xref: /linux/arch/x86/kvm/lapic.c (revision 4949009eb8d40a441dcddcd96e101e77d31cf1b2)
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 <linux/atomic.h>
37 #include <linux/jump_label.h>
38 #include "kvm_cache_regs.h"
39 #include "irq.h"
40 #include "trace.h"
41 #include "x86.h"
42 #include "cpuid.h"
43 
44 #ifndef CONFIG_X86_64
45 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
46 #else
47 #define mod_64(x, y) ((x) % (y))
48 #endif
49 
50 #define PRId64 "d"
51 #define PRIx64 "llx"
52 #define PRIu64 "u"
53 #define PRIo64 "o"
54 
55 #define APIC_BUS_CYCLE_NS 1
56 
57 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
58 #define apic_debug(fmt, arg...)
59 
60 #define APIC_LVT_NUM			6
61 /* 14 is the version for Xeon and Pentium 8.4.8*/
62 #define APIC_VERSION			(0x14UL | ((APIC_LVT_NUM - 1) << 16))
63 #define LAPIC_MMIO_LENGTH		(1 << 12)
64 /* followed define is not in apicdef.h */
65 #define APIC_SHORT_MASK			0xc0000
66 #define APIC_DEST_NOSHORT		0x0
67 #define APIC_DEST_MASK			0x800
68 #define MAX_APIC_VECTOR			256
69 #define APIC_VECTORS_PER_REG		32
70 
71 #define APIC_BROADCAST			0xFF
72 #define X2APIC_BROADCAST		0xFFFFFFFFul
73 
74 #define VEC_POS(v) ((v) & (32 - 1))
75 #define REG_POS(v) (((v) >> 5) << 4)
76 
77 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
78 {
79 	*((u32 *) (apic->regs + reg_off)) = val;
80 }
81 
82 static inline int apic_test_vector(int vec, void *bitmap)
83 {
84 	return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
85 }
86 
87 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
88 {
89 	struct kvm_lapic *apic = vcpu->arch.apic;
90 
91 	return apic_test_vector(vector, apic->regs + APIC_ISR) ||
92 		apic_test_vector(vector, apic->regs + APIC_IRR);
93 }
94 
95 static inline void apic_set_vector(int vec, void *bitmap)
96 {
97 	set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
98 }
99 
100 static inline void apic_clear_vector(int vec, void *bitmap)
101 {
102 	clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
103 }
104 
105 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
106 {
107 	return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
108 }
109 
110 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
111 {
112 	return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
113 }
114 
115 struct static_key_deferred apic_hw_disabled __read_mostly;
116 struct static_key_deferred apic_sw_disabled __read_mostly;
117 
118 static inline int apic_enabled(struct kvm_lapic *apic)
119 {
120 	return kvm_apic_sw_enabled(apic) &&	kvm_apic_hw_enabled(apic);
121 }
122 
123 #define LVT_MASK	\
124 	(APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
125 
126 #define LINT_MASK	\
127 	(LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
128 	 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
129 
130 static inline int kvm_apic_id(struct kvm_lapic *apic)
131 {
132 	return (kvm_apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
133 }
134 
135 static void recalculate_apic_map(struct kvm *kvm)
136 {
137 	struct kvm_apic_map *new, *old = NULL;
138 	struct kvm_vcpu *vcpu;
139 	int i;
140 
141 	new = kzalloc(sizeof(struct kvm_apic_map), GFP_KERNEL);
142 
143 	mutex_lock(&kvm->arch.apic_map_lock);
144 
145 	if (!new)
146 		goto out;
147 
148 	new->ldr_bits = 8;
149 	/* flat mode is default */
150 	new->cid_shift = 8;
151 	new->cid_mask = 0;
152 	new->lid_mask = 0xff;
153 	new->broadcast = APIC_BROADCAST;
154 
155 	kvm_for_each_vcpu(i, vcpu, kvm) {
156 		struct kvm_lapic *apic = vcpu->arch.apic;
157 
158 		if (!kvm_apic_present(vcpu))
159 			continue;
160 
161 		if (apic_x2apic_mode(apic)) {
162 			new->ldr_bits = 32;
163 			new->cid_shift = 16;
164 			new->cid_mask = new->lid_mask = 0xffff;
165 			new->broadcast = X2APIC_BROADCAST;
166 		} else if (kvm_apic_get_reg(apic, APIC_LDR)) {
167 			if (kvm_apic_get_reg(apic, APIC_DFR) ==
168 							APIC_DFR_CLUSTER) {
169 				new->cid_shift = 4;
170 				new->cid_mask = 0xf;
171 				new->lid_mask = 0xf;
172 			} else {
173 				new->cid_shift = 8;
174 				new->cid_mask = 0;
175 				new->lid_mask = 0xff;
176 			}
177 		}
178 
179 		/*
180 		 * All APICs have to be configured in the same mode by an OS.
181 		 * We take advatage of this while building logical id loockup
182 		 * table. After reset APICs are in software disabled mode, so if
183 		 * we find apic with different setting we assume this is the mode
184 		 * OS wants all apics to be in; build lookup table accordingly.
185 		 */
186 		if (kvm_apic_sw_enabled(apic))
187 			break;
188 	}
189 
190 	kvm_for_each_vcpu(i, vcpu, kvm) {
191 		struct kvm_lapic *apic = vcpu->arch.apic;
192 		u16 cid, lid;
193 		u32 ldr, aid;
194 
195 		if (!kvm_apic_present(vcpu))
196 			continue;
197 
198 		aid = kvm_apic_id(apic);
199 		ldr = kvm_apic_get_reg(apic, APIC_LDR);
200 		cid = apic_cluster_id(new, ldr);
201 		lid = apic_logical_id(new, ldr);
202 
203 		if (aid < ARRAY_SIZE(new->phys_map))
204 			new->phys_map[aid] = apic;
205 		if (lid && cid < ARRAY_SIZE(new->logical_map))
206 			new->logical_map[cid][ffs(lid) - 1] = apic;
207 	}
208 out:
209 	old = rcu_dereference_protected(kvm->arch.apic_map,
210 			lockdep_is_held(&kvm->arch.apic_map_lock));
211 	rcu_assign_pointer(kvm->arch.apic_map, new);
212 	mutex_unlock(&kvm->arch.apic_map_lock);
213 
214 	if (old)
215 		kfree_rcu(old, rcu);
216 
217 	kvm_vcpu_request_scan_ioapic(kvm);
218 }
219 
220 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
221 {
222 	bool enabled = val & APIC_SPIV_APIC_ENABLED;
223 
224 	apic_set_reg(apic, APIC_SPIV, val);
225 
226 	if (enabled != apic->sw_enabled) {
227 		apic->sw_enabled = enabled;
228 		if (enabled) {
229 			static_key_slow_dec_deferred(&apic_sw_disabled);
230 			recalculate_apic_map(apic->vcpu->kvm);
231 		} else
232 			static_key_slow_inc(&apic_sw_disabled.key);
233 	}
234 }
235 
236 static inline void kvm_apic_set_id(struct kvm_lapic *apic, u8 id)
237 {
238 	apic_set_reg(apic, APIC_ID, id << 24);
239 	recalculate_apic_map(apic->vcpu->kvm);
240 }
241 
242 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
243 {
244 	apic_set_reg(apic, APIC_LDR, id);
245 	recalculate_apic_map(apic->vcpu->kvm);
246 }
247 
248 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
249 {
250 	return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
251 }
252 
253 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
254 {
255 	return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
256 }
257 
258 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
259 {
260 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
261 }
262 
263 static inline int apic_lvtt_period(struct kvm_lapic *apic)
264 {
265 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
266 }
267 
268 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
269 {
270 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
271 }
272 
273 static inline int apic_lvt_nmi_mode(u32 lvt_val)
274 {
275 	return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
276 }
277 
278 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
279 {
280 	struct kvm_lapic *apic = vcpu->arch.apic;
281 	struct kvm_cpuid_entry2 *feat;
282 	u32 v = APIC_VERSION;
283 
284 	if (!kvm_vcpu_has_lapic(vcpu))
285 		return;
286 
287 	feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
288 	if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
289 		v |= APIC_LVR_DIRECTED_EOI;
290 	apic_set_reg(apic, APIC_LVR, v);
291 }
292 
293 static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
294 	LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
295 	LVT_MASK | APIC_MODE_MASK,	/* LVTTHMR */
296 	LVT_MASK | APIC_MODE_MASK,	/* LVTPC */
297 	LINT_MASK, LINT_MASK,	/* LVT0-1 */
298 	LVT_MASK		/* LVTERR */
299 };
300 
301 static int find_highest_vector(void *bitmap)
302 {
303 	int vec;
304 	u32 *reg;
305 
306 	for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
307 	     vec >= 0; vec -= APIC_VECTORS_PER_REG) {
308 		reg = bitmap + REG_POS(vec);
309 		if (*reg)
310 			return fls(*reg) - 1 + vec;
311 	}
312 
313 	return -1;
314 }
315 
316 static u8 count_vectors(void *bitmap)
317 {
318 	int vec;
319 	u32 *reg;
320 	u8 count = 0;
321 
322 	for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
323 		reg = bitmap + REG_POS(vec);
324 		count += hweight32(*reg);
325 	}
326 
327 	return count;
328 }
329 
330 void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
331 {
332 	u32 i, pir_val;
333 	struct kvm_lapic *apic = vcpu->arch.apic;
334 
335 	for (i = 0; i <= 7; i++) {
336 		pir_val = xchg(&pir[i], 0);
337 		if (pir_val)
338 			*((u32 *)(apic->regs + APIC_IRR + i * 0x10)) |= pir_val;
339 	}
340 }
341 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
342 
343 static inline void apic_set_irr(int vec, struct kvm_lapic *apic)
344 {
345 	apic_set_vector(vec, apic->regs + APIC_IRR);
346 	/*
347 	 * irr_pending must be true if any interrupt is pending; set it after
348 	 * APIC_IRR to avoid race with apic_clear_irr
349 	 */
350 	apic->irr_pending = true;
351 }
352 
353 static inline int apic_search_irr(struct kvm_lapic *apic)
354 {
355 	return find_highest_vector(apic->regs + APIC_IRR);
356 }
357 
358 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
359 {
360 	int result;
361 
362 	/*
363 	 * Note that irr_pending is just a hint. It will be always
364 	 * true with virtual interrupt delivery enabled.
365 	 */
366 	if (!apic->irr_pending)
367 		return -1;
368 
369 	kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
370 	result = apic_search_irr(apic);
371 	ASSERT(result == -1 || result >= 16);
372 
373 	return result;
374 }
375 
376 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
377 {
378 	struct kvm_vcpu *vcpu;
379 
380 	vcpu = apic->vcpu;
381 
382 	if (unlikely(kvm_apic_vid_enabled(vcpu->kvm))) {
383 		/* try to update RVI */
384 		apic_clear_vector(vec, apic->regs + APIC_IRR);
385 		kvm_make_request(KVM_REQ_EVENT, vcpu);
386 	} else {
387 		apic->irr_pending = false;
388 		apic_clear_vector(vec, apic->regs + APIC_IRR);
389 		if (apic_search_irr(apic) != -1)
390 			apic->irr_pending = true;
391 	}
392 }
393 
394 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
395 {
396 	struct kvm_vcpu *vcpu;
397 
398 	if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
399 		return;
400 
401 	vcpu = apic->vcpu;
402 
403 	/*
404 	 * With APIC virtualization enabled, all caching is disabled
405 	 * because the processor can modify ISR under the hood.  Instead
406 	 * just set SVI.
407 	 */
408 	if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
409 		kvm_x86_ops->hwapic_isr_update(vcpu->kvm, vec);
410 	else {
411 		++apic->isr_count;
412 		BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
413 		/*
414 		 * ISR (in service register) bit is set when injecting an interrupt.
415 		 * The highest vector is injected. Thus the latest bit set matches
416 		 * the highest bit in ISR.
417 		 */
418 		apic->highest_isr_cache = vec;
419 	}
420 }
421 
422 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
423 {
424 	int result;
425 
426 	/*
427 	 * Note that isr_count is always 1, and highest_isr_cache
428 	 * is always -1, with APIC virtualization enabled.
429 	 */
430 	if (!apic->isr_count)
431 		return -1;
432 	if (likely(apic->highest_isr_cache != -1))
433 		return apic->highest_isr_cache;
434 
435 	result = find_highest_vector(apic->regs + APIC_ISR);
436 	ASSERT(result == -1 || result >= 16);
437 
438 	return result;
439 }
440 
441 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
442 {
443 	struct kvm_vcpu *vcpu;
444 	if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
445 		return;
446 
447 	vcpu = apic->vcpu;
448 
449 	/*
450 	 * We do get here for APIC virtualization enabled if the guest
451 	 * uses the Hyper-V APIC enlightenment.  In this case we may need
452 	 * to trigger a new interrupt delivery by writing the SVI field;
453 	 * on the other hand isr_count and highest_isr_cache are unused
454 	 * and must be left alone.
455 	 */
456 	if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
457 		kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
458 					       apic_find_highest_isr(apic));
459 	else {
460 		--apic->isr_count;
461 		BUG_ON(apic->isr_count < 0);
462 		apic->highest_isr_cache = -1;
463 	}
464 }
465 
466 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
467 {
468 	int highest_irr;
469 
470 	/* This may race with setting of irr in __apic_accept_irq() and
471 	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
472 	 * will cause vmexit immediately and the value will be recalculated
473 	 * on the next vmentry.
474 	 */
475 	if (!kvm_vcpu_has_lapic(vcpu))
476 		return 0;
477 	highest_irr = apic_find_highest_irr(vcpu->arch.apic);
478 
479 	return highest_irr;
480 }
481 
482 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
483 			     int vector, int level, int trig_mode,
484 			     unsigned long *dest_map);
485 
486 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
487 		unsigned long *dest_map)
488 {
489 	struct kvm_lapic *apic = vcpu->arch.apic;
490 
491 	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
492 			irq->level, irq->trig_mode, dest_map);
493 }
494 
495 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
496 {
497 
498 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
499 				      sizeof(val));
500 }
501 
502 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
503 {
504 
505 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
506 				      sizeof(*val));
507 }
508 
509 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
510 {
511 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
512 }
513 
514 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
515 {
516 	u8 val;
517 	if (pv_eoi_get_user(vcpu, &val) < 0)
518 		apic_debug("Can't read EOI MSR value: 0x%llx\n",
519 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
520 	return val & 0x1;
521 }
522 
523 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
524 {
525 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
526 		apic_debug("Can't set EOI MSR value: 0x%llx\n",
527 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
528 		return;
529 	}
530 	__set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
531 }
532 
533 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
534 {
535 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
536 		apic_debug("Can't clear EOI MSR value: 0x%llx\n",
537 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
538 		return;
539 	}
540 	__clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
541 }
542 
543 void kvm_apic_update_tmr(struct kvm_vcpu *vcpu, u32 *tmr)
544 {
545 	struct kvm_lapic *apic = vcpu->arch.apic;
546 	int i;
547 
548 	for (i = 0; i < 8; i++)
549 		apic_set_reg(apic, APIC_TMR + 0x10 * i, tmr[i]);
550 }
551 
552 static void apic_update_ppr(struct kvm_lapic *apic)
553 {
554 	u32 tpr, isrv, ppr, old_ppr;
555 	int isr;
556 
557 	old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI);
558 	tpr = kvm_apic_get_reg(apic, APIC_TASKPRI);
559 	isr = apic_find_highest_isr(apic);
560 	isrv = (isr != -1) ? isr : 0;
561 
562 	if ((tpr & 0xf0) >= (isrv & 0xf0))
563 		ppr = tpr & 0xff;
564 	else
565 		ppr = isrv & 0xf0;
566 
567 	apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
568 		   apic, ppr, isr, isrv);
569 
570 	if (old_ppr != ppr) {
571 		apic_set_reg(apic, APIC_PROCPRI, ppr);
572 		if (ppr < old_ppr)
573 			kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
574 	}
575 }
576 
577 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
578 {
579 	apic_set_reg(apic, APIC_TASKPRI, tpr);
580 	apic_update_ppr(apic);
581 }
582 
583 static int kvm_apic_broadcast(struct kvm_lapic *apic, u32 dest)
584 {
585 	return dest == (apic_x2apic_mode(apic) ?
586 			X2APIC_BROADCAST : APIC_BROADCAST);
587 }
588 
589 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 dest)
590 {
591 	return kvm_apic_id(apic) == dest || kvm_apic_broadcast(apic, dest);
592 }
593 
594 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
595 {
596 	int result = 0;
597 	u32 logical_id;
598 
599 	if (kvm_apic_broadcast(apic, mda))
600 		return 1;
601 
602 	if (apic_x2apic_mode(apic)) {
603 		logical_id = kvm_apic_get_reg(apic, APIC_LDR);
604 		return logical_id & mda;
605 	}
606 
607 	logical_id = GET_APIC_LOGICAL_ID(kvm_apic_get_reg(apic, APIC_LDR));
608 
609 	switch (kvm_apic_get_reg(apic, APIC_DFR)) {
610 	case APIC_DFR_FLAT:
611 		if (logical_id & mda)
612 			result = 1;
613 		break;
614 	case APIC_DFR_CLUSTER:
615 		if (((logical_id >> 4) == (mda >> 0x4))
616 		    && (logical_id & mda & 0xf))
617 			result = 1;
618 		break;
619 	default:
620 		apic_debug("Bad DFR vcpu %d: %08x\n",
621 			   apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR));
622 		break;
623 	}
624 
625 	return result;
626 }
627 
628 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
629 			   int short_hand, unsigned int dest, int dest_mode)
630 {
631 	int result = 0;
632 	struct kvm_lapic *target = vcpu->arch.apic;
633 
634 	apic_debug("target %p, source %p, dest 0x%x, "
635 		   "dest_mode 0x%x, short_hand 0x%x\n",
636 		   target, source, dest, dest_mode, short_hand);
637 
638 	ASSERT(target);
639 	switch (short_hand) {
640 	case APIC_DEST_NOSHORT:
641 		if (dest_mode == 0)
642 			/* Physical mode. */
643 			result = kvm_apic_match_physical_addr(target, dest);
644 		else
645 			/* Logical mode. */
646 			result = kvm_apic_match_logical_addr(target, dest);
647 		break;
648 	case APIC_DEST_SELF:
649 		result = (target == source);
650 		break;
651 	case APIC_DEST_ALLINC:
652 		result = 1;
653 		break;
654 	case APIC_DEST_ALLBUT:
655 		result = (target != source);
656 		break;
657 	default:
658 		apic_debug("kvm: apic: Bad dest shorthand value %x\n",
659 			   short_hand);
660 		break;
661 	}
662 
663 	return result;
664 }
665 
666 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
667 		struct kvm_lapic_irq *irq, int *r, unsigned long *dest_map)
668 {
669 	struct kvm_apic_map *map;
670 	unsigned long bitmap = 1;
671 	struct kvm_lapic **dst;
672 	int i;
673 	bool ret = false;
674 
675 	*r = -1;
676 
677 	if (irq->shorthand == APIC_DEST_SELF) {
678 		*r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
679 		return true;
680 	}
681 
682 	if (irq->shorthand)
683 		return false;
684 
685 	rcu_read_lock();
686 	map = rcu_dereference(kvm->arch.apic_map);
687 
688 	if (!map)
689 		goto out;
690 
691 	if (irq->dest_id == map->broadcast)
692 		goto out;
693 
694 	ret = true;
695 
696 	if (irq->dest_mode == 0) { /* physical mode */
697 		if (irq->dest_id >= ARRAY_SIZE(map->phys_map))
698 			goto out;
699 
700 		dst = &map->phys_map[irq->dest_id];
701 	} else {
702 		u32 mda = irq->dest_id << (32 - map->ldr_bits);
703 		u16 cid = apic_cluster_id(map, mda);
704 
705 		if (cid >= ARRAY_SIZE(map->logical_map))
706 			goto out;
707 
708 		dst = map->logical_map[cid];
709 
710 		bitmap = apic_logical_id(map, mda);
711 
712 		if (irq->delivery_mode == APIC_DM_LOWEST) {
713 			int l = -1;
714 			for_each_set_bit(i, &bitmap, 16) {
715 				if (!dst[i])
716 					continue;
717 				if (l < 0)
718 					l = i;
719 				else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0)
720 					l = i;
721 			}
722 
723 			bitmap = (l >= 0) ? 1 << l : 0;
724 		}
725 	}
726 
727 	for_each_set_bit(i, &bitmap, 16) {
728 		if (!dst[i])
729 			continue;
730 		if (*r < 0)
731 			*r = 0;
732 		*r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
733 	}
734 out:
735 	rcu_read_unlock();
736 	return ret;
737 }
738 
739 /*
740  * Add a pending IRQ into lapic.
741  * Return 1 if successfully added and 0 if discarded.
742  */
743 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
744 			     int vector, int level, int trig_mode,
745 			     unsigned long *dest_map)
746 {
747 	int result = 0;
748 	struct kvm_vcpu *vcpu = apic->vcpu;
749 
750 	trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
751 				  trig_mode, vector);
752 	switch (delivery_mode) {
753 	case APIC_DM_LOWEST:
754 		vcpu->arch.apic_arb_prio++;
755 	case APIC_DM_FIXED:
756 		/* FIXME add logic for vcpu on reset */
757 		if (unlikely(!apic_enabled(apic)))
758 			break;
759 
760 		result = 1;
761 
762 		if (dest_map)
763 			__set_bit(vcpu->vcpu_id, dest_map);
764 
765 		if (kvm_x86_ops->deliver_posted_interrupt)
766 			kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
767 		else {
768 			apic_set_irr(vector, apic);
769 
770 			kvm_make_request(KVM_REQ_EVENT, vcpu);
771 			kvm_vcpu_kick(vcpu);
772 		}
773 		break;
774 
775 	case APIC_DM_REMRD:
776 		result = 1;
777 		vcpu->arch.pv.pv_unhalted = 1;
778 		kvm_make_request(KVM_REQ_EVENT, vcpu);
779 		kvm_vcpu_kick(vcpu);
780 		break;
781 
782 	case APIC_DM_SMI:
783 		apic_debug("Ignoring guest SMI\n");
784 		break;
785 
786 	case APIC_DM_NMI:
787 		result = 1;
788 		kvm_inject_nmi(vcpu);
789 		kvm_vcpu_kick(vcpu);
790 		break;
791 
792 	case APIC_DM_INIT:
793 		if (!trig_mode || level) {
794 			result = 1;
795 			/* assumes that there are only KVM_APIC_INIT/SIPI */
796 			apic->pending_events = (1UL << KVM_APIC_INIT);
797 			/* make sure pending_events is visible before sending
798 			 * the request */
799 			smp_wmb();
800 			kvm_make_request(KVM_REQ_EVENT, vcpu);
801 			kvm_vcpu_kick(vcpu);
802 		} else {
803 			apic_debug("Ignoring de-assert INIT to vcpu %d\n",
804 				   vcpu->vcpu_id);
805 		}
806 		break;
807 
808 	case APIC_DM_STARTUP:
809 		apic_debug("SIPI to vcpu %d vector 0x%02x\n",
810 			   vcpu->vcpu_id, vector);
811 		result = 1;
812 		apic->sipi_vector = vector;
813 		/* make sure sipi_vector is visible for the receiver */
814 		smp_wmb();
815 		set_bit(KVM_APIC_SIPI, &apic->pending_events);
816 		kvm_make_request(KVM_REQ_EVENT, vcpu);
817 		kvm_vcpu_kick(vcpu);
818 		break;
819 
820 	case APIC_DM_EXTINT:
821 		/*
822 		 * Should only be called by kvm_apic_local_deliver() with LVT0,
823 		 * before NMI watchdog was enabled. Already handled by
824 		 * kvm_apic_accept_pic_intr().
825 		 */
826 		break;
827 
828 	default:
829 		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
830 		       delivery_mode);
831 		break;
832 	}
833 	return result;
834 }
835 
836 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
837 {
838 	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
839 }
840 
841 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
842 {
843 	if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI) &&
844 	    kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
845 		int trigger_mode;
846 		if (apic_test_vector(vector, apic->regs + APIC_TMR))
847 			trigger_mode = IOAPIC_LEVEL_TRIG;
848 		else
849 			trigger_mode = IOAPIC_EDGE_TRIG;
850 		kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
851 	}
852 }
853 
854 static int apic_set_eoi(struct kvm_lapic *apic)
855 {
856 	int vector = apic_find_highest_isr(apic);
857 
858 	trace_kvm_eoi(apic, vector);
859 
860 	/*
861 	 * Not every write EOI will has corresponding ISR,
862 	 * one example is when Kernel check timer on setup_IO_APIC
863 	 */
864 	if (vector == -1)
865 		return vector;
866 
867 	apic_clear_isr(vector, apic);
868 	apic_update_ppr(apic);
869 
870 	kvm_ioapic_send_eoi(apic, vector);
871 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
872 	return vector;
873 }
874 
875 /*
876  * this interface assumes a trap-like exit, which has already finished
877  * desired side effect including vISR and vPPR update.
878  */
879 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
880 {
881 	struct kvm_lapic *apic = vcpu->arch.apic;
882 
883 	trace_kvm_eoi(apic, vector);
884 
885 	kvm_ioapic_send_eoi(apic, vector);
886 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
887 }
888 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
889 
890 static void apic_send_ipi(struct kvm_lapic *apic)
891 {
892 	u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR);
893 	u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2);
894 	struct kvm_lapic_irq irq;
895 
896 	irq.vector = icr_low & APIC_VECTOR_MASK;
897 	irq.delivery_mode = icr_low & APIC_MODE_MASK;
898 	irq.dest_mode = icr_low & APIC_DEST_MASK;
899 	irq.level = icr_low & APIC_INT_ASSERT;
900 	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
901 	irq.shorthand = icr_low & APIC_SHORT_MASK;
902 	if (apic_x2apic_mode(apic))
903 		irq.dest_id = icr_high;
904 	else
905 		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
906 
907 	trace_kvm_apic_ipi(icr_low, irq.dest_id);
908 
909 	apic_debug("icr_high 0x%x, icr_low 0x%x, "
910 		   "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
911 		   "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
912 		   icr_high, icr_low, irq.shorthand, irq.dest_id,
913 		   irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
914 		   irq.vector);
915 
916 	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
917 }
918 
919 static u32 apic_get_tmcct(struct kvm_lapic *apic)
920 {
921 	ktime_t remaining;
922 	s64 ns;
923 	u32 tmcct;
924 
925 	ASSERT(apic != NULL);
926 
927 	/* if initial count is 0, current count should also be 0 */
928 	if (kvm_apic_get_reg(apic, APIC_TMICT) == 0 ||
929 		apic->lapic_timer.period == 0)
930 		return 0;
931 
932 	remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
933 	if (ktime_to_ns(remaining) < 0)
934 		remaining = ktime_set(0, 0);
935 
936 	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
937 	tmcct = div64_u64(ns,
938 			 (APIC_BUS_CYCLE_NS * apic->divide_count));
939 
940 	return tmcct;
941 }
942 
943 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
944 {
945 	struct kvm_vcpu *vcpu = apic->vcpu;
946 	struct kvm_run *run = vcpu->run;
947 
948 	kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
949 	run->tpr_access.rip = kvm_rip_read(vcpu);
950 	run->tpr_access.is_write = write;
951 }
952 
953 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
954 {
955 	if (apic->vcpu->arch.tpr_access_reporting)
956 		__report_tpr_access(apic, write);
957 }
958 
959 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
960 {
961 	u32 val = 0;
962 
963 	if (offset >= LAPIC_MMIO_LENGTH)
964 		return 0;
965 
966 	switch (offset) {
967 	case APIC_ID:
968 		if (apic_x2apic_mode(apic))
969 			val = kvm_apic_id(apic);
970 		else
971 			val = kvm_apic_id(apic) << 24;
972 		break;
973 	case APIC_ARBPRI:
974 		apic_debug("Access APIC ARBPRI register which is for P6\n");
975 		break;
976 
977 	case APIC_TMCCT:	/* Timer CCR */
978 		if (apic_lvtt_tscdeadline(apic))
979 			return 0;
980 
981 		val = apic_get_tmcct(apic);
982 		break;
983 	case APIC_PROCPRI:
984 		apic_update_ppr(apic);
985 		val = kvm_apic_get_reg(apic, offset);
986 		break;
987 	case APIC_TASKPRI:
988 		report_tpr_access(apic, false);
989 		/* fall thru */
990 	default:
991 		val = kvm_apic_get_reg(apic, offset);
992 		break;
993 	}
994 
995 	return val;
996 }
997 
998 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
999 {
1000 	return container_of(dev, struct kvm_lapic, dev);
1001 }
1002 
1003 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1004 		void *data)
1005 {
1006 	unsigned char alignment = offset & 0xf;
1007 	u32 result;
1008 	/* this bitmask has a bit cleared for each reserved register */
1009 	static const u64 rmask = 0x43ff01ffffffe70cULL;
1010 
1011 	if ((alignment + len) > 4) {
1012 		apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1013 			   offset, len);
1014 		return 1;
1015 	}
1016 
1017 	if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1018 		apic_debug("KVM_APIC_READ: read reserved register %x\n",
1019 			   offset);
1020 		return 1;
1021 	}
1022 
1023 	result = __apic_read(apic, offset & ~0xf);
1024 
1025 	trace_kvm_apic_read(offset, result);
1026 
1027 	switch (len) {
1028 	case 1:
1029 	case 2:
1030 	case 4:
1031 		memcpy(data, (char *)&result + alignment, len);
1032 		break;
1033 	default:
1034 		printk(KERN_ERR "Local APIC read with len = %x, "
1035 		       "should be 1,2, or 4 instead\n", len);
1036 		break;
1037 	}
1038 	return 0;
1039 }
1040 
1041 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1042 {
1043 	return kvm_apic_hw_enabled(apic) &&
1044 	    addr >= apic->base_address &&
1045 	    addr < apic->base_address + LAPIC_MMIO_LENGTH;
1046 }
1047 
1048 static int apic_mmio_read(struct kvm_io_device *this,
1049 			   gpa_t address, int len, void *data)
1050 {
1051 	struct kvm_lapic *apic = to_lapic(this);
1052 	u32 offset = address - apic->base_address;
1053 
1054 	if (!apic_mmio_in_range(apic, address))
1055 		return -EOPNOTSUPP;
1056 
1057 	apic_reg_read(apic, offset, len, data);
1058 
1059 	return 0;
1060 }
1061 
1062 static void update_divide_count(struct kvm_lapic *apic)
1063 {
1064 	u32 tmp1, tmp2, tdcr;
1065 
1066 	tdcr = kvm_apic_get_reg(apic, APIC_TDCR);
1067 	tmp1 = tdcr & 0xf;
1068 	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1069 	apic->divide_count = 0x1 << (tmp2 & 0x7);
1070 
1071 	apic_debug("timer divide count is 0x%x\n",
1072 				   apic->divide_count);
1073 }
1074 
1075 static void apic_timer_expired(struct kvm_lapic *apic)
1076 {
1077 	struct kvm_vcpu *vcpu = apic->vcpu;
1078 	wait_queue_head_t *q = &vcpu->wq;
1079 
1080 	/*
1081 	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1082 	 * vcpu_enter_guest.
1083 	 */
1084 	if (atomic_read(&apic->lapic_timer.pending))
1085 		return;
1086 
1087 	atomic_inc(&apic->lapic_timer.pending);
1088 	/* FIXME: this code should not know anything about vcpus */
1089 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1090 
1091 	if (waitqueue_active(q))
1092 		wake_up_interruptible(q);
1093 }
1094 
1095 static void start_apic_timer(struct kvm_lapic *apic)
1096 {
1097 	ktime_t now;
1098 	atomic_set(&apic->lapic_timer.pending, 0);
1099 
1100 	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1101 		/* lapic timer in oneshot or periodic mode */
1102 		now = apic->lapic_timer.timer.base->get_time();
1103 		apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT)
1104 			    * APIC_BUS_CYCLE_NS * apic->divide_count;
1105 
1106 		if (!apic->lapic_timer.period)
1107 			return;
1108 		/*
1109 		 * Do not allow the guest to program periodic timers with small
1110 		 * interval, since the hrtimers are not throttled by the host
1111 		 * scheduler.
1112 		 */
1113 		if (apic_lvtt_period(apic)) {
1114 			s64 min_period = min_timer_period_us * 1000LL;
1115 
1116 			if (apic->lapic_timer.period < min_period) {
1117 				pr_info_ratelimited(
1118 				    "kvm: vcpu %i: requested %lld ns "
1119 				    "lapic timer period limited to %lld ns\n",
1120 				    apic->vcpu->vcpu_id,
1121 				    apic->lapic_timer.period, min_period);
1122 				apic->lapic_timer.period = min_period;
1123 			}
1124 		}
1125 
1126 		hrtimer_start(&apic->lapic_timer.timer,
1127 			      ktime_add_ns(now, apic->lapic_timer.period),
1128 			      HRTIMER_MODE_ABS);
1129 
1130 		apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1131 			   PRIx64 ", "
1132 			   "timer initial count 0x%x, period %lldns, "
1133 			   "expire @ 0x%016" PRIx64 ".\n", __func__,
1134 			   APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1135 			   kvm_apic_get_reg(apic, APIC_TMICT),
1136 			   apic->lapic_timer.period,
1137 			   ktime_to_ns(ktime_add_ns(now,
1138 					apic->lapic_timer.period)));
1139 	} else if (apic_lvtt_tscdeadline(apic)) {
1140 		/* lapic timer in tsc deadline mode */
1141 		u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1142 		u64 ns = 0;
1143 		struct kvm_vcpu *vcpu = apic->vcpu;
1144 		unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1145 		unsigned long flags;
1146 
1147 		if (unlikely(!tscdeadline || !this_tsc_khz))
1148 			return;
1149 
1150 		local_irq_save(flags);
1151 
1152 		now = apic->lapic_timer.timer.base->get_time();
1153 		guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1154 		if (likely(tscdeadline > guest_tsc)) {
1155 			ns = (tscdeadline - guest_tsc) * 1000000ULL;
1156 			do_div(ns, this_tsc_khz);
1157 			hrtimer_start(&apic->lapic_timer.timer,
1158 				ktime_add_ns(now, ns), HRTIMER_MODE_ABS);
1159 		} else
1160 			apic_timer_expired(apic);
1161 
1162 		local_irq_restore(flags);
1163 	}
1164 }
1165 
1166 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1167 {
1168 	int nmi_wd_enabled = apic_lvt_nmi_mode(kvm_apic_get_reg(apic, APIC_LVT0));
1169 
1170 	if (apic_lvt_nmi_mode(lvt0_val)) {
1171 		if (!nmi_wd_enabled) {
1172 			apic_debug("Receive NMI setting on APIC_LVT0 "
1173 				   "for cpu %d\n", apic->vcpu->vcpu_id);
1174 			apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
1175 		}
1176 	} else if (nmi_wd_enabled)
1177 		apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
1178 }
1179 
1180 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1181 {
1182 	int ret = 0;
1183 
1184 	trace_kvm_apic_write(reg, val);
1185 
1186 	switch (reg) {
1187 	case APIC_ID:		/* Local APIC ID */
1188 		if (!apic_x2apic_mode(apic))
1189 			kvm_apic_set_id(apic, val >> 24);
1190 		else
1191 			ret = 1;
1192 		break;
1193 
1194 	case APIC_TASKPRI:
1195 		report_tpr_access(apic, true);
1196 		apic_set_tpr(apic, val & 0xff);
1197 		break;
1198 
1199 	case APIC_EOI:
1200 		apic_set_eoi(apic);
1201 		break;
1202 
1203 	case APIC_LDR:
1204 		if (!apic_x2apic_mode(apic))
1205 			kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1206 		else
1207 			ret = 1;
1208 		break;
1209 
1210 	case APIC_DFR:
1211 		if (!apic_x2apic_mode(apic)) {
1212 			apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1213 			recalculate_apic_map(apic->vcpu->kvm);
1214 		} else
1215 			ret = 1;
1216 		break;
1217 
1218 	case APIC_SPIV: {
1219 		u32 mask = 0x3ff;
1220 		if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1221 			mask |= APIC_SPIV_DIRECTED_EOI;
1222 		apic_set_spiv(apic, val & mask);
1223 		if (!(val & APIC_SPIV_APIC_ENABLED)) {
1224 			int i;
1225 			u32 lvt_val;
1226 
1227 			for (i = 0; i < APIC_LVT_NUM; i++) {
1228 				lvt_val = kvm_apic_get_reg(apic,
1229 						       APIC_LVTT + 0x10 * i);
1230 				apic_set_reg(apic, APIC_LVTT + 0x10 * i,
1231 					     lvt_val | APIC_LVT_MASKED);
1232 			}
1233 			atomic_set(&apic->lapic_timer.pending, 0);
1234 
1235 		}
1236 		break;
1237 	}
1238 	case APIC_ICR:
1239 		/* No delay here, so we always clear the pending bit */
1240 		apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1241 		apic_send_ipi(apic);
1242 		break;
1243 
1244 	case APIC_ICR2:
1245 		if (!apic_x2apic_mode(apic))
1246 			val &= 0xff000000;
1247 		apic_set_reg(apic, APIC_ICR2, val);
1248 		break;
1249 
1250 	case APIC_LVT0:
1251 		apic_manage_nmi_watchdog(apic, val);
1252 	case APIC_LVTTHMR:
1253 	case APIC_LVTPC:
1254 	case APIC_LVT1:
1255 	case APIC_LVTERR:
1256 		/* TODO: Check vector */
1257 		if (!kvm_apic_sw_enabled(apic))
1258 			val |= APIC_LVT_MASKED;
1259 
1260 		val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1261 		apic_set_reg(apic, reg, val);
1262 
1263 		break;
1264 
1265 	case APIC_LVTT: {
1266 		u32 timer_mode = val & apic->lapic_timer.timer_mode_mask;
1267 
1268 		if (apic->lapic_timer.timer_mode != timer_mode) {
1269 			apic->lapic_timer.timer_mode = timer_mode;
1270 			hrtimer_cancel(&apic->lapic_timer.timer);
1271 		}
1272 
1273 		if (!kvm_apic_sw_enabled(apic))
1274 			val |= APIC_LVT_MASKED;
1275 		val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1276 		apic_set_reg(apic, APIC_LVTT, val);
1277 		break;
1278 	}
1279 
1280 	case APIC_TMICT:
1281 		if (apic_lvtt_tscdeadline(apic))
1282 			break;
1283 
1284 		hrtimer_cancel(&apic->lapic_timer.timer);
1285 		apic_set_reg(apic, APIC_TMICT, val);
1286 		start_apic_timer(apic);
1287 		break;
1288 
1289 	case APIC_TDCR:
1290 		if (val & 4)
1291 			apic_debug("KVM_WRITE:TDCR %x\n", val);
1292 		apic_set_reg(apic, APIC_TDCR, val);
1293 		update_divide_count(apic);
1294 		break;
1295 
1296 	case APIC_ESR:
1297 		if (apic_x2apic_mode(apic) && val != 0) {
1298 			apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1299 			ret = 1;
1300 		}
1301 		break;
1302 
1303 	case APIC_SELF_IPI:
1304 		if (apic_x2apic_mode(apic)) {
1305 			apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1306 		} else
1307 			ret = 1;
1308 		break;
1309 	default:
1310 		ret = 1;
1311 		break;
1312 	}
1313 	if (ret)
1314 		apic_debug("Local APIC Write to read-only register %x\n", reg);
1315 	return ret;
1316 }
1317 
1318 static int apic_mmio_write(struct kvm_io_device *this,
1319 			    gpa_t address, int len, const void *data)
1320 {
1321 	struct kvm_lapic *apic = to_lapic(this);
1322 	unsigned int offset = address - apic->base_address;
1323 	u32 val;
1324 
1325 	if (!apic_mmio_in_range(apic, address))
1326 		return -EOPNOTSUPP;
1327 
1328 	/*
1329 	 * APIC register must be aligned on 128-bits boundary.
1330 	 * 32/64/128 bits registers must be accessed thru 32 bits.
1331 	 * Refer SDM 8.4.1
1332 	 */
1333 	if (len != 4 || (offset & 0xf)) {
1334 		/* Don't shout loud, $infamous_os would cause only noise. */
1335 		apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1336 		return 0;
1337 	}
1338 
1339 	val = *(u32*)data;
1340 
1341 	/* too common printing */
1342 	if (offset != APIC_EOI)
1343 		apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1344 			   "0x%x\n", __func__, offset, len, val);
1345 
1346 	apic_reg_write(apic, offset & 0xff0, val);
1347 
1348 	return 0;
1349 }
1350 
1351 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1352 {
1353 	if (kvm_vcpu_has_lapic(vcpu))
1354 		apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1355 }
1356 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1357 
1358 /* emulate APIC access in a trap manner */
1359 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1360 {
1361 	u32 val = 0;
1362 
1363 	/* hw has done the conditional check and inst decode */
1364 	offset &= 0xff0;
1365 
1366 	apic_reg_read(vcpu->arch.apic, offset, 4, &val);
1367 
1368 	/* TODO: optimize to just emulate side effect w/o one more write */
1369 	apic_reg_write(vcpu->arch.apic, offset, val);
1370 }
1371 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1372 
1373 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1374 {
1375 	struct kvm_lapic *apic = vcpu->arch.apic;
1376 
1377 	if (!vcpu->arch.apic)
1378 		return;
1379 
1380 	hrtimer_cancel(&apic->lapic_timer.timer);
1381 
1382 	if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1383 		static_key_slow_dec_deferred(&apic_hw_disabled);
1384 
1385 	if (!apic->sw_enabled)
1386 		static_key_slow_dec_deferred(&apic_sw_disabled);
1387 
1388 	if (apic->regs)
1389 		free_page((unsigned long)apic->regs);
1390 
1391 	kfree(apic);
1392 }
1393 
1394 /*
1395  *----------------------------------------------------------------------
1396  * LAPIC interface
1397  *----------------------------------------------------------------------
1398  */
1399 
1400 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1401 {
1402 	struct kvm_lapic *apic = vcpu->arch.apic;
1403 
1404 	if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1405 			apic_lvtt_period(apic))
1406 		return 0;
1407 
1408 	return apic->lapic_timer.tscdeadline;
1409 }
1410 
1411 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1412 {
1413 	struct kvm_lapic *apic = vcpu->arch.apic;
1414 
1415 	if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1416 			apic_lvtt_period(apic))
1417 		return;
1418 
1419 	hrtimer_cancel(&apic->lapic_timer.timer);
1420 	apic->lapic_timer.tscdeadline = data;
1421 	start_apic_timer(apic);
1422 }
1423 
1424 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1425 {
1426 	struct kvm_lapic *apic = vcpu->arch.apic;
1427 
1428 	if (!kvm_vcpu_has_lapic(vcpu))
1429 		return;
1430 
1431 	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1432 		     | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4));
1433 }
1434 
1435 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1436 {
1437 	u64 tpr;
1438 
1439 	if (!kvm_vcpu_has_lapic(vcpu))
1440 		return 0;
1441 
1442 	tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1443 
1444 	return (tpr & 0xf0) >> 4;
1445 }
1446 
1447 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1448 {
1449 	u64 old_value = vcpu->arch.apic_base;
1450 	struct kvm_lapic *apic = vcpu->arch.apic;
1451 
1452 	if (!apic) {
1453 		value |= MSR_IA32_APICBASE_BSP;
1454 		vcpu->arch.apic_base = value;
1455 		return;
1456 	}
1457 
1458 	if (!kvm_vcpu_is_bsp(apic->vcpu))
1459 		value &= ~MSR_IA32_APICBASE_BSP;
1460 	vcpu->arch.apic_base = value;
1461 
1462 	/* update jump label if enable bit changes */
1463 	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1464 		if (value & MSR_IA32_APICBASE_ENABLE)
1465 			static_key_slow_dec_deferred(&apic_hw_disabled);
1466 		else
1467 			static_key_slow_inc(&apic_hw_disabled.key);
1468 		recalculate_apic_map(vcpu->kvm);
1469 	}
1470 
1471 	if ((old_value ^ value) & X2APIC_ENABLE) {
1472 		if (value & X2APIC_ENABLE) {
1473 			u32 id = kvm_apic_id(apic);
1474 			u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
1475 			kvm_apic_set_ldr(apic, ldr);
1476 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1477 		} else
1478 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1479 	}
1480 
1481 	apic->base_address = apic->vcpu->arch.apic_base &
1482 			     MSR_IA32_APICBASE_BASE;
1483 
1484 	if ((value & MSR_IA32_APICBASE_ENABLE) &&
1485 	     apic->base_address != APIC_DEFAULT_PHYS_BASE)
1486 		pr_warn_once("APIC base relocation is unsupported by KVM");
1487 
1488 	/* with FSB delivery interrupt, we can restart APIC functionality */
1489 	apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1490 		   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1491 
1492 }
1493 
1494 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1495 {
1496 	struct kvm_lapic *apic;
1497 	int i;
1498 
1499 	apic_debug("%s\n", __func__);
1500 
1501 	ASSERT(vcpu);
1502 	apic = vcpu->arch.apic;
1503 	ASSERT(apic != NULL);
1504 
1505 	/* Stop the timer in case it's a reset to an active apic */
1506 	hrtimer_cancel(&apic->lapic_timer.timer);
1507 
1508 	kvm_apic_set_id(apic, vcpu->vcpu_id);
1509 	kvm_apic_set_version(apic->vcpu);
1510 
1511 	for (i = 0; i < APIC_LVT_NUM; i++)
1512 		apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1513 	apic->lapic_timer.timer_mode = 0;
1514 	apic_set_reg(apic, APIC_LVT0,
1515 		     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1516 
1517 	apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1518 	apic_set_spiv(apic, 0xff);
1519 	apic_set_reg(apic, APIC_TASKPRI, 0);
1520 	kvm_apic_set_ldr(apic, 0);
1521 	apic_set_reg(apic, APIC_ESR, 0);
1522 	apic_set_reg(apic, APIC_ICR, 0);
1523 	apic_set_reg(apic, APIC_ICR2, 0);
1524 	apic_set_reg(apic, APIC_TDCR, 0);
1525 	apic_set_reg(apic, APIC_TMICT, 0);
1526 	for (i = 0; i < 8; i++) {
1527 		apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1528 		apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1529 		apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1530 	}
1531 	apic->irr_pending = kvm_apic_vid_enabled(vcpu->kvm);
1532 	apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm);
1533 	apic->highest_isr_cache = -1;
1534 	update_divide_count(apic);
1535 	atomic_set(&apic->lapic_timer.pending, 0);
1536 	if (kvm_vcpu_is_bsp(vcpu))
1537 		kvm_lapic_set_base(vcpu,
1538 				vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1539 	vcpu->arch.pv_eoi.msr_val = 0;
1540 	apic_update_ppr(apic);
1541 
1542 	vcpu->arch.apic_arb_prio = 0;
1543 	vcpu->arch.apic_attention = 0;
1544 
1545 	apic_debug("%s: vcpu=%p, id=%d, base_msr="
1546 		   "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1547 		   vcpu, kvm_apic_id(apic),
1548 		   vcpu->arch.apic_base, apic->base_address);
1549 }
1550 
1551 /*
1552  *----------------------------------------------------------------------
1553  * timer interface
1554  *----------------------------------------------------------------------
1555  */
1556 
1557 static bool lapic_is_periodic(struct kvm_lapic *apic)
1558 {
1559 	return apic_lvtt_period(apic);
1560 }
1561 
1562 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1563 {
1564 	struct kvm_lapic *apic = vcpu->arch.apic;
1565 
1566 	if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) &&
1567 			apic_lvt_enabled(apic, APIC_LVTT))
1568 		return atomic_read(&apic->lapic_timer.pending);
1569 
1570 	return 0;
1571 }
1572 
1573 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1574 {
1575 	u32 reg = kvm_apic_get_reg(apic, lvt_type);
1576 	int vector, mode, trig_mode;
1577 
1578 	if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1579 		vector = reg & APIC_VECTOR_MASK;
1580 		mode = reg & APIC_MODE_MASK;
1581 		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1582 		return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1583 					NULL);
1584 	}
1585 	return 0;
1586 }
1587 
1588 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1589 {
1590 	struct kvm_lapic *apic = vcpu->arch.apic;
1591 
1592 	if (apic)
1593 		kvm_apic_local_deliver(apic, APIC_LVT0);
1594 }
1595 
1596 static const struct kvm_io_device_ops apic_mmio_ops = {
1597 	.read     = apic_mmio_read,
1598 	.write    = apic_mmio_write,
1599 };
1600 
1601 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1602 {
1603 	struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1604 	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1605 
1606 	apic_timer_expired(apic);
1607 
1608 	if (lapic_is_periodic(apic)) {
1609 		hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1610 		return HRTIMER_RESTART;
1611 	} else
1612 		return HRTIMER_NORESTART;
1613 }
1614 
1615 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1616 {
1617 	struct kvm_lapic *apic;
1618 
1619 	ASSERT(vcpu != NULL);
1620 	apic_debug("apic_init %d\n", vcpu->vcpu_id);
1621 
1622 	apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1623 	if (!apic)
1624 		goto nomem;
1625 
1626 	vcpu->arch.apic = apic;
1627 
1628 	apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1629 	if (!apic->regs) {
1630 		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1631 		       vcpu->vcpu_id);
1632 		goto nomem_free_apic;
1633 	}
1634 	apic->vcpu = vcpu;
1635 
1636 	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1637 		     HRTIMER_MODE_ABS);
1638 	apic->lapic_timer.timer.function = apic_timer_fn;
1639 
1640 	/*
1641 	 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1642 	 * thinking that APIC satet has changed.
1643 	 */
1644 	vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1645 	kvm_lapic_set_base(vcpu,
1646 			APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE);
1647 
1648 	static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1649 	kvm_lapic_reset(vcpu);
1650 	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1651 
1652 	return 0;
1653 nomem_free_apic:
1654 	kfree(apic);
1655 nomem:
1656 	return -ENOMEM;
1657 }
1658 
1659 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1660 {
1661 	struct kvm_lapic *apic = vcpu->arch.apic;
1662 	int highest_irr;
1663 
1664 	if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic))
1665 		return -1;
1666 
1667 	apic_update_ppr(apic);
1668 	highest_irr = apic_find_highest_irr(apic);
1669 	if ((highest_irr == -1) ||
1670 	    ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI)))
1671 		return -1;
1672 	return highest_irr;
1673 }
1674 
1675 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1676 {
1677 	u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1678 	int r = 0;
1679 
1680 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1681 		r = 1;
1682 	if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1683 	    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1684 		r = 1;
1685 	return r;
1686 }
1687 
1688 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1689 {
1690 	struct kvm_lapic *apic = vcpu->arch.apic;
1691 
1692 	if (!kvm_vcpu_has_lapic(vcpu))
1693 		return;
1694 
1695 	if (atomic_read(&apic->lapic_timer.pending) > 0) {
1696 		kvm_apic_local_deliver(apic, APIC_LVTT);
1697 		if (apic_lvtt_tscdeadline(apic))
1698 			apic->lapic_timer.tscdeadline = 0;
1699 		atomic_set(&apic->lapic_timer.pending, 0);
1700 	}
1701 }
1702 
1703 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1704 {
1705 	int vector = kvm_apic_has_interrupt(vcpu);
1706 	struct kvm_lapic *apic = vcpu->arch.apic;
1707 
1708 	if (vector == -1)
1709 		return -1;
1710 
1711 	/*
1712 	 * We get here even with APIC virtualization enabled, if doing
1713 	 * nested virtualization and L1 runs with the "acknowledge interrupt
1714 	 * on exit" mode.  Then we cannot inject the interrupt via RVI,
1715 	 * because the process would deliver it through the IDT.
1716 	 */
1717 
1718 	apic_set_isr(vector, apic);
1719 	apic_update_ppr(apic);
1720 	apic_clear_irr(vector, apic);
1721 	return vector;
1722 }
1723 
1724 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
1725 		struct kvm_lapic_state *s)
1726 {
1727 	struct kvm_lapic *apic = vcpu->arch.apic;
1728 
1729 	kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
1730 	/* set SPIV separately to get count of SW disabled APICs right */
1731 	apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
1732 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1733 	/* call kvm_apic_set_id() to put apic into apic_map */
1734 	kvm_apic_set_id(apic, kvm_apic_id(apic));
1735 	kvm_apic_set_version(vcpu);
1736 
1737 	apic_update_ppr(apic);
1738 	hrtimer_cancel(&apic->lapic_timer.timer);
1739 	update_divide_count(apic);
1740 	start_apic_timer(apic);
1741 	apic->irr_pending = true;
1742 	apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm) ?
1743 				1 : count_vectors(apic->regs + APIC_ISR);
1744 	apic->highest_isr_cache = -1;
1745 	if (kvm_x86_ops->hwapic_irr_update)
1746 		kvm_x86_ops->hwapic_irr_update(vcpu,
1747 				apic_find_highest_irr(apic));
1748 	kvm_x86_ops->hwapic_isr_update(vcpu->kvm, apic_find_highest_isr(apic));
1749 	kvm_make_request(KVM_REQ_EVENT, vcpu);
1750 	kvm_rtc_eoi_tracking_restore_one(vcpu);
1751 }
1752 
1753 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1754 {
1755 	struct hrtimer *timer;
1756 
1757 	if (!kvm_vcpu_has_lapic(vcpu))
1758 		return;
1759 
1760 	timer = &vcpu->arch.apic->lapic_timer.timer;
1761 	if (hrtimer_cancel(timer))
1762 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1763 }
1764 
1765 /*
1766  * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1767  *
1768  * Detect whether guest triggered PV EOI since the
1769  * last entry. If yes, set EOI on guests's behalf.
1770  * Clear PV EOI in guest memory in any case.
1771  */
1772 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1773 					struct kvm_lapic *apic)
1774 {
1775 	bool pending;
1776 	int vector;
1777 	/*
1778 	 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1779 	 * and KVM_PV_EOI_ENABLED in guest memory as follows:
1780 	 *
1781 	 * KVM_APIC_PV_EOI_PENDING is unset:
1782 	 * 	-> host disabled PV EOI.
1783 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1784 	 * 	-> host enabled PV EOI, guest did not execute EOI yet.
1785 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1786 	 * 	-> host enabled PV EOI, guest executed EOI.
1787 	 */
1788 	BUG_ON(!pv_eoi_enabled(vcpu));
1789 	pending = pv_eoi_get_pending(vcpu);
1790 	/*
1791 	 * Clear pending bit in any case: it will be set again on vmentry.
1792 	 * While this might not be ideal from performance point of view,
1793 	 * this makes sure pv eoi is only enabled when we know it's safe.
1794 	 */
1795 	pv_eoi_clr_pending(vcpu);
1796 	if (pending)
1797 		return;
1798 	vector = apic_set_eoi(apic);
1799 	trace_kvm_pv_eoi(apic, vector);
1800 }
1801 
1802 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1803 {
1804 	u32 data;
1805 
1806 	if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1807 		apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1808 
1809 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1810 		return;
1811 
1812 	kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1813 				sizeof(u32));
1814 
1815 	apic_set_tpr(vcpu->arch.apic, data & 0xff);
1816 }
1817 
1818 /*
1819  * apic_sync_pv_eoi_to_guest - called before vmentry
1820  *
1821  * Detect whether it's safe to enable PV EOI and
1822  * if yes do so.
1823  */
1824 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
1825 					struct kvm_lapic *apic)
1826 {
1827 	if (!pv_eoi_enabled(vcpu) ||
1828 	    /* IRR set or many bits in ISR: could be nested. */
1829 	    apic->irr_pending ||
1830 	    /* Cache not set: could be safe but we don't bother. */
1831 	    apic->highest_isr_cache == -1 ||
1832 	    /* Need EOI to update ioapic. */
1833 	    kvm_ioapic_handles_vector(vcpu->kvm, apic->highest_isr_cache)) {
1834 		/*
1835 		 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1836 		 * so we need not do anything here.
1837 		 */
1838 		return;
1839 	}
1840 
1841 	pv_eoi_set_pending(apic->vcpu);
1842 }
1843 
1844 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1845 {
1846 	u32 data, tpr;
1847 	int max_irr, max_isr;
1848 	struct kvm_lapic *apic = vcpu->arch.apic;
1849 
1850 	apic_sync_pv_eoi_to_guest(vcpu, apic);
1851 
1852 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1853 		return;
1854 
1855 	tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1856 	max_irr = apic_find_highest_irr(apic);
1857 	if (max_irr < 0)
1858 		max_irr = 0;
1859 	max_isr = apic_find_highest_isr(apic);
1860 	if (max_isr < 0)
1861 		max_isr = 0;
1862 	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1863 
1864 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1865 				sizeof(u32));
1866 }
1867 
1868 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1869 {
1870 	if (vapic_addr) {
1871 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1872 					&vcpu->arch.apic->vapic_cache,
1873 					vapic_addr, sizeof(u32)))
1874 			return -EINVAL;
1875 		__set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1876 	} else {
1877 		__clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1878 	}
1879 
1880 	vcpu->arch.apic->vapic_addr = vapic_addr;
1881 	return 0;
1882 }
1883 
1884 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1885 {
1886 	struct kvm_lapic *apic = vcpu->arch.apic;
1887 	u32 reg = (msr - APIC_BASE_MSR) << 4;
1888 
1889 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1890 		return 1;
1891 
1892 	if (reg == APIC_ICR2)
1893 		return 1;
1894 
1895 	/* if this is ICR write vector before command */
1896 	if (reg == APIC_ICR)
1897 		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1898 	return apic_reg_write(apic, reg, (u32)data);
1899 }
1900 
1901 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1902 {
1903 	struct kvm_lapic *apic = vcpu->arch.apic;
1904 	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1905 
1906 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1907 		return 1;
1908 
1909 	if (reg == APIC_DFR || reg == APIC_ICR2) {
1910 		apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
1911 			   reg);
1912 		return 1;
1913 	}
1914 
1915 	if (apic_reg_read(apic, reg, 4, &low))
1916 		return 1;
1917 	if (reg == APIC_ICR)
1918 		apic_reg_read(apic, APIC_ICR2, 4, &high);
1919 
1920 	*data = (((u64)high) << 32) | low;
1921 
1922 	return 0;
1923 }
1924 
1925 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1926 {
1927 	struct kvm_lapic *apic = vcpu->arch.apic;
1928 
1929 	if (!kvm_vcpu_has_lapic(vcpu))
1930 		return 1;
1931 
1932 	/* if this is ICR write vector before command */
1933 	if (reg == APIC_ICR)
1934 		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1935 	return apic_reg_write(apic, reg, (u32)data);
1936 }
1937 
1938 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
1939 {
1940 	struct kvm_lapic *apic = vcpu->arch.apic;
1941 	u32 low, high = 0;
1942 
1943 	if (!kvm_vcpu_has_lapic(vcpu))
1944 		return 1;
1945 
1946 	if (apic_reg_read(apic, reg, 4, &low))
1947 		return 1;
1948 	if (reg == APIC_ICR)
1949 		apic_reg_read(apic, APIC_ICR2, 4, &high);
1950 
1951 	*data = (((u64)high) << 32) | low;
1952 
1953 	return 0;
1954 }
1955 
1956 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
1957 {
1958 	u64 addr = data & ~KVM_MSR_ENABLED;
1959 	if (!IS_ALIGNED(addr, 4))
1960 		return 1;
1961 
1962 	vcpu->arch.pv_eoi.msr_val = data;
1963 	if (!pv_eoi_enabled(vcpu))
1964 		return 0;
1965 	return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
1966 					 addr, sizeof(u8));
1967 }
1968 
1969 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
1970 {
1971 	struct kvm_lapic *apic = vcpu->arch.apic;
1972 	u8 sipi_vector;
1973 	unsigned long pe;
1974 
1975 	if (!kvm_vcpu_has_lapic(vcpu) || !apic->pending_events)
1976 		return;
1977 
1978 	pe = xchg(&apic->pending_events, 0);
1979 
1980 	if (test_bit(KVM_APIC_INIT, &pe)) {
1981 		kvm_lapic_reset(vcpu);
1982 		kvm_vcpu_reset(vcpu);
1983 		if (kvm_vcpu_is_bsp(apic->vcpu))
1984 			vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1985 		else
1986 			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
1987 	}
1988 	if (test_bit(KVM_APIC_SIPI, &pe) &&
1989 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
1990 		/* evaluate pending_events before reading the vector */
1991 		smp_rmb();
1992 		sipi_vector = apic->sipi_vector;
1993 		apic_debug("vcpu %d received sipi with vector # %x\n",
1994 			 vcpu->vcpu_id, sipi_vector);
1995 		kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
1996 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1997 	}
1998 }
1999 
2000 void kvm_lapic_init(void)
2001 {
2002 	/* do not patch jump label more than once per second */
2003 	jump_label_rate_limit(&apic_hw_disabled, HZ);
2004 	jump_label_rate_limit(&apic_sw_disabled, HZ);
2005 }
2006