest_isr_cache
453 * is always -1, with APIC virtualization enabled.
455 if (!apic
->isr_count
)
457 if (likely(apic
->highest_isr_cache
!= -1))
458 return apic
->highest_isr_cache
;
460 result
= find_highest_vector(apic
->regs
+ APIC_ISR
);
461 ASSERT(result
== -1 || result
>= 16);
466 static inline void apic_clear_isr(int vec
, struct kvm_lapic
*apic
)
468 struct kvm_vcpu
*vcpu
;
469 if (!__apic_test_and_clear_vector(vec
, apic
->regs
+ APIC_ISR
))
475 * We do get here for APIC virtualization enabled if the guest
476 * uses the Hyper-V APIC enlightenment. In this case we may need
477 * to trigger a new interrupt delivery by writing the SVI field;
478 * on the other hand isr_count and highest_isr_cache are unused
479 * and must be left alone.
481 if (unlikely(vcpu
->arch
.apicv_active
))
482 kvm_x86_ops
->hwapic_isr_update(vcpu
,
483 apic_find_highest_isr(apic
));
486 BUG_ON(apic
->isr_count
< 0);
487 apic
->highest_isr_cache
= -1;
491 int kvm_lapic_find_highest_irr(struct kvm_vcpu
*vcpu
)
493 /* This may race with setting of irr in __apic_accept_irq() and
494 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
495 * will cause vmexit immediately and the value will be recalculated
496 * on the next vmentry.
498 return apic_find_highest_irr(vcpu
->arch
.apic
);
501 static int __apic_accept_irq(struct kvm_lapic
*apic
, int delivery_mode
,
502 int vector
, int level
, int trig_mode
,
503 struct dest_map
*dest_map
);
505 int kvm_apic_set_irq(struct kvm_vcpu
*vcpu
, struct kvm_lapic_irq
*irq
,
506 struct dest_map
*dest_map
)
508 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
510 return __apic_accept_irq(apic
, irq
->delivery_mode
, irq
->vector
,
511 irq
->level
, irq
->trig_mode
, dest_map
);
514 static int pv_eoi_put_user(struct kvm_vcpu
*vcpu
, u8 val
)
517 return kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.pv_eoi
.data
, &val
,
521 static int pv_eoi_get_user(struct kvm_vcpu
*vcpu
, u8
*val
)
524 return kvm_read_guest_cached(vcpu
->kvm
, &vcpu
->arch
.pv_eoi
.data
, val
,
528 static inline bool pv_eoi_enabled(struct kvm_vcpu
*vcpu
)
530 return vcpu
->arch
.pv_eoi
.msr_val
& KVM_MSR_ENABLED
;
533 static bool pv_eoi_get_pending(struct kvm_vcpu
*vcpu
)
536 if (pv_eoi_get_user(vcpu
, &val
) < 0)
537 apic_debug("Can't read EOI MSR value: 0x%llx\n",
538 (unsigned long long)vcpu
->arch
.pv_eoi
.msr_val
);
542 static void pv_eoi_set_pending(struct kvm_vcpu
*vcpu
)
544 if (pv_eoi_put_user(vcpu
, KVM_PV_EOI_ENABLED
) < 0) {
545 apic_debug("Can't set EOI MSR value: 0x%llx\n",
546 (unsigned long long)vcpu
->arch
.pv_eoi
.msr_val
);
549 __set_bit(KVM_APIC_PV_EOI_PENDING
, &vcpu
->arch
.apic_attention
);
552 static void pv_eoi_clr_pending(struct kvm_vcpu
*vcpu
)
554 if (pv_eoi_put_user(vcpu
, KVM_PV_EOI_DISABLED
) < 0) {
555 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
556 (unsigned long long)vcpu
->arch
.pv_eoi
.msr_val
);
559 __clear_bit(KVM_APIC_PV_EOI_PENDING
, &vcpu
->arch
.apic_attention
);
562 static void apic_update_ppr(struct kvm_lapic
*apic
)
564 u32 tpr
, isrv
, ppr
, old_ppr
;
567 old_ppr
= kvm_lapic_get_reg(apic
, APIC_PROCPRI
);
568 tpr
= kvm_lapic_get_reg(apic
, APIC_TASKPRI
);
569 isr
= apic_find_highest_isr(apic
);
570 isrv
= (isr
!= -1) ? isr
: 0;
572 if ((tpr
& 0xf0) >= (isrv
& 0xf0))
577 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
578 apic
, ppr
, isr
, isrv
);
580 if (old_ppr
!= ppr
) {
581 kvm_lapic_set_reg(apic
, APIC_PROCPRI
, ppr
);
583 kvm_make_request(KVM_REQ_EVENT
, apic
->vcpu
);
587 static void apic_set_tpr(struct kvm_lapic
*apic
, u32 tpr
)
589 kvm_lapic_set_reg(apic
, APIC_TASKPRI
, tpr
);
590 apic_update_ppr(apic
);
593 static bool kvm_apic_broadcast(struct kvm_lapic
*apic
, u32 mda
)
595 if (apic_x2apic_mode(apic
))
596 return mda
== X2APIC_BROADCAST
;
598 return GET_APIC_DEST_FIELD(mda
) == APIC_BROADCAST
;
601 static bool kvm_apic_match_physical_addr(struct kvm_lapic
*apic
, u32 mda
)
603 if (kvm_apic_broadcast(apic
, mda
))
606 if (apic_x2apic_mode(apic
))
607 return mda
== kvm_x2apic_id(apic
);
609 return mda
== SET_APIC_DEST_FIELD(kvm_xapic_id(apic
));
612 static bool kvm_apic_match_logical_addr(struct kvm_lapic
*apic
, u32 mda
)
616 if (kvm_apic_broadcast(apic
, mda
))
619 logical_id
= kvm_lapic_get_reg(apic
, APIC_LDR
);
621 if (apic_x2apic_mode(apic
))
622 return ((logical_id
>> 16) == (mda
>> 16))
623 && (logical_id
& mda
& 0xffff) != 0;
625 logical_id
= GET_APIC_LOGICAL_ID(logical_id
);
626 mda
= GET_APIC_DEST_FIELD(mda
);
628 switch (kvm_lapic_get_reg(apic
, APIC_DFR
)) {
630 return (logical_id
& mda
) != 0;
631 case APIC_DFR_CLUSTER
:
632 return ((logical_id
>> 4) == (mda
>> 4))
633 && (logical_id
& mda
& 0xf) != 0;
635 apic_debug("Bad DFR vcpu %d: %08x\n",
636 apic
->vcpu
->vcpu_id
, kvm_lapic_get_reg(apic
, APIC_DFR
));
641 /* The KVM local APIC implementation has two quirks:
643 * - the xAPIC MDA stores the destination at bits 24-31, while this
644 * is not true of struct kvm_lapic_irq's dest_id field. This is
645 * just a quirk in the API and is not problematic.
647 * - in-kernel IOAPIC messages have to be delivered directly to
648 * x2APIC, because the kernel does not support interrupt remapping.
649 * In order to support broadcast without interrupt remapping, x2APIC
650 * rewrites the destination of non-IPI messages from APIC_BROADCAST
651 * to X2APIC_BROADCAST.
653 * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API. This is
654 * important when userspace wants to use x2APIC-format MSIs, because
655 * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
657 static u32
kvm_apic_mda(struct kvm_vcpu
*vcpu
, unsigned int dest_id
,
658 struct kvm_lapic
*source
, struct kvm_lapic
*target
)
660 bool ipi
= source
!= NULL
;
661 bool x2apic_mda
= apic_x2apic_mode(ipi
? source
: target
);
663 if (!vcpu
->kvm
->arch
.x2apic_broadcast_quirk_disabled
&&
664 !ipi
&& dest_id
== APIC_BROADCAST
&& x2apic_mda
)
665 return X2APIC_BROADCAST
;
667 return x2apic_mda
? dest_id
: SET_APIC_DEST_FIELD(dest_id
);
670 bool kvm_apic_match_dest(struct kvm_vcpu
*vcpu
, struct kvm_lapic
*source
,
671 int short_hand
, unsigned int dest
, int dest_mode
)
673 struct kvm_lapic
*target
= vcpu
->arch
.apic
;
674 u32 mda
= kvm_apic_mda(vcpu
, dest
, source
, target
);
676 apic_debug("target %p, source %p, dest 0x%x, "
677 "dest_mode 0x%x, short_hand 0x%x\n",
678 target
, source
, dest
, dest_mode
, short_hand
);
681 switch (short_hand
) {
682 case APIC_DEST_NOSHORT
:
683 if (dest_mode
== APIC_DEST_PHYSICAL
)
684 return kvm_apic_match_physical_addr(target
, mda
);
686 return kvm_apic_match_logical_addr(target
, mda
);
688 return target
== source
;
689 case APIC_DEST_ALLINC
:
691 case APIC_DEST_ALLBUT
:
692 return target
!= source
;
694 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
699 EXPORT_SYMBOL_GPL(kvm_apic_match_dest
);
701 int kvm_vector_to_index(u32 vector
, u32 dest_vcpus
,
702 const unsigned long *bitmap
, u32 bitmap_size
)
707 mod
= vector
% dest_vcpus
;
709 for (i
= 0; i
<= mod
; i
++) {
710 idx
= find_next_bit(bitmap
, bitmap_size
, idx
+ 1);
711 BUG_ON(idx
== bitmap_size
);
717 static void kvm_apic_disabled_lapic_found(struct kvm
*kvm
)
719 if (!kvm
->arch
.disabled_lapic_found
) {
720 kvm
->arch
.disabled_lapic_found
= true;
722 "Disabled LAPIC found during irq injection\n");
726 static bool kvm_apic_is_broadcast_dest(struct kvm
*kvm
, struct kvm_lapic
**src
,
727 struct kvm_lapic_irq
*irq
, struct kvm_apic_map
*map
)
729 if (kvm
->arch
.x2apic_broadcast_quirk_disabled
) {
730 if ((irq
->dest_id
== APIC_BROADCAST
&&
731 map
->mode
!= KVM_APIC_MODE_X2APIC
))
733 if (irq
->dest_id
== X2APIC_BROADCAST
)
736 bool x2apic_ipi
= src
&& *src
&& apic_x2apic_mode(*src
);
737 if (irq
->dest_id
== (x2apic_ipi
?
738 X2APIC_BROADCAST
: APIC_BROADCAST
))
745 /* Return true if the interrupt can be handled by using *bitmap as index mask
746 * for valid destinations in *dst array.
747 * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
748 * Note: we may have zero kvm_lapic destinations when we return true, which
749 * means that the interrupt should be dropped. In this case, *bitmap would be
750 * zero and *dst undefined.
752 static inline bool kvm_apic_map_get_dest_lapic(struct kvm
*kvm
,
753 struct kvm_lapic
**src
, struct kvm_lapic_irq
*irq
,
754 struct kvm_apic_map
*map
, struct kvm_lapic
***dst
,
755 unsigned long *bitmap
)
759 if (irq
->shorthand
== APIC_DEST_SELF
&& src
) {
763 } else if (irq
->shorthand
)
766 if (!map
|| kvm_apic_is_broadcast_dest(kvm
, src
, irq
, map
))
769 if (irq
->dest_mode
== APIC_DEST_PHYSICAL
) {
770 if (irq
->dest_id
> map
->max_apic_id
) {
773 *dst
= &map
->phys_map
[irq
->dest_id
];
780 if (!kvm_apic_map_get_logical_dest(map
, irq
->dest_id
, dst
,
784 if (!kvm_lowest_prio_delivery(irq
))
787 if (!kvm_vector_hashing_enabled()) {
789 for_each_set_bit(i
, bitmap
, 16) {
794 else if (kvm_apic_compare_prio((*dst
)[i
]->vcpu
,
795 (*dst
)[lowest
]->vcpu
) < 0)
802 lowest
= kvm_vector_to_index(irq
->vector
, hweight16(*bitmap
),
805 if (!(*dst
)[lowest
]) {
806 kvm_apic_disabled_lapic_found(kvm
);
812 *bitmap
= (lowest
>= 0) ? 1 << lowest
: 0;
817 bool kvm_irq_delivery_to_apic_fast(struct kvm
*kvm
, struct kvm_lapic
*src
,
818 struct kvm_lapic_irq
*irq
, int *r
, struct dest_map
*dest_map
)
820 struct kvm_apic_map
*map
;
821 unsigned long bitmap
;
822 struct kvm_lapic
**dst
= NULL
;
828 if (irq
->shorthand
== APIC_DEST_SELF
) {
829 *r
= kvm_apic_set_irq(src
->vcpu
, irq
, dest_map
);
834 map
= rcu_dereference(kvm
->arch
.apic_map
);
836 ret
= kvm_apic_map_get_dest_lapic(kvm
, &src
, irq
, map
, &dst
, &bitmap
);
838 for_each_set_bit(i
, &bitmap
, 16) {
843 *r
+= kvm_apic_set_irq(dst
[i
]->vcpu
, irq
, dest_map
);
851 * This routine tries to handler interrupts in posted mode, here is how
852 * it deals with different cases:
853 * - For single-destination interrupts, handle it in posted mode
854 * - Else if vector hashing is enabled and it is a lowest-priority
855 * interrupt, handle it in posted mode and use the following mechanism
856 * to find the destinaiton vCPU.
857 * 1. For lowest-priority interrupts, store all the possible
858 * destination vCPUs in an array.
859 * 2. Use "guest vector % max number of destination vCPUs" to find
860 * the right destination vCPU in the array for the lowest-priority
862 * - Otherwise, use remapped mode to inject the interrupt.
864 bool kvm_intr_is_single_vcpu_fast(struct kvm
*kvm
, struct kvm_lapic_irq
*irq
,
865 struct kvm_vcpu
**dest_vcpu
)
867 struct kvm_apic_map
*map
;
868 unsigned long bitmap
;
869 struct kvm_lapic
**dst
= NULL
;
876 map
= rcu_dereference(kvm
->arch
.apic_map
);
878 if (kvm_apic_map_get_dest_lapic(kvm
, NULL
, irq
, map
, &dst
, &bitmap
) &&
879 hweight16(bitmap
) == 1) {
880 unsigned long i
= find_first_bit(&bitmap
, 16);
883 *dest_vcpu
= dst
[i
]->vcpu
;
893 * Add a pending IRQ into lapic.
894 * Return 1 if successfully added and 0 if discarded.
896 static int __apic_accept_irq(struct kvm_lapic
*apic
, int delivery_mode
,
897 int vector
, int level
, int trig_mode
,
898 struct dest_map
*dest_map
)
901 struct kvm_vcpu
*vcpu
= apic
->vcpu
;
903 trace_kvm_apic_accept_irq(vcpu
->vcpu_id
, delivery_mode
,
905 switch (delivery_mode
) {
907 vcpu
->arch
.apic_arb_prio
++;
909 if (unlikely(trig_mode
&& !level
))
912 /* FIXME add logic for vcpu on reset */
913 if (unlikely(!apic_enabled(apic
)))
919 __set_bit(vcpu
->vcpu_id
, dest_map
->map
);
920 dest_map
->vectors
[vcpu
->vcpu_id
] = vector
;
923 if (apic_test_vector(vector
, apic
->regs
+ APIC_TMR
) != !!trig_mode
) {
925 kvm_lapic_set_vector(vector
, apic
->regs
+ APIC_TMR
);
927 apic_clear_vector(vector
, apic
->regs
+ APIC_TMR
);
930 if (vcpu
->arch
.apicv_active
)
931 kvm_x86_ops
->deliver_posted_interrupt(vcpu
, vector
);
933 kvm_lapic_set_irr(vector
, apic
);
935 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
942 vcpu
->arch
.pv
.pv_unhalted
= 1;
943 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
949 kvm_make_request(KVM_REQ_SMI
, vcpu
);
955 kvm_inject_nmi(vcpu
);
960 if (!trig_mode
|| level
) {
962 /* assumes that there are only KVM_APIC_INIT/SIPI */
963 apic
->pending_events
= (1UL << KVM_APIC_INIT
);
964 /* make sure pending_events is visible before sending
967 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
970 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
975 case APIC_DM_STARTUP
:
976 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
977 vcpu
->vcpu_id
, vector
);
979 apic
->sipi_vector
= vector
;
980 /* make sure sipi_vector is visible for the receiver */
982 set_bit(KVM_APIC_SIPI
, &apic
->pending_events
);
983 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
989 * Should only be called by kvm_apic_local_deliver() with LVT0,
990 * before NMI watchdog was enabled. Already handled by
991 * kvm_apic_accept_pic_intr().
996 printk(KERN_ERR
"TODO: unsupported delivery mode %x\n",
1003 int kvm_apic_compare_prio(struct kvm_vcpu
*vcpu1
, struct kvm_vcpu
*vcpu2
)
1005 return vcpu1
->arch
.apic_arb_prio
- vcpu2
->arch
.apic_arb_prio
;
1008 static bool kvm_ioapic_handles_vector(struct kvm_lapic
*apic
, int vector
)
1010 return test_bit(vector
, apic
->vcpu
->arch
.ioapic_handled_vectors
);
1013 static void kvm_ioapic_send_eoi(struct kvm_lapic
*apic
, int vector
)
1017 /* Eoi the ioapic only if the ioapic doesn't own the vector. */
1018 if (!kvm_ioapic_handles_vector(apic
, vector
))
1021 /* Request a KVM exit to inform the userspace IOAPIC. */
1022 if (irqchip_split(apic
->vcpu
->kvm
)) {
1023 apic
->vcpu
->arch
.pending_ioapic_eoi
= vector
;
1024 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT
, apic
->vcpu
);
1028 if (apic_test_vector(vector
, apic
->regs
+ APIC_TMR
))
1029 trigger_mode
= IOAPIC_LEVEL_TRIG
;
1031 trigger_mode
= IOAPIC_EDGE_TRIG
;
1033 kvm_ioapic_update_eoi(apic
->vcpu
, vector
, trigger_mode
);
1036 static int apic_set_eoi(struct kvm_lapic
*apic
)
1038 int vector
= apic_find_highest_isr(apic
);
1040 trace_kvm_eoi(apic
, vector
);
1043 * Not every write EOI will has corresponding ISR,
1044 * one example is when Kernel check timer on setup_IO_APIC
1049 apic_clear_isr(vector
, apic
);
1050 apic_update_ppr(apic
);
1052 if (test_bit(vector
, vcpu_to_synic(apic
->vcpu
)->vec_bitmap
))
1053 kvm_hv_synic_send_eoi(apic
->vcpu
, vector
);
1055 kvm_ioapic_send_eoi(apic
, vector
);
1056 kvm_make_request(KVM_REQ_EVENT
, apic
->vcpu
);
1061 * this interface assumes a trap-like exit, which has already finished
1062 * desired side effect including vISR and vPPR update.
1064 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu
*vcpu
, int vector
)
1066 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1068 trace_kvm_eoi(apic
, vector
);
1070 kvm_ioapic_send_eoi(apic
, vector
);
1071 kvm_make_request(KVM_REQ_EVENT
, apic
->vcpu
);
1073 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated
);
1075 static void apic_send_ipi(struct kvm_lapic
*apic
)
1077 u32 icr_low
= kvm_lapic_get_reg(apic
, APIC_ICR
);
1078 u32 icr_high
= kvm_lapic_get_reg(apic
, APIC_ICR2
);
1079 struct kvm_lapic_irq irq
;
1081 irq
.vector
= icr_low
& APIC_VECTOR_MASK
;
1082 irq
.delivery_mode
= icr_low
& APIC_MODE_MASK
;
1083 irq
.dest_mode
= icr_low
& APIC_DEST_MASK
;
1084 irq
.level
= (icr_low
& APIC_INT_ASSERT
) != 0;
1085 irq
.trig_mode
= icr_low
& APIC_INT_LEVELTRIG
;
1086 irq
.shorthand
= icr_low
& APIC_SHORT_MASK
;
1087 irq
.msi_redir_hint
= false;
1088 if (apic_x2apic_mode(apic
))
1089 irq
.dest_id
= icr_high
;
1091 irq
.dest_id
= GET_APIC_DEST_FIELD(icr_high
);
1093 trace_kvm_apic_ipi(icr_low
, irq
.dest_id
);
1095 apic_debug("icr_high 0x%x, icr_low 0x%x, "
1096 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
1097 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x, "
1098 "msi_redir_hint 0x%x\n",
1099 icr_high
, icr_low
, irq
.shorthand
, irq
.dest_id
,
1100 irq
.trig_mode
, irq
.level
, irq
.dest_mode
, irq
.delivery_mode
,
1101 irq
.vector
, irq
.msi_redir_hint
);
1103 kvm_irq_delivery_to_apic(apic
->vcpu
->kvm
, apic
, &irq
, NULL
);
1106 static u32
apic_get_tmcct(struct kvm_lapic
*apic
)
1108 ktime_t remaining
, now
;
1112 ASSERT(apic
!= NULL
);
1114 /* if initial count is 0, current count should also be 0 */
1115 if (kvm_lapic_get_reg(apic
, APIC_TMICT
) == 0 ||
1116 apic
->lapic_timer
.period
== 0)
1120 remaining
= ktime_sub(apic
->lapic_timer
.target_expiration
, now
);
1121 if (ktime_to_ns(remaining
) < 0)
1124 ns
= mod_64(ktime_to_ns(remaining
), apic
->lapic_timer
.period
);
1125 tmcct
= div64_u64(ns
,
1126 (APIC_BUS_CYCLE_NS
* apic
->divide_count
));
1131 static void __report_tpr_access(struct kvm_lapic
*apic
, bool write
)
1133 struct kvm_vcpu
*vcpu
= apic
->vcpu
;
1134 struct kvm_run
*run
= vcpu
->run
;
1136 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS
, vcpu
);
1137 run
->tpr_access
.rip
= kvm_rip_read(vcpu
);
1138 run
->tpr_access
.is_write
= write
;
1141 static inline void report_tpr_access(struct kvm_lapic
*apic
, bool write
)
1143 if (apic
->vcpu
->arch
.tpr_access_reporting
)
1144 __report_tpr_access(apic
, write
);
1147 static u32
__apic_read(struct kvm_lapic
*apic
, unsigned int offset
)
1151 if (offset
>= LAPIC_MMIO_LENGTH
)
1156 apic_debug("Access APIC ARBPRI register which is for P6\n");
1159 case APIC_TMCCT
: /* Timer CCR */
1160 if (apic_lvtt_tscdeadline(apic
))
1163 val
= apic_get_tmcct(apic
);
1166 apic_update_ppr(apic
);
1167 val
= kvm_lapic_get_reg(apic
, offset
);
1170 report_tpr_access(apic
, false);
1173 val
= kvm_lapic_get_reg(apic
, offset
);
1180 static inline struct kvm_lapic
*to_lapic(struct kvm_io_device
*dev
)
1182 return container_of(dev
, struct kvm_lapic
, dev
);
1185 int kvm_lapic_reg_read(struct kvm_lapic
*apic
, u32 offset
, int len
,
1188 unsigned char alignment
= offset
& 0xf;
1190 /* this bitmask has a bit cleared for each reserved register */
1191 static const u64 rmask
= 0x43ff01ffffffe70cULL
;
1193 if ((alignment
+ len
) > 4) {
1194 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1199 if (offset
> 0x3f0 || !(rmask
& (1ULL << (offset
>> 4)))) {
1200 apic_debug("KVM_APIC_READ: read reserved register %x\n",
1205 result
= __apic_read(apic
, offset
& ~0xf);
1207 trace_kvm_apic_read(offset
, result
);
1213 memcpy(data
, (char *)&result
+ alignment
, len
);
1216 printk(KERN_ERR
"Local APIC read with len = %x, "
1217 "should be 1,2, or 4 instead\n", len
);
1222 EXPORT_SYMBOL_GPL(kvm_lapic_reg_read
);
1224 static int apic_mmio_in_range(struct kvm_lapic
*apic
, gpa_t addr
)
1226 return kvm_apic_hw_enabled(apic
) &&
1227 addr
>= apic
->base_address
&&
1228 addr
< apic
->base_address
+ LAPIC_MMIO_LENGTH
;
1231 static int apic_mmio_read(struct kvm_vcpu
*vcpu
, struct kvm_io_device
*this,
1232 gpa_t address
, int len
, void *data
)
1234 struct kvm_lapic
*apic
= to_lapic(this);
1235 u32 offset
= address
- apic
->base_address
;
1237 if (!apic_mmio_in_range(apic
, address
))
1240 kvm_lapic_reg_read(apic
, offset
, len
, data
);
1245 static void update_divide_count(struct kvm_lapic
*apic
)
1247 u32 tmp1
, tmp2
, tdcr
;
1249 tdcr
= kvm_lapic_get_reg(apic
, APIC_TDCR
);
1251 tmp2
= ((tmp1
& 0x3) | ((tmp1
& 0x8) >> 1)) + 1;
1252 apic
->divide_count
= 0x1 << (tmp2
& 0x7);
1254 apic_debug("timer divide count is 0x%x\n",
1255 apic
->divide_count
);
1258 static void apic_update_lvtt(struct kvm_lapic
*apic
)
1260 u32 timer_mode
= kvm_lapic_get_reg(apic
, APIC_LVTT
) &
1261 apic
->lapic_timer
.timer_mode_mask
;
1263 if (apic
->lapic_timer
.timer_mode
!= timer_mode
) {
1264 apic
->lapic_timer
.timer_mode
= timer_mode
;
1265 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1269 static void apic_timer_expired(struct kvm_lapic
*apic
)
1271 struct kvm_vcpu
*vcpu
= apic
->vcpu
;
1272 struct swait_queue_head
*q
= &vcpu
->wq
;
1273 struct kvm_timer
*ktimer
= &apic
->lapic_timer
;
1275 if (atomic_read(&apic
->lapic_timer
.pending
))
1278 atomic_inc(&apic
->lapic_timer
.pending
);
1279 kvm_set_pending_timer(vcpu
);
1281 if (swait_active(q
))
1284 if (apic_lvtt_tscdeadline(apic
))
1285 ktimer
->expired_tscdeadline
= ktimer
->tscdeadline
;
1289 * On APICv, this test will cause a busy wait
1290 * during a higher-priority task.
1293 static bool lapic_timer_int_injected(struct kvm_vcpu
*vcpu
)
1295 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1296 u32 reg
= kvm_lapic_get_reg(apic
, APIC_LVTT
);
1298 if (kvm_apic_hw_enabled(apic
)) {
1299 int vec
= reg
& APIC_VECTOR_MASK
;
1300 void *bitmap
= apic
->regs
+ APIC_ISR
;
1302 if (vcpu
->arch
.apicv_active
)
1303 bitmap
= apic
->regs
+ APIC_IRR
;
1305 if (apic_test_vector(vec
, bitmap
))
1311 void wait_lapic_expire(struct kvm_vcpu
*vcpu
)
1313 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1314 u64 guest_tsc
, tsc_deadline
;
1316 if (!lapic_in_kernel(vcpu
))
1319 if (apic
->lapic_timer
.expired_tscdeadline
== 0)
1322 if (!lapic_timer_int_injected(vcpu
))
1325 tsc_deadline
= apic
->lapic_timer
.expired_tscdeadline
;
1326 apic
->lapic_timer
.expired_tscdeadline
= 0;
1327 guest_tsc
= kvm_read_l1_tsc(vcpu
, rdtsc());
1328 trace_kvm_wait_lapic_expire(vcpu
->vcpu_id
, guest_tsc
- tsc_deadline
);
1330 /* __delay is delay_tsc whenever the hardware has TSC, thus always. */
1331 if (guest_tsc
< tsc_deadline
)
1332 __delay(min(tsc_deadline
- guest_tsc
,
1333 nsec_to_cycles(vcpu
, lapic_timer_advance_ns
)));
1336 static void start_sw_tscdeadline(struct kvm_lapic
*apic
)
1338 u64 guest_tsc
, tscdeadline
= apic
->lapic_timer
.tscdeadline
;
1341 struct kvm_vcpu
*vcpu
= apic
->vcpu
;
1342 unsigned long this_tsc_khz
= vcpu
->arch
.virtual_tsc_khz
;
1343 unsigned long flags
;
1346 if (unlikely(!tscdeadline
|| !this_tsc_khz
))
1349 local_irq_save(flags
);
1352 guest_tsc
= kvm_read_l1_tsc(vcpu
, rdtsc());
1353 if (likely(tscdeadline
> guest_tsc
)) {
1354 ns
= (tscdeadline
- guest_tsc
) * 1000000ULL;
1355 do_div(ns
, this_tsc_khz
);
1356 expire
= ktime_add_ns(now
, ns
);
1357 expire
= ktime_sub_ns(expire
, lapic_timer_advance_ns
);
1358 hrtimer_start(&apic
->lapic_timer
.timer
,
1359 expire
, HRTIMER_MODE_ABS_PINNED
);
1361 apic_timer_expired(apic
);
1363 local_irq_restore(flags
);
1366 static void start_sw_period(struct kvm_lapic
*apic
)
1368 if (!apic
->lapic_timer
.period
)
1371 if (apic_lvtt_oneshot(apic
) &&
1372 ktime_after(ktime_get(),
1373 apic
->lapic_timer
.target_expiration
)) {
1374 apic_timer_expired(apic
);
1378 hrtimer_start(&apic
->lapic_timer
.timer
,
1379 apic
->lapic_timer
.target_expiration
,
1380 HRTIMER_MODE_ABS_PINNED
);
1383 static bool set_target_expiration(struct kvm_lapic
*apic
)
1389 apic
->lapic_timer
.period
= (u64
)kvm_lapic_get_reg(apic
, APIC_TMICT
)
1390 * APIC_BUS_CYCLE_NS
* apic
->divide_count
;
1392 if (!apic
->lapic_timer
.period
)
1396 * Do not allow the guest to program periodic timers with small
1397 * interval, since the hrtimers are not throttled by the host
1400 if (apic_lvtt_period(apic
)) {
1401 s64 min_period
= min_timer_period_us
* 1000LL;
1403 if (apic
->lapic_timer
.period
< min_period
) {
1404 pr_info_ratelimited(
1405 "kvm: vcpu %i: requested %lld ns "
1406 "lapic timer period limited to %lld ns\n",
1407 apic
->vcpu
->vcpu_id
,
1408 apic
->lapic_timer
.period
, min_period
);
1409 apic
->lapic_timer
.period
= min_period
;
1413 apic_debug("%s: bus cycle is %" PRId64
"ns, now 0x%016"
1415 "timer initial count 0x%x, period %lldns, "
1416 "expire @ 0x%016" PRIx64
".\n", __func__
,
1417 APIC_BUS_CYCLE_NS
, ktime_to_ns(now
),
1418 kvm_lapic_get_reg(apic
, APIC_TMICT
),
1419 apic
->lapic_timer
.period
,
1420 ktime_to_ns(ktime_add_ns(now
,
1421 apic
->lapic_timer
.period
)));
1423 apic
->lapic_timer
.tscdeadline
= kvm_read_l1_tsc(apic
->vcpu
, tscl
) +
1424 nsec_to_cycles(apic
->vcpu
, apic
->lapic_timer
.period
);
1425 apic
->lapic_timer
.target_expiration
= ktime_add_ns(now
, apic
->lapic_timer
.period
);
1430 static void advance_periodic_target_expiration(struct kvm_lapic
*apic
)
1432 apic
->lapic_timer
.tscdeadline
+=
1433 nsec_to_cycles(apic
->vcpu
, apic
->lapic_timer
.period
);
1434 apic
->lapic_timer
.target_expiration
=
1435 ktime_add_ns(apic
->lapic_timer
.target_expiration
,
1436 apic
->lapic_timer
.period
);
1439 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu
*vcpu
)
1441 if (!lapic_in_kernel(vcpu
))
1444 return vcpu
->arch
.apic
->lapic_timer
.hv_timer_in_use
;
1446 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use
);
1448 static void cancel_hv_timer(struct kvm_lapic
*apic
)
1450 kvm_x86_ops
->cancel_hv_timer(apic
->vcpu
);
1451 apic
->lapic_timer
.hv_timer_in_use
= false;
1454 static bool start_hv_timer(struct kvm_lapic
*apic
)
1456 u64 tscdeadline
= apic
->lapic_timer
.tscdeadline
;
1458 if ((atomic_read(&apic
->lapic_timer
.pending
) &&
1459 !apic_lvtt_period(apic
)) ||
1460 kvm_x86_ops
->set_hv_timer(apic
->vcpu
, tscdeadline
)) {
1461 if (apic
->lapic_timer
.hv_timer_in_use
)
1462 cancel_hv_timer(apic
);
1464 apic
->lapic_timer
.hv_timer_in_use
= true;
1465 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1467 /* In case the sw timer triggered in the window */
1468 if (atomic_read(&apic
->lapic_timer
.pending
) &&
1469 !apic_lvtt_period(apic
))
1470 cancel_hv_timer(apic
);
1472 trace_kvm_hv_timer_state(apic
->vcpu
->vcpu_id
,
1473 apic
->lapic_timer
.hv_timer_in_use
);
1474 return apic
->lapic_timer
.hv_timer_in_use
;
1477 void kvm_lapic_expired_hv_timer(struct kvm_vcpu
*vcpu
)
1479 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1481 WARN_ON(!apic
->lapic_timer
.hv_timer_in_use
);
1482 WARN_ON(swait_active(&vcpu
->wq
));
1483 cancel_hv_timer(apic
);
1484 apic_timer_expired(apic
);
1486 if (apic_lvtt_period(apic
) && apic
->lapic_timer
.period
) {
1487 advance_periodic_target_expiration(apic
);
1488 if (!start_hv_timer(apic
))
1489 start_sw_period(apic
);
1492 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer
);
1494 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu
*vcpu
)
1496 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1498 WARN_ON(apic
->lapic_timer
.hv_timer_in_use
);
1500 start_hv_timer(apic
);
1502 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer
);
1504 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu
*vcpu
)
1506 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1508 /* Possibly the TSC deadline timer is not enabled yet */
1509 if (!apic
->lapic_timer
.hv_timer_in_use
)
1512 cancel_hv_timer(apic
);
1514 if (atomic_read(&apic
->lapic_timer
.pending
))
1517 if (apic_lvtt_period(apic
) || apic_lvtt_oneshot(apic
))
1518 start_sw_period(apic
);
1519 else if (apic_lvtt_tscdeadline(apic
))
1520 start_sw_tscdeadline(apic
);
1522 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer
);
1524 static void start_apic_timer(struct kvm_lapic
*apic
)
1526 atomic_set(&apic
->lapic_timer
.pending
, 0);
1528 if (apic_lvtt_period(apic
) || apic_lvtt_oneshot(apic
)) {
1529 if (set_target_expiration(apic
) &&
1530 !(kvm_x86_ops
->set_hv_timer
&& start_hv_timer(apic
)))
1531 start_sw_period(apic
);
1532 } else if (apic_lvtt_tscdeadline(apic
)) {
1533 if (!(kvm_x86_ops
->set_hv_timer
&& start_hv_timer(apic
)))
1534 start_sw_tscdeadline(apic
);
1538 static void apic_manage_nmi_watchdog(struct kvm_lapic
*apic
, u32 lvt0_val
)
1540 bool lvt0_in_nmi_mode
= apic_lvt_nmi_mode(lvt0_val
);
1542 if (apic
->lvt0_in_nmi_mode
!= lvt0_in_nmi_mode
) {
1543 apic
->lvt0_in_nmi_mode
= lvt0_in_nmi_mode
;
1544 if (lvt0_in_nmi_mode
) {
1545 apic_debug("Receive NMI setting on APIC_LVT0 "
1546 "for cpu %d\n", apic
->vcpu
->vcpu_id
);
1547 atomic_inc(&apic
->vcpu
->kvm
->arch
.vapics_in_nmi_mode
);
1549 atomic_dec(&apic
->vcpu
->kvm
->arch
.vapics_in_nmi_mode
);
1553 int kvm_lapic_reg_write(struct kvm_lapic
*apic
, u32 reg
, u32 val
)
1557 trace_kvm_apic_write(reg
, val
);
1560 case APIC_ID
: /* Local APIC ID */
1561 if (!apic_x2apic_mode(apic
))
1562 kvm_apic_set_xapic_id(apic
, val
>> 24);
1568 report_tpr_access(apic
, true);
1569 apic_set_tpr(apic
, val
& 0xff);
1577 if (!apic_x2apic_mode(apic
))
1578 kvm_apic_set_ldr(apic
, val
& APIC_LDR_MASK
);
1584 if (!apic_x2apic_mode(apic
)) {
1585 kvm_lapic_set_reg(apic
, APIC_DFR
, val
| 0x0FFFFFFF);
1586 recalculate_apic_map(apic
->vcpu
->kvm
);
1593 if (kvm_lapic_get_reg(apic
, APIC_LVR
) & APIC_LVR_DIRECTED_EOI
)
1594 mask
|= APIC_SPIV_DIRECTED_EOI
;
1595 apic_set_spiv(apic
, val
& mask
);
1596 if (!(val
& APIC_SPIV_APIC_ENABLED
)) {
1600 for (i
= 0; i
< KVM_APIC_LVT_NUM
; i
++) {
1601 lvt_val
= kvm_lapic_get_reg(apic
,
1602 APIC_LVTT
+ 0x10 * i
);
1603 kvm_lapic_set_reg(apic
, APIC_LVTT
+ 0x10 * i
,
1604 lvt_val
| APIC_LVT_MASKED
);
1606 apic_update_lvtt(apic
);
1607 atomic_set(&apic
->lapic_timer
.pending
, 0);
1613 /* No delay here, so we always clear the pending bit */
1614 kvm_lapic_set_reg(apic
, APIC_ICR
, val
& ~(1 << 12));
1615 apic_send_ipi(apic
);
1619 if (!apic_x2apic_mode(apic
))
1621 kvm_lapic_set_reg(apic
, APIC_ICR2
, val
);
1625 apic_manage_nmi_watchdog(apic
, val
);
1630 /* TODO: Check vector */
1631 if (!kvm_apic_sw_enabled(apic
))
1632 val
|= APIC_LVT_MASKED
;
1634 val
&= apic_lvt_mask
[(reg
- APIC_LVTT
) >> 4];
1635 kvm_lapic_set_reg(apic
, reg
, val
);
1640 if (!kvm_apic_sw_enabled(apic
))
1641 val
|= APIC_LVT_MASKED
;
1642 val
&= (apic_lvt_mask
[0] | apic
->lapic_timer
.timer_mode_mask
);
1643 kvm_lapic_set_reg(apic
, APIC_LVTT
, val
);
1644 apic_update_lvtt(apic
);
1648 if (apic_lvtt_tscdeadline(apic
))
1651 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1652 kvm_lapic_set_reg(apic
, APIC_TMICT
, val
);
1653 start_apic_timer(apic
);
1658 apic_debug("KVM_WRITE:TDCR %x\n", val
);
1659 kvm_lapic_set_reg(apic
, APIC_TDCR
, val
);
1660 update_divide_count(apic
);
1664 if (apic_x2apic_mode(apic
) && val
!= 0) {
1665 apic_debug("KVM_WRITE:ESR not zero %x\n", val
);
1671 if (apic_x2apic_mode(apic
)) {
1672 kvm_lapic_reg_write(apic
, APIC_ICR
, 0x40000 | (val
& 0xff));
1681 apic_debug("Local APIC Write to read-only register %x\n", reg
);
1684 EXPORT_SYMBOL_GPL(kvm_lapic_reg_write
);
1686 static int apic_mmio_write(struct kvm_vcpu
*vcpu
, struct kvm_io_device
*this,
1687 gpa_t address
, int len
, const void *data
)
1689 struct kvm_lapic
*apic
= to_lapic(this);
1690 unsigned int offset
= address
- apic
->base_address
;
1693 if (!apic_mmio_in_range(apic
, address
))
1697 * APIC register must be aligned on 128-bits boundary.
1698 * 32/64/128 bits registers must be accessed thru 32 bits.
1701 if (len
!= 4 || (offset
& 0xf)) {
1702 /* Don't shout loud, $infamous_os would cause only noise. */
1703 apic_debug("apic write: bad size=%d %lx\n", len
, (long)address
);
1709 /* too common printing */
1710 if (offset
!= APIC_EOI
)
1711 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1712 "0x%x\n", __func__
, offset
, len
, val
);
1714 kvm_lapic_reg_write(apic
, offset
& 0xff0, val
);
1719 void kvm_lapic_set_eoi(struct kvm_vcpu
*vcpu
)
1721 kvm_lapic_reg_write(vcpu
->arch
.apic
, APIC_EOI
, 0);
1723 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi
);
1725 /* emulate APIC access in a trap manner */
1726 void kvm_apic_write_nodecode(struct kvm_vcpu
*vcpu
, u32 offset
)
1730 /* hw has done the conditional check and inst decode */
1733 kvm_lapic_reg_read(vcpu
->arch
.apic
, offset
, 4, &val
);
1735 /* TODO: optimize to just emulate side effect w/o one more write */
1736 kvm_lapic_reg_write(vcpu
->arch
.apic
, offset
, val
);
1738 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode
);
1740 void kvm_free_lapic(struct kvm_vcpu
*vcpu
)
1742 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1744 if (!vcpu
->arch
.apic
)
1747 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1749 if (!(vcpu
->arch
.apic_base
& MSR_IA32_APICBASE_ENABLE
))
1750 static_key_slow_dec_deferred(&apic_hw_disabled
);
1752 if (!apic
->sw_enabled
)
1753 static_key_slow_dec_deferred(&apic_sw_disabled
);
1756 free_page((unsigned long)apic
->regs
);
1762 *----------------------------------------------------------------------
1764 *----------------------------------------------------------------------
1766 u64
kvm_get_lapic_target_expiration_tsc(struct kvm_vcpu
*vcpu
)
1768 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1770 if (!lapic_in_kernel(vcpu
))
1773 return apic
->lapic_timer
.tscdeadline
;
1776 u64
kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu
*vcpu
)
1778 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1780 if (!lapic_in_kernel(vcpu
) ||
1781 !apic_lvtt_tscdeadline(apic
))
1784 return apic
->lapic_timer
.tscdeadline
;
1787 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu
*vcpu
, u64 data
)
1789 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1791 if (!lapic_in_kernel(vcpu
) || apic_lvtt_oneshot(apic
) ||
1792 apic_lvtt_period(apic
))
1795 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1796 apic
->lapic_timer
.tscdeadline
= data
;
1797 start_apic_timer(apic
);
1800 void kvm_lapic_set_tpr(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
1802 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1804 apic_set_tpr(apic
, ((cr8
& 0x0f) << 4)
1805 | (kvm_lapic_get_reg(apic
, APIC_TASKPRI
) & 4));
1808 u64
kvm_lapic_get_cr8(struct kvm_vcpu
*vcpu
)
1812 tpr
= (u64
) kvm_lapic_get_reg(vcpu
->arch
.apic
, APIC_TASKPRI
);
1814 return (tpr
& 0xf0) >> 4;
1817 void kvm_lapic_set_base(struct kvm_vcpu
*vcpu
, u64 value
)
1819 u64 old_value
= vcpu
->arch
.apic_base
;
1820 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1823 value
|= MSR_IA32_APICBASE_BSP
;
1825 vcpu
->arch
.apic_base
= value
;
1827 if ((old_value
^ value
) & MSR_IA32_APICBASE_ENABLE
)
1828 kvm_update_cpuid(vcpu
);
1833 /* update jump label if enable bit changes */
1834 if ((old_value
^ value
) & MSR_IA32_APICBASE_ENABLE
) {
1835 if (value
& MSR_IA32_APICBASE_ENABLE
) {
1836 kvm_apic_set_xapic_id(apic
, vcpu
->vcpu_id
);
1837 static_key_slow_dec_deferred(&apic_hw_disabled
);
1839 static_key_slow_inc(&apic_hw_disabled
.key
);
1840 recalculate_apic_map(vcpu
->kvm
);
1844 if ((old_value
^ value
) & X2APIC_ENABLE
) {
1845 if (value
& X2APIC_ENABLE
) {
1846 kvm_apic_set_x2apic_id(apic
, vcpu
->vcpu_id
);
1847 kvm_x86_ops
->set_virtual_x2apic_mode(vcpu
, true);
1849 kvm_x86_ops
->set_virtual_x2apic_mode(vcpu
, false);
1852 apic
->base_address
= apic
->vcpu
->arch
.apic_base
&
1853 MSR_IA32_APICBASE_BASE
;
1855 if ((value
& MSR_IA32_APICBASE_ENABLE
) &&
1856 apic
->base_address
!= APIC_DEFAULT_PHYS_BASE
)
1857 pr_warn_once("APIC base relocation is unsupported by KVM");
1859 /* with FSB delivery interrupt, we can restart APIC functionality */
1860 apic_debug("apic base msr is 0x%016" PRIx64
", and base address is "
1861 "0x%lx.\n", apic
->vcpu
->arch
.apic_base
, apic
->base_address
);
1865 void kvm_lapic_reset(struct kvm_vcpu
*vcpu
, bool init_event
)
1867 struct kvm_lapic
*apic
;
1870 apic_debug("%s\n", __func__
);
1873 apic
= vcpu
->arch
.apic
;
1874 ASSERT(apic
!= NULL
);
1876 /* Stop the timer in case it's a reset to an active apic */
1877 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1880 kvm_lapic_set_base(vcpu
, APIC_DEFAULT_PHYS_BASE
|
1881 MSR_IA32_APICBASE_ENABLE
);
1882 kvm_apic_set_xapic_id(apic
, vcpu
->vcpu_id
);
1884 kvm_apic_set_version(apic
->vcpu
);
1886 for (i
= 0; i
< KVM_APIC_LVT_NUM
; i
++)
1887 kvm_lapic_set_reg(apic
, APIC_LVTT
+ 0x10 * i
, APIC_LVT_MASKED
);
1888 apic_update_lvtt(apic
);
1889 if (kvm_check_has_quirk(vcpu
->kvm
, KVM_X86_QUIRK_LINT0_REENABLED
))
1890 kvm_lapic_set_reg(apic
, APIC_LVT0
,
1891 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT
));
1892 apic_manage_nmi_watchdog(apic
, kvm_lapic_get_reg(apic
, APIC_LVT0
));
1894 kvm_lapic_set_reg(apic
, APIC_DFR
, 0xffffffffU
);
1895 apic_set_spiv(apic
, 0xff);
1896 kvm_lapic_set_reg(apic
, APIC_TASKPRI
, 0);
1897 if (!apic_x2apic_mode(apic
))
1898 kvm_apic_set_ldr(apic
, 0);
1899 kvm_lapic_set_reg(apic
, APIC_ESR
, 0);
1900 kvm_lapic_set_reg(apic
, APIC_ICR
, 0);
1901 kvm_lapic_set_reg(apic
, APIC_ICR2
, 0);
1902 kvm_lapic_set_reg(apic
, APIC_TDCR
, 0);
1903 kvm_lapic_set_reg(apic
, APIC_TMICT
, 0);
1904 for (i
= 0; i
< 8; i
++) {
1905 kvm_lapic_set_reg(apic
, APIC_IRR
+ 0x10 * i
, 0);
1906 kvm_lapic_set_reg(apic
, APIC_ISR
+ 0x10 * i
, 0);
1907 kvm_lapic_set_reg(apic
, APIC_TMR
+ 0x10 * i
, 0);
1909 apic
->irr_pending
= vcpu
->arch
.apicv_active
;
1910 apic
->isr_count
= vcpu
->arch
.apicv_active
? 1 : 0;
1911 apic
->highest_isr_cache
= -1;
1912 update_divide_count(apic
);
1913 atomic_set(&apic
->lapic_timer
.pending
, 0);
1914 if (kvm_vcpu_is_bsp(vcpu
))
1915 kvm_lapic_set_base(vcpu
,
1916 vcpu
->arch
.apic_base
| MSR_IA32_APICBASE_BSP
);
1917 vcpu
->arch
.pv_eoi
.msr_val
= 0;
1918 apic_update_ppr(apic
);
1920 vcpu
->arch
.apic_arb_prio
= 0;
1921 vcpu
->arch
.apic_attention
= 0;
1923 apic_debug("%s: vcpu=%p, id=0x%x, base_msr="
1924 "0x%016" PRIx64
", base_address=0x%0lx.\n", __func__
,
1925 vcpu
, kvm_lapic_get_reg(apic
, APIC_ID
),
1926 vcpu
->arch
.apic_base
, apic
->base_address
);
1930 *----------------------------------------------------------------------
1932 *----------------------------------------------------------------------
1935 static bool lapic_is_periodic(struct kvm_lapic
*apic
)
1937 return apic_lvtt_period(apic
);
1940 int apic_has_pending_timer(struct kvm_vcpu
*vcpu
)
1942 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1944 if (apic_enabled(apic
) && apic_lvt_enabled(apic
, APIC_LVTT
))
1945 return atomic_read(&apic
->lapic_timer
.pending
);
1950 int kvm_apic_local_deliver(struct kvm_lapic
*apic
, int lvt_type
)
1952 u32 reg
= kvm_lapic_get_reg(apic
, lvt_type
);
1953 int vector
, mode
, trig_mode
;
1955 if (kvm_apic_hw_enabled(apic
) && !(reg
& APIC_LVT_MASKED
)) {
1956 vector
= reg
& APIC_VECTOR_MASK
;
1957 mode
= reg
& APIC_MODE_MASK
;
1958 trig_mode
= reg
& APIC_LVT_LEVEL_TRIGGER
;
1959 return __apic_accept_irq(apic
, mode
, vector
, 1, trig_mode
,
1965 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu
*vcpu
)
1967 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1970 kvm_apic_local_deliver(apic
, APIC_LVT0
);
1973 static const struct kvm_io_device_ops apic_mmio_ops
= {
1974 .read
= apic_mmio_read
,
1975 .write
= apic_mmio_write
,
1978 static enum hrtimer_restart
apic_timer_fn(struct hrtimer
*data
)
1980 struct kvm_timer
*ktimer
= container_of(data
, struct kvm_timer
, timer
);
1981 struct kvm_lapic
*apic
= container_of(ktimer
, struct kvm_lapic
, lapic_timer
);
1983 apic_timer_expired(apic
);
1985 if (lapic_is_periodic(apic
)) {
1986 advance_periodic_target_expiration(apic
);
1987 hrtimer_add_expires_ns(&ktimer
->timer
, ktimer
->period
);
1988 return HRTIMER_RESTART
;
1990 return HRTIMER_NORESTART
;
1993 int kvm_create_lapic(struct kvm_vcpu
*vcpu
)
1995 struct kvm_lapic
*apic
;
1997 ASSERT(vcpu
!= NULL
);
1998 apic_debug("apic_init %d\n", vcpu
->vcpu_id
);
2000 apic
= kzalloc(sizeof(*apic
), GFP_KERNEL
);
2004 vcpu
->arch
.apic
= apic
;
2006 apic
->regs
= (void *)get_zeroed_page(GFP_KERNEL
);
2008 printk(KERN_ERR
"malloc apic regs error for vcpu %x\n",
2010 goto nomem_free_apic
;
2014 hrtimer_init(&apic
->lapic_timer
.timer
, CLOCK_MONOTONIC
,
2015 HRTIMER_MODE_ABS_PINNED
);
2016 apic
->lapic_timer
.timer
.function
= apic_timer_fn
;
2019 * APIC is created enabled. This will prevent kvm_lapic_set_base from
2020 * thinking that APIC satet has changed.
2022 vcpu
->arch
.apic_base
= MSR_IA32_APICBASE_ENABLE
;
2023 static_key_slow_inc(&apic_sw_disabled
.key
); /* sw disabled at reset */
2024 kvm_lapic_reset(vcpu
, false);
2025 kvm_iodevice_init(&apic
->dev
, &apic_mmio_ops
);
2034 int kvm_apic_has_interrupt(struct kvm_vcpu
*vcpu
)
2036 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2039 if (!apic_enabled(apic
))
2042 apic_update_ppr(apic
);
2043 highest_irr
= apic_find_highest_irr(apic
);
2044 if ((highest_irr
== -1) ||
2045 ((highest_irr
& 0xF0) <= kvm_lapic_get_reg(apic
, APIC_PROCPRI
)))
2050 int kvm_apic_accept_pic_intr(struct kvm_vcpu
*vcpu
)
2052 u32 lvt0
= kvm_lapic_get_reg(vcpu
->arch
.apic
, APIC_LVT0
);
2055 if (!kvm_apic_hw_enabled(vcpu
->arch
.apic
))
2057 if ((lvt0
& APIC_LVT_MASKED
) == 0 &&
2058 GET_APIC_DELIVERY_MODE(lvt0
) == APIC_MODE_EXTINT
)
2063 void kvm_inject_apic_timer_irqs(struct kvm_vcpu
*vcpu
)
2065 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2067 if (atomic_read(&apic
->lapic_timer
.pending
) > 0) {
2068 kvm_apic_local_deliver(apic
, APIC_LVTT
);
2069 if (apic_lvtt_tscdeadline(apic
))
2070 apic
->lapic_timer
.tscdeadline
= 0;
2071 if (apic_lvtt_oneshot(apic
)) {
2072 apic
->lapic_timer
.tscdeadline
= 0;
2073 apic
->lapic_timer
.target_expiration
= 0;
2075 atomic_set(&apic
->lapic_timer
.pending
, 0);
2079 int kvm_get_apic_interrupt(struct kvm_vcpu
*vcpu
)
2081 int vector
= kvm_apic_has_interrupt(vcpu
);
2082 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2088 * We get here even with APIC virtualization enabled, if doing
2089 * nested virtualization and L1 runs with the "acknowledge interrupt
2090 * on exit" mode. Then we cannot inject the interrupt via RVI,
2091 * because the process would deliver it through the IDT.
2094 apic_set_isr(vector
, apic
);
2095 apic_update_ppr(apic
);
2096 apic_clear_irr(vector
, apic
);
2098 if (test_bit(vector
, vcpu_to_synic(vcpu
)->auto_eoi_bitmap
)) {
2099 apic_clear_isr(vector
, apic
);
2100 apic_update_ppr(apic
);
2106 static int kvm_apic_state_fixup(struct kvm_vcpu
*vcpu
,
2107 struct kvm_lapic_state
*s
, bool set
)
2109 if (apic_x2apic_mode(vcpu
->arch
.apic
)) {
2110 u32
*id
= (u32
*)(s
->regs
+ APIC_ID
);
2112 if (vcpu
->kvm
->arch
.x2apic_format
) {
2113 if (*id
!= vcpu
->vcpu_id
)
2126 int kvm_apic_get_state(struct kvm_vcpu
*vcpu
, struct kvm_lapic_state
*s
)
2128 memcpy(s
->regs
, vcpu
->arch
.apic
->regs
, sizeof(*s
));
2129 return kvm_apic_state_fixup(vcpu
, s
, false);
2132 int kvm_apic_set_state(struct kvm_vcpu
*vcpu
, struct kvm_lapic_state
*s
)
2134 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2138 kvm_lapic_set_base(vcpu
, vcpu
->arch
.apic_base
);
2139 /* set SPIV separately to get count of SW disabled APICs right */
2140 apic_set_spiv(apic
, *((u32
*)(s
->regs
+ APIC_SPIV
)));
2142 r
= kvm_apic_state_fixup(vcpu
, s
, true);
2145 memcpy(vcpu
->arch
.apic
->regs
, s
->regs
, sizeof *s
);
2147 recalculate_apic_map(vcpu
->kvm
);
2148 kvm_apic_set_version(vcpu
);
2150 apic_update_ppr(apic
);
2151 hrtimer_cancel(&apic
->lapic_timer
.timer
);
2152 apic_update_lvtt(apic
);
2153 apic_manage_nmi_watchdog(apic
, kvm_lapic_get_reg(apic
, APIC_LVT0
));
2154 update_divide_count(apic
);
2155 start_apic_timer(apic
);
2156 apic
->irr_pending
= true;
2157 apic
->isr_count
= vcpu
->arch
.apicv_active
?
2158 1 : count_vectors(apic
->regs
+ APIC_ISR
);
2159 apic
->highest_isr_cache
= -1;
2160 if (vcpu
->arch
.apicv_active
) {
2161 if (kvm_x86_ops
->apicv_post_state_restore
)
2162 kvm_x86_ops
->apicv_post_state_restore(vcpu
);
2163 kvm_x86_ops
->hwapic_irr_update(vcpu
,
2164 apic_find_highest_irr(apic
));
2165 kvm_x86_ops
->hwapic_isr_update(vcpu
,
2166 apic_find_highest_isr(apic
));
2168 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
2169 if (ioapic_in_kernel(vcpu
->kvm
))
2170 kvm_rtc_eoi_tracking_restore_one(vcpu
);
2172 vcpu
->arch
.apic_arb_prio
= 0;
2177 void __kvm_migrate_apic_timer(struct kvm_vcpu
*vcpu
)
2179 struct hrtimer
*timer
;
2181 if (!lapic_in_kernel(vcpu
))
2184 timer
= &vcpu
->arch
.apic
->lapic_timer
.timer
;
2185 if (hrtimer_cancel(timer
))
2186 hrtimer_start_expires(timer
, HRTIMER_MODE_ABS_PINNED
);
2190 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2192 * Detect whether guest triggered PV EOI since the
2193 * last entry. If yes, set EOI on guests's behalf.
2194 * Clear PV EOI in guest memory in any case.
2196 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu
*vcpu
,
2197 struct kvm_lapic
*apic
)
2202 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2203 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2205 * KVM_APIC_PV_EOI_PENDING is unset:
2206 * -> host disabled PV EOI.
2207 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2208 * -> host enabled PV EOI, guest did not execute EOI yet.
2209 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2210 * -> host enabled PV EOI, guest executed EOI.
2212 BUG_ON(!pv_eoi_enabled(vcpu
));
2213 pending
= pv_eoi_get_pending(vcpu
);
2215 * Clear pending bit in any case: it will be set again on vmentry.
2216 * While this might not be ideal from performance point of view,
2217 * this makes sure pv eoi is only enabled when we know it's safe.
2219 pv_eoi_clr_pending(vcpu
);
2222 vector
= apic_set_eoi(apic
);
2223 trace_kvm_pv_eoi(apic
, vector
);
2226 void kvm_lapic_sync_from_vapic(struct kvm_vcpu
*vcpu
)
2230 if (test_bit(KVM_APIC_PV_EOI_PENDING
, &vcpu
->arch
.apic_attention
))
2231 apic_sync_pv_eoi_from_guest(vcpu
, vcpu
->arch
.apic
);
2233 if (!test_bit(KVM_APIC_CHECK_VAPIC
, &vcpu
->arch
.apic_attention
))
2236 if (kvm_read_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apic
->vapic_cache
, &data
,
2240 apic_set_tpr(vcpu
->arch
.apic
, data
& 0xff);
2244 * apic_sync_pv_eoi_to_guest - called before vmentry
2246 * Detect whether it's safe to enable PV EOI and
2249 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu
*vcpu
,
2250 struct kvm_lapic
*apic
)
2252 if (!pv_eoi_enabled(vcpu
) ||
2253 /* IRR set or many bits in ISR: could be nested. */
2254 apic
->irr_pending
||
2255 /* Cache not set: could be safe but we don't bother. */
2256 apic
->highest_isr_cache
== -1 ||
2257 /* Need EOI to update ioapic. */
2258 kvm_ioapic_handles_vector(apic
, apic
->highest_isr_cache
)) {
2260 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2261 * so we need not do anything here.
2266 pv_eoi_set_pending(apic
->vcpu
);
2269 void kvm_lapic_sync_to_vapic(struct kvm_vcpu
*vcpu
)
2272 int max_irr
, max_isr
;
2273 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2275 apic_sync_pv_eoi_to_guest(vcpu
, apic
);
2277 if (!test_bit(KVM_APIC_CHECK_VAPIC
, &vcpu
->arch
.apic_attention
))
2280 tpr
= kvm_lapic_get_reg(apic
, APIC_TASKPRI
) & 0xff;
2281 max_irr
= apic_find_highest_irr(apic
);
2284 max_isr
= apic_find_highest_isr(apic
);
2287 data
= (tpr
& 0xff) | ((max_isr
& 0xf0) << 8) | (max_irr
<< 24);
2289 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apic
->vapic_cache
, &data
,
2293 int kvm_lapic_set_vapic_addr(struct kvm_vcpu
*vcpu
, gpa_t vapic_addr
)
2296 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
,
2297 &vcpu
->arch
.apic
->vapic_cache
,
2298 vapic_addr
, sizeof(u32
)))
2300 __set_bit(KVM_APIC_CHECK_VAPIC
, &vcpu
->arch
.apic_attention
);
2302 __clear_bit(KVM_APIC_CHECK_VAPIC
, &vcpu
->arch
.apic_attention
);
2305 vcpu
->arch
.apic
->vapic_addr
= vapic_addr
;
2309 int kvm_x2apic_msr_write(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
2311 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2312 u32 reg
= (msr
- APIC_BASE_MSR
) << 4;
2314 if (!lapic_in_kernel(vcpu
) || !apic_x2apic_mode(apic
))
2317 if (reg
== APIC_ICR2
)
2320 /* if this is ICR write vector before command */
2321 if (reg
== APIC_ICR
)
2322 kvm_lapic_reg_write(apic
, APIC_ICR2
, (u32
)(data
>> 32));
2323 return kvm_lapic_reg_write(apic
, reg
, (u32
)data
);
2326 int kvm_x2apic_msr_read(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*data
)
2328 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2329 u32 reg
= (msr
- APIC_BASE_MSR
) << 4, low
, high
= 0;
2331 if (!lapic_in_kernel(vcpu
) || !apic_x2apic_mode(apic
))
2334 if (reg
== APIC_DFR
|| reg
== APIC_ICR2
) {
2335 apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
2340 if (kvm_lapic_reg_read(apic
, reg
, 4, &low
))
2342 if (reg
== APIC_ICR
)
2343 kvm_lapic_reg_read(apic
, APIC_ICR2
, 4, &high
);
2345 *data
= (((u64
)high
) << 32) | low
;
2350 int kvm_hv_vapic_msr_write(struct kvm_vcpu
*vcpu
, u32 reg
, u64 data
)
2352 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2354 if (!lapic_in_kernel(vcpu
))
2357 /* if this is ICR write vector before command */
2358 if (reg
== APIC_ICR
)
2359 kvm_lapic_reg_write(apic
, APIC_ICR2
, (u32
)(data
>> 32));
2360 return kvm_lapic_reg_write(apic
, reg
, (u32
)data
);
2363 int kvm_hv_vapic_msr_read(struct kvm_vcpu
*vcpu
, u32 reg
, u64
*data
)
2365 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2368 if (!lapic_in_kernel(vcpu
))
2371 if (kvm_lapic_reg_read(apic
, reg
, 4, &low
))
2373 if (reg
== APIC_ICR
)
2374 kvm_lapic_reg_read(apic
, APIC_ICR2
, 4, &high
);
2376 *data
= (((u64
)high
) << 32) | low
;
2381 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu
*vcpu
, u64 data
)
2383 u64 addr
= data
& ~KVM_MSR_ENABLED
;
2384 if (!IS_ALIGNED(addr
, 4))
2387 vcpu
->arch
.pv_eoi
.msr_val
= data
;
2388 if (!pv_eoi_enabled(vcpu
))
2390 return kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.pv_eoi
.data
,
2394 void kvm_apic_accept_events(struct kvm_vcpu
*vcpu
)
2396 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2400 if (!lapic_in_kernel(vcpu
) || !apic
->pending_events
)
2404 * INITs are latched while in SMM. Because an SMM CPU cannot
2405 * be in KVM_MP_STATE_INIT_RECEIVED state, just eat SIPIs
2406 * and delay processing of INIT until the next RSM.
2409 WARN_ON_ONCE(vcpu
->arch
.mp_state
== KVM_MP_STATE_INIT_RECEIVED
);
2410 if (test_bit(KVM_APIC_SIPI
, &apic
->pending_events
))
2411 clear_bit(KVM_APIC_SIPI
, &apic
->pending_events
);
2415 pe
= xchg(&apic
->pending_events
, 0);
2416 if (test_bit(KVM_APIC_INIT
, &pe
)) {
2417 kvm_lapic_reset(vcpu
, true);
2418 kvm_vcpu_reset(vcpu
, true);
2419 if (kvm_vcpu_is_bsp(apic
->vcpu
))
2420 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
2422 vcpu
->arch
.mp_state
= KVM_MP_STATE_INIT_RECEIVED
;
2424 if (test_bit(KVM_APIC_SIPI
, &pe
) &&
2425 vcpu
->arch
.mp_state
== KVM_MP_STATE_INIT_RECEIVED
) {
2426 /* evaluate pending_events before reading the vector */
2428 sipi_vector
= apic
->sipi_vector
;
2429 apic_debug("vcpu %d received sipi with vector # %x\n",
2430 vcpu
->vcpu_id
, sipi_vector
);
2431 kvm_vcpu_deliver_sipi_vector(vcpu
, sipi_vector
);
2432 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
2436 void kvm_lapic_init(void)
2438 /* do not patch jump label more than once per second */
2439 jump_label_rate_limit(&apic_hw_disabled
, HZ
);
2440 jump_label_rate_limit(&apic_sw_disabled
, HZ
);