xref: /linux/arch/x86/kvm/lapic.c (revision b6ebbac51bedf9e98e837688bc838f400196da5e)
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/export.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 #include "hyperv.h"
45 
46 #ifndef CONFIG_X86_64
47 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
48 #else
49 #define mod_64(x, y) ((x) % (y))
50 #endif
51 
52 #define PRId64 "d"
53 #define PRIx64 "llx"
54 #define PRIu64 "u"
55 #define PRIo64 "o"
56 
57 #define APIC_BUS_CYCLE_NS 1
58 
59 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
60 #define apic_debug(fmt, arg...)
61 
62 /* 14 is the version for Xeon and Pentium 8.4.8*/
63 #define APIC_VERSION			(0x14UL | ((KVM_APIC_LVT_NUM - 1) << 16))
64 #define LAPIC_MMIO_LENGTH		(1 << 12)
65 /* followed define is not in apicdef.h */
66 #define APIC_SHORT_MASK			0xc0000
67 #define APIC_DEST_NOSHORT		0x0
68 #define APIC_DEST_MASK			0x800
69 #define MAX_APIC_VECTOR			256
70 #define APIC_VECTORS_PER_REG		32
71 
72 #define APIC_BROADCAST			0xFF
73 #define X2APIC_BROADCAST		0xFFFFFFFFul
74 
75 static inline int apic_test_vector(int vec, void *bitmap)
76 {
77 	return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
78 }
79 
80 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
81 {
82 	struct kvm_lapic *apic = vcpu->arch.apic;
83 
84 	return apic_test_vector(vector, apic->regs + APIC_ISR) ||
85 		apic_test_vector(vector, apic->regs + APIC_IRR);
86 }
87 
88 static inline void apic_clear_vector(int vec, void *bitmap)
89 {
90 	clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
91 }
92 
93 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
94 {
95 	return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
96 }
97 
98 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
99 {
100 	return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
101 }
102 
103 struct static_key_deferred apic_hw_disabled __read_mostly;
104 struct static_key_deferred apic_sw_disabled __read_mostly;
105 
106 static inline int apic_enabled(struct kvm_lapic *apic)
107 {
108 	return kvm_apic_sw_enabled(apic) &&	kvm_apic_hw_enabled(apic);
109 }
110 
111 #define LVT_MASK	\
112 	(APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
113 
114 #define LINT_MASK	\
115 	(LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
116 	 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
117 
118 static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
119 		u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
120 	switch (map->mode) {
121 	case KVM_APIC_MODE_X2APIC: {
122 		u32 offset = (dest_id >> 16) * 16;
123 		u32 max_apic_id = map->max_apic_id;
124 
125 		if (offset <= max_apic_id) {
126 			u8 cluster_size = min(max_apic_id - offset + 1, 16U);
127 
128 			*cluster = &map->phys_map[offset];
129 			*mask = dest_id & (0xffff >> (16 - cluster_size));
130 		} else {
131 			*mask = 0;
132 		}
133 
134 		return true;
135 		}
136 	case KVM_APIC_MODE_XAPIC_FLAT:
137 		*cluster = map->xapic_flat_map;
138 		*mask = dest_id & 0xff;
139 		return true;
140 	case KVM_APIC_MODE_XAPIC_CLUSTER:
141 		*cluster = map->xapic_cluster_map[dest_id >> 4];
142 		*mask = dest_id & 0xf;
143 		return true;
144 	default:
145 		/* Not optimized. */
146 		return false;
147 	}
148 }
149 
150 static void kvm_apic_map_free(struct rcu_head *rcu)
151 {
152 	struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
153 
154 	kvfree(map);
155 }
156 
157 static void recalculate_apic_map(struct kvm *kvm)
158 {
159 	struct kvm_apic_map *new, *old = NULL;
160 	struct kvm_vcpu *vcpu;
161 	int i;
162 	u32 max_id = 255;
163 
164 	mutex_lock(&kvm->arch.apic_map_lock);
165 
166 	kvm_for_each_vcpu(i, vcpu, kvm)
167 		if (kvm_apic_present(vcpu))
168 			max_id = max(max_id, kvm_apic_id(vcpu->arch.apic));
169 
170 	new = kvm_kvzalloc(sizeof(struct kvm_apic_map) +
171 	                   sizeof(struct kvm_lapic *) * ((u64)max_id + 1));
172 
173 	if (!new)
174 		goto out;
175 
176 	new->max_apic_id = max_id;
177 
178 	kvm_for_each_vcpu(i, vcpu, kvm) {
179 		struct kvm_lapic *apic = vcpu->arch.apic;
180 		struct kvm_lapic **cluster;
181 		u16 mask;
182 		u32 ldr, aid;
183 
184 		if (!kvm_apic_present(vcpu))
185 			continue;
186 
187 		aid = kvm_apic_id(apic);
188 		ldr = kvm_lapic_get_reg(apic, APIC_LDR);
189 
190 		if (aid <= new->max_apic_id)
191 			new->phys_map[aid] = apic;
192 
193 		if (apic_x2apic_mode(apic)) {
194 			new->mode |= KVM_APIC_MODE_X2APIC;
195 		} else if (ldr) {
196 			ldr = GET_APIC_LOGICAL_ID(ldr);
197 			if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
198 				new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
199 			else
200 				new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
201 		}
202 
203 		if (!kvm_apic_map_get_logical_dest(new, ldr, &cluster, &mask))
204 			continue;
205 
206 		if (mask)
207 			cluster[ffs(mask) - 1] = apic;
208 	}
209 out:
210 	old = rcu_dereference_protected(kvm->arch.apic_map,
211 			lockdep_is_held(&kvm->arch.apic_map_lock));
212 	rcu_assign_pointer(kvm->arch.apic_map, new);
213 	mutex_unlock(&kvm->arch.apic_map_lock);
214 
215 	if (old)
216 		call_rcu(&old->rcu, kvm_apic_map_free);
217 
218 	kvm_make_scan_ioapic_request(kvm);
219 }
220 
221 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
222 {
223 	bool enabled = val & APIC_SPIV_APIC_ENABLED;
224 
225 	kvm_lapic_set_reg(apic, APIC_SPIV, val);
226 
227 	if (enabled != apic->sw_enabled) {
228 		apic->sw_enabled = enabled;
229 		if (enabled) {
230 			static_key_slow_dec_deferred(&apic_sw_disabled);
231 			recalculate_apic_map(apic->vcpu->kvm);
232 		} else
233 			static_key_slow_inc(&apic_sw_disabled.key);
234 	}
235 }
236 
237 static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
238 {
239 	kvm_lapic_set_reg(apic, APIC_ID, id << 24);
240 	recalculate_apic_map(apic->vcpu->kvm);
241 }
242 
243 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
244 {
245 	kvm_lapic_set_reg(apic, APIC_LDR, id);
246 	recalculate_apic_map(apic->vcpu->kvm);
247 }
248 
249 static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
250 {
251 	u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
252 
253 	kvm_lapic_set_reg(apic, APIC_ID, id);
254 	kvm_lapic_set_reg(apic, APIC_LDR, ldr);
255 	recalculate_apic_map(apic->vcpu->kvm);
256 }
257 
258 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
259 {
260 	return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
261 }
262 
263 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
264 {
265 	return kvm_lapic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
266 }
267 
268 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
269 {
270 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
271 }
272 
273 static inline int apic_lvtt_period(struct kvm_lapic *apic)
274 {
275 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
276 }
277 
278 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
279 {
280 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
281 }
282 
283 static inline int apic_lvt_nmi_mode(u32 lvt_val)
284 {
285 	return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
286 }
287 
288 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
289 {
290 	struct kvm_lapic *apic = vcpu->arch.apic;
291 	struct kvm_cpuid_entry2 *feat;
292 	u32 v = APIC_VERSION;
293 
294 	if (!lapic_in_kernel(vcpu))
295 		return;
296 
297 	feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
298 	if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
299 		v |= APIC_LVR_DIRECTED_EOI;
300 	kvm_lapic_set_reg(apic, APIC_LVR, v);
301 }
302 
303 static const unsigned int apic_lvt_mask[KVM_APIC_LVT_NUM] = {
304 	LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
305 	LVT_MASK | APIC_MODE_MASK,	/* LVTTHMR */
306 	LVT_MASK | APIC_MODE_MASK,	/* LVTPC */
307 	LINT_MASK, LINT_MASK,	/* LVT0-1 */
308 	LVT_MASK		/* LVTERR */
309 };
310 
311 static int find_highest_vector(void *bitmap)
312 {
313 	int vec;
314 	u32 *reg;
315 
316 	for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
317 	     vec >= 0; vec -= APIC_VECTORS_PER_REG) {
318 		reg = bitmap + REG_POS(vec);
319 		if (*reg)
320 			return fls(*reg) - 1 + vec;
321 	}
322 
323 	return -1;
324 }
325 
326 static u8 count_vectors(void *bitmap)
327 {
328 	int vec;
329 	u32 *reg;
330 	u8 count = 0;
331 
332 	for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
333 		reg = bitmap + REG_POS(vec);
334 		count += hweight32(*reg);
335 	}
336 
337 	return count;
338 }
339 
340 void __kvm_apic_update_irr(u32 *pir, void *regs)
341 {
342 	u32 i, pir_val;
343 
344 	for (i = 0; i <= 7; i++) {
345 		pir_val = xchg(&pir[i], 0);
346 		if (pir_val)
347 			*((u32 *)(regs + APIC_IRR + i * 0x10)) |= pir_val;
348 	}
349 }
350 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
351 
352 void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
353 {
354 	struct kvm_lapic *apic = vcpu->arch.apic;
355 
356 	__kvm_apic_update_irr(pir, apic->regs);
357 
358 	kvm_make_request(KVM_REQ_EVENT, vcpu);
359 }
360 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
361 
362 static inline int apic_search_irr(struct kvm_lapic *apic)
363 {
364 	return find_highest_vector(apic->regs + APIC_IRR);
365 }
366 
367 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
368 {
369 	int result;
370 
371 	/*
372 	 * Note that irr_pending is just a hint. It will be always
373 	 * true with virtual interrupt delivery enabled.
374 	 */
375 	if (!apic->irr_pending)
376 		return -1;
377 
378 	if (apic->vcpu->arch.apicv_active)
379 		kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
380 	result = apic_search_irr(apic);
381 	ASSERT(result == -1 || result >= 16);
382 
383 	return result;
384 }
385 
386 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
387 {
388 	struct kvm_vcpu *vcpu;
389 
390 	vcpu = apic->vcpu;
391 
392 	if (unlikely(vcpu->arch.apicv_active)) {
393 		/* try to update RVI */
394 		apic_clear_vector(vec, apic->regs + APIC_IRR);
395 		kvm_make_request(KVM_REQ_EVENT, vcpu);
396 	} else {
397 		apic->irr_pending = false;
398 		apic_clear_vector(vec, apic->regs + APIC_IRR);
399 		if (apic_search_irr(apic) != -1)
400 			apic->irr_pending = true;
401 	}
402 }
403 
404 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
405 {
406 	struct kvm_vcpu *vcpu;
407 
408 	if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
409 		return;
410 
411 	vcpu = apic->vcpu;
412 
413 	/*
414 	 * With APIC virtualization enabled, all caching is disabled
415 	 * because the processor can modify ISR under the hood.  Instead
416 	 * just set SVI.
417 	 */
418 	if (unlikely(vcpu->arch.apicv_active))
419 		kvm_x86_ops->hwapic_isr_update(vcpu, vec);
420 	else {
421 		++apic->isr_count;
422 		BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
423 		/*
424 		 * ISR (in service register) bit is set when injecting an interrupt.
425 		 * The highest vector is injected. Thus the latest bit set matches
426 		 * the highest bit in ISR.
427 		 */
428 		apic->highest_isr_cache = vec;
429 	}
430 }
431 
432 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
433 {
434 	int result;
435 
436 	/*
437 	 * Note that isr_count is always 1, and highest_isr_cache
438 	 * is always -1, with APIC virtualization enabled.
439 	 */
440 	if (!apic->isr_count)
441 		return -1;
442 	if (likely(apic->highest_isr_cache != -1))
443 		return apic->highest_isr_cache;
444 
445 	result = find_highest_vector(apic->regs + APIC_ISR);
446 	ASSERT(result == -1 || result >= 16);
447 
448 	return result;
449 }
450 
451 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
452 {
453 	struct kvm_vcpu *vcpu;
454 	if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
455 		return;
456 
457 	vcpu = apic->vcpu;
458 
459 	/*
460 	 * We do get here for APIC virtualization enabled if the guest
461 	 * uses the Hyper-V APIC enlightenment.  In this case we may need
462 	 * to trigger a new interrupt delivery by writing the SVI field;
463 	 * on the other hand isr_count and highest_isr_cache are unused
464 	 * and must be left alone.
465 	 */
466 	if (unlikely(vcpu->arch.apicv_active))
467 		kvm_x86_ops->hwapic_isr_update(vcpu,
468 					       apic_find_highest_isr(apic));
469 	else {
470 		--apic->isr_count;
471 		BUG_ON(apic->isr_count < 0);
472 		apic->highest_isr_cache = -1;
473 	}
474 }
475 
476 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
477 {
478 	/* This may race with setting of irr in __apic_accept_irq() and
479 	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
480 	 * will cause vmexit immediately and the value will be recalculated
481 	 * on the next vmentry.
482 	 */
483 	return apic_find_highest_irr(vcpu->arch.apic);
484 }
485 
486 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
487 			     int vector, int level, int trig_mode,
488 			     struct dest_map *dest_map);
489 
490 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
491 		     struct dest_map *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 
499 static 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 
506 static 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 
513 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
514 {
515 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
516 }
517 
518 static 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 
527 static 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 
537 static 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 
547 static void apic_update_ppr(struct kvm_lapic *apic)
548 {
549 	u32 tpr, isrv, ppr, old_ppr;
550 	int isr;
551 
552 	old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
553 	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
554 	isr = apic_find_highest_isr(apic);
555 	isrv = (isr != -1) ? isr : 0;
556 
557 	if ((tpr & 0xf0) >= (isrv & 0xf0))
558 		ppr = tpr & 0xff;
559 	else
560 		ppr = isrv & 0xf0;
561 
562 	apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
563 		   apic, ppr, isr, isrv);
564 
565 	if (old_ppr != ppr) {
566 		kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
567 		if (ppr < old_ppr)
568 			kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
569 	}
570 }
571 
572 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
573 {
574 	kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
575 	apic_update_ppr(apic);
576 }
577 
578 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
579 {
580 	if (apic_x2apic_mode(apic))
581 		return mda == X2APIC_BROADCAST;
582 
583 	return GET_APIC_DEST_FIELD(mda) == APIC_BROADCAST;
584 }
585 
586 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
587 {
588 	if (kvm_apic_broadcast(apic, mda))
589 		return true;
590 
591 	if (apic_x2apic_mode(apic))
592 		return mda == kvm_apic_id(apic);
593 
594 	return mda == SET_APIC_DEST_FIELD(kvm_apic_id(apic));
595 }
596 
597 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
598 {
599 	u32 logical_id;
600 
601 	if (kvm_apic_broadcast(apic, mda))
602 		return true;
603 
604 	logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
605 
606 	if (apic_x2apic_mode(apic))
607 		return ((logical_id >> 16) == (mda >> 16))
608 		       && (logical_id & mda & 0xffff) != 0;
609 
610 	logical_id = GET_APIC_LOGICAL_ID(logical_id);
611 	mda = GET_APIC_DEST_FIELD(mda);
612 
613 	switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
614 	case APIC_DFR_FLAT:
615 		return (logical_id & mda) != 0;
616 	case APIC_DFR_CLUSTER:
617 		return ((logical_id >> 4) == (mda >> 4))
618 		       && (logical_id & mda & 0xf) != 0;
619 	default:
620 		apic_debug("Bad DFR vcpu %d: %08x\n",
621 			   apic->vcpu->vcpu_id, kvm_lapic_get_reg(apic, APIC_DFR));
622 		return false;
623 	}
624 }
625 
626 /* The KVM local APIC implementation has two quirks:
627  *
628  *  - the xAPIC MDA stores the destination at bits 24-31, while this
629  *    is not true of struct kvm_lapic_irq's dest_id field.  This is
630  *    just a quirk in the API and is not problematic.
631  *
632  *  - in-kernel IOAPIC messages have to be delivered directly to
633  *    x2APIC, because the kernel does not support interrupt remapping.
634  *    In order to support broadcast without interrupt remapping, x2APIC
635  *    rewrites the destination of non-IPI messages from APIC_BROADCAST
636  *    to X2APIC_BROADCAST.
637  *
638  * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API.  This is
639  * important when userspace wants to use x2APIC-format MSIs, because
640  * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
641  */
642 static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
643 		struct kvm_lapic *source, struct kvm_lapic *target)
644 {
645 	bool ipi = source != NULL;
646 	bool x2apic_mda = apic_x2apic_mode(ipi ? source : target);
647 
648 	if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
649 	    !ipi && dest_id == APIC_BROADCAST && x2apic_mda)
650 		return X2APIC_BROADCAST;
651 
652 	return x2apic_mda ? dest_id : SET_APIC_DEST_FIELD(dest_id);
653 }
654 
655 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
656 			   int short_hand, unsigned int dest, int dest_mode)
657 {
658 	struct kvm_lapic *target = vcpu->arch.apic;
659 	u32 mda = kvm_apic_mda(vcpu, dest, source, target);
660 
661 	apic_debug("target %p, source %p, dest 0x%x, "
662 		   "dest_mode 0x%x, short_hand 0x%x\n",
663 		   target, source, dest, dest_mode, short_hand);
664 
665 	ASSERT(target);
666 	switch (short_hand) {
667 	case APIC_DEST_NOSHORT:
668 		if (dest_mode == APIC_DEST_PHYSICAL)
669 			return kvm_apic_match_physical_addr(target, mda);
670 		else
671 			return kvm_apic_match_logical_addr(target, mda);
672 	case APIC_DEST_SELF:
673 		return target == source;
674 	case APIC_DEST_ALLINC:
675 		return true;
676 	case APIC_DEST_ALLBUT:
677 		return target != source;
678 	default:
679 		apic_debug("kvm: apic: Bad dest shorthand value %x\n",
680 			   short_hand);
681 		return false;
682 	}
683 }
684 EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
685 
686 int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
687 		       const unsigned long *bitmap, u32 bitmap_size)
688 {
689 	u32 mod;
690 	int i, idx = -1;
691 
692 	mod = vector % dest_vcpus;
693 
694 	for (i = 0; i <= mod; i++) {
695 		idx = find_next_bit(bitmap, bitmap_size, idx + 1);
696 		BUG_ON(idx == bitmap_size);
697 	}
698 
699 	return idx;
700 }
701 
702 static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
703 {
704 	if (!kvm->arch.disabled_lapic_found) {
705 		kvm->arch.disabled_lapic_found = true;
706 		printk(KERN_INFO
707 		       "Disabled LAPIC found during irq injection\n");
708 	}
709 }
710 
711 static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
712 		struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
713 {
714 	if (kvm->arch.x2apic_broadcast_quirk_disabled) {
715 		if ((irq->dest_id == APIC_BROADCAST &&
716 				map->mode != KVM_APIC_MODE_X2APIC))
717 			return true;
718 		if (irq->dest_id == X2APIC_BROADCAST)
719 			return true;
720 	} else {
721 		bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
722 		if (irq->dest_id == (x2apic_ipi ?
723 		                     X2APIC_BROADCAST : APIC_BROADCAST))
724 			return true;
725 	}
726 
727 	return false;
728 }
729 
730 /* Return true if the interrupt can be handled by using *bitmap as index mask
731  * for valid destinations in *dst array.
732  * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
733  * Note: we may have zero kvm_lapic destinations when we return true, which
734  * means that the interrupt should be dropped.  In this case, *bitmap would be
735  * zero and *dst undefined.
736  */
737 static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
738 		struct kvm_lapic **src, struct kvm_lapic_irq *irq,
739 		struct kvm_apic_map *map, struct kvm_lapic ***dst,
740 		unsigned long *bitmap)
741 {
742 	int i, lowest;
743 
744 	if (irq->shorthand == APIC_DEST_SELF && src) {
745 		*dst = src;
746 		*bitmap = 1;
747 		return true;
748 	} else if (irq->shorthand)
749 		return false;
750 
751 	if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
752 		return false;
753 
754 	if (irq->dest_mode == APIC_DEST_PHYSICAL) {
755 		if (irq->dest_id > map->max_apic_id) {
756 			*bitmap = 0;
757 		} else {
758 			*dst = &map->phys_map[irq->dest_id];
759 			*bitmap = 1;
760 		}
761 		return true;
762 	}
763 
764 	*bitmap = 0;
765 	if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
766 				(u16 *)bitmap))
767 		return false;
768 
769 	if (!kvm_lowest_prio_delivery(irq))
770 		return true;
771 
772 	if (!kvm_vector_hashing_enabled()) {
773 		lowest = -1;
774 		for_each_set_bit(i, bitmap, 16) {
775 			if (!(*dst)[i])
776 				continue;
777 			if (lowest < 0)
778 				lowest = i;
779 			else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
780 						(*dst)[lowest]->vcpu) < 0)
781 				lowest = i;
782 		}
783 	} else {
784 		if (!*bitmap)
785 			return true;
786 
787 		lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
788 				bitmap, 16);
789 
790 		if (!(*dst)[lowest]) {
791 			kvm_apic_disabled_lapic_found(kvm);
792 			*bitmap = 0;
793 			return true;
794 		}
795 	}
796 
797 	*bitmap = (lowest >= 0) ? 1 << lowest : 0;
798 
799 	return true;
800 }
801 
802 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
803 		struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
804 {
805 	struct kvm_apic_map *map;
806 	unsigned long bitmap;
807 	struct kvm_lapic **dst = NULL;
808 	int i;
809 	bool ret;
810 
811 	*r = -1;
812 
813 	if (irq->shorthand == APIC_DEST_SELF) {
814 		*r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
815 		return true;
816 	}
817 
818 	rcu_read_lock();
819 	map = rcu_dereference(kvm->arch.apic_map);
820 
821 	ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
822 	if (ret)
823 		for_each_set_bit(i, &bitmap, 16) {
824 			if (!dst[i])
825 				continue;
826 			if (*r < 0)
827 				*r = 0;
828 			*r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
829 		}
830 
831 	rcu_read_unlock();
832 	return ret;
833 }
834 
835 /*
836  * This routine tries to handler interrupts in posted mode, here is how
837  * it deals with different cases:
838  * - For single-destination interrupts, handle it in posted mode
839  * - Else if vector hashing is enabled and it is a lowest-priority
840  *   interrupt, handle it in posted mode and use the following mechanism
841  *   to find the destinaiton vCPU.
842  *	1. For lowest-priority interrupts, store all the possible
843  *	   destination vCPUs in an array.
844  *	2. Use "guest vector % max number of destination vCPUs" to find
845  *	   the right destination vCPU in the array for the lowest-priority
846  *	   interrupt.
847  * - Otherwise, use remapped mode to inject the interrupt.
848  */
849 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
850 			struct kvm_vcpu **dest_vcpu)
851 {
852 	struct kvm_apic_map *map;
853 	unsigned long bitmap;
854 	struct kvm_lapic **dst = NULL;
855 	bool ret = false;
856 
857 	if (irq->shorthand)
858 		return false;
859 
860 	rcu_read_lock();
861 	map = rcu_dereference(kvm->arch.apic_map);
862 
863 	if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
864 			hweight16(bitmap) == 1) {
865 		unsigned long i = find_first_bit(&bitmap, 16);
866 
867 		if (dst[i]) {
868 			*dest_vcpu = dst[i]->vcpu;
869 			ret = true;
870 		}
871 	}
872 
873 	rcu_read_unlock();
874 	return ret;
875 }
876 
877 /*
878  * Add a pending IRQ into lapic.
879  * Return 1 if successfully added and 0 if discarded.
880  */
881 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
882 			     int vector, int level, int trig_mode,
883 			     struct dest_map *dest_map)
884 {
885 	int result = 0;
886 	struct kvm_vcpu *vcpu = apic->vcpu;
887 
888 	trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
889 				  trig_mode, vector);
890 	switch (delivery_mode) {
891 	case APIC_DM_LOWEST:
892 		vcpu->arch.apic_arb_prio++;
893 	case APIC_DM_FIXED:
894 		if (unlikely(trig_mode && !level))
895 			break;
896 
897 		/* FIXME add logic for vcpu on reset */
898 		if (unlikely(!apic_enabled(apic)))
899 			break;
900 
901 		result = 1;
902 
903 		if (dest_map) {
904 			__set_bit(vcpu->vcpu_id, dest_map->map);
905 			dest_map->vectors[vcpu->vcpu_id] = vector;
906 		}
907 
908 		if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
909 			if (trig_mode)
910 				kvm_lapic_set_vector(vector, apic->regs + APIC_TMR);
911 			else
912 				apic_clear_vector(vector, apic->regs + APIC_TMR);
913 		}
914 
915 		if (vcpu->arch.apicv_active)
916 			kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
917 		else {
918 			kvm_lapic_set_irr(vector, apic);
919 
920 			kvm_make_request(KVM_REQ_EVENT, vcpu);
921 			kvm_vcpu_kick(vcpu);
922 		}
923 		break;
924 
925 	case APIC_DM_REMRD:
926 		result = 1;
927 		vcpu->arch.pv.pv_unhalted = 1;
928 		kvm_make_request(KVM_REQ_EVENT, vcpu);
929 		kvm_vcpu_kick(vcpu);
930 		break;
931 
932 	case APIC_DM_SMI:
933 		result = 1;
934 		kvm_make_request(KVM_REQ_SMI, vcpu);
935 		kvm_vcpu_kick(vcpu);
936 		break;
937 
938 	case APIC_DM_NMI:
939 		result = 1;
940 		kvm_inject_nmi(vcpu);
941 		kvm_vcpu_kick(vcpu);
942 		break;
943 
944 	case APIC_DM_INIT:
945 		if (!trig_mode || level) {
946 			result = 1;
947 			/* assumes that there are only KVM_APIC_INIT/SIPI */
948 			apic->pending_events = (1UL << KVM_APIC_INIT);
949 			/* make sure pending_events is visible before sending
950 			 * the request */
951 			smp_wmb();
952 			kvm_make_request(KVM_REQ_EVENT, vcpu);
953 			kvm_vcpu_kick(vcpu);
954 		} else {
955 			apic_debug("Ignoring de-assert INIT to vcpu %d\n",
956 				   vcpu->vcpu_id);
957 		}
958 		break;
959 
960 	case APIC_DM_STARTUP:
961 		apic_debug("SIPI to vcpu %d vector 0x%02x\n",
962 			   vcpu->vcpu_id, vector);
963 		result = 1;
964 		apic->sipi_vector = vector;
965 		/* make sure sipi_vector is visible for the receiver */
966 		smp_wmb();
967 		set_bit(KVM_APIC_SIPI, &apic->pending_events);
968 		kvm_make_request(KVM_REQ_EVENT, vcpu);
969 		kvm_vcpu_kick(vcpu);
970 		break;
971 
972 	case APIC_DM_EXTINT:
973 		/*
974 		 * Should only be called by kvm_apic_local_deliver() with LVT0,
975 		 * before NMI watchdog was enabled. Already handled by
976 		 * kvm_apic_accept_pic_intr().
977 		 */
978 		break;
979 
980 	default:
981 		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
982 		       delivery_mode);
983 		break;
984 	}
985 	return result;
986 }
987 
988 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
989 {
990 	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
991 }
992 
993 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
994 {
995 	return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
996 }
997 
998 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
999 {
1000 	int trigger_mode;
1001 
1002 	/* Eoi the ioapic only if the ioapic doesn't own the vector. */
1003 	if (!kvm_ioapic_handles_vector(apic, vector))
1004 		return;
1005 
1006 	/* Request a KVM exit to inform the userspace IOAPIC. */
1007 	if (irqchip_split(apic->vcpu->kvm)) {
1008 		apic->vcpu->arch.pending_ioapic_eoi = vector;
1009 		kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1010 		return;
1011 	}
1012 
1013 	if (apic_test_vector(vector, apic->regs + APIC_TMR))
1014 		trigger_mode = IOAPIC_LEVEL_TRIG;
1015 	else
1016 		trigger_mode = IOAPIC_EDGE_TRIG;
1017 
1018 	kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1019 }
1020 
1021 static int apic_set_eoi(struct kvm_lapic *apic)
1022 {
1023 	int vector = apic_find_highest_isr(apic);
1024 
1025 	trace_kvm_eoi(apic, vector);
1026 
1027 	/*
1028 	 * Not every write EOI will has corresponding ISR,
1029 	 * one example is when Kernel check timer on setup_IO_APIC
1030 	 */
1031 	if (vector == -1)
1032 		return vector;
1033 
1034 	apic_clear_isr(vector, apic);
1035 	apic_update_ppr(apic);
1036 
1037 	if (test_bit(vector, vcpu_to_synic(apic->vcpu)->vec_bitmap))
1038 		kvm_hv_synic_send_eoi(apic->vcpu, vector);
1039 
1040 	kvm_ioapic_send_eoi(apic, vector);
1041 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1042 	return vector;
1043 }
1044 
1045 /*
1046  * this interface assumes a trap-like exit, which has already finished
1047  * desired side effect including vISR and vPPR update.
1048  */
1049 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1050 {
1051 	struct kvm_lapic *apic = vcpu->arch.apic;
1052 
1053 	trace_kvm_eoi(apic, vector);
1054 
1055 	kvm_ioapic_send_eoi(apic, vector);
1056 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1057 }
1058 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1059 
1060 static void apic_send_ipi(struct kvm_lapic *apic)
1061 {
1062 	u32 icr_low = kvm_lapic_get_reg(apic, APIC_ICR);
1063 	u32 icr_high = kvm_lapic_get_reg(apic, APIC_ICR2);
1064 	struct kvm_lapic_irq irq;
1065 
1066 	irq.vector = icr_low & APIC_VECTOR_MASK;
1067 	irq.delivery_mode = icr_low & APIC_MODE_MASK;
1068 	irq.dest_mode = icr_low & APIC_DEST_MASK;
1069 	irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1070 	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1071 	irq.shorthand = icr_low & APIC_SHORT_MASK;
1072 	irq.msi_redir_hint = false;
1073 	if (apic_x2apic_mode(apic))
1074 		irq.dest_id = icr_high;
1075 	else
1076 		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1077 
1078 	trace_kvm_apic_ipi(icr_low, irq.dest_id);
1079 
1080 	apic_debug("icr_high 0x%x, icr_low 0x%x, "
1081 		   "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
1082 		   "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x, "
1083 		   "msi_redir_hint 0x%x\n",
1084 		   icr_high, icr_low, irq.shorthand, irq.dest_id,
1085 		   irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
1086 		   irq.vector, irq.msi_redir_hint);
1087 
1088 	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1089 }
1090 
1091 static u32 apic_get_tmcct(struct kvm_lapic *apic)
1092 {
1093 	ktime_t remaining;
1094 	s64 ns;
1095 	u32 tmcct;
1096 
1097 	ASSERT(apic != NULL);
1098 
1099 	/* if initial count is 0, current count should also be 0 */
1100 	if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1101 		apic->lapic_timer.period == 0)
1102 		return 0;
1103 
1104 	remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
1105 	if (ktime_to_ns(remaining) < 0)
1106 		remaining = ktime_set(0, 0);
1107 
1108 	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1109 	tmcct = div64_u64(ns,
1110 			 (APIC_BUS_CYCLE_NS * apic->divide_count));
1111 
1112 	return tmcct;
1113 }
1114 
1115 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1116 {
1117 	struct kvm_vcpu *vcpu = apic->vcpu;
1118 	struct kvm_run *run = vcpu->run;
1119 
1120 	kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1121 	run->tpr_access.rip = kvm_rip_read(vcpu);
1122 	run->tpr_access.is_write = write;
1123 }
1124 
1125 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1126 {
1127 	if (apic->vcpu->arch.tpr_access_reporting)
1128 		__report_tpr_access(apic, write);
1129 }
1130 
1131 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1132 {
1133 	u32 val = 0;
1134 
1135 	if (offset >= LAPIC_MMIO_LENGTH)
1136 		return 0;
1137 
1138 	switch (offset) {
1139 	case APIC_ARBPRI:
1140 		apic_debug("Access APIC ARBPRI register which is for P6\n");
1141 		break;
1142 
1143 	case APIC_TMCCT:	/* Timer CCR */
1144 		if (apic_lvtt_tscdeadline(apic))
1145 			return 0;
1146 
1147 		val = apic_get_tmcct(apic);
1148 		break;
1149 	case APIC_PROCPRI:
1150 		apic_update_ppr(apic);
1151 		val = kvm_lapic_get_reg(apic, offset);
1152 		break;
1153 	case APIC_TASKPRI:
1154 		report_tpr_access(apic, false);
1155 		/* fall thru */
1156 	default:
1157 		val = kvm_lapic_get_reg(apic, offset);
1158 		break;
1159 	}
1160 
1161 	return val;
1162 }
1163 
1164 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1165 {
1166 	return container_of(dev, struct kvm_lapic, dev);
1167 }
1168 
1169 int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1170 		void *data)
1171 {
1172 	unsigned char alignment = offset & 0xf;
1173 	u32 result;
1174 	/* this bitmask has a bit cleared for each reserved register */
1175 	static const u64 rmask = 0x43ff01ffffffe70cULL;
1176 
1177 	if ((alignment + len) > 4) {
1178 		apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1179 			   offset, len);
1180 		return 1;
1181 	}
1182 
1183 	if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1184 		apic_debug("KVM_APIC_READ: read reserved register %x\n",
1185 			   offset);
1186 		return 1;
1187 	}
1188 
1189 	result = __apic_read(apic, offset & ~0xf);
1190 
1191 	trace_kvm_apic_read(offset, result);
1192 
1193 	switch (len) {
1194 	case 1:
1195 	case 2:
1196 	case 4:
1197 		memcpy(data, (char *)&result + alignment, len);
1198 		break;
1199 	default:
1200 		printk(KERN_ERR "Local APIC read with len = %x, "
1201 		       "should be 1,2, or 4 instead\n", len);
1202 		break;
1203 	}
1204 	return 0;
1205 }
1206 EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1207 
1208 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1209 {
1210 	return kvm_apic_hw_enabled(apic) &&
1211 	    addr >= apic->base_address &&
1212 	    addr < apic->base_address + LAPIC_MMIO_LENGTH;
1213 }
1214 
1215 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1216 			   gpa_t address, int len, void *data)
1217 {
1218 	struct kvm_lapic *apic = to_lapic(this);
1219 	u32 offset = address - apic->base_address;
1220 
1221 	if (!apic_mmio_in_range(apic, address))
1222 		return -EOPNOTSUPP;
1223 
1224 	kvm_lapic_reg_read(apic, offset, len, data);
1225 
1226 	return 0;
1227 }
1228 
1229 static void update_divide_count(struct kvm_lapic *apic)
1230 {
1231 	u32 tmp1, tmp2, tdcr;
1232 
1233 	tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1234 	tmp1 = tdcr & 0xf;
1235 	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1236 	apic->divide_count = 0x1 << (tmp2 & 0x7);
1237 
1238 	apic_debug("timer divide count is 0x%x\n",
1239 				   apic->divide_count);
1240 }
1241 
1242 static void apic_update_lvtt(struct kvm_lapic *apic)
1243 {
1244 	u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1245 			apic->lapic_timer.timer_mode_mask;
1246 
1247 	if (apic->lapic_timer.timer_mode != timer_mode) {
1248 		apic->lapic_timer.timer_mode = timer_mode;
1249 		hrtimer_cancel(&apic->lapic_timer.timer);
1250 	}
1251 }
1252 
1253 static void apic_timer_expired(struct kvm_lapic *apic)
1254 {
1255 	struct kvm_vcpu *vcpu = apic->vcpu;
1256 	struct swait_queue_head *q = &vcpu->wq;
1257 	struct kvm_timer *ktimer = &apic->lapic_timer;
1258 
1259 	if (atomic_read(&apic->lapic_timer.pending))
1260 		return;
1261 
1262 	atomic_inc(&apic->lapic_timer.pending);
1263 	kvm_set_pending_timer(vcpu);
1264 
1265 	if (swait_active(q))
1266 		swake_up(q);
1267 
1268 	if (apic_lvtt_tscdeadline(apic))
1269 		ktimer->expired_tscdeadline = ktimer->tscdeadline;
1270 }
1271 
1272 /*
1273  * On APICv, this test will cause a busy wait
1274  * during a higher-priority task.
1275  */
1276 
1277 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1278 {
1279 	struct kvm_lapic *apic = vcpu->arch.apic;
1280 	u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1281 
1282 	if (kvm_apic_hw_enabled(apic)) {
1283 		int vec = reg & APIC_VECTOR_MASK;
1284 		void *bitmap = apic->regs + APIC_ISR;
1285 
1286 		if (vcpu->arch.apicv_active)
1287 			bitmap = apic->regs + APIC_IRR;
1288 
1289 		if (apic_test_vector(vec, bitmap))
1290 			return true;
1291 	}
1292 	return false;
1293 }
1294 
1295 void wait_lapic_expire(struct kvm_vcpu *vcpu)
1296 {
1297 	struct kvm_lapic *apic = vcpu->arch.apic;
1298 	u64 guest_tsc, tsc_deadline;
1299 
1300 	if (!lapic_in_kernel(vcpu))
1301 		return;
1302 
1303 	if (apic->lapic_timer.expired_tscdeadline == 0)
1304 		return;
1305 
1306 	if (!lapic_timer_int_injected(vcpu))
1307 		return;
1308 
1309 	tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1310 	apic->lapic_timer.expired_tscdeadline = 0;
1311 	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1312 	trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1313 
1314 	/* __delay is delay_tsc whenever the hardware has TSC, thus always.  */
1315 	if (guest_tsc < tsc_deadline)
1316 		__delay(min(tsc_deadline - guest_tsc,
1317 			nsec_to_cycles(vcpu, lapic_timer_advance_ns)));
1318 }
1319 
1320 static void start_sw_tscdeadline(struct kvm_lapic *apic)
1321 {
1322 	u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1323 	u64 ns = 0;
1324 	ktime_t expire;
1325 	struct kvm_vcpu *vcpu = apic->vcpu;
1326 	unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1327 	unsigned long flags;
1328 	ktime_t now;
1329 
1330 	if (unlikely(!tscdeadline || !this_tsc_khz))
1331 		return;
1332 
1333 	local_irq_save(flags);
1334 
1335 	now = apic->lapic_timer.timer.base->get_time();
1336 	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1337 	if (likely(tscdeadline > guest_tsc)) {
1338 		ns = (tscdeadline - guest_tsc) * 1000000ULL;
1339 		do_div(ns, this_tsc_khz);
1340 		expire = ktime_add_ns(now, ns);
1341 		expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1342 		hrtimer_start(&apic->lapic_timer.timer,
1343 				expire, HRTIMER_MODE_ABS_PINNED);
1344 	} else
1345 		apic_timer_expired(apic);
1346 
1347 	local_irq_restore(flags);
1348 }
1349 
1350 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1351 {
1352 	if (!lapic_in_kernel(vcpu))
1353 		return false;
1354 
1355 	return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1356 }
1357 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1358 
1359 static void cancel_hv_tscdeadline(struct kvm_lapic *apic)
1360 {
1361 	kvm_x86_ops->cancel_hv_timer(apic->vcpu);
1362 	apic->lapic_timer.hv_timer_in_use = false;
1363 }
1364 
1365 void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1366 {
1367 	struct kvm_lapic *apic = vcpu->arch.apic;
1368 
1369 	WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1370 	WARN_ON(swait_active(&vcpu->wq));
1371 	cancel_hv_tscdeadline(apic);
1372 	apic_timer_expired(apic);
1373 }
1374 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1375 
1376 static bool start_hv_tscdeadline(struct kvm_lapic *apic)
1377 {
1378 	u64 tscdeadline = apic->lapic_timer.tscdeadline;
1379 
1380 	if (atomic_read(&apic->lapic_timer.pending) ||
1381 		kvm_x86_ops->set_hv_timer(apic->vcpu, tscdeadline)) {
1382 		if (apic->lapic_timer.hv_timer_in_use)
1383 			cancel_hv_tscdeadline(apic);
1384 	} else {
1385 		apic->lapic_timer.hv_timer_in_use = true;
1386 		hrtimer_cancel(&apic->lapic_timer.timer);
1387 
1388 		/* In case the sw timer triggered in the window */
1389 		if (atomic_read(&apic->lapic_timer.pending))
1390 			cancel_hv_tscdeadline(apic);
1391 	}
1392 	trace_kvm_hv_timer_state(apic->vcpu->vcpu_id,
1393 			apic->lapic_timer.hv_timer_in_use);
1394 	return apic->lapic_timer.hv_timer_in_use;
1395 }
1396 
1397 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1398 {
1399 	struct kvm_lapic *apic = vcpu->arch.apic;
1400 
1401 	WARN_ON(apic->lapic_timer.hv_timer_in_use);
1402 
1403 	if (apic_lvtt_tscdeadline(apic))
1404 		start_hv_tscdeadline(apic);
1405 }
1406 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1407 
1408 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1409 {
1410 	struct kvm_lapic *apic = vcpu->arch.apic;
1411 
1412 	/* Possibly the TSC deadline timer is not enabled yet */
1413 	if (!apic->lapic_timer.hv_timer_in_use)
1414 		return;
1415 
1416 	cancel_hv_tscdeadline(apic);
1417 
1418 	if (atomic_read(&apic->lapic_timer.pending))
1419 		return;
1420 
1421 	start_sw_tscdeadline(apic);
1422 }
1423 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1424 
1425 static void start_apic_timer(struct kvm_lapic *apic)
1426 {
1427 	ktime_t now;
1428 
1429 	atomic_set(&apic->lapic_timer.pending, 0);
1430 
1431 	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1432 		/* lapic timer in oneshot or periodic mode */
1433 		now = apic->lapic_timer.timer.base->get_time();
1434 		apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1435 			    * APIC_BUS_CYCLE_NS * apic->divide_count;
1436 
1437 		if (!apic->lapic_timer.period)
1438 			return;
1439 		/*
1440 		 * Do not allow the guest to program periodic timers with small
1441 		 * interval, since the hrtimers are not throttled by the host
1442 		 * scheduler.
1443 		 */
1444 		if (apic_lvtt_period(apic)) {
1445 			s64 min_period = min_timer_period_us * 1000LL;
1446 
1447 			if (apic->lapic_timer.period < min_period) {
1448 				pr_info_ratelimited(
1449 				    "kvm: vcpu %i: requested %lld ns "
1450 				    "lapic timer period limited to %lld ns\n",
1451 				    apic->vcpu->vcpu_id,
1452 				    apic->lapic_timer.period, min_period);
1453 				apic->lapic_timer.period = min_period;
1454 			}
1455 		}
1456 
1457 		hrtimer_start(&apic->lapic_timer.timer,
1458 			      ktime_add_ns(now, apic->lapic_timer.period),
1459 			      HRTIMER_MODE_ABS_PINNED);
1460 
1461 		apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1462 			   PRIx64 ", "
1463 			   "timer initial count 0x%x, period %lldns, "
1464 			   "expire @ 0x%016" PRIx64 ".\n", __func__,
1465 			   APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1466 			   kvm_lapic_get_reg(apic, APIC_TMICT),
1467 			   apic->lapic_timer.period,
1468 			   ktime_to_ns(ktime_add_ns(now,
1469 					apic->lapic_timer.period)));
1470 	} else if (apic_lvtt_tscdeadline(apic)) {
1471 		if (!(kvm_x86_ops->set_hv_timer && start_hv_tscdeadline(apic)))
1472 			start_sw_tscdeadline(apic);
1473 	}
1474 }
1475 
1476 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1477 {
1478 	bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1479 
1480 	if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1481 		apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1482 		if (lvt0_in_nmi_mode) {
1483 			apic_debug("Receive NMI setting on APIC_LVT0 "
1484 				   "for cpu %d\n", apic->vcpu->vcpu_id);
1485 			atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1486 		} else
1487 			atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1488 	}
1489 }
1490 
1491 int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1492 {
1493 	int ret = 0;
1494 
1495 	trace_kvm_apic_write(reg, val);
1496 
1497 	switch (reg) {
1498 	case APIC_ID:		/* Local APIC ID */
1499 		if (!apic_x2apic_mode(apic))
1500 			kvm_apic_set_xapic_id(apic, val >> 24);
1501 		else
1502 			ret = 1;
1503 		break;
1504 
1505 	case APIC_TASKPRI:
1506 		report_tpr_access(apic, true);
1507 		apic_set_tpr(apic, val & 0xff);
1508 		break;
1509 
1510 	case APIC_EOI:
1511 		apic_set_eoi(apic);
1512 		break;
1513 
1514 	case APIC_LDR:
1515 		if (!apic_x2apic_mode(apic))
1516 			kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1517 		else
1518 			ret = 1;
1519 		break;
1520 
1521 	case APIC_DFR:
1522 		if (!apic_x2apic_mode(apic)) {
1523 			kvm_lapic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1524 			recalculate_apic_map(apic->vcpu->kvm);
1525 		} else
1526 			ret = 1;
1527 		break;
1528 
1529 	case APIC_SPIV: {
1530 		u32 mask = 0x3ff;
1531 		if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1532 			mask |= APIC_SPIV_DIRECTED_EOI;
1533 		apic_set_spiv(apic, val & mask);
1534 		if (!(val & APIC_SPIV_APIC_ENABLED)) {
1535 			int i;
1536 			u32 lvt_val;
1537 
1538 			for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
1539 				lvt_val = kvm_lapic_get_reg(apic,
1540 						       APIC_LVTT + 0x10 * i);
1541 				kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
1542 					     lvt_val | APIC_LVT_MASKED);
1543 			}
1544 			apic_update_lvtt(apic);
1545 			atomic_set(&apic->lapic_timer.pending, 0);
1546 
1547 		}
1548 		break;
1549 	}
1550 	case APIC_ICR:
1551 		/* No delay here, so we always clear the pending bit */
1552 		kvm_lapic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1553 		apic_send_ipi(apic);
1554 		break;
1555 
1556 	case APIC_ICR2:
1557 		if (!apic_x2apic_mode(apic))
1558 			val &= 0xff000000;
1559 		kvm_lapic_set_reg(apic, APIC_ICR2, val);
1560 		break;
1561 
1562 	case APIC_LVT0:
1563 		apic_manage_nmi_watchdog(apic, val);
1564 	case APIC_LVTTHMR:
1565 	case APIC_LVTPC:
1566 	case APIC_LVT1:
1567 	case APIC_LVTERR:
1568 		/* TODO: Check vector */
1569 		if (!kvm_apic_sw_enabled(apic))
1570 			val |= APIC_LVT_MASKED;
1571 
1572 		val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1573 		kvm_lapic_set_reg(apic, reg, val);
1574 
1575 		break;
1576 
1577 	case APIC_LVTT:
1578 		if (!kvm_apic_sw_enabled(apic))
1579 			val |= APIC_LVT_MASKED;
1580 		val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1581 		kvm_lapic_set_reg(apic, APIC_LVTT, val);
1582 		apic_update_lvtt(apic);
1583 		break;
1584 
1585 	case APIC_TMICT:
1586 		if (apic_lvtt_tscdeadline(apic))
1587 			break;
1588 
1589 		hrtimer_cancel(&apic->lapic_timer.timer);
1590 		kvm_lapic_set_reg(apic, APIC_TMICT, val);
1591 		start_apic_timer(apic);
1592 		break;
1593 
1594 	case APIC_TDCR:
1595 		if (val & 4)
1596 			apic_debug("KVM_WRITE:TDCR %x\n", val);
1597 		kvm_lapic_set_reg(apic, APIC_TDCR, val);
1598 		update_divide_count(apic);
1599 		break;
1600 
1601 	case APIC_ESR:
1602 		if (apic_x2apic_mode(apic) && val != 0) {
1603 			apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1604 			ret = 1;
1605 		}
1606 		break;
1607 
1608 	case APIC_SELF_IPI:
1609 		if (apic_x2apic_mode(apic)) {
1610 			kvm_lapic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1611 		} else
1612 			ret = 1;
1613 		break;
1614 	default:
1615 		ret = 1;
1616 		break;
1617 	}
1618 	if (ret)
1619 		apic_debug("Local APIC Write to read-only register %x\n", reg);
1620 	return ret;
1621 }
1622 EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
1623 
1624 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1625 			    gpa_t address, int len, const void *data)
1626 {
1627 	struct kvm_lapic *apic = to_lapic(this);
1628 	unsigned int offset = address - apic->base_address;
1629 	u32 val;
1630 
1631 	if (!apic_mmio_in_range(apic, address))
1632 		return -EOPNOTSUPP;
1633 
1634 	/*
1635 	 * APIC register must be aligned on 128-bits boundary.
1636 	 * 32/64/128 bits registers must be accessed thru 32 bits.
1637 	 * Refer SDM 8.4.1
1638 	 */
1639 	if (len != 4 || (offset & 0xf)) {
1640 		/* Don't shout loud, $infamous_os would cause only noise. */
1641 		apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1642 		return 0;
1643 	}
1644 
1645 	val = *(u32*)data;
1646 
1647 	/* too common printing */
1648 	if (offset != APIC_EOI)
1649 		apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1650 			   "0x%x\n", __func__, offset, len, val);
1651 
1652 	kvm_lapic_reg_write(apic, offset & 0xff0, val);
1653 
1654 	return 0;
1655 }
1656 
1657 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1658 {
1659 	kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1660 }
1661 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1662 
1663 /* emulate APIC access in a trap manner */
1664 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1665 {
1666 	u32 val = 0;
1667 
1668 	/* hw has done the conditional check and inst decode */
1669 	offset &= 0xff0;
1670 
1671 	kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
1672 
1673 	/* TODO: optimize to just emulate side effect w/o one more write */
1674 	kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
1675 }
1676 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1677 
1678 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1679 {
1680 	struct kvm_lapic *apic = vcpu->arch.apic;
1681 
1682 	if (!vcpu->arch.apic)
1683 		return;
1684 
1685 	hrtimer_cancel(&apic->lapic_timer.timer);
1686 
1687 	if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1688 		static_key_slow_dec_deferred(&apic_hw_disabled);
1689 
1690 	if (!apic->sw_enabled)
1691 		static_key_slow_dec_deferred(&apic_sw_disabled);
1692 
1693 	if (apic->regs)
1694 		free_page((unsigned long)apic->regs);
1695 
1696 	kfree(apic);
1697 }
1698 
1699 /*
1700  *----------------------------------------------------------------------
1701  * LAPIC interface
1702  *----------------------------------------------------------------------
1703  */
1704 
1705 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1706 {
1707 	struct kvm_lapic *apic = vcpu->arch.apic;
1708 
1709 	if (!lapic_in_kernel(vcpu) || apic_lvtt_oneshot(apic) ||
1710 			apic_lvtt_period(apic))
1711 		return 0;
1712 
1713 	return apic->lapic_timer.tscdeadline;
1714 }
1715 
1716 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1717 {
1718 	struct kvm_lapic *apic = vcpu->arch.apic;
1719 
1720 	if (!lapic_in_kernel(vcpu) || apic_lvtt_oneshot(apic) ||
1721 			apic_lvtt_period(apic))
1722 		return;
1723 
1724 	hrtimer_cancel(&apic->lapic_timer.timer);
1725 	apic->lapic_timer.tscdeadline = data;
1726 	start_apic_timer(apic);
1727 }
1728 
1729 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1730 {
1731 	struct kvm_lapic *apic = vcpu->arch.apic;
1732 
1733 	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1734 		     | (kvm_lapic_get_reg(apic, APIC_TASKPRI) & 4));
1735 }
1736 
1737 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1738 {
1739 	u64 tpr;
1740 
1741 	tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1742 
1743 	return (tpr & 0xf0) >> 4;
1744 }
1745 
1746 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1747 {
1748 	u64 old_value = vcpu->arch.apic_base;
1749 	struct kvm_lapic *apic = vcpu->arch.apic;
1750 
1751 	if (!apic) {
1752 		value |= MSR_IA32_APICBASE_BSP;
1753 		vcpu->arch.apic_base = value;
1754 		return;
1755 	}
1756 
1757 	vcpu->arch.apic_base = value;
1758 
1759 	/* update jump label if enable bit changes */
1760 	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1761 		if (value & MSR_IA32_APICBASE_ENABLE) {
1762 			kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
1763 			static_key_slow_dec_deferred(&apic_hw_disabled);
1764 		} else
1765 			static_key_slow_inc(&apic_hw_disabled.key);
1766 		recalculate_apic_map(vcpu->kvm);
1767 	}
1768 
1769 	if ((old_value ^ value) & X2APIC_ENABLE) {
1770 		if (value & X2APIC_ENABLE) {
1771 			kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
1772 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1773 		} else
1774 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1775 	}
1776 
1777 	apic->base_address = apic->vcpu->arch.apic_base &
1778 			     MSR_IA32_APICBASE_BASE;
1779 
1780 	if ((value & MSR_IA32_APICBASE_ENABLE) &&
1781 	     apic->base_address != APIC_DEFAULT_PHYS_BASE)
1782 		pr_warn_once("APIC base relocation is unsupported by KVM");
1783 
1784 	/* with FSB delivery interrupt, we can restart APIC functionality */
1785 	apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1786 		   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1787 
1788 }
1789 
1790 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
1791 {
1792 	struct kvm_lapic *apic;
1793 	int i;
1794 
1795 	apic_debug("%s\n", __func__);
1796 
1797 	ASSERT(vcpu);
1798 	apic = vcpu->arch.apic;
1799 	ASSERT(apic != NULL);
1800 
1801 	/* Stop the timer in case it's a reset to an active apic */
1802 	hrtimer_cancel(&apic->lapic_timer.timer);
1803 
1804 	if (!init_event) {
1805 		kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
1806 		                         MSR_IA32_APICBASE_ENABLE);
1807 		kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
1808 	}
1809 	kvm_apic_set_version(apic->vcpu);
1810 
1811 	for (i = 0; i < KVM_APIC_LVT_NUM; i++)
1812 		kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1813 	apic_update_lvtt(apic);
1814 	if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
1815 		kvm_lapic_set_reg(apic, APIC_LVT0,
1816 			     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1817 	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
1818 
1819 	kvm_lapic_set_reg(apic, APIC_DFR, 0xffffffffU);
1820 	apic_set_spiv(apic, 0xff);
1821 	kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
1822 	if (!apic_x2apic_mode(apic))
1823 		kvm_apic_set_ldr(apic, 0);
1824 	kvm_lapic_set_reg(apic, APIC_ESR, 0);
1825 	kvm_lapic_set_reg(apic, APIC_ICR, 0);
1826 	kvm_lapic_set_reg(apic, APIC_ICR2, 0);
1827 	kvm_lapic_set_reg(apic, APIC_TDCR, 0);
1828 	kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1829 	for (i = 0; i < 8; i++) {
1830 		kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1831 		kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1832 		kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1833 	}
1834 	apic->irr_pending = vcpu->arch.apicv_active;
1835 	apic->isr_count = vcpu->arch.apicv_active ? 1 : 0;
1836 	apic->highest_isr_cache = -1;
1837 	update_divide_count(apic);
1838 	atomic_set(&apic->lapic_timer.pending, 0);
1839 	if (kvm_vcpu_is_bsp(vcpu))
1840 		kvm_lapic_set_base(vcpu,
1841 				vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1842 	vcpu->arch.pv_eoi.msr_val = 0;
1843 	apic_update_ppr(apic);
1844 
1845 	vcpu->arch.apic_arb_prio = 0;
1846 	vcpu->arch.apic_attention = 0;
1847 
1848 	apic_debug("%s: vcpu=%p, id=%d, base_msr="
1849 		   "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1850 		   vcpu, kvm_apic_id(apic),
1851 		   vcpu->arch.apic_base, apic->base_address);
1852 }
1853 
1854 /*
1855  *----------------------------------------------------------------------
1856  * timer interface
1857  *----------------------------------------------------------------------
1858  */
1859 
1860 static bool lapic_is_periodic(struct kvm_lapic *apic)
1861 {
1862 	return apic_lvtt_period(apic);
1863 }
1864 
1865 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1866 {
1867 	struct kvm_lapic *apic = vcpu->arch.apic;
1868 
1869 	if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
1870 		return atomic_read(&apic->lapic_timer.pending);
1871 
1872 	return 0;
1873 }
1874 
1875 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1876 {
1877 	u32 reg = kvm_lapic_get_reg(apic, lvt_type);
1878 	int vector, mode, trig_mode;
1879 
1880 	if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1881 		vector = reg & APIC_VECTOR_MASK;
1882 		mode = reg & APIC_MODE_MASK;
1883 		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1884 		return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1885 					NULL);
1886 	}
1887 	return 0;
1888 }
1889 
1890 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1891 {
1892 	struct kvm_lapic *apic = vcpu->arch.apic;
1893 
1894 	if (apic)
1895 		kvm_apic_local_deliver(apic, APIC_LVT0);
1896 }
1897 
1898 static const struct kvm_io_device_ops apic_mmio_ops = {
1899 	.read     = apic_mmio_read,
1900 	.write    = apic_mmio_write,
1901 };
1902 
1903 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1904 {
1905 	struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1906 	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1907 
1908 	apic_timer_expired(apic);
1909 
1910 	if (lapic_is_periodic(apic)) {
1911 		hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1912 		return HRTIMER_RESTART;
1913 	} else
1914 		return HRTIMER_NORESTART;
1915 }
1916 
1917 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1918 {
1919 	struct kvm_lapic *apic;
1920 
1921 	ASSERT(vcpu != NULL);
1922 	apic_debug("apic_init %d\n", vcpu->vcpu_id);
1923 
1924 	apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1925 	if (!apic)
1926 		goto nomem;
1927 
1928 	vcpu->arch.apic = apic;
1929 
1930 	apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1931 	if (!apic->regs) {
1932 		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1933 		       vcpu->vcpu_id);
1934 		goto nomem_free_apic;
1935 	}
1936 	apic->vcpu = vcpu;
1937 
1938 	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1939 		     HRTIMER_MODE_ABS_PINNED);
1940 	apic->lapic_timer.timer.function = apic_timer_fn;
1941 
1942 	/*
1943 	 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1944 	 * thinking that APIC satet has changed.
1945 	 */
1946 	vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1947 	static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1948 	kvm_lapic_reset(vcpu, false);
1949 	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1950 
1951 	return 0;
1952 nomem_free_apic:
1953 	kfree(apic);
1954 nomem:
1955 	return -ENOMEM;
1956 }
1957 
1958 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1959 {
1960 	struct kvm_lapic *apic = vcpu->arch.apic;
1961 	int highest_irr;
1962 
1963 	if (!apic_enabled(apic))
1964 		return -1;
1965 
1966 	apic_update_ppr(apic);
1967 	highest_irr = apic_find_highest_irr(apic);
1968 	if ((highest_irr == -1) ||
1969 	    ((highest_irr & 0xF0) <= kvm_lapic_get_reg(apic, APIC_PROCPRI)))
1970 		return -1;
1971 	return highest_irr;
1972 }
1973 
1974 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1975 {
1976 	u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
1977 	int r = 0;
1978 
1979 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1980 		r = 1;
1981 	if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1982 	    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1983 		r = 1;
1984 	return r;
1985 }
1986 
1987 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1988 {
1989 	struct kvm_lapic *apic = vcpu->arch.apic;
1990 
1991 	if (atomic_read(&apic->lapic_timer.pending) > 0) {
1992 		kvm_apic_local_deliver(apic, APIC_LVTT);
1993 		if (apic_lvtt_tscdeadline(apic))
1994 			apic->lapic_timer.tscdeadline = 0;
1995 		atomic_set(&apic->lapic_timer.pending, 0);
1996 	}
1997 }
1998 
1999 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2000 {
2001 	int vector = kvm_apic_has_interrupt(vcpu);
2002 	struct kvm_lapic *apic = vcpu->arch.apic;
2003 
2004 	if (vector == -1)
2005 		return -1;
2006 
2007 	/*
2008 	 * We get here even with APIC virtualization enabled, if doing
2009 	 * nested virtualization and L1 runs with the "acknowledge interrupt
2010 	 * on exit" mode.  Then we cannot inject the interrupt via RVI,
2011 	 * because the process would deliver it through the IDT.
2012 	 */
2013 
2014 	apic_set_isr(vector, apic);
2015 	apic_update_ppr(apic);
2016 	apic_clear_irr(vector, apic);
2017 
2018 	if (test_bit(vector, vcpu_to_synic(vcpu)->auto_eoi_bitmap)) {
2019 		apic_clear_isr(vector, apic);
2020 		apic_update_ppr(apic);
2021 	}
2022 
2023 	return vector;
2024 }
2025 
2026 static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2027 		struct kvm_lapic_state *s, bool set)
2028 {
2029 	if (apic_x2apic_mode(vcpu->arch.apic)) {
2030 		u32 *id = (u32 *)(s->regs + APIC_ID);
2031 
2032 		if (vcpu->kvm->arch.x2apic_format) {
2033 			if (*id != vcpu->vcpu_id)
2034 				return -EINVAL;
2035 		} else {
2036 			if (set)
2037 				*id >>= 24;
2038 			else
2039 				*id <<= 24;
2040 		}
2041 	}
2042 
2043 	return 0;
2044 }
2045 
2046 int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2047 {
2048 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2049 	return kvm_apic_state_fixup(vcpu, s, false);
2050 }
2051 
2052 int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2053 {
2054 	struct kvm_lapic *apic = vcpu->arch.apic;
2055 	int r;
2056 
2057 
2058 	kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2059 	/* set SPIV separately to get count of SW disabled APICs right */
2060 	apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2061 
2062 	r = kvm_apic_state_fixup(vcpu, s, true);
2063 	if (r)
2064 		return r;
2065 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2066 
2067 	recalculate_apic_map(vcpu->kvm);
2068 	kvm_apic_set_version(vcpu);
2069 
2070 	apic_update_ppr(apic);
2071 	hrtimer_cancel(&apic->lapic_timer.timer);
2072 	apic_update_lvtt(apic);
2073 	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2074 	update_divide_count(apic);
2075 	start_apic_timer(apic);
2076 	apic->irr_pending = true;
2077 	apic->isr_count = vcpu->arch.apicv_active ?
2078 				1 : count_vectors(apic->regs + APIC_ISR);
2079 	apic->highest_isr_cache = -1;
2080 	if (vcpu->arch.apicv_active) {
2081 		if (kvm_x86_ops->apicv_post_state_restore)
2082 			kvm_x86_ops->apicv_post_state_restore(vcpu);
2083 		kvm_x86_ops->hwapic_irr_update(vcpu,
2084 				apic_find_highest_irr(apic));
2085 		kvm_x86_ops->hwapic_isr_update(vcpu,
2086 				apic_find_highest_isr(apic));
2087 	}
2088 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2089 	if (ioapic_in_kernel(vcpu->kvm))
2090 		kvm_rtc_eoi_tracking_restore_one(vcpu);
2091 
2092 	vcpu->arch.apic_arb_prio = 0;
2093 
2094 	return 0;
2095 }
2096 
2097 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2098 {
2099 	struct hrtimer *timer;
2100 
2101 	if (!lapic_in_kernel(vcpu))
2102 		return;
2103 
2104 	timer = &vcpu->arch.apic->lapic_timer.timer;
2105 	if (hrtimer_cancel(timer))
2106 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
2107 }
2108 
2109 /*
2110  * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2111  *
2112  * Detect whether guest triggered PV EOI since the
2113  * last entry. If yes, set EOI on guests's behalf.
2114  * Clear PV EOI in guest memory in any case.
2115  */
2116 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2117 					struct kvm_lapic *apic)
2118 {
2119 	bool pending;
2120 	int vector;
2121 	/*
2122 	 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2123 	 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2124 	 *
2125 	 * KVM_APIC_PV_EOI_PENDING is unset:
2126 	 * 	-> host disabled PV EOI.
2127 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2128 	 * 	-> host enabled PV EOI, guest did not execute EOI yet.
2129 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2130 	 * 	-> host enabled PV EOI, guest executed EOI.
2131 	 */
2132 	BUG_ON(!pv_eoi_enabled(vcpu));
2133 	pending = pv_eoi_get_pending(vcpu);
2134 	/*
2135 	 * Clear pending bit in any case: it will be set again on vmentry.
2136 	 * While this might not be ideal from performance point of view,
2137 	 * this makes sure pv eoi is only enabled when we know it's safe.
2138 	 */
2139 	pv_eoi_clr_pending(vcpu);
2140 	if (pending)
2141 		return;
2142 	vector = apic_set_eoi(apic);
2143 	trace_kvm_pv_eoi(apic, vector);
2144 }
2145 
2146 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2147 {
2148 	u32 data;
2149 
2150 	if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2151 		apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2152 
2153 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2154 		return;
2155 
2156 	if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2157 				  sizeof(u32)))
2158 		return;
2159 
2160 	apic_set_tpr(vcpu->arch.apic, data & 0xff);
2161 }
2162 
2163 /*
2164  * apic_sync_pv_eoi_to_guest - called before vmentry
2165  *
2166  * Detect whether it's safe to enable PV EOI and
2167  * if yes do so.
2168  */
2169 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2170 					struct kvm_lapic *apic)
2171 {
2172 	if (!pv_eoi_enabled(vcpu) ||
2173 	    /* IRR set or many bits in ISR: could be nested. */
2174 	    apic->irr_pending ||
2175 	    /* Cache not set: could be safe but we don't bother. */
2176 	    apic->highest_isr_cache == -1 ||
2177 	    /* Need EOI to update ioapic. */
2178 	    kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2179 		/*
2180 		 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2181 		 * so we need not do anything here.
2182 		 */
2183 		return;
2184 	}
2185 
2186 	pv_eoi_set_pending(apic->vcpu);
2187 }
2188 
2189 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2190 {
2191 	u32 data, tpr;
2192 	int max_irr, max_isr;
2193 	struct kvm_lapic *apic = vcpu->arch.apic;
2194 
2195 	apic_sync_pv_eoi_to_guest(vcpu, apic);
2196 
2197 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2198 		return;
2199 
2200 	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2201 	max_irr = apic_find_highest_irr(apic);
2202 	if (max_irr < 0)
2203 		max_irr = 0;
2204 	max_isr = apic_find_highest_isr(apic);
2205 	if (max_isr < 0)
2206 		max_isr = 0;
2207 	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2208 
2209 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2210 				sizeof(u32));
2211 }
2212 
2213 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2214 {
2215 	if (vapic_addr) {
2216 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2217 					&vcpu->arch.apic->vapic_cache,
2218 					vapic_addr, sizeof(u32)))
2219 			return -EINVAL;
2220 		__set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2221 	} else {
2222 		__clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2223 	}
2224 
2225 	vcpu->arch.apic->vapic_addr = vapic_addr;
2226 	return 0;
2227 }
2228 
2229 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2230 {
2231 	struct kvm_lapic *apic = vcpu->arch.apic;
2232 	u32 reg = (msr - APIC_BASE_MSR) << 4;
2233 
2234 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2235 		return 1;
2236 
2237 	if (reg == APIC_ICR2)
2238 		return 1;
2239 
2240 	/* if this is ICR write vector before command */
2241 	if (reg == APIC_ICR)
2242 		kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2243 	return kvm_lapic_reg_write(apic, reg, (u32)data);
2244 }
2245 
2246 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2247 {
2248 	struct kvm_lapic *apic = vcpu->arch.apic;
2249 	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2250 
2251 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2252 		return 1;
2253 
2254 	if (reg == APIC_DFR || reg == APIC_ICR2) {
2255 		apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
2256 			   reg);
2257 		return 1;
2258 	}
2259 
2260 	if (kvm_lapic_reg_read(apic, reg, 4, &low))
2261 		return 1;
2262 	if (reg == APIC_ICR)
2263 		kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2264 
2265 	*data = (((u64)high) << 32) | low;
2266 
2267 	return 0;
2268 }
2269 
2270 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2271 {
2272 	struct kvm_lapic *apic = vcpu->arch.apic;
2273 
2274 	if (!lapic_in_kernel(vcpu))
2275 		return 1;
2276 
2277 	/* if this is ICR write vector before command */
2278 	if (reg == APIC_ICR)
2279 		kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2280 	return kvm_lapic_reg_write(apic, reg, (u32)data);
2281 }
2282 
2283 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2284 {
2285 	struct kvm_lapic *apic = vcpu->arch.apic;
2286 	u32 low, high = 0;
2287 
2288 	if (!lapic_in_kernel(vcpu))
2289 		return 1;
2290 
2291 	if (kvm_lapic_reg_read(apic, reg, 4, &low))
2292 		return 1;
2293 	if (reg == APIC_ICR)
2294 		kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2295 
2296 	*data = (((u64)high) << 32) | low;
2297 
2298 	return 0;
2299 }
2300 
2301 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2302 {
2303 	u64 addr = data & ~KVM_MSR_ENABLED;
2304 	if (!IS_ALIGNED(addr, 4))
2305 		return 1;
2306 
2307 	vcpu->arch.pv_eoi.msr_val = data;
2308 	if (!pv_eoi_enabled(vcpu))
2309 		return 0;
2310 	return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
2311 					 addr, sizeof(u8));
2312 }
2313 
2314 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2315 {
2316 	struct kvm_lapic *apic = vcpu->arch.apic;
2317 	u8 sipi_vector;
2318 	unsigned long pe;
2319 
2320 	if (!lapic_in_kernel(vcpu) || !apic->pending_events)
2321 		return;
2322 
2323 	/*
2324 	 * INITs are latched while in SMM.  Because an SMM CPU cannot
2325 	 * be in KVM_MP_STATE_INIT_RECEIVED state, just eat SIPIs
2326 	 * and delay processing of INIT until the next RSM.
2327 	 */
2328 	if (is_smm(vcpu)) {
2329 		WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2330 		if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2331 			clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2332 		return;
2333 	}
2334 
2335 	pe = xchg(&apic->pending_events, 0);
2336 	if (test_bit(KVM_APIC_INIT, &pe)) {
2337 		kvm_lapic_reset(vcpu, true);
2338 		kvm_vcpu_reset(vcpu, true);
2339 		if (kvm_vcpu_is_bsp(apic->vcpu))
2340 			vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2341 		else
2342 			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2343 	}
2344 	if (test_bit(KVM_APIC_SIPI, &pe) &&
2345 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2346 		/* evaluate pending_events before reading the vector */
2347 		smp_rmb();
2348 		sipi_vector = apic->sipi_vector;
2349 		apic_debug("vcpu %d received sipi with vector # %x\n",
2350 			 vcpu->vcpu_id, sipi_vector);
2351 		kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2352 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2353 	}
2354 }
2355 
2356 void kvm_lapic_init(void)
2357 {
2358 	/* do not patch jump label more than once per second */
2359 	jump_label_rate_limit(&apic_hw_disabled, HZ);
2360 	jump_label_rate_limit(&apic_sw_disabled, HZ);
2361 }
2362