1
2/*
3 * Local APIC virtualization
4 *
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.
9 *
10 * Authors:
11 *   Dor Laor <dor.laor@qumranet.com>
12 *   Gregory Haskins <ghaskins@novell.com>
13 *   Yaozu (Eddie) Dong <eddie.dong@intel.com>
14 *
15 * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
16 *
17 * This work is licensed under the terms of the GNU GPL, version 2.  See
18 * the COPYING file in the top-level directory.
19 */
20
21#include <linux/kvm_host.h>
22#include <linux/kvm.h>
23#include <linux/mm.h>
24#include <linux/highmem.h>
25#include <linux/smp.h>
26#include <linux/hrtimer.h>
27#include <linux/io.h>
28#include <linux/module.h>
29#include <linux/math64.h>
30#include <linux/slab.h>
31#include <asm/processor.h>
32#include <asm/msr.h>
33#include <asm/page.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"
40#include "irq.h"
41#include "trace.h"
42#include "x86.h"
43#include "cpuid.h"
44
45#ifndef CONFIG_X86_64
46#define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
47#else
48#define mod_64(x, y) ((x) % (y))
49#endif
50
51#define PRId64 "d"
52#define PRIx64 "llx"
53#define PRIu64 "u"
54#define PRIo64 "o"
55
56#define APIC_BUS_CYCLE_NS 1
57
58/* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
59#define apic_debug(fmt, arg...)
60
61#define APIC_LVT_NUM			6
62/* 14 is the version for Xeon and Pentium 8.4.8*/
63#define APIC_VERSION			(0x14UL | ((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
71
72#define APIC_BROADCAST			0xFF
73#define X2APIC_BROADCAST		0xFFFFFFFFul
74
75#define VEC_POS(v) ((v) & (32 - 1))
76#define REG_POS(v) (((v) >> 5) << 4)
77
78static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
79{
80	*((u32 *) (apic->regs + reg_off)) = val;
81}
82
83static inline int apic_test_vector(int vec, void *bitmap)
84{
85	return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
86}
87
88bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
89{
90	struct kvm_lapic *apic = vcpu->arch.apic;
91
92	return apic_test_vector(vector, apic->regs + APIC_ISR) ||
93		apic_test_vector(vector, apic->regs + APIC_IRR);
94}
95
96static inline void apic_set_vector(int vec, void *bitmap)
97{
98	set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
99}
100
101static inline void apic_clear_vector(int vec, void *bitmap)
102{
103	clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
104}
105
106static inline int __apic_test_and_set_vector(int vec, void *bitmap)
107{
108	return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
109}
110
111static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
112{
113	return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
114}
115
116struct static_key_deferred apic_hw_disabled __read_mostly;
117struct static_key_deferred apic_sw_disabled __read_mostly;
118
119static inline int apic_enabled(struct kvm_lapic *apic)
120{
121	return kvm_apic_sw_enabled(apic) &&	kvm_apic_hw_enabled(apic);
122}
123
124#define LVT_MASK	\
125	(APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
126
127#define LINT_MASK	\
128	(LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
129	 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
130
131static inline int kvm_apic_id(struct kvm_lapic *apic)
132{
133	return (kvm_apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
134}
135
136/* The logical map is definitely wrong if we have multiple
137 * modes at the same time.  (Physical map is always right.)
138 */
139static inline bool kvm_apic_logical_map_valid(struct kvm_apic_map *map)
140{
141	return !(map->mode & (map->mode - 1));
142}
143
144static inline void
145apic_logical_id(struct kvm_apic_map *map, u32 dest_id, u16 *cid, u16 *lid)
146{
147	unsigned lid_bits;
148
149	BUILD_BUG_ON(KVM_APIC_MODE_XAPIC_CLUSTER !=  4);
150	BUILD_BUG_ON(KVM_APIC_MODE_XAPIC_FLAT    !=  8);
151	BUILD_BUG_ON(KVM_APIC_MODE_X2APIC        != 16);
152	lid_bits = map->mode;
153
154	*cid = dest_id >> lid_bits;
155	*lid = dest_id & ((1 << lid_bits) - 1);
156}
157
158static void recalculate_apic_map(struct kvm *kvm)
159{
160	struct kvm_apic_map *new, *old = NULL;
161	struct kvm_vcpu *vcpu;
162	int i;
163
164	new = kzalloc(sizeof(struct kvm_apic_map), GFP_KERNEL);
165
166	mutex_lock(&kvm->arch.apic_map_lock);
167
168	if (!new)
169		goto out;
170
171	kvm_for_each_vcpu(i, vcpu, kvm) {
172		struct kvm_lapic *apic = vcpu->arch.apic;
173		u16 cid, lid;
174		u32 ldr, aid;
175
176		if (!kvm_apic_present(vcpu))
177			continue;
178
179		aid = kvm_apic_id(apic);
180		ldr = kvm_apic_get_reg(apic, APIC_LDR);
181
182		if (aid < ARRAY_SIZE(new->phys_map))
183			new->phys_map[aid] = apic;
184
185		if (apic_x2apic_mode(apic)) {
186			new->mode |= KVM_APIC_MODE_X2APIC;
187		} else if (ldr) {
188			ldr = GET_APIC_LOGICAL_ID(ldr);
189			if (kvm_apic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
190				new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
191			else
192				new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
193		}
194
195		if (!kvm_apic_logical_map_valid(new))
196			continue;
197
198		apic_logical_id(new, ldr, &cid, &lid);
199
200		if (lid && cid < ARRAY_SIZE(new->logical_map))
201			new->logical_map[cid][ffs(lid) - 1] = apic;
202	}
203out:
204	old = rcu_dereference_protected(kvm->arch.apic_map,
205			lockdep_is_held(&kvm->arch.apic_map_lock));
206	rcu_assign_pointer(kvm->arch.apic_map, new);
207	mutex_unlock(&kvm->arch.apic_map_lock);
208
209	if (old)
210		kfree_rcu(old, rcu);
211
212	kvm_vcpu_request_scan_ioapic(kvm);
213}
214
215static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
216{
217	bool enabled = val & APIC_SPIV_APIC_ENABLED;
218
219	apic_set_reg(apic, APIC_SPIV, val);
220
221	if (enabled != apic->sw_enabled) {
222		apic->sw_enabled = enabled;
223		if (enabled) {
224			static_key_slow_dec_deferred(&apic_sw_disabled);
225			recalculate_apic_map(apic->vcpu->kvm);
226		} else
227			static_key_slow_inc(&apic_sw_disabled.key);
228	}
229}
230
231static inline void kvm_apic_set_id(struct kvm_lapic *apic, u8 id)
232{
233	apic_set_reg(apic, APIC_ID, id << 24);
234	recalculate_apic_map(apic->vcpu->kvm);
235}
236
237static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
238{
239	apic_set_reg(apic, APIC_LDR, id);
240	recalculate_apic_map(apic->vcpu->kvm);
241}
242
243static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
244{
245	return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
246}
247
248static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
249{
250	return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
251}
252
253static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
254{
255	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
256}
257
258static inline int apic_lvtt_period(struct kvm_lapic *apic)
259{
260	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
261}
262
263static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
264{
265	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
266}
267
268static inline int apic_lvt_nmi_mode(u32 lvt_val)
269{
270	return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
271}
272
273void kvm_apic_set_version(struct kvm_vcpu *vcpu)
274{
275	struct kvm_lapic *apic = vcpu->arch.apic;
276	struct kvm_cpuid_entry2 *feat;
277	u32 v = APIC_VERSION;
278
279	if (!kvm_vcpu_has_lapic(vcpu))
280		return;
281
282	feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
283	if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
284		v |= APIC_LVR_DIRECTED_EOI;
285	apic_set_reg(apic, APIC_LVR, v);
286}
287
288static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
289	LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
290	LVT_MASK | APIC_MODE_MASK,	/* LVTTHMR */
291	LVT_MASK | APIC_MODE_MASK,	/* LVTPC */
292	LINT_MASK, LINT_MASK,	/* LVT0-1 */
293	LVT_MASK		/* LVTERR */
294};
295
296static int find_highest_vector(void *bitmap)
297{
298	int vec;
299	u32 *reg;
300
301	for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
302	     vec >= 0; vec -= APIC_VECTORS_PER_REG) {
303		reg = bitmap + REG_POS(vec);
304		if (*reg)
305			return fls(*reg) - 1 + vec;
306	}
307
308	return -1;
309}
310
311static u8 count_vectors(void *bitmap)
312{
313	int vec;
314	u32 *reg;
315	u8 count = 0;
316
317	for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
318		reg = bitmap + REG_POS(vec);
319		count += hweight32(*reg);
320	}
321
322	return count;
323}
324
325void __kvm_apic_update_irr(u32 *pir, void *regs)
326{
327	u32 i, pir_val;
328
329	for (i = 0; i <= 7; i++) {
330		pir_val = xchg(&pir[i], 0);
331		if (pir_val)
332			*((u32 *)(regs + APIC_IRR + i * 0x10)) |= pir_val;
333	}
334}
335EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
336
337void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
338{
339	struct kvm_lapic *apic = vcpu->arch.apic;
340
341	__kvm_apic_update_irr(pir, apic->regs);
342
343	kvm_make_request(KVM_REQ_EVENT, vcpu);
344}
345EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
346
347static inline void apic_set_irr(int vec, struct kvm_lapic *apic)
348{
349	apic_set_vector(vec, apic->regs + APIC_IRR);
350	/*
351	 * irr_pending must be true if any interrupt is pending; set it after
352	 * APIC_IRR to avoid race with apic_clear_irr
353	 */
354	apic->irr_pending = true;
355}
356
357static inline int apic_search_irr(struct kvm_lapic *apic)
358{
359	return find_highest_vector(apic->regs + APIC_IRR);
360}
361
362static inline int apic_find_highest_irr(struct kvm_lapic *apic)
363{
364	int result;
365
366	/*
367	 * Note that irr_pending is just a hint. It will be always
368	 * true with virtual interrupt delivery enabled.
369	 */
370	if (!apic->irr_pending)
371		return -1;
372
373	kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
374	result = apic_search_irr(apic);
375	ASSERT(result == -1 || result >= 16);
376
377	return result;
378}
379
380static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
381{
382	struct kvm_vcpu *vcpu;
383
384	vcpu = apic->vcpu;
385
386	if (unlikely(kvm_apic_vid_enabled(vcpu->kvm))) {
387		/* try to update RVI */
388		apic_clear_vector(vec, apic->regs + APIC_IRR);
389		kvm_make_request(KVM_REQ_EVENT, vcpu);
390	} else {
391		apic->irr_pending = false;
392		apic_clear_vector(vec, apic->regs + APIC_IRR);
393		if (apic_search_irr(apic) != -1)
394			apic->irr_pending = true;
395	}
396}
397
398static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
399{
400	struct kvm_vcpu *vcpu;
401
402	if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
403		return;
404
405	vcpu = apic->vcpu;
406
407	/*
408	 * With APIC virtualization enabled, all caching is disabled
409	 * because the processor can modify ISR under the hood.  Instead
410	 * just set SVI.
411	 */
412	if (unlikely(kvm_x86_ops->hwapic_isr_update))
413		kvm_x86_ops->hwapic_isr_update(vcpu->kvm, vec);
414	else {
415		++apic->isr_count;
416		BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
417		/*
418		 * ISR (in service register) bit is set when injecting an interrupt.
419		 * The highest vector is injected. Thus the latest bit set matches
420		 * the highest bit in ISR.
421		 */
422		apic->highest_isr_cache = vec;
423	}
424}
425
426static inline int apic_find_highest_isr(struct kvm_lapic *apic)
427{
428	int result;
429
430	/*
431	 * Note that isr_count is always 1, and highest_isr_cache
432	 * is always -1, with APIC virtualization enabled.
433	 */
434	if (!apic->isr_count)
435		return -1;
436	if (likely(apic->highest_isr_cache != -1))
437		return apic->highest_isr_cache;
438
439	result = find_highest_vector(apic->regs + APIC_ISR);
440	ASSERT(result == -1 || result >= 16);
441
442	return result;
443}
444
445static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
446{
447	struct kvm_vcpu *vcpu;
448	if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
449		return;
450
451	vcpu = apic->vcpu;
452
453	/*
454	 * We do get here for APIC virtualization enabled if the guest
455	 * uses the Hyper-V APIC enlightenment.  In this case we may need
456	 * to trigger a new interrupt delivery by writing the SVI field;
457	 * on the other hand isr_count and highest_isr_cache are unused
458	 * and must be left alone.
459	 */
460	if (unlikely(kvm_x86_ops->hwapic_isr_update))
461		kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
462					       apic_find_highest_isr(apic));
463	else {
464		--apic->isr_count;
465		BUG_ON(apic->isr_count < 0);
466		apic->highest_isr_cache = -1;
467	}
468}
469
470int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
471{
472	int highest_irr;
473
474	/* This may race with setting of irr in __apic_accept_irq() and
475	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
476	 * will cause vmexit immediately and the value will be recalculated
477	 * on the next vmentry.
478	 */
479	if (!kvm_vcpu_has_lapic(vcpu))
480		return 0;
481	highest_irr = apic_find_highest_irr(vcpu->arch.apic);
482
483	return highest_irr;
484}
485
486static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
487			     int vector, int level, int trig_mode,
488			     unsigned long *dest_map);
489
490int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
491		unsigned long *dest_map)
492{
493	struct kvm_lapic *apic = vcpu->arch.apic;
494
495	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
496			irq->level, irq->trig_mode, dest_map);
497}
498
499static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
500{
501
502	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
503				      sizeof(val));
504}
505
506static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
507{
508
509	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
510				      sizeof(*val));
511}
512
513static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
514{
515	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
516}
517
518static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
519{
520	u8 val;
521	if (pv_eoi_get_user(vcpu, &val) < 0)
522		apic_debug("Can't read EOI MSR value: 0x%llx\n",
523			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
524	return val & 0x1;
525}
526
527static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
528{
529	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
530		apic_debug("Can't set EOI MSR value: 0x%llx\n",
531			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
532		return;
533	}
534	__set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
535}
536
537static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
538{
539	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
540		apic_debug("Can't clear EOI MSR value: 0x%llx\n",
541			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
542		return;
543	}
544	__clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
545}
546
547void kvm_apic_update_tmr(struct kvm_vcpu *vcpu, u32 *tmr)
548{
549	struct kvm_lapic *apic = vcpu->arch.apic;
550	int i;
551
552	for (i = 0; i < 8; i++)
553		apic_set_reg(apic, APIC_TMR + 0x10 * i, tmr[i]);
554}
555
556static void apic_update_ppr(struct kvm_lapic *apic)
557{
558	u32 tpr, isrv, ppr, old_ppr;
559	int isr;
560
561	old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI);
562	tpr = kvm_apic_get_reg(apic, APIC_TASKPRI);
563	isr = apic_find_highest_isr(apic);
564	isrv = (isr != -1) ? isr : 0;
565
566	if ((tpr & 0xf0) >= (isrv & 0xf0))
567		ppr = tpr & 0xff;
568	else
569		ppr = isrv & 0xf0;
570
571	apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
572		   apic, ppr, isr, isrv);
573
574	if (old_ppr != ppr) {
575		apic_set_reg(apic, APIC_PROCPRI, ppr);
576		if (ppr < old_ppr)
577			kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
578	}
579}
580
581static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
582{
583	apic_set_reg(apic, APIC_TASKPRI, tpr);
584	apic_update_ppr(apic);
585}
586
587static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
588{
589	if (apic_x2apic_mode(apic))
590		return mda == X2APIC_BROADCAST;
591
592	return GET_APIC_DEST_FIELD(mda) == APIC_BROADCAST;
593}
594
595static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
596{
597	if (kvm_apic_broadcast(apic, mda))
598		return true;
599
600	if (apic_x2apic_mode(apic))
601		return mda == kvm_apic_id(apic);
602
603	return mda == SET_APIC_DEST_FIELD(kvm_apic_id(apic));
604}
605
606static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
607{
608	u32 logical_id;
609
610	if (kvm_apic_broadcast(apic, mda))
611		return true;
612
613	logical_id = kvm_apic_get_reg(apic, APIC_LDR);
614
615	if (apic_x2apic_mode(apic))
616		return ((logical_id >> 16) == (mda >> 16))
617		       && (logical_id & mda & 0xffff) != 0;
618
619	logical_id = GET_APIC_LOGICAL_ID(logical_id);
620	mda = GET_APIC_DEST_FIELD(mda);
621
622	switch (kvm_apic_get_reg(apic, APIC_DFR)) {
623	case APIC_DFR_FLAT:
624		return (logical_id & mda) != 0;
625	case APIC_DFR_CLUSTER:
626		return ((logical_id >> 4) == (mda >> 4))
627		       && (logical_id & mda & 0xf) != 0;
628	default:
629		apic_debug("Bad DFR vcpu %d: %08x\n",
630			   apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR));
631		return false;
632	}
633}
634
635/* KVM APIC implementation has two quirks
636 *  - dest always begins at 0 while xAPIC MDA has offset 24,
637 *  - IOxAPIC messages have to be delivered (directly) to x2APIC.
638 */
639static u32 kvm_apic_mda(unsigned int dest_id, struct kvm_lapic *source,
640                                              struct kvm_lapic *target)
641{
642	bool ipi = source != NULL;
643	bool x2apic_mda = apic_x2apic_mode(ipi ? source : target);
644
645	if (!ipi && dest_id == APIC_BROADCAST && x2apic_mda)
646		return X2APIC_BROADCAST;
647
648	return x2apic_mda ? dest_id : SET_APIC_DEST_FIELD(dest_id);
649}
650
651bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
652			   int short_hand, unsigned int dest, int dest_mode)
653{
654	struct kvm_lapic *target = vcpu->arch.apic;
655	u32 mda = kvm_apic_mda(dest, source, target);
656
657	apic_debug("target %p, source %p, dest 0x%x, "
658		   "dest_mode 0x%x, short_hand 0x%x\n",
659		   target, source, dest, dest_mode, short_hand);
660
661	ASSERT(target);
662	switch (short_hand) {
663	case APIC_DEST_NOSHORT:
664		if (dest_mode == APIC_DEST_PHYSICAL)
665			return kvm_apic_match_physical_addr(target, mda);
666		else
667			return kvm_apic_match_logical_addr(target, mda);
668	case APIC_DEST_SELF:
669		return target == source;
670	case APIC_DEST_ALLINC:
671		return true;
672	case APIC_DEST_ALLBUT:
673		return target != source;
674	default:
675		apic_debug("kvm: apic: Bad dest shorthand value %x\n",
676			   short_hand);
677		return false;
678	}
679}
680
681bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
682		struct kvm_lapic_irq *irq, int *r, unsigned long *dest_map)
683{
684	struct kvm_apic_map *map;
685	unsigned long bitmap = 1;
686	struct kvm_lapic **dst;
687	int i;
688	bool ret, x2apic_ipi;
689
690	*r = -1;
691
692	if (irq->shorthand == APIC_DEST_SELF) {
693		*r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
694		return true;
695	}
696
697	if (irq->shorthand)
698		return false;
699
700	x2apic_ipi = src && apic_x2apic_mode(src);
701	if (irq->dest_id == (x2apic_ipi ? X2APIC_BROADCAST : APIC_BROADCAST))
702		return false;
703
704	ret = true;
705	rcu_read_lock();
706	map = rcu_dereference(kvm->arch.apic_map);
707
708	if (!map) {
709		ret = false;
710		goto out;
711	}
712
713	if (irq->dest_mode == APIC_DEST_PHYSICAL) {
714		if (irq->dest_id >= ARRAY_SIZE(map->phys_map))
715			goto out;
716
717		dst = &map->phys_map[irq->dest_id];
718	} else {
719		u16 cid;
720
721		if (!kvm_apic_logical_map_valid(map)) {
722			ret = false;
723			goto out;
724		}
725
726		apic_logical_id(map, irq->dest_id, &cid, (u16 *)&bitmap);
727
728		if (cid >= ARRAY_SIZE(map->logical_map))
729			goto out;
730
731		dst = map->logical_map[cid];
732
733		if (irq->delivery_mode == APIC_DM_LOWEST) {
734			int l = -1;
735			for_each_set_bit(i, &bitmap, 16) {
736				if (!dst[i])
737					continue;
738				if (l < 0)
739					l = i;
740				else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0)
741					l = i;
742			}
743
744			bitmap = (l >= 0) ? 1 << l : 0;
745		}
746	}
747
748	for_each_set_bit(i, &bitmap, 16) {
749		if (!dst[i])
750			continue;
751		if (*r < 0)
752			*r = 0;
753		*r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
754	}
755out:
756	rcu_read_unlock();
757	return ret;
758}
759
760/*
761 * Add a pending IRQ into lapic.
762 * Return 1 if successfully added and 0 if discarded.
763 */
764static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
765			     int vector, int level, int trig_mode,
766			     unsigned long *dest_map)
767{
768	int result = 0;
769	struct kvm_vcpu *vcpu = apic->vcpu;
770
771	trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
772				  trig_mode, vector);
773	switch (delivery_mode) {
774	case APIC_DM_LOWEST:
775		vcpu->arch.apic_arb_prio++;
776	case APIC_DM_FIXED:
777		/* FIXME add logic for vcpu on reset */
778		if (unlikely(!apic_enabled(apic)))
779			break;
780
781		result = 1;
782
783		if (dest_map)
784			__set_bit(vcpu->vcpu_id, dest_map);
785
786		if (kvm_x86_ops->deliver_posted_interrupt)
787			kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
788		else {
789			apic_set_irr(vector, apic);
790
791			kvm_make_request(KVM_REQ_EVENT, vcpu);
792			kvm_vcpu_kick(vcpu);
793		}
794		break;
795
796	case APIC_DM_REMRD:
797		result = 1;
798		vcpu->arch.pv.pv_unhalted = 1;
799		kvm_make_request(KVM_REQ_EVENT, vcpu);
800		kvm_vcpu_kick(vcpu);
801		break;
802
803	case APIC_DM_SMI:
804		apic_debug("Ignoring guest SMI\n");
805		break;
806
807	case APIC_DM_NMI:
808		result = 1;
809		kvm_inject_nmi(vcpu);
810		kvm_vcpu_kick(vcpu);
811		break;
812
813	case APIC_DM_INIT:
814		if (!trig_mode || level) {
815			result = 1;
816			/* assumes that there are only KVM_APIC_INIT/SIPI */
817			apic->pending_events = (1UL << KVM_APIC_INIT);
818			/* make sure pending_events is visible before sending
819			 * the request */
820			smp_wmb();
821			kvm_make_request(KVM_REQ_EVENT, vcpu);
822			kvm_vcpu_kick(vcpu);
823		} else {
824			apic_debug("Ignoring de-assert INIT to vcpu %d\n",
825				   vcpu->vcpu_id);
826		}
827		break;
828
829	case APIC_DM_STARTUP:
830		apic_debug("SIPI to vcpu %d vector 0x%02x\n",
831			   vcpu->vcpu_id, vector);
832		result = 1;
833		apic->sipi_vector = vector;
834		/* make sure sipi_vector is visible for the receiver */
835		smp_wmb();
836		set_bit(KVM_APIC_SIPI, &apic->pending_events);
837		kvm_make_request(KVM_REQ_EVENT, vcpu);
838		kvm_vcpu_kick(vcpu);
839		break;
840
841	case APIC_DM_EXTINT:
842		/*
843		 * Should only be called by kvm_apic_local_deliver() with LVT0,
844		 * before NMI watchdog was enabled. Already handled by
845		 * kvm_apic_accept_pic_intr().
846		 */
847		break;
848
849	default:
850		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
851		       delivery_mode);
852		break;
853	}
854	return result;
855}
856
857int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
858{
859	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
860}
861
862static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
863{
864	if (kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
865		int trigger_mode;
866		if (apic_test_vector(vector, apic->regs + APIC_TMR))
867			trigger_mode = IOAPIC_LEVEL_TRIG;
868		else
869			trigger_mode = IOAPIC_EDGE_TRIG;
870		kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
871	}
872}
873
874static int apic_set_eoi(struct kvm_lapic *apic)
875{
876	int vector = apic_find_highest_isr(apic);
877
878	trace_kvm_eoi(apic, vector);
879
880	/*
881	 * Not every write EOI will has corresponding ISR,
882	 * one example is when Kernel check timer on setup_IO_APIC
883	 */
884	if (vector == -1)
885		return vector;
886
887	apic_clear_isr(vector, apic);
888	apic_update_ppr(apic);
889
890	kvm_ioapic_send_eoi(apic, vector);
891	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
892	return vector;
893}
894
895/*
896 * this interface assumes a trap-like exit, which has already finished
897 * desired side effect including vISR and vPPR update.
898 */
899void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
900{
901	struct kvm_lapic *apic = vcpu->arch.apic;
902
903	trace_kvm_eoi(apic, vector);
904
905	kvm_ioapic_send_eoi(apic, vector);
906	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
907}
908EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
909
910static void apic_send_ipi(struct kvm_lapic *apic)
911{
912	u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR);
913	u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2);
914	struct kvm_lapic_irq irq;
915
916	irq.vector = icr_low & APIC_VECTOR_MASK;
917	irq.delivery_mode = icr_low & APIC_MODE_MASK;
918	irq.dest_mode = icr_low & APIC_DEST_MASK;
919	irq.level = icr_low & APIC_INT_ASSERT;
920	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
921	irq.shorthand = icr_low & APIC_SHORT_MASK;
922	if (apic_x2apic_mode(apic))
923		irq.dest_id = icr_high;
924	else
925		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
926
927	trace_kvm_apic_ipi(icr_low, irq.dest_id);
928
929	apic_debug("icr_high 0x%x, icr_low 0x%x, "
930		   "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
931		   "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
932		   icr_high, icr_low, irq.shorthand, irq.dest_id,
933		   irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
934		   irq.vector);
935
936	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
937}
938
939static u32 apic_get_tmcct(struct kvm_lapic *apic)
940{
941	ktime_t remaining;
942	s64 ns;
943	u32 tmcct;
944
945	ASSERT(apic != NULL);
946
947	/* if initial count is 0, current count should also be 0 */
948	if (kvm_apic_get_reg(apic, APIC_TMICT) == 0 ||
949		apic->lapic_timer.period == 0)
950		return 0;
951
952	remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
953	if (ktime_to_ns(remaining) < 0)
954		remaining = ktime_set(0, 0);
955
956	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
957	tmcct = div64_u64(ns,
958			 (APIC_BUS_CYCLE_NS * apic->divide_count));
959
960	return tmcct;
961}
962
963static void __report_tpr_access(struct kvm_lapic *apic, bool write)
964{
965	struct kvm_vcpu *vcpu = apic->vcpu;
966	struct kvm_run *run = vcpu->run;
967
968	kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
969	run->tpr_access.rip = kvm_rip_read(vcpu);
970	run->tpr_access.is_write = write;
971}
972
973static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
974{
975	if (apic->vcpu->arch.tpr_access_reporting)
976		__report_tpr_access(apic, write);
977}
978
979static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
980{
981	u32 val = 0;
982
983	if (offset >= LAPIC_MMIO_LENGTH)
984		return 0;
985
986	switch (offset) {
987	case APIC_ID:
988		if (apic_x2apic_mode(apic))
989			val = kvm_apic_id(apic);
990		else
991			val = kvm_apic_id(apic) << 24;
992		break;
993	case APIC_ARBPRI:
994		apic_debug("Access APIC ARBPRI register which is for P6\n");
995		break;
996
997	case APIC_TMCCT:	/* Timer CCR */
998		if (apic_lvtt_tscdeadline(apic))
999			return 0;
1000
1001		val = apic_get_tmcct(apic);
1002		break;
1003	case APIC_PROCPRI:
1004		apic_update_ppr(apic);
1005		val = kvm_apic_get_reg(apic, offset);
1006		break;
1007	case APIC_TASKPRI:
1008		report_tpr_access(apic, false);
1009		/* fall thru */
1010	default:
1011		val = kvm_apic_get_reg(apic, offset);
1012		break;
1013	}
1014
1015	return val;
1016}
1017
1018static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1019{
1020	return container_of(dev, struct kvm_lapic, dev);
1021}
1022
1023static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1024		void *data)
1025{
1026	unsigned char alignment = offset & 0xf;
1027	u32 result;
1028	/* this bitmask has a bit cleared for each reserved register */
1029	static const u64 rmask = 0x43ff01ffffffe70cULL;
1030
1031	if ((alignment + len) > 4) {
1032		apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1033			   offset, len);
1034		return 1;
1035	}
1036
1037	if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1038		apic_debug("KVM_APIC_READ: read reserved register %x\n",
1039			   offset);
1040		return 1;
1041	}
1042
1043	result = __apic_read(apic, offset & ~0xf);
1044
1045	trace_kvm_apic_read(offset, result);
1046
1047	switch (len) {
1048	case 1:
1049	case 2:
1050	case 4:
1051		memcpy(data, (char *)&result + alignment, len);
1052		break;
1053	default:
1054		printk(KERN_ERR "Local APIC read with len = %x, "
1055		       "should be 1,2, or 4 instead\n", len);
1056		break;
1057	}
1058	return 0;
1059}
1060
1061static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1062{
1063	return kvm_apic_hw_enabled(apic) &&
1064	    addr >= apic->base_address &&
1065	    addr < apic->base_address + LAPIC_MMIO_LENGTH;
1066}
1067
1068static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1069			   gpa_t address, int len, void *data)
1070{
1071	struct kvm_lapic *apic = to_lapic(this);
1072	u32 offset = address - apic->base_address;
1073
1074	if (!apic_mmio_in_range(apic, address))
1075		return -EOPNOTSUPP;
1076
1077	apic_reg_read(apic, offset, len, data);
1078
1079	return 0;
1080}
1081
1082static void update_divide_count(struct kvm_lapic *apic)
1083{
1084	u32 tmp1, tmp2, tdcr;
1085
1086	tdcr = kvm_apic_get_reg(apic, APIC_TDCR);
1087	tmp1 = tdcr & 0xf;
1088	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1089	apic->divide_count = 0x1 << (tmp2 & 0x7);
1090
1091	apic_debug("timer divide count is 0x%x\n",
1092				   apic->divide_count);
1093}
1094
1095static void apic_update_lvtt(struct kvm_lapic *apic)
1096{
1097	u32 timer_mode = kvm_apic_get_reg(apic, APIC_LVTT) &
1098			apic->lapic_timer.timer_mode_mask;
1099
1100	if (apic->lapic_timer.timer_mode != timer_mode) {
1101		apic->lapic_timer.timer_mode = timer_mode;
1102		hrtimer_cancel(&apic->lapic_timer.timer);
1103	}
1104}
1105
1106static void apic_timer_expired(struct kvm_lapic *apic)
1107{
1108	struct kvm_vcpu *vcpu = apic->vcpu;
1109	wait_queue_head_t *q = &vcpu->wq;
1110	struct kvm_timer *ktimer = &apic->lapic_timer;
1111
1112	if (atomic_read(&apic->lapic_timer.pending))
1113		return;
1114
1115	atomic_inc(&apic->lapic_timer.pending);
1116	kvm_set_pending_timer(vcpu);
1117
1118	if (waitqueue_active(q))
1119		wake_up_interruptible(q);
1120
1121	if (apic_lvtt_tscdeadline(apic))
1122		ktimer->expired_tscdeadline = ktimer->tscdeadline;
1123}
1124
1125/*
1126 * On APICv, this test will cause a busy wait
1127 * during a higher-priority task.
1128 */
1129
1130static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1131{
1132	struct kvm_lapic *apic = vcpu->arch.apic;
1133	u32 reg = kvm_apic_get_reg(apic, APIC_LVTT);
1134
1135	if (kvm_apic_hw_enabled(apic)) {
1136		int vec = reg & APIC_VECTOR_MASK;
1137		void *bitmap = apic->regs + APIC_ISR;
1138
1139		if (kvm_x86_ops->deliver_posted_interrupt)
1140			bitmap = apic->regs + APIC_IRR;
1141
1142		if (apic_test_vector(vec, bitmap))
1143			return true;
1144	}
1145	return false;
1146}
1147
1148void wait_lapic_expire(struct kvm_vcpu *vcpu)
1149{
1150	struct kvm_lapic *apic = vcpu->arch.apic;
1151	u64 guest_tsc, tsc_deadline;
1152
1153	if (!kvm_vcpu_has_lapic(vcpu))
1154		return;
1155
1156	if (apic->lapic_timer.expired_tscdeadline == 0)
1157		return;
1158
1159	if (!lapic_timer_int_injected(vcpu))
1160		return;
1161
1162	tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1163	apic->lapic_timer.expired_tscdeadline = 0;
1164	guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1165	trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1166
1167	/* __delay is delay_tsc whenever the hardware has TSC, thus always.  */
1168	if (guest_tsc < tsc_deadline)
1169		__delay(tsc_deadline - guest_tsc);
1170}
1171
1172static void start_apic_timer(struct kvm_lapic *apic)
1173{
1174	ktime_t now;
1175
1176	atomic_set(&apic->lapic_timer.pending, 0);
1177
1178	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1179		/* lapic timer in oneshot or periodic mode */
1180		now = apic->lapic_timer.timer.base->get_time();
1181		apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT)
1182			    * APIC_BUS_CYCLE_NS * apic->divide_count;
1183
1184		if (!apic->lapic_timer.period)
1185			return;
1186		/*
1187		 * Do not allow the guest to program periodic timers with small
1188		 * interval, since the hrtimers are not throttled by the host
1189		 * scheduler.
1190		 */
1191		if (apic_lvtt_period(apic)) {
1192			s64 min_period = min_timer_period_us * 1000LL;
1193
1194			if (apic->lapic_timer.period < min_period) {
1195				pr_info_ratelimited(
1196				    "kvm: vcpu %i: requested %lld ns "
1197				    "lapic timer period limited to %lld ns\n",
1198				    apic->vcpu->vcpu_id,
1199				    apic->lapic_timer.period, min_period);
1200				apic->lapic_timer.period = min_period;
1201			}
1202		}
1203
1204		hrtimer_start(&apic->lapic_timer.timer,
1205			      ktime_add_ns(now, apic->lapic_timer.period),
1206			      HRTIMER_MODE_ABS);
1207
1208		apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1209			   PRIx64 ", "
1210			   "timer initial count 0x%x, period %lldns, "
1211			   "expire @ 0x%016" PRIx64 ".\n", __func__,
1212			   APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1213			   kvm_apic_get_reg(apic, APIC_TMICT),
1214			   apic->lapic_timer.period,
1215			   ktime_to_ns(ktime_add_ns(now,
1216					apic->lapic_timer.period)));
1217	} else if (apic_lvtt_tscdeadline(apic)) {
1218		/* lapic timer in tsc deadline mode */
1219		u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1220		u64 ns = 0;
1221		ktime_t expire;
1222		struct kvm_vcpu *vcpu = apic->vcpu;
1223		unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1224		unsigned long flags;
1225
1226		if (unlikely(!tscdeadline || !this_tsc_khz))
1227			return;
1228
1229		local_irq_save(flags);
1230
1231		now = apic->lapic_timer.timer.base->get_time();
1232		guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1233		if (likely(tscdeadline > guest_tsc)) {
1234			ns = (tscdeadline - guest_tsc) * 1000000ULL;
1235			do_div(ns, this_tsc_khz);
1236			expire = ktime_add_ns(now, ns);
1237			expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1238			hrtimer_start(&apic->lapic_timer.timer,
1239				      expire, HRTIMER_MODE_ABS);
1240		} else
1241			apic_timer_expired(apic);
1242
1243		local_irq_restore(flags);
1244	}
1245}
1246
1247static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1248{
1249	int nmi_wd_enabled = apic_lvt_nmi_mode(kvm_apic_get_reg(apic, APIC_LVT0));
1250
1251	if (apic_lvt_nmi_mode(lvt0_val)) {
1252		if (!nmi_wd_enabled) {
1253			apic_debug("Receive NMI setting on APIC_LVT0 "
1254				   "for cpu %d\n", apic->vcpu->vcpu_id);
1255			atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1256		}
1257	} else if (nmi_wd_enabled)
1258		atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1259}
1260
1261static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1262{
1263	int ret = 0;
1264
1265	trace_kvm_apic_write(reg, val);
1266
1267	switch (reg) {
1268	case APIC_ID:		/* Local APIC ID */
1269		if (!apic_x2apic_mode(apic))
1270			kvm_apic_set_id(apic, val >> 24);
1271		else
1272			ret = 1;
1273		break;
1274
1275	case APIC_TASKPRI:
1276		report_tpr_access(apic, true);
1277		apic_set_tpr(apic, val & 0xff);
1278		break;
1279
1280	case APIC_EOI:
1281		apic_set_eoi(apic);
1282		break;
1283
1284	case APIC_LDR:
1285		if (!apic_x2apic_mode(apic))
1286			kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1287		else
1288			ret = 1;
1289		break;
1290
1291	case APIC_DFR:
1292		if (!apic_x2apic_mode(apic)) {
1293			apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1294			recalculate_apic_map(apic->vcpu->kvm);
1295		} else
1296			ret = 1;
1297		break;
1298
1299	case APIC_SPIV: {
1300		u32 mask = 0x3ff;
1301		if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1302			mask |= APIC_SPIV_DIRECTED_EOI;
1303		apic_set_spiv(apic, val & mask);
1304		if (!(val & APIC_SPIV_APIC_ENABLED)) {
1305			int i;
1306			u32 lvt_val;
1307
1308			for (i = 0; i < APIC_LVT_NUM; i++) {
1309				lvt_val = kvm_apic_get_reg(apic,
1310						       APIC_LVTT + 0x10 * i);
1311				apic_set_reg(apic, APIC_LVTT + 0x10 * i,
1312					     lvt_val | APIC_LVT_MASKED);
1313			}
1314			apic_update_lvtt(apic);
1315			atomic_set(&apic->lapic_timer.pending, 0);
1316
1317		}
1318		break;
1319	}
1320	case APIC_ICR:
1321		/* No delay here, so we always clear the pending bit */
1322		apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1323		apic_send_ipi(apic);
1324		break;
1325
1326	case APIC_ICR2:
1327		if (!apic_x2apic_mode(apic))
1328			val &= 0xff000000;
1329		apic_set_reg(apic, APIC_ICR2, val);
1330		break;
1331
1332	case APIC_LVT0:
1333		apic_manage_nmi_watchdog(apic, val);
1334	case APIC_LVTTHMR:
1335	case APIC_LVTPC:
1336	case APIC_LVT1:
1337	case APIC_LVTERR:
1338		/* TODO: Check vector */
1339		if (!kvm_apic_sw_enabled(apic))
1340			val |= APIC_LVT_MASKED;
1341
1342		val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1343		apic_set_reg(apic, reg, val);
1344
1345		break;
1346
1347	case APIC_LVTT:
1348		if (!kvm_apic_sw_enabled(apic))
1349			val |= APIC_LVT_MASKED;
1350		val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1351		apic_set_reg(apic, APIC_LVTT, val);
1352		apic_update_lvtt(apic);
1353		break;
1354
1355	case APIC_TMICT:
1356		if (apic_lvtt_tscdeadline(apic))
1357			break;
1358
1359		hrtimer_cancel(&apic->lapic_timer.timer);
1360		apic_set_reg(apic, APIC_TMICT, val);
1361		start_apic_timer(apic);
1362		break;
1363
1364	case APIC_TDCR:
1365		if (val & 4)
1366			apic_debug("KVM_WRITE:TDCR %x\n", val);
1367		apic_set_reg(apic, APIC_TDCR, val);
1368		update_divide_count(apic);
1369		break;
1370
1371	case APIC_ESR:
1372		if (apic_x2apic_mode(apic) && val != 0) {
1373			apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1374			ret = 1;
1375		}
1376		break;
1377
1378	case APIC_SELF_IPI:
1379		if (apic_x2apic_mode(apic)) {
1380			apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1381		} else
1382			ret = 1;
1383		break;
1384	default:
1385		ret = 1;
1386		break;
1387	}
1388	if (ret)
1389		apic_debug("Local APIC Write to read-only register %x\n", reg);
1390	return ret;
1391}
1392
1393static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1394			    gpa_t address, int len, const void *data)
1395{
1396	struct kvm_lapic *apic = to_lapic(this);
1397	unsigned int offset = address - apic->base_address;
1398	u32 val;
1399
1400	if (!apic_mmio_in_range(apic, address))
1401		return -EOPNOTSUPP;
1402
1403	/*
1404	 * APIC register must be aligned on 128-bits boundary.
1405	 * 32/64/128 bits registers must be accessed thru 32 bits.
1406	 * Refer SDM 8.4.1
1407	 */
1408	if (len != 4 || (offset & 0xf)) {
1409		/* Don't shout loud, $infamous_os would cause only noise. */
1410		apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1411		return 0;
1412	}
1413
1414	val = *(u32*)data;
1415
1416	/* too common printing */
1417	if (offset != APIC_EOI)
1418		apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1419			   "0x%x\n", __func__, offset, len, val);
1420
1421	apic_reg_write(apic, offset & 0xff0, val);
1422
1423	return 0;
1424}
1425
1426void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1427{
1428	if (kvm_vcpu_has_lapic(vcpu))
1429		apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1430}
1431EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1432
1433/* emulate APIC access in a trap manner */
1434void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1435{
1436	u32 val = 0;
1437
1438	/* hw has done the conditional check and inst decode */
1439	offset &= 0xff0;
1440
1441	apic_reg_read(vcpu->arch.apic, offset, 4, &val);
1442
1443	/* TODO: optimize to just emulate side effect w/o one more write */
1444	apic_reg_write(vcpu->arch.apic, offset, val);
1445}
1446EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1447
1448void kvm_free_lapic(struct kvm_vcpu *vcpu)
1449{
1450	struct kvm_lapic *apic = vcpu->arch.apic;
1451
1452	if (!vcpu->arch.apic)
1453		return;
1454
1455	hrtimer_cancel(&apic->lapic_timer.timer);
1456
1457	if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1458		static_key_slow_dec_deferred(&apic_hw_disabled);
1459
1460	if (!apic->sw_enabled)
1461		static_key_slow_dec_deferred(&apic_sw_disabled);
1462
1463	if (apic->regs)
1464		free_page((unsigned long)apic->regs);
1465
1466	kfree(apic);
1467}
1468
1469/*
1470 *----------------------------------------------------------------------
1471 * LAPIC interface
1472 *----------------------------------------------------------------------
1473 */
1474
1475u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1476{
1477	struct kvm_lapic *apic = vcpu->arch.apic;
1478
1479	if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1480			apic_lvtt_period(apic))
1481		return 0;
1482
1483	return apic->lapic_timer.tscdeadline;
1484}
1485
1486void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1487{
1488	struct kvm_lapic *apic = vcpu->arch.apic;
1489
1490	if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1491			apic_lvtt_period(apic))
1492		return;
1493
1494	hrtimer_cancel(&apic->lapic_timer.timer);
1495	apic->lapic_timer.tscdeadline = data;
1496	start_apic_timer(apic);
1497}
1498
1499void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1500{
1501	struct kvm_lapic *apic = vcpu->arch.apic;
1502
1503	if (!kvm_vcpu_has_lapic(vcpu))
1504		return;
1505
1506	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1507		     | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4));
1508}
1509
1510u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1511{
1512	u64 tpr;
1513
1514	if (!kvm_vcpu_has_lapic(vcpu))
1515		return 0;
1516
1517	tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1518
1519	return (tpr & 0xf0) >> 4;
1520}
1521
1522void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1523{
1524	u64 old_value = vcpu->arch.apic_base;
1525	struct kvm_lapic *apic = vcpu->arch.apic;
1526
1527	if (!apic) {
1528		value |= MSR_IA32_APICBASE_BSP;
1529		vcpu->arch.apic_base = value;
1530		return;
1531	}
1532
1533	vcpu->arch.apic_base = value;
1534
1535	/* update jump label if enable bit changes */
1536	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1537		if (value & MSR_IA32_APICBASE_ENABLE)
1538			static_key_slow_dec_deferred(&apic_hw_disabled);
1539		else
1540			static_key_slow_inc(&apic_hw_disabled.key);
1541		recalculate_apic_map(vcpu->kvm);
1542	}
1543
1544	if ((old_value ^ value) & X2APIC_ENABLE) {
1545		if (value & X2APIC_ENABLE) {
1546			u32 id = kvm_apic_id(apic);
1547			u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
1548			kvm_apic_set_ldr(apic, ldr);
1549			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1550		} else
1551			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1552	}
1553
1554	apic->base_address = apic->vcpu->arch.apic_base &
1555			     MSR_IA32_APICBASE_BASE;
1556
1557	if ((value & MSR_IA32_APICBASE_ENABLE) &&
1558	     apic->base_address != APIC_DEFAULT_PHYS_BASE)
1559		pr_warn_once("APIC base relocation is unsupported by KVM");
1560
1561	/* with FSB delivery interrupt, we can restart APIC functionality */
1562	apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1563		   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1564
1565}
1566
1567void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1568{
1569	struct kvm_lapic *apic;
1570	int i;
1571
1572	apic_debug("%s\n", __func__);
1573
1574	ASSERT(vcpu);
1575	apic = vcpu->arch.apic;
1576	ASSERT(apic != NULL);
1577
1578	/* Stop the timer in case it's a reset to an active apic */
1579	hrtimer_cancel(&apic->lapic_timer.timer);
1580
1581	kvm_apic_set_id(apic, vcpu->vcpu_id);
1582	kvm_apic_set_version(apic->vcpu);
1583
1584	for (i = 0; i < APIC_LVT_NUM; i++)
1585		apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1586	apic_update_lvtt(apic);
1587	apic_set_reg(apic, APIC_LVT0,
1588		     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1589
1590	apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1591	apic_set_spiv(apic, 0xff);
1592	apic_set_reg(apic, APIC_TASKPRI, 0);
1593	kvm_apic_set_ldr(apic, 0);
1594	apic_set_reg(apic, APIC_ESR, 0);
1595	apic_set_reg(apic, APIC_ICR, 0);
1596	apic_set_reg(apic, APIC_ICR2, 0);
1597	apic_set_reg(apic, APIC_TDCR, 0);
1598	apic_set_reg(apic, APIC_TMICT, 0);
1599	for (i = 0; i < 8; i++) {
1600		apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1601		apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1602		apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1603	}
1604	apic->irr_pending = kvm_apic_vid_enabled(vcpu->kvm);
1605	apic->isr_count = kvm_x86_ops->hwapic_isr_update ? 1 : 0;
1606	apic->highest_isr_cache = -1;
1607	update_divide_count(apic);
1608	atomic_set(&apic->lapic_timer.pending, 0);
1609	if (kvm_vcpu_is_bsp(vcpu))
1610		kvm_lapic_set_base(vcpu,
1611				vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1612	vcpu->arch.pv_eoi.msr_val = 0;
1613	apic_update_ppr(apic);
1614
1615	vcpu->arch.apic_arb_prio = 0;
1616	vcpu->arch.apic_attention = 0;
1617
1618	apic_debug("%s: vcpu=%p, id=%d, base_msr="
1619		   "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1620		   vcpu, kvm_apic_id(apic),
1621		   vcpu->arch.apic_base, apic->base_address);
1622}
1623
1624/*
1625 *----------------------------------------------------------------------
1626 * timer interface
1627 *----------------------------------------------------------------------
1628 */
1629
1630static bool lapic_is_periodic(struct kvm_lapic *apic)
1631{
1632	return apic_lvtt_period(apic);
1633}
1634
1635int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1636{
1637	struct kvm_lapic *apic = vcpu->arch.apic;
1638
1639	if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) &&
1640			apic_lvt_enabled(apic, APIC_LVTT))
1641		return atomic_read(&apic->lapic_timer.pending);
1642
1643	return 0;
1644}
1645
1646int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1647{
1648	u32 reg = kvm_apic_get_reg(apic, lvt_type);
1649	int vector, mode, trig_mode;
1650
1651	if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1652		vector = reg & APIC_VECTOR_MASK;
1653		mode = reg & APIC_MODE_MASK;
1654		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1655		return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1656					NULL);
1657	}
1658	return 0;
1659}
1660
1661void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1662{
1663	struct kvm_lapic *apic = vcpu->arch.apic;
1664
1665	if (apic)
1666		kvm_apic_local_deliver(apic, APIC_LVT0);
1667}
1668
1669static const struct kvm_io_device_ops apic_mmio_ops = {
1670	.read     = apic_mmio_read,
1671	.write    = apic_mmio_write,
1672};
1673
1674static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1675{
1676	struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1677	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1678
1679	apic_timer_expired(apic);
1680
1681	if (lapic_is_periodic(apic)) {
1682		hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1683		return HRTIMER_RESTART;
1684	} else
1685		return HRTIMER_NORESTART;
1686}
1687
1688int kvm_create_lapic(struct kvm_vcpu *vcpu)
1689{
1690	struct kvm_lapic *apic;
1691
1692	ASSERT(vcpu != NULL);
1693	apic_debug("apic_init %d\n", vcpu->vcpu_id);
1694
1695	apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1696	if (!apic)
1697		goto nomem;
1698
1699	vcpu->arch.apic = apic;
1700
1701	apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1702	if (!apic->regs) {
1703		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1704		       vcpu->vcpu_id);
1705		goto nomem_free_apic;
1706	}
1707	apic->vcpu = vcpu;
1708
1709	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1710		     HRTIMER_MODE_ABS);
1711	apic->lapic_timer.timer.function = apic_timer_fn;
1712
1713	/*
1714	 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1715	 * thinking that APIC satet has changed.
1716	 */
1717	vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1718	kvm_lapic_set_base(vcpu,
1719			APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE);
1720
1721	static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1722	kvm_lapic_reset(vcpu);
1723	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1724
1725	return 0;
1726nomem_free_apic:
1727	kfree(apic);
1728nomem:
1729	return -ENOMEM;
1730}
1731
1732int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1733{
1734	struct kvm_lapic *apic = vcpu->arch.apic;
1735	int highest_irr;
1736
1737	if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic))
1738		return -1;
1739
1740	apic_update_ppr(apic);
1741	highest_irr = apic_find_highest_irr(apic);
1742	if ((highest_irr == -1) ||
1743	    ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI)))
1744		return -1;
1745	return highest_irr;
1746}
1747
1748int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1749{
1750	u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1751	int r = 0;
1752
1753	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1754		r = 1;
1755	if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1756	    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1757		r = 1;
1758	return r;
1759}
1760
1761void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1762{
1763	struct kvm_lapic *apic = vcpu->arch.apic;
1764
1765	if (!kvm_vcpu_has_lapic(vcpu))
1766		return;
1767
1768	if (atomic_read(&apic->lapic_timer.pending) > 0) {
1769		kvm_apic_local_deliver(apic, APIC_LVTT);
1770		if (apic_lvtt_tscdeadline(apic))
1771			apic->lapic_timer.tscdeadline = 0;
1772		atomic_set(&apic->lapic_timer.pending, 0);
1773	}
1774}
1775
1776int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1777{
1778	int vector = kvm_apic_has_interrupt(vcpu);
1779	struct kvm_lapic *apic = vcpu->arch.apic;
1780
1781	if (vector == -1)
1782		return -1;
1783
1784	/*
1785	 * We get here even with APIC virtualization enabled, if doing
1786	 * nested virtualization and L1 runs with the "acknowledge interrupt
1787	 * on exit" mode.  Then we cannot inject the interrupt via RVI,
1788	 * because the process would deliver it through the IDT.
1789	 */
1790
1791	apic_set_isr(vector, apic);
1792	apic_update_ppr(apic);
1793	apic_clear_irr(vector, apic);
1794	return vector;
1795}
1796
1797void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
1798		struct kvm_lapic_state *s)
1799{
1800	struct kvm_lapic *apic = vcpu->arch.apic;
1801
1802	kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
1803	/* set SPIV separately to get count of SW disabled APICs right */
1804	apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
1805	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1806	/* call kvm_apic_set_id() to put apic into apic_map */
1807	kvm_apic_set_id(apic, kvm_apic_id(apic));
1808	kvm_apic_set_version(vcpu);
1809
1810	apic_update_ppr(apic);
1811	hrtimer_cancel(&apic->lapic_timer.timer);
1812	apic_update_lvtt(apic);
1813	apic_manage_nmi_watchdog(apic, kvm_apic_get_reg(apic, APIC_LVT0));
1814	update_divide_count(apic);
1815	start_apic_timer(apic);
1816	apic->irr_pending = true;
1817	apic->isr_count = kvm_x86_ops->hwapic_isr_update ?
1818				1 : count_vectors(apic->regs + APIC_ISR);
1819	apic->highest_isr_cache = -1;
1820	if (kvm_x86_ops->hwapic_irr_update)
1821		kvm_x86_ops->hwapic_irr_update(vcpu,
1822				apic_find_highest_irr(apic));
1823	if (unlikely(kvm_x86_ops->hwapic_isr_update))
1824		kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
1825				apic_find_highest_isr(apic));
1826	kvm_make_request(KVM_REQ_EVENT, vcpu);
1827	kvm_rtc_eoi_tracking_restore_one(vcpu);
1828}
1829
1830void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1831{
1832	struct hrtimer *timer;
1833
1834	if (!kvm_vcpu_has_lapic(vcpu))
1835		return;
1836
1837	timer = &vcpu->arch.apic->lapic_timer.timer;
1838	if (hrtimer_cancel(timer))
1839		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1840}
1841
1842/*
1843 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1844 *
1845 * Detect whether guest triggered PV EOI since the
1846 * last entry. If yes, set EOI on guests's behalf.
1847 * Clear PV EOI in guest memory in any case.
1848 */
1849static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1850					struct kvm_lapic *apic)
1851{
1852	bool pending;
1853	int vector;
1854	/*
1855	 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1856	 * and KVM_PV_EOI_ENABLED in guest memory as follows:
1857	 *
1858	 * KVM_APIC_PV_EOI_PENDING is unset:
1859	 * 	-> host disabled PV EOI.
1860	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1861	 * 	-> host enabled PV EOI, guest did not execute EOI yet.
1862	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1863	 * 	-> host enabled PV EOI, guest executed EOI.
1864	 */
1865	BUG_ON(!pv_eoi_enabled(vcpu));
1866	pending = pv_eoi_get_pending(vcpu);
1867	/*
1868	 * Clear pending bit in any case: it will be set again on vmentry.
1869	 * While this might not be ideal from performance point of view,
1870	 * this makes sure pv eoi is only enabled when we know it's safe.
1871	 */
1872	pv_eoi_clr_pending(vcpu);
1873	if (pending)
1874		return;
1875	vector = apic_set_eoi(apic);
1876	trace_kvm_pv_eoi(apic, vector);
1877}
1878
1879void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1880{
1881	u32 data;
1882
1883	if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1884		apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1885
1886	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1887		return;
1888
1889	kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1890				sizeof(u32));
1891
1892	apic_set_tpr(vcpu->arch.apic, data & 0xff);
1893}
1894
1895/*
1896 * apic_sync_pv_eoi_to_guest - called before vmentry
1897 *
1898 * Detect whether it's safe to enable PV EOI and
1899 * if yes do so.
1900 */
1901static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
1902					struct kvm_lapic *apic)
1903{
1904	if (!pv_eoi_enabled(vcpu) ||
1905	    /* IRR set or many bits in ISR: could be nested. */
1906	    apic->irr_pending ||
1907	    /* Cache not set: could be safe but we don't bother. */
1908	    apic->highest_isr_cache == -1 ||
1909	    /* Need EOI to update ioapic. */
1910	    kvm_ioapic_handles_vector(vcpu->kvm, apic->highest_isr_cache)) {
1911		/*
1912		 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1913		 * so we need not do anything here.
1914		 */
1915		return;
1916	}
1917
1918	pv_eoi_set_pending(apic->vcpu);
1919}
1920
1921void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1922{
1923	u32 data, tpr;
1924	int max_irr, max_isr;
1925	struct kvm_lapic *apic = vcpu->arch.apic;
1926
1927	apic_sync_pv_eoi_to_guest(vcpu, apic);
1928
1929	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1930		return;
1931
1932	tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1933	max_irr = apic_find_highest_irr(apic);
1934	if (max_irr < 0)
1935		max_irr = 0;
1936	max_isr = apic_find_highest_isr(apic);
1937	if (max_isr < 0)
1938		max_isr = 0;
1939	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1940
1941	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1942				sizeof(u32));
1943}
1944
1945int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1946{
1947	if (vapic_addr) {
1948		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1949					&vcpu->arch.apic->vapic_cache,
1950					vapic_addr, sizeof(u32)))
1951			return -EINVAL;
1952		__set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1953	} else {
1954		__clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1955	}
1956
1957	vcpu->arch.apic->vapic_addr = vapic_addr;
1958	return 0;
1959}
1960
1961int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1962{
1963	struct kvm_lapic *apic = vcpu->arch.apic;
1964	u32 reg = (msr - APIC_BASE_MSR) << 4;
1965
1966	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1967		return 1;
1968
1969	if (reg == APIC_ICR2)
1970		return 1;
1971
1972	/* if this is ICR write vector before command */
1973	if (reg == APIC_ICR)
1974		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1975	return apic_reg_write(apic, reg, (u32)data);
1976}
1977
1978int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1979{
1980	struct kvm_lapic *apic = vcpu->arch.apic;
1981	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1982
1983	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1984		return 1;
1985
1986	if (reg == APIC_DFR || reg == APIC_ICR2) {
1987		apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
1988			   reg);
1989		return 1;
1990	}
1991
1992	if (apic_reg_read(apic, reg, 4, &low))
1993		return 1;
1994	if (reg == APIC_ICR)
1995		apic_reg_read(apic, APIC_ICR2, 4, &high);
1996
1997	*data = (((u64)high) << 32) | low;
1998
1999	return 0;
2000}
2001
2002int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2003{
2004	struct kvm_lapic *apic = vcpu->arch.apic;
2005
2006	if (!kvm_vcpu_has_lapic(vcpu))
2007		return 1;
2008
2009	/* if this is ICR write vector before command */
2010	if (reg == APIC_ICR)
2011		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2012	return apic_reg_write(apic, reg, (u32)data);
2013}
2014
2015int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2016{
2017	struct kvm_lapic *apic = vcpu->arch.apic;
2018	u32 low, high = 0;
2019
2020	if (!kvm_vcpu_has_lapic(vcpu))
2021		return 1;
2022
2023	if (apic_reg_read(apic, reg, 4, &low))
2024		return 1;
2025	if (reg == APIC_ICR)
2026		apic_reg_read(apic, APIC_ICR2, 4, &high);
2027
2028	*data = (((u64)high) << 32) | low;
2029
2030	return 0;
2031}
2032
2033int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2034{
2035	u64 addr = data & ~KVM_MSR_ENABLED;
2036	if (!IS_ALIGNED(addr, 4))
2037		return 1;
2038
2039	vcpu->arch.pv_eoi.msr_val = data;
2040	if (!pv_eoi_enabled(vcpu))
2041		return 0;
2042	return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
2043					 addr, sizeof(u8));
2044}
2045
2046void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2047{
2048	struct kvm_lapic *apic = vcpu->arch.apic;
2049	u8 sipi_vector;
2050	unsigned long pe;
2051
2052	if (!kvm_vcpu_has_lapic(vcpu) || !apic->pending_events)
2053		return;
2054
2055	pe = xchg(&apic->pending_events, 0);
2056
2057	if (test_bit(KVM_APIC_INIT, &pe)) {
2058		kvm_lapic_reset(vcpu);
2059		kvm_vcpu_reset(vcpu);
2060		if (kvm_vcpu_is_bsp(apic->vcpu))
2061			vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2062		else
2063			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2064	}
2065	if (test_bit(KVM_APIC_SIPI, &pe) &&
2066	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2067		/* evaluate pending_events before reading the vector */
2068		smp_rmb();
2069		sipi_vector = apic->sipi_vector;
2070		apic_debug("vcpu %d received sipi with vector # %x\n",
2071			 vcpu->vcpu_id, sipi_vector);
2072		kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2073		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2074	}
2075}
2076
2077void kvm_lapic_init(void)
2078{
2079	/* do not patch jump label more than once per second */
2080	jump_label_rate_limit(&apic_hw_disabled, HZ);
2081	jump_label_rate_limit(&apic_sw_disabled, HZ);
2082}
2083