3 * Local APIC virtualization
5 * Copyright (C) 2006 Qumranet, Inc.
6 * Copyright (C) 2007 Novell
7 * Copyright (C) 2007 Intel
8 * Copyright 2009 Red Hat, Inc. and/or its affiliates.
11 * Dor Laor <dor.laor@qumranet.com>
12 * Gregory Haskins <ghaskins@novell.com>
13 * Yaozu (Eddie) Dong <eddie.dong@intel.com>
15 * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
24 #include <linux/highmem.h>
25 #include <linux/smp.h>
26 #include <linux/hrtimer.h>
28 #include <linux/module.h>
29 #include <linux/math64.h>
30 #include <linux/slab.h>
31 #include <asm/processor.h>
34 #include <asm/current.h>
35 #include <asm/apicdef.h>
36 #include <linux/atomic.h>
37 #include "kvm_cache_regs.h"
43 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
45 #define mod_64(x, y) ((x) % (y))
53 #define APIC_BUS_CYCLE_NS 1
55 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
56 #define apic_debug(fmt, arg...)
58 #define APIC_LVT_NUM 6
59 /* 14 is the version for Xeon and Pentium 8.4.8*/
60 #define APIC_VERSION (0x14UL | ((APIC_LVT_NUM - 1) << 16))
61 #define LAPIC_MMIO_LENGTH (1 << 12)
62 /* followed define is not in apicdef.h */
63 #define APIC_SHORT_MASK 0xc0000
64 #define APIC_DEST_NOSHORT 0x0
65 #define APIC_DEST_MASK 0x800
66 #define MAX_APIC_VECTOR 256
68 #define VEC_POS(v) ((v) & (32 - 1))
69 #define REG_POS(v) (((v) >> 5) << 4)
71 static unsigned int min_timer_period_us
= 500;
72 module_param(min_timer_period_us
, uint
, S_IRUGO
| S_IWUSR
);
74 static inline u32
apic_get_reg(struct kvm_lapic
*apic
, int reg_off
)
76 return *((u32
*) (apic
->regs
+ reg_off
));
79 static inline void apic_set_reg(struct kvm_lapic
*apic
, int reg_off
, u32 val
)
81 *((u32
*) (apic
->regs
+ reg_off
)) = val
;
84 static inline int apic_test_and_set_vector(int vec
, void *bitmap
)
86 return test_and_set_bit(VEC_POS(vec
), (bitmap
) + REG_POS(vec
));
89 static inline int apic_test_and_clear_vector(int vec
, void *bitmap
)
91 return test_and_clear_bit(VEC_POS(vec
), (bitmap
) + REG_POS(vec
));
94 static inline void apic_set_vector(int vec
, void *bitmap
)
96 set_bit(VEC_POS(vec
), (bitmap
) + REG_POS(vec
));
99 static inline void apic_clear_vector(int vec
, void *bitmap
)
101 clear_bit(VEC_POS(vec
), (bitmap
) + REG_POS(vec
));
104 static inline int apic_hw_enabled(struct kvm_lapic
*apic
)
106 return (apic
)->vcpu
->arch
.apic_base
& MSR_IA32_APICBASE_ENABLE
;
109 static inline int apic_sw_enabled(struct kvm_lapic
*apic
)
111 return apic_get_reg(apic
, APIC_SPIV
) & APIC_SPIV_APIC_ENABLED
;
114 static inline int apic_enabled(struct kvm_lapic
*apic
)
116 return apic_sw_enabled(apic
) && apic_hw_enabled(apic
);
120 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
123 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
124 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
126 static inline int kvm_apic_id(struct kvm_lapic
*apic
)
128 return (apic_get_reg(apic
, APIC_ID
) >> 24) & 0xff;
131 static inline int apic_lvt_enabled(struct kvm_lapic
*apic
, int lvt_type
)
133 return !(apic_get_reg(apic
, lvt_type
) & APIC_LVT_MASKED
);
136 static inline int apic_lvt_vector(struct kvm_lapic
*apic
, int lvt_type
)
138 return apic_get_reg(apic
, lvt_type
) & APIC_VECTOR_MASK
;
141 static inline int apic_lvtt_oneshot(struct kvm_lapic
*apic
)
143 return ((apic_get_reg(apic
, APIC_LVTT
) &
144 apic
->lapic_timer
.timer_mode_mask
) == APIC_LVT_TIMER_ONESHOT
);
147 static inline int apic_lvtt_period(struct kvm_lapic
*apic
)
149 return ((apic_get_reg(apic
, APIC_LVTT
) &
150 apic
->lapic_timer
.timer_mode_mask
) == APIC_LVT_TIMER_PERIODIC
);
153 static inline int apic_lvtt_tscdeadline(struct kvm_lapic
*apic
)
155 return ((apic_get_reg(apic
, APIC_LVTT
) &
156 apic
->lapic_timer
.timer_mode_mask
) ==
157 APIC_LVT_TIMER_TSCDEADLINE
);
160 static inline int apic_lvt_nmi_mode(u32 lvt_val
)
162 return (lvt_val
& (APIC_MODE_MASK
| APIC_LVT_MASKED
)) == APIC_DM_NMI
;
165 void kvm_apic_set_version(struct kvm_vcpu
*vcpu
)
167 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
168 struct kvm_cpuid_entry2
*feat
;
169 u32 v
= APIC_VERSION
;
171 if (!irqchip_in_kernel(vcpu
->kvm
))
174 feat
= kvm_find_cpuid_entry(apic
->vcpu
, 0x1, 0);
175 if (feat
&& (feat
->ecx
& (1 << (X86_FEATURE_X2APIC
& 31))))
176 v
|= APIC_LVR_DIRECTED_EOI
;
177 apic_set_reg(apic
, APIC_LVR
, v
);
180 static inline int apic_x2apic_mode(struct kvm_lapic
*apic
)
182 return apic
->vcpu
->arch
.apic_base
& X2APIC_ENABLE
;
185 static unsigned int apic_lvt_mask
[APIC_LVT_NUM
] = {
186 LVT_MASK
, /* part LVTT mask, timer mode mask added at runtime */
187 LVT_MASK
| APIC_MODE_MASK
, /* LVTTHMR */
188 LVT_MASK
| APIC_MODE_MASK
, /* LVTPC */
189 LINT_MASK
, LINT_MASK
, /* LVT0-1 */
190 LVT_MASK
/* LVTERR */
193 static int find_highest_vector(void *bitmap
)
196 int word_offset
= MAX_APIC_VECTOR
>> 5;
198 while ((word_offset
!= 0) && (word
[(--word_offset
) << 2] == 0))
201 if (likely(!word_offset
&& !word
[0]))
204 return fls(word
[word_offset
<< 2]) - 1 + (word_offset
<< 5);
207 static inline int apic_test_and_set_irr(int vec
, struct kvm_lapic
*apic
)
209 apic
->irr_pending
= true;
210 return apic_test_and_set_vector(vec
, apic
->regs
+ APIC_IRR
);
213 static inline int apic_search_irr(struct kvm_lapic
*apic
)
215 return find_highest_vector(apic
->regs
+ APIC_IRR
);
218 static inline int apic_find_highest_irr(struct kvm_lapic
*apic
)
222 if (!apic
->irr_pending
)
225 result
= apic_search_irr(apic
);
226 ASSERT(result
== -1 || result
>= 16);
231 static inline void apic_clear_irr(int vec
, struct kvm_lapic
*apic
)
233 apic
->irr_pending
= false;
234 apic_clear_vector(vec
, apic
->regs
+ APIC_IRR
);
235 if (apic_search_irr(apic
) != -1)
236 apic
->irr_pending
= true;
239 int kvm_lapic_find_highest_irr(struct kvm_vcpu
*vcpu
)
241 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
244 /* This may race with setting of irr in __apic_accept_irq() and
245 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
246 * will cause vmexit immediately and the value will be recalculated
247 * on the next vmentry.
251 highest_irr
= apic_find_highest_irr(apic
);
256 static int __apic_accept_irq(struct kvm_lapic
*apic
, int delivery_mode
,
257 int vector
, int level
, int trig_mode
);
259 int kvm_apic_set_irq(struct kvm_vcpu
*vcpu
, struct kvm_lapic_irq
*irq
)
261 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
263 return __apic_accept_irq(apic
, irq
->delivery_mode
, irq
->vector
,
264 irq
->level
, irq
->trig_mode
);
267 static inline int apic_find_highest_isr(struct kvm_lapic
*apic
)
271 result
= find_highest_vector(apic
->regs
+ APIC_ISR
);
272 ASSERT(result
== -1 || result
>= 16);
277 static void apic_update_ppr(struct kvm_lapic
*apic
)
279 u32 tpr
, isrv
, ppr
, old_ppr
;
282 old_ppr
= apic_get_reg(apic
, APIC_PROCPRI
);
283 tpr
= apic_get_reg(apic
, APIC_TASKPRI
);
284 isr
= apic_find_highest_isr(apic
);
285 isrv
= (isr
!= -1) ? isr
: 0;
287 if ((tpr
& 0xf0) >= (isrv
& 0xf0))
292 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
293 apic
, ppr
, isr
, isrv
);
295 if (old_ppr
!= ppr
) {
296 apic_set_reg(apic
, APIC_PROCPRI
, ppr
);
298 kvm_make_request(KVM_REQ_EVENT
, apic
->vcpu
);
302 static void apic_set_tpr(struct kvm_lapic
*apic
, u32 tpr
)
304 apic_set_reg(apic
, APIC_TASKPRI
, tpr
);
305 apic_update_ppr(apic
);
308 int kvm_apic_match_physical_addr(struct kvm_lapic
*apic
, u16 dest
)
310 return dest
== 0xff || kvm_apic_id(apic
) == dest
;
313 int kvm_apic_match_logical_addr(struct kvm_lapic
*apic
, u8 mda
)
318 if (apic_x2apic_mode(apic
)) {
319 logical_id
= apic_get_reg(apic
, APIC_LDR
);
320 return logical_id
& mda
;
323 logical_id
= GET_APIC_LOGICAL_ID(apic_get_reg(apic
, APIC_LDR
));
325 switch (apic_get_reg(apic
, APIC_DFR
)) {
327 if (logical_id
& mda
)
330 case APIC_DFR_CLUSTER
:
331 if (((logical_id
>> 4) == (mda
>> 0x4))
332 && (logical_id
& mda
& 0xf))
336 apic_debug("Bad DFR vcpu %d: %08x\n",
337 apic
->vcpu
->vcpu_id
, apic_get_reg(apic
, APIC_DFR
));
344 int kvm_apic_match_dest(struct kvm_vcpu
*vcpu
, struct kvm_lapic
*source
,
345 int short_hand
, int dest
, int dest_mode
)
348 struct kvm_lapic
*target
= vcpu
->arch
.apic
;
350 apic_debug("target %p, source %p, dest 0x%x, "
351 "dest_mode 0x%x, short_hand 0x%x\n",
352 target
, source
, dest
, dest_mode
, short_hand
);
355 switch (short_hand
) {
356 case APIC_DEST_NOSHORT
:
359 result
= kvm_apic_match_physical_addr(target
, dest
);
362 result
= kvm_apic_match_logical_addr(target
, dest
);
365 result
= (target
== source
);
367 case APIC_DEST_ALLINC
:
370 case APIC_DEST_ALLBUT
:
371 result
= (target
!= source
);
374 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
383 * Add a pending IRQ into lapic.
384 * Return 1 if successfully added and 0 if discarded.
386 static int __apic_accept_irq(struct kvm_lapic
*apic
, int delivery_mode
,
387 int vector
, int level
, int trig_mode
)
390 struct kvm_vcpu
*vcpu
= apic
->vcpu
;
392 switch (delivery_mode
) {
394 vcpu
->arch
.apic_arb_prio
++;
396 /* FIXME add logic for vcpu on reset */
397 if (unlikely(!apic_enabled(apic
)))
401 apic_debug("level trig mode for vector %d", vector
);
402 apic_set_vector(vector
, apic
->regs
+ APIC_TMR
);
404 apic_clear_vector(vector
, apic
->regs
+ APIC_TMR
);
406 result
= !apic_test_and_set_irr(vector
, apic
);
407 trace_kvm_apic_accept_irq(vcpu
->vcpu_id
, delivery_mode
,
408 trig_mode
, vector
, !result
);
411 apic_debug("level trig mode repeatedly for "
412 "vector %d", vector
);
416 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
421 apic_debug("Ignoring delivery mode 3\n");
425 apic_debug("Ignoring guest SMI\n");
430 kvm_inject_nmi(vcpu
);
437 vcpu
->arch
.mp_state
= KVM_MP_STATE_INIT_RECEIVED
;
438 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
441 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
446 case APIC_DM_STARTUP
:
447 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
448 vcpu
->vcpu_id
, vector
);
449 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_INIT_RECEIVED
) {
451 vcpu
->arch
.sipi_vector
= vector
;
452 vcpu
->arch
.mp_state
= KVM_MP_STATE_SIPI_RECEIVED
;
453 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
460 * Should only be called by kvm_apic_local_deliver() with LVT0,
461 * before NMI watchdog was enabled. Already handled by
462 * kvm_apic_accept_pic_intr().
467 printk(KERN_ERR
"TODO: unsupported delivery mode %x\n",
474 int kvm_apic_compare_prio(struct kvm_vcpu
*vcpu1
, struct kvm_vcpu
*vcpu2
)
476 return vcpu1
->arch
.apic_arb_prio
- vcpu2
->arch
.apic_arb_prio
;
479 static void apic_set_eoi(struct kvm_lapic
*apic
)
481 int vector
= apic_find_highest_isr(apic
);
484 * Not every write EOI will has corresponding ISR,
485 * one example is when Kernel check timer on setup_IO_APIC
490 apic_clear_vector(vector
, apic
->regs
+ APIC_ISR
);
491 apic_update_ppr(apic
);
493 if (apic_test_and_clear_vector(vector
, apic
->regs
+ APIC_TMR
))
494 trigger_mode
= IOAPIC_LEVEL_TRIG
;
496 trigger_mode
= IOAPIC_EDGE_TRIG
;
497 if (!(apic_get_reg(apic
, APIC_SPIV
) & APIC_SPIV_DIRECTED_EOI
))
498 kvm_ioapic_update_eoi(apic
->vcpu
->kvm
, vector
, trigger_mode
);
499 kvm_make_request(KVM_REQ_EVENT
, apic
->vcpu
);
502 static void apic_send_ipi(struct kvm_lapic
*apic
)
504 u32 icr_low
= apic_get_reg(apic
, APIC_ICR
);
505 u32 icr_high
= apic_get_reg(apic
, APIC_ICR2
);
506 struct kvm_lapic_irq irq
;
508 irq
.vector
= icr_low
& APIC_VECTOR_MASK
;
509 irq
.delivery_mode
= icr_low
& APIC_MODE_MASK
;
510 irq
.dest_mode
= icr_low
& APIC_DEST_MASK
;
511 irq
.level
= icr_low
& APIC_INT_ASSERT
;
512 irq
.trig_mode
= icr_low
& APIC_INT_LEVELTRIG
;
513 irq
.shorthand
= icr_low
& APIC_SHORT_MASK
;
514 if (apic_x2apic_mode(apic
))
515 irq
.dest_id
= icr_high
;
517 irq
.dest_id
= GET_APIC_DEST_FIELD(icr_high
);
519 trace_kvm_apic_ipi(icr_low
, irq
.dest_id
);
521 apic_debug("icr_high 0x%x, icr_low 0x%x, "
522 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
523 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
524 icr_high
, icr_low
, irq
.shorthand
, irq
.dest_id
,
525 irq
.trig_mode
, irq
.level
, irq
.dest_mode
, irq
.delivery_mode
,
528 kvm_irq_delivery_to_apic(apic
->vcpu
->kvm
, apic
, &irq
);
531 static u32
apic_get_tmcct(struct kvm_lapic
*apic
)
537 ASSERT(apic
!= NULL
);
539 /* if initial count is 0, current count should also be 0 */
540 if (apic_get_reg(apic
, APIC_TMICT
) == 0)
543 remaining
= hrtimer_get_remaining(&apic
->lapic_timer
.timer
);
544 if (ktime_to_ns(remaining
) < 0)
545 remaining
= ktime_set(0, 0);
547 ns
= mod_64(ktime_to_ns(remaining
), apic
->lapic_timer
.period
);
548 tmcct
= div64_u64(ns
,
549 (APIC_BUS_CYCLE_NS
* apic
->divide_count
));
554 static void __report_tpr_access(struct kvm_lapic
*apic
, bool write
)
556 struct kvm_vcpu
*vcpu
= apic
->vcpu
;
557 struct kvm_run
*run
= vcpu
->run
;
559 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS
, vcpu
);
560 run
->tpr_access
.rip
= kvm_rip_read(vcpu
);
561 run
->tpr_access
.is_write
= write
;
564 static inline void report_tpr_access(struct kvm_lapic
*apic
, bool write
)
566 if (apic
->vcpu
->arch
.tpr_access_reporting
)
567 __report_tpr_access(apic
, write
);
570 static u32
__apic_read(struct kvm_lapic
*apic
, unsigned int offset
)
574 if (offset
>= LAPIC_MMIO_LENGTH
)
579 if (apic_x2apic_mode(apic
))
580 val
= kvm_apic_id(apic
);
582 val
= kvm_apic_id(apic
) << 24;
585 apic_debug("Access APIC ARBPRI register which is for P6\n");
588 case APIC_TMCCT
: /* Timer CCR */
589 if (apic_lvtt_tscdeadline(apic
))
592 val
= apic_get_tmcct(apic
);
596 report_tpr_access(apic
, false);
599 apic_update_ppr(apic
);
600 val
= apic_get_reg(apic
, offset
);
607 static inline struct kvm_lapic
*to_lapic(struct kvm_io_device
*dev
)
609 return container_of(dev
, struct kvm_lapic
, dev
);
612 static int apic_reg_read(struct kvm_lapic
*apic
, u32 offset
, int len
,
615 unsigned char alignment
= offset
& 0xf;
617 /* this bitmask has a bit cleared for each reserver register */
618 static const u64 rmask
= 0x43ff01ffffffe70cULL
;
620 if ((alignment
+ len
) > 4) {
621 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
626 if (offset
> 0x3f0 || !(rmask
& (1ULL << (offset
>> 4)))) {
627 apic_debug("KVM_APIC_READ: read reserved register %x\n",
632 result
= __apic_read(apic
, offset
& ~0xf);
634 trace_kvm_apic_read(offset
, result
);
640 memcpy(data
, (char *)&result
+ alignment
, len
);
643 printk(KERN_ERR
"Local APIC read with len = %x, "
644 "should be 1,2, or 4 instead\n", len
);
650 static int apic_mmio_in_range(struct kvm_lapic
*apic
, gpa_t addr
)
652 return apic_hw_enabled(apic
) &&
653 addr
>= apic
->base_address
&&
654 addr
< apic
->base_address
+ LAPIC_MMIO_LENGTH
;
657 static int apic_mmio_read(struct kvm_io_device
*this,
658 gpa_t address
, int len
, void *data
)
660 struct kvm_lapic
*apic
= to_lapic(this);
661 u32 offset
= address
- apic
->base_address
;
663 if (!apic_mmio_in_range(apic
, address
))
666 apic_reg_read(apic
, offset
, len
, data
);
671 static void update_divide_count(struct kvm_lapic
*apic
)
673 u32 tmp1
, tmp2
, tdcr
;
675 tdcr
= apic_get_reg(apic
, APIC_TDCR
);
677 tmp2
= ((tmp1
& 0x3) | ((tmp1
& 0x8) >> 1)) + 1;
678 apic
->divide_count
= 0x1 << (tmp2
& 0x7);
680 apic_debug("timer divide count is 0x%x\n",
684 static void start_apic_timer(struct kvm_lapic
*apic
)
687 atomic_set(&apic
->lapic_timer
.pending
, 0);
689 if (apic_lvtt_period(apic
) || apic_lvtt_oneshot(apic
)) {
690 /* lapic timer in oneshot or peroidic mode */
691 now
= apic
->lapic_timer
.timer
.base
->get_time();
692 apic
->lapic_timer
.period
= (u64
)apic_get_reg(apic
, APIC_TMICT
)
693 * APIC_BUS_CYCLE_NS
* apic
->divide_count
;
695 if (!apic
->lapic_timer
.period
)
698 * Do not allow the guest to program periodic timers with small
699 * interval, since the hrtimers are not throttled by the host
702 if (apic_lvtt_period(apic
)) {
703 s64 min_period
= min_timer_period_us
* 1000LL;
705 if (apic
->lapic_timer
.period
< min_period
) {
707 "kvm: vcpu %i: requested %lld ns "
708 "lapic timer period limited to %lld ns\n",
710 apic
->lapic_timer
.period
, min_period
);
711 apic
->lapic_timer
.period
= min_period
;
715 hrtimer_start(&apic
->lapic_timer
.timer
,
716 ktime_add_ns(now
, apic
->lapic_timer
.period
),
719 apic_debug("%s: bus cycle is %" PRId64
"ns, now 0x%016"
721 "timer initial count 0x%x, period %lldns, "
722 "expire @ 0x%016" PRIx64
".\n", __func__
,
723 APIC_BUS_CYCLE_NS
, ktime_to_ns(now
),
724 apic_get_reg(apic
, APIC_TMICT
),
725 apic
->lapic_timer
.period
,
726 ktime_to_ns(ktime_add_ns(now
,
727 apic
->lapic_timer
.period
)));
728 } else if (apic_lvtt_tscdeadline(apic
)) {
729 /* lapic timer in tsc deadline mode */
730 u64 guest_tsc
, tscdeadline
= apic
->lapic_timer
.tscdeadline
;
732 struct kvm_vcpu
*vcpu
= apic
->vcpu
;
733 unsigned long this_tsc_khz
= vcpu_tsc_khz(vcpu
);
736 if (unlikely(!tscdeadline
|| !this_tsc_khz
))
739 local_irq_save(flags
);
741 now
= apic
->lapic_timer
.timer
.base
->get_time();
742 guest_tsc
= kvm_x86_ops
->read_l1_tsc(vcpu
);
743 if (likely(tscdeadline
> guest_tsc
)) {
744 ns
= (tscdeadline
- guest_tsc
) * 1000000ULL;
745 do_div(ns
, this_tsc_khz
);
747 hrtimer_start(&apic
->lapic_timer
.timer
,
748 ktime_add_ns(now
, ns
), HRTIMER_MODE_ABS
);
750 local_irq_restore(flags
);
754 static void apic_manage_nmi_watchdog(struct kvm_lapic
*apic
, u32 lvt0_val
)
756 int nmi_wd_enabled
= apic_lvt_nmi_mode(apic_get_reg(apic
, APIC_LVT0
));
758 if (apic_lvt_nmi_mode(lvt0_val
)) {
759 if (!nmi_wd_enabled
) {
760 apic_debug("Receive NMI setting on APIC_LVT0 "
761 "for cpu %d\n", apic
->vcpu
->vcpu_id
);
762 apic
->vcpu
->kvm
->arch
.vapics_in_nmi_mode
++;
764 } else if (nmi_wd_enabled
)
765 apic
->vcpu
->kvm
->arch
.vapics_in_nmi_mode
--;
768 static int apic_reg_write(struct kvm_lapic
*apic
, u32 reg
, u32 val
)
772 trace_kvm_apic_write(reg
, val
);
775 case APIC_ID
: /* Local APIC ID */
776 if (!apic_x2apic_mode(apic
))
777 apic_set_reg(apic
, APIC_ID
, val
);
783 report_tpr_access(apic
, true);
784 apic_set_tpr(apic
, val
& 0xff);
792 if (!apic_x2apic_mode(apic
))
793 apic_set_reg(apic
, APIC_LDR
, val
& APIC_LDR_MASK
);
799 if (!apic_x2apic_mode(apic
))
800 apic_set_reg(apic
, APIC_DFR
, val
| 0x0FFFFFFF);
807 if (apic_get_reg(apic
, APIC_LVR
) & APIC_LVR_DIRECTED_EOI
)
808 mask
|= APIC_SPIV_DIRECTED_EOI
;
809 apic_set_reg(apic
, APIC_SPIV
, val
& mask
);
810 if (!(val
& APIC_SPIV_APIC_ENABLED
)) {
814 for (i
= 0; i
< APIC_LVT_NUM
; i
++) {
815 lvt_val
= apic_get_reg(apic
,
816 APIC_LVTT
+ 0x10 * i
);
817 apic_set_reg(apic
, APIC_LVTT
+ 0x10 * i
,
818 lvt_val
| APIC_LVT_MASKED
);
820 atomic_set(&apic
->lapic_timer
.pending
, 0);
826 /* No delay here, so we always clear the pending bit */
827 apic_set_reg(apic
, APIC_ICR
, val
& ~(1 << 12));
832 if (!apic_x2apic_mode(apic
))
834 apic_set_reg(apic
, APIC_ICR2
, val
);
838 apic_manage_nmi_watchdog(apic
, val
);
843 /* TODO: Check vector */
844 if (!apic_sw_enabled(apic
))
845 val
|= APIC_LVT_MASKED
;
847 val
&= apic_lvt_mask
[(reg
- APIC_LVTT
) >> 4];
848 apic_set_reg(apic
, reg
, val
);
853 if ((apic_get_reg(apic
, APIC_LVTT
) &
854 apic
->lapic_timer
.timer_mode_mask
) !=
855 (val
& apic
->lapic_timer
.timer_mode_mask
))
856 hrtimer_cancel(&apic
->lapic_timer
.timer
);
858 if (!apic_sw_enabled(apic
))
859 val
|= APIC_LVT_MASKED
;
860 val
&= (apic_lvt_mask
[0] | apic
->lapic_timer
.timer_mode_mask
);
861 apic_set_reg(apic
, APIC_LVTT
, val
);
865 if (apic_lvtt_tscdeadline(apic
))
868 hrtimer_cancel(&apic
->lapic_timer
.timer
);
869 apic_set_reg(apic
, APIC_TMICT
, val
);
870 start_apic_timer(apic
);
875 apic_debug("KVM_WRITE:TDCR %x\n", val
);
876 apic_set_reg(apic
, APIC_TDCR
, val
);
877 update_divide_count(apic
);
881 if (apic_x2apic_mode(apic
) && val
!= 0) {
882 apic_debug("KVM_WRITE:ESR not zero %x\n", val
);
888 if (apic_x2apic_mode(apic
)) {
889 apic_reg_write(apic
, APIC_ICR
, 0x40000 | (val
& 0xff));
898 apic_debug("Local APIC Write to read-only register %x\n", reg
);
902 static int apic_mmio_write(struct kvm_io_device
*this,
903 gpa_t address
, int len
, const void *data
)
905 struct kvm_lapic
*apic
= to_lapic(this);
906 unsigned int offset
= address
- apic
->base_address
;
909 if (!apic_mmio_in_range(apic
, address
))
913 * APIC register must be aligned on 128-bits boundary.
914 * 32/64/128 bits registers must be accessed thru 32 bits.
917 if (len
!= 4 || (offset
& 0xf)) {
918 /* Don't shout loud, $infamous_os would cause only noise. */
919 apic_debug("apic write: bad size=%d %lx\n", len
, (long)address
);
925 /* too common printing */
926 if (offset
!= APIC_EOI
)
927 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
928 "0x%x\n", __func__
, offset
, len
, val
);
930 apic_reg_write(apic
, offset
& 0xff0, val
);
935 void kvm_lapic_set_eoi(struct kvm_vcpu
*vcpu
)
937 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
940 apic_reg_write(vcpu
->arch
.apic
, APIC_EOI
, 0);
942 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi
);
944 void kvm_free_lapic(struct kvm_vcpu
*vcpu
)
946 if (!vcpu
->arch
.apic
)
949 hrtimer_cancel(&vcpu
->arch
.apic
->lapic_timer
.timer
);
951 if (vcpu
->arch
.apic
->regs
)
952 free_page((unsigned long)vcpu
->arch
.apic
->regs
);
954 kfree(vcpu
->arch
.apic
);
958 *----------------------------------------------------------------------
960 *----------------------------------------------------------------------
963 u64
kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu
*vcpu
)
965 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
969 if (apic_lvtt_oneshot(apic
) || apic_lvtt_period(apic
))
972 return apic
->lapic_timer
.tscdeadline
;
975 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu
*vcpu
, u64 data
)
977 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
981 if (apic_lvtt_oneshot(apic
) || apic_lvtt_period(apic
))
984 hrtimer_cancel(&apic
->lapic_timer
.timer
);
985 apic
->lapic_timer
.tscdeadline
= data
;
986 start_apic_timer(apic
);
989 void kvm_lapic_set_tpr(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
991 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
995 apic_set_tpr(apic
, ((cr8
& 0x0f) << 4)
996 | (apic_get_reg(apic
, APIC_TASKPRI
) & 4));
999 u64
kvm_lapic_get_cr8(struct kvm_vcpu
*vcpu
)
1001 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1006 tpr
= (u64
) apic_get_reg(apic
, APIC_TASKPRI
);
1008 return (tpr
& 0xf0) >> 4;
1011 void kvm_lapic_set_base(struct kvm_vcpu
*vcpu
, u64 value
)
1013 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1016 value
|= MSR_IA32_APICBASE_BSP
;
1017 vcpu
->arch
.apic_base
= value
;
1021 if (!kvm_vcpu_is_bsp(apic
->vcpu
))
1022 value
&= ~MSR_IA32_APICBASE_BSP
;
1024 vcpu
->arch
.apic_base
= value
;
1025 if (apic_x2apic_mode(apic
)) {
1026 u32 id
= kvm_apic_id(apic
);
1027 u32 ldr
= ((id
& ~0xf) << 16) | (1 << (id
& 0xf));
1028 apic_set_reg(apic
, APIC_LDR
, ldr
);
1030 apic
->base_address
= apic
->vcpu
->arch
.apic_base
&
1031 MSR_IA32_APICBASE_BASE
;
1033 /* with FSB delivery interrupt, we can restart APIC functionality */
1034 apic_debug("apic base msr is 0x%016" PRIx64
", and base address is "
1035 "0x%lx.\n", apic
->vcpu
->arch
.apic_base
, apic
->base_address
);
1039 void kvm_lapic_reset(struct kvm_vcpu
*vcpu
)
1041 struct kvm_lapic
*apic
;
1044 apic_debug("%s\n", __func__
);
1047 apic
= vcpu
->arch
.apic
;
1048 ASSERT(apic
!= NULL
);
1050 /* Stop the timer in case it's a reset to an active apic */
1051 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1053 apic_set_reg(apic
, APIC_ID
, vcpu
->vcpu_id
<< 24);
1054 kvm_apic_set_version(apic
->vcpu
);
1056 for (i
= 0; i
< APIC_LVT_NUM
; i
++)
1057 apic_set_reg(apic
, APIC_LVTT
+ 0x10 * i
, APIC_LVT_MASKED
);
1058 apic_set_reg(apic
, APIC_LVT0
,
1059 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT
));
1061 apic_set_reg(apic
, APIC_DFR
, 0xffffffffU
);
1062 apic_set_reg(apic
, APIC_SPIV
, 0xff);
1063 apic_set_reg(apic
, APIC_TASKPRI
, 0);
1064 apic_set_reg(apic
, APIC_LDR
, 0);
1065 apic_set_reg(apic
, APIC_ESR
, 0);
1066 apic_set_reg(apic
, APIC_ICR
, 0);
1067 apic_set_reg(apic
, APIC_ICR2
, 0);
1068 apic_set_reg(apic
, APIC_TDCR
, 0);
1069 apic_set_reg(apic
, APIC_TMICT
, 0);
1070 for (i
= 0; i
< 8; i
++) {
1071 apic_set_reg(apic
, APIC_IRR
+ 0x10 * i
, 0);
1072 apic_set_reg(apic
, APIC_ISR
+ 0x10 * i
, 0);
1073 apic_set_reg(apic
, APIC_TMR
+ 0x10 * i
, 0);
1075 apic
->irr_pending
= false;
1076 update_divide_count(apic
);
1077 atomic_set(&apic
->lapic_timer
.pending
, 0);
1078 if (kvm_vcpu_is_bsp(vcpu
))
1079 vcpu
->arch
.apic_base
|= MSR_IA32_APICBASE_BSP
;
1080 apic_update_ppr(apic
);
1082 vcpu
->arch
.apic_arb_prio
= 0;
1084 apic_debug(KERN_INFO
"%s: vcpu=%p, id=%d, base_msr="
1085 "0x%016" PRIx64
", base_address=0x%0lx.\n", __func__
,
1086 vcpu
, kvm_apic_id(apic
),
1087 vcpu
->arch
.apic_base
, apic
->base_address
);
1090 bool kvm_apic_present(struct kvm_vcpu
*vcpu
)
1092 return vcpu
->arch
.apic
&& apic_hw_enabled(vcpu
->arch
.apic
);
1095 int kvm_lapic_enabled(struct kvm_vcpu
*vcpu
)
1097 return kvm_apic_present(vcpu
) && apic_sw_enabled(vcpu
->arch
.apic
);
1101 *----------------------------------------------------------------------
1103 *----------------------------------------------------------------------
1106 static bool lapic_is_periodic(struct kvm_timer
*ktimer
)
1108 struct kvm_lapic
*apic
= container_of(ktimer
, struct kvm_lapic
,
1110 return apic_lvtt_period(apic
);
1113 int apic_has_pending_timer(struct kvm_vcpu
*vcpu
)
1115 struct kvm_lapic
*lapic
= vcpu
->arch
.apic
;
1117 if (lapic
&& apic_enabled(lapic
) && apic_lvt_enabled(lapic
, APIC_LVTT
))
1118 return atomic_read(&lapic
->lapic_timer
.pending
);
1123 static int kvm_apic_local_deliver(struct kvm_lapic
*apic
, int lvt_type
)
1125 u32 reg
= apic_get_reg(apic
, lvt_type
);
1126 int vector
, mode
, trig_mode
;
1128 if (apic_hw_enabled(apic
) && !(reg
& APIC_LVT_MASKED
)) {
1129 vector
= reg
& APIC_VECTOR_MASK
;
1130 mode
= reg
& APIC_MODE_MASK
;
1131 trig_mode
= reg
& APIC_LVT_LEVEL_TRIGGER
;
1132 return __apic_accept_irq(apic
, mode
, vector
, 1, trig_mode
);
1137 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu
*vcpu
)
1139 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1142 kvm_apic_local_deliver(apic
, APIC_LVT0
);
1145 static struct kvm_timer_ops lapic_timer_ops
= {
1146 .is_periodic
= lapic_is_periodic
,
1149 static const struct kvm_io_device_ops apic_mmio_ops
= {
1150 .read
= apic_mmio_read
,
1151 .write
= apic_mmio_write
,
1154 int kvm_create_lapic(struct kvm_vcpu
*vcpu
)
1156 struct kvm_lapic
*apic
;
1158 ASSERT(vcpu
!= NULL
);
1159 apic_debug("apic_init %d\n", vcpu
->vcpu_id
);
1161 apic
= kzalloc(sizeof(*apic
), GFP_KERNEL
);
1165 vcpu
->arch
.apic
= apic
;
1167 apic
->regs
= (void *)get_zeroed_page(GFP_KERNEL
);
1169 printk(KERN_ERR
"malloc apic regs error for vcpu %x\n",
1171 goto nomem_free_apic
;
1175 hrtimer_init(&apic
->lapic_timer
.timer
, CLOCK_MONOTONIC
,
1177 apic
->lapic_timer
.timer
.function
= kvm_timer_fn
;
1178 apic
->lapic_timer
.t_ops
= &lapic_timer_ops
;
1179 apic
->lapic_timer
.kvm
= vcpu
->kvm
;
1180 apic
->lapic_timer
.vcpu
= vcpu
;
1182 apic
->base_address
= APIC_DEFAULT_PHYS_BASE
;
1183 vcpu
->arch
.apic_base
= APIC_DEFAULT_PHYS_BASE
;
1185 kvm_lapic_reset(vcpu
);
1186 kvm_iodevice_init(&apic
->dev
, &apic_mmio_ops
);
1195 int kvm_apic_has_interrupt(struct kvm_vcpu
*vcpu
)
1197 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1200 if (!apic
|| !apic_enabled(apic
))
1203 apic_update_ppr(apic
);
1204 highest_irr
= apic_find_highest_irr(apic
);
1205 if ((highest_irr
== -1) ||
1206 ((highest_irr
& 0xF0) <= apic_get_reg(apic
, APIC_PROCPRI
)))
1211 int kvm_apic_accept_pic_intr(struct kvm_vcpu
*vcpu
)
1213 u32 lvt0
= apic_get_reg(vcpu
->arch
.apic
, APIC_LVT0
);
1216 if (!apic_hw_enabled(vcpu
->arch
.apic
))
1218 if ((lvt0
& APIC_LVT_MASKED
) == 0 &&
1219 GET_APIC_DELIVERY_MODE(lvt0
) == APIC_MODE_EXTINT
)
1224 void kvm_inject_apic_timer_irqs(struct kvm_vcpu
*vcpu
)
1226 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1228 if (apic
&& atomic_read(&apic
->lapic_timer
.pending
) > 0) {
1229 if (kvm_apic_local_deliver(apic
, APIC_LVTT
))
1230 atomic_dec(&apic
->lapic_timer
.pending
);
1234 int kvm_get_apic_interrupt(struct kvm_vcpu
*vcpu
)
1236 int vector
= kvm_apic_has_interrupt(vcpu
);
1237 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1242 apic_set_vector(vector
, apic
->regs
+ APIC_ISR
);
1243 apic_update_ppr(apic
);
1244 apic_clear_irr(vector
, apic
);
1248 void kvm_apic_post_state_restore(struct kvm_vcpu
*vcpu
)
1250 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1252 apic
->base_address
= vcpu
->arch
.apic_base
&
1253 MSR_IA32_APICBASE_BASE
;
1254 kvm_apic_set_version(vcpu
);
1256 apic_update_ppr(apic
);
1257 hrtimer_cancel(&apic
->lapic_timer
.timer
);
1258 update_divide_count(apic
);
1259 start_apic_timer(apic
);
1260 apic
->irr_pending
= true;
1261 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
1264 void __kvm_migrate_apic_timer(struct kvm_vcpu
*vcpu
)
1266 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1267 struct hrtimer
*timer
;
1272 timer
= &apic
->lapic_timer
.timer
;
1273 if (hrtimer_cancel(timer
))
1274 hrtimer_start_expires(timer
, HRTIMER_MODE_ABS
);
1277 void kvm_lapic_sync_from_vapic(struct kvm_vcpu
*vcpu
)
1282 if (!irqchip_in_kernel(vcpu
->kvm
) || !vcpu
->arch
.apic
->vapic_addr
)
1285 vapic
= kmap_atomic(vcpu
->arch
.apic
->vapic_page
, KM_USER0
);
1286 data
= *(u32
*)(vapic
+ offset_in_page(vcpu
->arch
.apic
->vapic_addr
));
1287 kunmap_atomic(vapic
, KM_USER0
);
1289 apic_set_tpr(vcpu
->arch
.apic
, data
& 0xff);
1292 void kvm_lapic_sync_to_vapic(struct kvm_vcpu
*vcpu
)
1295 int max_irr
, max_isr
;
1296 struct kvm_lapic
*apic
;
1299 if (!irqchip_in_kernel(vcpu
->kvm
) || !vcpu
->arch
.apic
->vapic_addr
)
1302 apic
= vcpu
->arch
.apic
;
1303 tpr
= apic_get_reg(apic
, APIC_TASKPRI
) & 0xff;
1304 max_irr
= apic_find_highest_irr(apic
);
1307 max_isr
= apic_find_highest_isr(apic
);
1310 data
= (tpr
& 0xff) | ((max_isr
& 0xf0) << 8) | (max_irr
<< 24);
1312 vapic
= kmap_atomic(vcpu
->arch
.apic
->vapic_page
, KM_USER0
);
1313 *(u32
*)(vapic
+ offset_in_page(vcpu
->arch
.apic
->vapic_addr
)) = data
;
1314 kunmap_atomic(vapic
, KM_USER0
);
1317 void kvm_lapic_set_vapic_addr(struct kvm_vcpu
*vcpu
, gpa_t vapic_addr
)
1319 if (!irqchip_in_kernel(vcpu
->kvm
))
1322 vcpu
->arch
.apic
->vapic_addr
= vapic_addr
;
1325 int kvm_x2apic_msr_write(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1327 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1328 u32 reg
= (msr
- APIC_BASE_MSR
) << 4;
1330 if (!irqchip_in_kernel(vcpu
->kvm
) || !apic_x2apic_mode(apic
))
1333 /* if this is ICR write vector before command */
1335 apic_reg_write(apic
, APIC_ICR2
, (u32
)(data
>> 32));
1336 return apic_reg_write(apic
, reg
, (u32
)data
);
1339 int kvm_x2apic_msr_read(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*data
)
1341 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1342 u32 reg
= (msr
- APIC_BASE_MSR
) << 4, low
, high
= 0;
1344 if (!irqchip_in_kernel(vcpu
->kvm
) || !apic_x2apic_mode(apic
))
1347 if (apic_reg_read(apic
, reg
, 4, &low
))
1350 apic_reg_read(apic
, APIC_ICR2
, 4, &high
);
1352 *data
= (((u64
)high
) << 32) | low
;
1357 int kvm_hv_vapic_msr_write(struct kvm_vcpu
*vcpu
, u32 reg
, u64 data
)
1359 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1361 if (!irqchip_in_kernel(vcpu
->kvm
))
1364 /* if this is ICR write vector before command */
1365 if (reg
== APIC_ICR
)
1366 apic_reg_write(apic
, APIC_ICR2
, (u32
)(data
>> 32));
1367 return apic_reg_write(apic
, reg
, (u32
)data
);
1370 int kvm_hv_vapic_msr_read(struct kvm_vcpu
*vcpu
, u32 reg
, u64
*data
)
1372 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
1375 if (!irqchip_in_kernel(vcpu
->kvm
))
1378 if (apic_reg_read(apic
, reg
, 4, &low
))
1380 if (reg
== APIC_ICR
)
1381 apic_reg_read(apic
, APIC_ICR2
, 4, &high
);
1383 *data
= (((u64
)high
) << 32) | low
;