xref: /openbmc/linux/arch/x86/kvm/lapic.c (revision 7eec52db361a6ae6fbbd86c2299718586866b664)
1 
2 /*
3  * Local APIC virtualization
4  *
5  * Copyright (C) 2006 Qumranet, Inc.
6  * Copyright (C) 2007 Novell
7  * Copyright (C) 2007 Intel
8  * Copyright 2009 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Dor Laor <dor.laor@qumranet.com>
12  *   Gregory Haskins <ghaskins@novell.com>
13  *   Yaozu (Eddie) Dong <eddie.dong@intel.com>
14  *
15  * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
16  *
17  * This work is licensed under the terms of the GNU GPL, version 2.  See
18  * the COPYING file in the top-level directory.
19  */
20 
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/mm.h>
24 #include <linux/highmem.h>
25 #include <linux/smp.h>
26 #include <linux/hrtimer.h>
27 #include <linux/io.h>
28 #include <linux/module.h>
29 #include <linux/math64.h>
30 #include <linux/slab.h>
31 #include <asm/processor.h>
32 #include <asm/msr.h>
33 #include <asm/page.h>
34 #include <asm/current.h>
35 #include <asm/apicdef.h>
36 #include <asm/delay.h>
37 #include <linux/atomic.h>
38 #include <linux/jump_label.h>
39 #include "kvm_cache_regs.h"
40 #include "irq.h"
41 #include "trace.h"
42 #include "x86.h"
43 #include "cpuid.h"
44 
45 #ifndef CONFIG_X86_64
46 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
47 #else
48 #define mod_64(x, y) ((x) % (y))
49 #endif
50 
51 #define PRId64 "d"
52 #define PRIx64 "llx"
53 #define PRIu64 "u"
54 #define PRIo64 "o"
55 
56 #define APIC_BUS_CYCLE_NS 1
57 
58 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
59 #define apic_debug(fmt, arg...)
60 
61 #define APIC_LVT_NUM			6
62 /* 14 is the version for Xeon and Pentium 8.4.8*/
63 #define APIC_VERSION			(0x14UL | ((APIC_LVT_NUM - 1) << 16))
64 #define LAPIC_MMIO_LENGTH		(1 << 12)
65 /* followed define is not in apicdef.h */
66 #define APIC_SHORT_MASK			0xc0000
67 #define APIC_DEST_NOSHORT		0x0
68 #define APIC_DEST_MASK			0x800
69 #define MAX_APIC_VECTOR			256
70 #define APIC_VECTORS_PER_REG		32
71 
72 #define APIC_BROADCAST			0xFF
73 #define X2APIC_BROADCAST		0xFFFFFFFFul
74 
75 #define VEC_POS(v) ((v) & (32 - 1))
76 #define REG_POS(v) (((v) >> 5) << 4)
77 
78 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
79 {
80 	*((u32 *) (apic->regs + reg_off)) = val;
81 }
82 
83 static inline int apic_test_vector(int vec, void *bitmap)
84 {
85 	return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
86 }
87 
88 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
89 {
90 	struct kvm_lapic *apic = vcpu->arch.apic;
91 
92 	return apic_test_vector(vector, apic->regs + APIC_ISR) ||
93 		apic_test_vector(vector, apic->regs + APIC_IRR);
94 }
95 
96 static inline void apic_set_vector(int vec, void *bitmap)
97 {
98 	set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
99 }
100 
101 static inline void apic_clear_vector(int vec, void *bitmap)
102 {
103 	clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
104 }
105 
106 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
107 {
108 	return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
109 }
110 
111 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
112 {
113 	return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
114 }
115 
116 struct static_key_deferred apic_hw_disabled __read_mostly;
117 struct static_key_deferred apic_sw_disabled __read_mostly;
118 
119 static inline int apic_enabled(struct kvm_lapic *apic)
120 {
121 	return kvm_apic_sw_enabled(apic) &&	kvm_apic_hw_enabled(apic);
122 }
123 
124 #define LVT_MASK	\
125 	(APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
126 
127 #define LINT_MASK	\
128 	(LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
129 	 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
130 
131 static inline int kvm_apic_id(struct kvm_lapic *apic)
132 {
133 	return (kvm_apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
134 }
135 
136 static void recalculate_apic_map(struct kvm *kvm)
137 {
138 	struct kvm_apic_map *new, *old = NULL;
139 	struct kvm_vcpu *vcpu;
140 	int i;
141 
142 	new = kzalloc(sizeof(struct kvm_apic_map), GFP_KERNEL);
143 
144 	mutex_lock(&kvm->arch.apic_map_lock);
145 
146 	if (!new)
147 		goto out;
148 
149 	new->ldr_bits = 8;
150 	/* flat mode is default */
151 	new->cid_shift = 8;
152 	new->cid_mask = 0;
153 	new->lid_mask = 0xff;
154 	new->broadcast = APIC_BROADCAST;
155 
156 	kvm_for_each_vcpu(i, vcpu, kvm) {
157 		struct kvm_lapic *apic = vcpu->arch.apic;
158 
159 		if (!kvm_apic_present(vcpu))
160 			continue;
161 
162 		if (apic_x2apic_mode(apic)) {
163 			new->ldr_bits = 32;
164 			new->cid_shift = 16;
165 			new->cid_mask = new->lid_mask = 0xffff;
166 			new->broadcast = X2APIC_BROADCAST;
167 		} else if (kvm_apic_get_reg(apic, APIC_LDR)) {
168 			if (kvm_apic_get_reg(apic, APIC_DFR) ==
169 							APIC_DFR_CLUSTER) {
170 				new->cid_shift = 4;
171 				new->cid_mask = 0xf;
172 				new->lid_mask = 0xf;
173 			} else {
174 				new->cid_shift = 8;
175 				new->cid_mask = 0;
176 				new->lid_mask = 0xff;
177 			}
178 		}
179 
180 		/*
181 		 * All APICs have to be configured in the same mode by an OS.
182 		 * We take advatage of this while building logical id loockup
183 		 * table. After reset APICs are in software disabled mode, so if
184 		 * we find apic with different setting we assume this is the mode
185 		 * OS wants all apics to be in; build lookup table accordingly.
186 		 */
187 		if (kvm_apic_sw_enabled(apic))
188 			break;
189 	}
190 
191 	kvm_for_each_vcpu(i, vcpu, kvm) {
192 		struct kvm_lapic *apic = vcpu->arch.apic;
193 		u16 cid, lid;
194 		u32 ldr, aid;
195 
196 		if (!kvm_apic_present(vcpu))
197 			continue;
198 
199 		aid = kvm_apic_id(apic);
200 		ldr = kvm_apic_get_reg(apic, APIC_LDR);
201 		cid = apic_cluster_id(new, ldr);
202 		lid = apic_logical_id(new, ldr);
203 
204 		if (aid < ARRAY_SIZE(new->phys_map))
205 			new->phys_map[aid] = apic;
206 		if (lid && cid < ARRAY_SIZE(new->logical_map))
207 			new->logical_map[cid][ffs(lid) - 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 		kfree_rcu(old, rcu);
217 
218 	kvm_vcpu_request_scan_ioapic(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 	apic_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_id(struct kvm_lapic *apic, u8 id)
238 {
239 	apic_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 	apic_set_reg(apic, APIC_LDR, id);
246 	recalculate_apic_map(apic->vcpu->kvm);
247 }
248 
249 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
250 {
251 	return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
252 }
253 
254 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
255 {
256 	return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
257 }
258 
259 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
260 {
261 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
262 }
263 
264 static inline int apic_lvtt_period(struct kvm_lapic *apic)
265 {
266 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
267 }
268 
269 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
270 {
271 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
272 }
273 
274 static inline int apic_lvt_nmi_mode(u32 lvt_val)
275 {
276 	return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
277 }
278 
279 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
280 {
281 	struct kvm_lapic *apic = vcpu->arch.apic;
282 	struct kvm_cpuid_entry2 *feat;
283 	u32 v = APIC_VERSION;
284 
285 	if (!kvm_vcpu_has_lapic(vcpu))
286 		return;
287 
288 	feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
289 	if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
290 		v |= APIC_LVR_DIRECTED_EOI;
291 	apic_set_reg(apic, APIC_LVR, v);
292 }
293 
294 static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
295 	LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
296 	LVT_MASK | APIC_MODE_MASK,	/* LVTTHMR */
297 	LVT_MASK | APIC_MODE_MASK,	/* LVTPC */
298 	LINT_MASK, LINT_MASK,	/* LVT0-1 */
299 	LVT_MASK		/* LVTERR */
300 };
301 
302 static int find_highest_vector(void *bitmap)
303 {
304 	int vec;
305 	u32 *reg;
306 
307 	for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
308 	     vec >= 0; vec -= APIC_VECTORS_PER_REG) {
309 		reg = bitmap + REG_POS(vec);
310 		if (*reg)
311 			return fls(*reg) - 1 + vec;
312 	}
313 
314 	return -1;
315 }
316 
317 static u8 count_vectors(void *bitmap)
318 {
319 	int vec;
320 	u32 *reg;
321 	u8 count = 0;
322 
323 	for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
324 		reg = bitmap + REG_POS(vec);
325 		count += hweight32(*reg);
326 	}
327 
328 	return count;
329 }
330 
331 void __kvm_apic_update_irr(u32 *pir, void *regs)
332 {
333 	u32 i, pir_val;
334 
335 	for (i = 0; i <= 7; i++) {
336 		pir_val = xchg(&pir[i], 0);
337 		if (pir_val)
338 			*((u32 *)(regs + APIC_IRR + i * 0x10)) |= pir_val;
339 	}
340 }
341 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
342 
343 void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
344 {
345 	struct kvm_lapic *apic = vcpu->arch.apic;
346 
347 	__kvm_apic_update_irr(pir, apic->regs);
348 }
349 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
350 
351 static inline void apic_set_irr(int vec, struct kvm_lapic *apic)
352 {
353 	apic_set_vector(vec, apic->regs + APIC_IRR);
354 	/*
355 	 * irr_pending must be true if any interrupt is pending; set it after
356 	 * APIC_IRR to avoid race with apic_clear_irr
357 	 */
358 	apic->irr_pending = true;
359 }
360 
361 static inline int apic_search_irr(struct kvm_lapic *apic)
362 {
363 	return find_highest_vector(apic->regs + APIC_IRR);
364 }
365 
366 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
367 {
368 	int result;
369 
370 	/*
371 	 * Note that irr_pending is just a hint. It will be always
372 	 * true with virtual interrupt delivery enabled.
373 	 */
374 	if (!apic->irr_pending)
375 		return -1;
376 
377 	kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
378 	result = apic_search_irr(apic);
379 	ASSERT(result == -1 || result >= 16);
380 
381 	return result;
382 }
383 
384 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
385 {
386 	struct kvm_vcpu *vcpu;
387 
388 	vcpu = apic->vcpu;
389 
390 	if (unlikely(kvm_apic_vid_enabled(vcpu->kvm))) {
391 		/* try to update RVI */
392 		apic_clear_vector(vec, apic->regs + APIC_IRR);
393 		kvm_make_request(KVM_REQ_EVENT, vcpu);
394 	} else {
395 		apic->irr_pending = false;
396 		apic_clear_vector(vec, apic->regs + APIC_IRR);
397 		if (apic_search_irr(apic) != -1)
398 			apic->irr_pending = true;
399 	}
400 }
401 
402 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
403 {
404 	struct kvm_vcpu *vcpu;
405 
406 	if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
407 		return;
408 
409 	vcpu = apic->vcpu;
410 
411 	/*
412 	 * With APIC virtualization enabled, all caching is disabled
413 	 * because the processor can modify ISR under the hood.  Instead
414 	 * just set SVI.
415 	 */
416 	if (unlikely(kvm_x86_ops->hwapic_isr_update))
417 		kvm_x86_ops->hwapic_isr_update(vcpu->kvm, vec);
418 	else {
419 		++apic->isr_count;
420 		BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
421 		/*
422 		 * ISR (in service register) bit is set when injecting an interrupt.
423 		 * The highest vector is injected. Thus the latest bit set matches
424 		 * the highest bit in ISR.
425 		 */
426 		apic->highest_isr_cache = vec;
427 	}
428 }
429 
430 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
431 {
432 	int result;
433 
434 	/*
435 	 * Note that isr_count is always 1, and highest_isr_cache
436 	 * is always -1, with APIC virtualization enabled.
437 	 */
438 	if (!apic->isr_count)
439 		return -1;
440 	if (likely(apic->highest_isr_cache != -1))
441 		return apic->highest_isr_cache;
442 
443 	result = find_highest_vector(apic->regs + APIC_ISR);
444 	ASSERT(result == -1 || result >= 16);
445 
446 	return result;
447 }
448 
449 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
450 {
451 	struct kvm_vcpu *vcpu;
452 	if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
453 		return;
454 
455 	vcpu = apic->vcpu;
456 
457 	/*
458 	 * We do get here for APIC virtualization enabled if the guest
459 	 * uses the Hyper-V APIC enlightenment.  In this case we may need
460 	 * to trigger a new interrupt delivery by writing the SVI field;
461 	 * on the other hand isr_count and highest_isr_cache are unused
462 	 * and must be left alone.
463 	 */
464 	if (unlikely(kvm_x86_ops->hwapic_isr_update))
465 		kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
466 					       apic_find_highest_isr(apic));
467 	else {
468 		--apic->isr_count;
469 		BUG_ON(apic->isr_count < 0);
470 		apic->highest_isr_cache = -1;
471 	}
472 }
473 
474 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
475 {
476 	int highest_irr;
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 	if (!kvm_vcpu_has_lapic(vcpu))
484 		return 0;
485 	highest_irr = apic_find_highest_irr(vcpu->arch.apic);
486 
487 	return highest_irr;
488 }
489 
490 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
491 			     int vector, int level, int trig_mode,
492 			     unsigned long *dest_map);
493 
494 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
495 		unsigned long *dest_map)
496 {
497 	struct kvm_lapic *apic = vcpu->arch.apic;
498 
499 	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
500 			irq->level, irq->trig_mode, dest_map);
501 }
502 
503 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
504 {
505 
506 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
507 				      sizeof(val));
508 }
509 
510 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
511 {
512 
513 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
514 				      sizeof(*val));
515 }
516 
517 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
518 {
519 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
520 }
521 
522 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
523 {
524 	u8 val;
525 	if (pv_eoi_get_user(vcpu, &val) < 0)
526 		apic_debug("Can't read EOI MSR value: 0x%llx\n",
527 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
528 	return val & 0x1;
529 }
530 
531 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
532 {
533 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
534 		apic_debug("Can't set EOI MSR value: 0x%llx\n",
535 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
536 		return;
537 	}
538 	__set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
539 }
540 
541 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
542 {
543 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
544 		apic_debug("Can't clear EOI MSR value: 0x%llx\n",
545 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
546 		return;
547 	}
548 	__clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
549 }
550 
551 void kvm_apic_update_tmr(struct kvm_vcpu *vcpu, u32 *tmr)
552 {
553 	struct kvm_lapic *apic = vcpu->arch.apic;
554 	int i;
555 
556 	for (i = 0; i < 8; i++)
557 		apic_set_reg(apic, APIC_TMR + 0x10 * i, tmr[i]);
558 }
559 
560 static void apic_update_ppr(struct kvm_lapic *apic)
561 {
562 	u32 tpr, isrv, ppr, old_ppr;
563 	int isr;
564 
565 	old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI);
566 	tpr = kvm_apic_get_reg(apic, APIC_TASKPRI);
567 	isr = apic_find_highest_isr(apic);
568 	isrv = (isr != -1) ? isr : 0;
569 
570 	if ((tpr & 0xf0) >= (isrv & 0xf0))
571 		ppr = tpr & 0xff;
572 	else
573 		ppr = isrv & 0xf0;
574 
575 	apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
576 		   apic, ppr, isr, isrv);
577 
578 	if (old_ppr != ppr) {
579 		apic_set_reg(apic, APIC_PROCPRI, ppr);
580 		if (ppr < old_ppr)
581 			kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
582 	}
583 }
584 
585 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
586 {
587 	apic_set_reg(apic, APIC_TASKPRI, tpr);
588 	apic_update_ppr(apic);
589 }
590 
591 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 dest)
592 {
593 	return dest == (apic_x2apic_mode(apic) ?
594 			X2APIC_BROADCAST : APIC_BROADCAST);
595 }
596 
597 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 dest)
598 {
599 	return kvm_apic_id(apic) == dest || kvm_apic_broadcast(apic, dest);
600 }
601 
602 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
603 {
604 	u32 logical_id;
605 
606 	if (kvm_apic_broadcast(apic, mda))
607 		return true;
608 
609 	logical_id = kvm_apic_get_reg(apic, APIC_LDR);
610 
611 	if (apic_x2apic_mode(apic))
612 		return ((logical_id >> 16) == (mda >> 16))
613 		       && (logical_id & mda & 0xffff) != 0;
614 
615 	logical_id = GET_APIC_LOGICAL_ID(logical_id);
616 
617 	switch (kvm_apic_get_reg(apic, APIC_DFR)) {
618 	case APIC_DFR_FLAT:
619 		return (logical_id & mda) != 0;
620 	case APIC_DFR_CLUSTER:
621 		return ((logical_id >> 4) == (mda >> 4))
622 		       && (logical_id & mda & 0xf) != 0;
623 	default:
624 		apic_debug("Bad DFR vcpu %d: %08x\n",
625 			   apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR));
626 		return false;
627 	}
628 }
629 
630 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
631 			   int short_hand, unsigned int dest, int dest_mode)
632 {
633 	struct kvm_lapic *target = vcpu->arch.apic;
634 
635 	apic_debug("target %p, source %p, dest 0x%x, "
636 		   "dest_mode 0x%x, short_hand 0x%x\n",
637 		   target, source, dest, dest_mode, short_hand);
638 
639 	ASSERT(target);
640 	switch (short_hand) {
641 	case APIC_DEST_NOSHORT:
642 		if (dest_mode == APIC_DEST_PHYSICAL)
643 			return kvm_apic_match_physical_addr(target, dest);
644 		else
645 			return kvm_apic_match_logical_addr(target, dest);
646 	case APIC_DEST_SELF:
647 		return target == source;
648 	case APIC_DEST_ALLINC:
649 		return true;
650 	case APIC_DEST_ALLBUT:
651 		return target != source;
652 	default:
653 		apic_debug("kvm: apic: Bad dest shorthand value %x\n",
654 			   short_hand);
655 		return false;
656 	}
657 }
658 
659 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
660 		struct kvm_lapic_irq *irq, int *r, unsigned long *dest_map)
661 {
662 	struct kvm_apic_map *map;
663 	unsigned long bitmap = 1;
664 	struct kvm_lapic **dst;
665 	int i;
666 	bool ret = false;
667 
668 	*r = -1;
669 
670 	if (irq->shorthand == APIC_DEST_SELF) {
671 		*r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
672 		return true;
673 	}
674 
675 	if (irq->shorthand)
676 		return false;
677 
678 	rcu_read_lock();
679 	map = rcu_dereference(kvm->arch.apic_map);
680 
681 	if (!map)
682 		goto out;
683 
684 	if (irq->dest_id == map->broadcast)
685 		goto out;
686 
687 	ret = true;
688 
689 	if (irq->dest_mode == APIC_DEST_PHYSICAL) {
690 		if (irq->dest_id >= ARRAY_SIZE(map->phys_map))
691 			goto out;
692 
693 		dst = &map->phys_map[irq->dest_id];
694 	} else {
695 		u32 mda = irq->dest_id << (32 - map->ldr_bits);
696 		u16 cid = apic_cluster_id(map, mda);
697 
698 		if (cid >= ARRAY_SIZE(map->logical_map))
699 			goto out;
700 
701 		dst = map->logical_map[cid];
702 
703 		bitmap = apic_logical_id(map, mda);
704 
705 		if (irq->delivery_mode == APIC_DM_LOWEST) {
706 			int l = -1;
707 			for_each_set_bit(i, &bitmap, 16) {
708 				if (!dst[i])
709 					continue;
710 				if (l < 0)
711 					l = i;
712 				else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0)
713 					l = i;
714 			}
715 
716 			bitmap = (l >= 0) ? 1 << l : 0;
717 		}
718 	}
719 
720 	for_each_set_bit(i, &bitmap, 16) {
721 		if (!dst[i])
722 			continue;
723 		if (*r < 0)
724 			*r = 0;
725 		*r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
726 	}
727 out:
728 	rcu_read_unlock();
729 	return ret;
730 }
731 
732 /*
733  * Add a pending IRQ into lapic.
734  * Return 1 if successfully added and 0 if discarded.
735  */
736 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
737 			     int vector, int level, int trig_mode,
738 			     unsigned long *dest_map)
739 {
740 	int result = 0;
741 	struct kvm_vcpu *vcpu = apic->vcpu;
742 
743 	trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
744 				  trig_mode, vector);
745 	switch (delivery_mode) {
746 	case APIC_DM_LOWEST:
747 		vcpu->arch.apic_arb_prio++;
748 	case APIC_DM_FIXED:
749 		/* FIXME add logic for vcpu on reset */
750 		if (unlikely(!apic_enabled(apic)))
751 			break;
752 
753 		result = 1;
754 
755 		if (dest_map)
756 			__set_bit(vcpu->vcpu_id, dest_map);
757 
758 		if (kvm_x86_ops->deliver_posted_interrupt)
759 			kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
760 		else {
761 			apic_set_irr(vector, apic);
762 
763 			kvm_make_request(KVM_REQ_EVENT, vcpu);
764 			kvm_vcpu_kick(vcpu);
765 		}
766 		break;
767 
768 	case APIC_DM_REMRD:
769 		result = 1;
770 		vcpu->arch.pv.pv_unhalted = 1;
771 		kvm_make_request(KVM_REQ_EVENT, vcpu);
772 		kvm_vcpu_kick(vcpu);
773 		break;
774 
775 	case APIC_DM_SMI:
776 		apic_debug("Ignoring guest SMI\n");
777 		break;
778 
779 	case APIC_DM_NMI:
780 		result = 1;
781 		kvm_inject_nmi(vcpu);
782 		kvm_vcpu_kick(vcpu);
783 		break;
784 
785 	case APIC_DM_INIT:
786 		if (!trig_mode || level) {
787 			result = 1;
788 			/* assumes that there are only KVM_APIC_INIT/SIPI */
789 			apic->pending_events = (1UL << KVM_APIC_INIT);
790 			/* make sure pending_events is visible before sending
791 			 * the request */
792 			smp_wmb();
793 			kvm_make_request(KVM_REQ_EVENT, vcpu);
794 			kvm_vcpu_kick(vcpu);
795 		} else {
796 			apic_debug("Ignoring de-assert INIT to vcpu %d\n",
797 				   vcpu->vcpu_id);
798 		}
799 		break;
800 
801 	case APIC_DM_STARTUP:
802 		apic_debug("SIPI to vcpu %d vector 0x%02x\n",
803 			   vcpu->vcpu_id, vector);
804 		result = 1;
805 		apic->sipi_vector = vector;
806 		/* make sure sipi_vector is visible for the receiver */
807 		smp_wmb();
808 		set_bit(KVM_APIC_SIPI, &apic->pending_events);
809 		kvm_make_request(KVM_REQ_EVENT, vcpu);
810 		kvm_vcpu_kick(vcpu);
811 		break;
812 
813 	case APIC_DM_EXTINT:
814 		/*
815 		 * Should only be called by kvm_apic_local_deliver() with LVT0,
816 		 * before NMI watchdog was enabled. Already handled by
817 		 * kvm_apic_accept_pic_intr().
818 		 */
819 		break;
820 
821 	default:
822 		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
823 		       delivery_mode);
824 		break;
825 	}
826 	return result;
827 }
828 
829 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
830 {
831 	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
832 }
833 
834 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
835 {
836 	if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI) &&
837 	    kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
838 		int trigger_mode;
839 		if (apic_test_vector(vector, apic->regs + APIC_TMR))
840 			trigger_mode = IOAPIC_LEVEL_TRIG;
841 		else
842 			trigger_mode = IOAPIC_EDGE_TRIG;
843 		kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
844 	}
845 }
846 
847 static int apic_set_eoi(struct kvm_lapic *apic)
848 {
849 	int vector = apic_find_highest_isr(apic);
850 
851 	trace_kvm_eoi(apic, vector);
852 
853 	/*
854 	 * Not every write EOI will has corresponding ISR,
855 	 * one example is when Kernel check timer on setup_IO_APIC
856 	 */
857 	if (vector == -1)
858 		return vector;
859 
860 	apic_clear_isr(vector, apic);
861 	apic_update_ppr(apic);
862 
863 	kvm_ioapic_send_eoi(apic, vector);
864 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
865 	return vector;
866 }
867 
868 /*
869  * this interface assumes a trap-like exit, which has already finished
870  * desired side effect including vISR and vPPR update.
871  */
872 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
873 {
874 	struct kvm_lapic *apic = vcpu->arch.apic;
875 
876 	trace_kvm_eoi(apic, vector);
877 
878 	kvm_ioapic_send_eoi(apic, vector);
879 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
880 }
881 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
882 
883 static void apic_send_ipi(struct kvm_lapic *apic)
884 {
885 	u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR);
886 	u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2);
887 	struct kvm_lapic_irq irq;
888 
889 	irq.vector = icr_low & APIC_VECTOR_MASK;
890 	irq.delivery_mode = icr_low & APIC_MODE_MASK;
891 	irq.dest_mode = icr_low & APIC_DEST_MASK;
892 	irq.level = icr_low & APIC_INT_ASSERT;
893 	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
894 	irq.shorthand = icr_low & APIC_SHORT_MASK;
895 	if (apic_x2apic_mode(apic))
896 		irq.dest_id = icr_high;
897 	else
898 		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
899 
900 	trace_kvm_apic_ipi(icr_low, irq.dest_id);
901 
902 	apic_debug("icr_high 0x%x, icr_low 0x%x, "
903 		   "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
904 		   "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
905 		   icr_high, icr_low, irq.shorthand, irq.dest_id,
906 		   irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
907 		   irq.vector);
908 
909 	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
910 }
911 
912 static u32 apic_get_tmcct(struct kvm_lapic *apic)
913 {
914 	ktime_t remaining;
915 	s64 ns;
916 	u32 tmcct;
917 
918 	ASSERT(apic != NULL);
919 
920 	/* if initial count is 0, current count should also be 0 */
921 	if (kvm_apic_get_reg(apic, APIC_TMICT) == 0 ||
922 		apic->lapic_timer.period == 0)
923 		return 0;
924 
925 	remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
926 	if (ktime_to_ns(remaining) < 0)
927 		remaining = ktime_set(0, 0);
928 
929 	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
930 	tmcct = div64_u64(ns,
931 			 (APIC_BUS_CYCLE_NS * apic->divide_count));
932 
933 	return tmcct;
934 }
935 
936 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
937 {
938 	struct kvm_vcpu *vcpu = apic->vcpu;
939 	struct kvm_run *run = vcpu->run;
940 
941 	kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
942 	run->tpr_access.rip = kvm_rip_read(vcpu);
943 	run->tpr_access.is_write = write;
944 }
945 
946 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
947 {
948 	if (apic->vcpu->arch.tpr_access_reporting)
949 		__report_tpr_access(apic, write);
950 }
951 
952 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
953 {
954 	u32 val = 0;
955 
956 	if (offset >= LAPIC_MMIO_LENGTH)
957 		return 0;
958 
959 	switch (offset) {
960 	case APIC_ID:
961 		if (apic_x2apic_mode(apic))
962 			val = kvm_apic_id(apic);
963 		else
964 			val = kvm_apic_id(apic) << 24;
965 		break;
966 	case APIC_ARBPRI:
967 		apic_debug("Access APIC ARBPRI register which is for P6\n");
968 		break;
969 
970 	case APIC_TMCCT:	/* Timer CCR */
971 		if (apic_lvtt_tscdeadline(apic))
972 			return 0;
973 
974 		val = apic_get_tmcct(apic);
975 		break;
976 	case APIC_PROCPRI:
977 		apic_update_ppr(apic);
978 		val = kvm_apic_get_reg(apic, offset);
979 		break;
980 	case APIC_TASKPRI:
981 		report_tpr_access(apic, false);
982 		/* fall thru */
983 	default:
984 		val = kvm_apic_get_reg(apic, offset);
985 		break;
986 	}
987 
988 	return val;
989 }
990 
991 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
992 {
993 	return container_of(dev, struct kvm_lapic, dev);
994 }
995 
996 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
997 		void *data)
998 {
999 	unsigned char alignment = offset & 0xf;
1000 	u32 result;
1001 	/* this bitmask has a bit cleared for each reserved register */
1002 	static const u64 rmask = 0x43ff01ffffffe70cULL;
1003 
1004 	if ((alignment + len) > 4) {
1005 		apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1006 			   offset, len);
1007 		return 1;
1008 	}
1009 
1010 	if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1011 		apic_debug("KVM_APIC_READ: read reserved register %x\n",
1012 			   offset);
1013 		return 1;
1014 	}
1015 
1016 	result = __apic_read(apic, offset & ~0xf);
1017 
1018 	trace_kvm_apic_read(offset, result);
1019 
1020 	switch (len) {
1021 	case 1:
1022 	case 2:
1023 	case 4:
1024 		memcpy(data, (char *)&result + alignment, len);
1025 		break;
1026 	default:
1027 		printk(KERN_ERR "Local APIC read with len = %x, "
1028 		       "should be 1,2, or 4 instead\n", len);
1029 		break;
1030 	}
1031 	return 0;
1032 }
1033 
1034 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1035 {
1036 	return kvm_apic_hw_enabled(apic) &&
1037 	    addr >= apic->base_address &&
1038 	    addr < apic->base_address + LAPIC_MMIO_LENGTH;
1039 }
1040 
1041 static int apic_mmio_read(struct kvm_io_device *this,
1042 			   gpa_t address, int len, void *data)
1043 {
1044 	struct kvm_lapic *apic = to_lapic(this);
1045 	u32 offset = address - apic->base_address;
1046 
1047 	if (!apic_mmio_in_range(apic, address))
1048 		return -EOPNOTSUPP;
1049 
1050 	apic_reg_read(apic, offset, len, data);
1051 
1052 	return 0;
1053 }
1054 
1055 static void update_divide_count(struct kvm_lapic *apic)
1056 {
1057 	u32 tmp1, tmp2, tdcr;
1058 
1059 	tdcr = kvm_apic_get_reg(apic, APIC_TDCR);
1060 	tmp1 = tdcr & 0xf;
1061 	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1062 	apic->divide_count = 0x1 << (tmp2 & 0x7);
1063 
1064 	apic_debug("timer divide count is 0x%x\n",
1065 				   apic->divide_count);
1066 }
1067 
1068 static void apic_timer_expired(struct kvm_lapic *apic)
1069 {
1070 	struct kvm_vcpu *vcpu = apic->vcpu;
1071 	wait_queue_head_t *q = &vcpu->wq;
1072 	struct kvm_timer *ktimer = &apic->lapic_timer;
1073 
1074 	if (atomic_read(&apic->lapic_timer.pending))
1075 		return;
1076 
1077 	atomic_inc(&apic->lapic_timer.pending);
1078 	kvm_set_pending_timer(vcpu);
1079 
1080 	if (waitqueue_active(q))
1081 		wake_up_interruptible(q);
1082 
1083 	if (apic_lvtt_tscdeadline(apic))
1084 		ktimer->expired_tscdeadline = ktimer->tscdeadline;
1085 }
1086 
1087 /*
1088  * On APICv, this test will cause a busy wait
1089  * during a higher-priority task.
1090  */
1091 
1092 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1093 {
1094 	struct kvm_lapic *apic = vcpu->arch.apic;
1095 	u32 reg = kvm_apic_get_reg(apic, APIC_LVTT);
1096 
1097 	if (kvm_apic_hw_enabled(apic)) {
1098 		int vec = reg & APIC_VECTOR_MASK;
1099 		void *bitmap = apic->regs + APIC_ISR;
1100 
1101 		if (kvm_x86_ops->deliver_posted_interrupt)
1102 			bitmap = apic->regs + APIC_IRR;
1103 
1104 		if (apic_test_vector(vec, bitmap))
1105 			return true;
1106 	}
1107 	return false;
1108 }
1109 
1110 void wait_lapic_expire(struct kvm_vcpu *vcpu)
1111 {
1112 	struct kvm_lapic *apic = vcpu->arch.apic;
1113 	u64 guest_tsc, tsc_deadline;
1114 
1115 	if (!kvm_vcpu_has_lapic(vcpu))
1116 		return;
1117 
1118 	if (apic->lapic_timer.expired_tscdeadline == 0)
1119 		return;
1120 
1121 	if (!lapic_timer_int_injected(vcpu))
1122 		return;
1123 
1124 	tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1125 	apic->lapic_timer.expired_tscdeadline = 0;
1126 	guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1127 	trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1128 
1129 	/* __delay is delay_tsc whenever the hardware has TSC, thus always.  */
1130 	if (guest_tsc < tsc_deadline)
1131 		__delay(tsc_deadline - guest_tsc);
1132 }
1133 
1134 static void start_apic_timer(struct kvm_lapic *apic)
1135 {
1136 	ktime_t now;
1137 
1138 	atomic_set(&apic->lapic_timer.pending, 0);
1139 
1140 	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1141 		/* lapic timer in oneshot or periodic mode */
1142 		now = apic->lapic_timer.timer.base->get_time();
1143 		apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT)
1144 			    * APIC_BUS_CYCLE_NS * apic->divide_count;
1145 
1146 		if (!apic->lapic_timer.period)
1147 			return;
1148 		/*
1149 		 * Do not allow the guest to program periodic timers with small
1150 		 * interval, since the hrtimers are not throttled by the host
1151 		 * scheduler.
1152 		 */
1153 		if (apic_lvtt_period(apic)) {
1154 			s64 min_period = min_timer_period_us * 1000LL;
1155 
1156 			if (apic->lapic_timer.period < min_period) {
1157 				pr_info_ratelimited(
1158 				    "kvm: vcpu %i: requested %lld ns "
1159 				    "lapic timer period limited to %lld ns\n",
1160 				    apic->vcpu->vcpu_id,
1161 				    apic->lapic_timer.period, min_period);
1162 				apic->lapic_timer.period = min_period;
1163 			}
1164 		}
1165 
1166 		hrtimer_start(&apic->lapic_timer.timer,
1167 			      ktime_add_ns(now, apic->lapic_timer.period),
1168 			      HRTIMER_MODE_ABS);
1169 
1170 		apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1171 			   PRIx64 ", "
1172 			   "timer initial count 0x%x, period %lldns, "
1173 			   "expire @ 0x%016" PRIx64 ".\n", __func__,
1174 			   APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1175 			   kvm_apic_get_reg(apic, APIC_TMICT),
1176 			   apic->lapic_timer.period,
1177 			   ktime_to_ns(ktime_add_ns(now,
1178 					apic->lapic_timer.period)));
1179 	} else if (apic_lvtt_tscdeadline(apic)) {
1180 		/* lapic timer in tsc deadline mode */
1181 		u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1182 		u64 ns = 0;
1183 		ktime_t expire;
1184 		struct kvm_vcpu *vcpu = apic->vcpu;
1185 		unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1186 		unsigned long flags;
1187 
1188 		if (unlikely(!tscdeadline || !this_tsc_khz))
1189 			return;
1190 
1191 		local_irq_save(flags);
1192 
1193 		now = apic->lapic_timer.timer.base->get_time();
1194 		guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1195 		if (likely(tscdeadline > guest_tsc)) {
1196 			ns = (tscdeadline - guest_tsc) * 1000000ULL;
1197 			do_div(ns, this_tsc_khz);
1198 			expire = ktime_add_ns(now, ns);
1199 			expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1200 			hrtimer_start(&apic->lapic_timer.timer,
1201 				      expire, HRTIMER_MODE_ABS);
1202 		} else
1203 			apic_timer_expired(apic);
1204 
1205 		local_irq_restore(flags);
1206 	}
1207 }
1208 
1209 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1210 {
1211 	int nmi_wd_enabled = apic_lvt_nmi_mode(kvm_apic_get_reg(apic, APIC_LVT0));
1212 
1213 	if (apic_lvt_nmi_mode(lvt0_val)) {
1214 		if (!nmi_wd_enabled) {
1215 			apic_debug("Receive NMI setting on APIC_LVT0 "
1216 				   "for cpu %d\n", apic->vcpu->vcpu_id);
1217 			apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
1218 		}
1219 	} else if (nmi_wd_enabled)
1220 		apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
1221 }
1222 
1223 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1224 {
1225 	int ret = 0;
1226 
1227 	trace_kvm_apic_write(reg, val);
1228 
1229 	switch (reg) {
1230 	case APIC_ID:		/* Local APIC ID */
1231 		if (!apic_x2apic_mode(apic))
1232 			kvm_apic_set_id(apic, val >> 24);
1233 		else
1234 			ret = 1;
1235 		break;
1236 
1237 	case APIC_TASKPRI:
1238 		report_tpr_access(apic, true);
1239 		apic_set_tpr(apic, val & 0xff);
1240 		break;
1241 
1242 	case APIC_EOI:
1243 		apic_set_eoi(apic);
1244 		break;
1245 
1246 	case APIC_LDR:
1247 		if (!apic_x2apic_mode(apic))
1248 			kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1249 		else
1250 			ret = 1;
1251 		break;
1252 
1253 	case APIC_DFR:
1254 		if (!apic_x2apic_mode(apic)) {
1255 			apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1256 			recalculate_apic_map(apic->vcpu->kvm);
1257 		} else
1258 			ret = 1;
1259 		break;
1260 
1261 	case APIC_SPIV: {
1262 		u32 mask = 0x3ff;
1263 		if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1264 			mask |= APIC_SPIV_DIRECTED_EOI;
1265 		apic_set_spiv(apic, val & mask);
1266 		if (!(val & APIC_SPIV_APIC_ENABLED)) {
1267 			int i;
1268 			u32 lvt_val;
1269 
1270 			for (i = 0; i < APIC_LVT_NUM; i++) {
1271 				lvt_val = kvm_apic_get_reg(apic,
1272 						       APIC_LVTT + 0x10 * i);
1273 				apic_set_reg(apic, APIC_LVTT + 0x10 * i,
1274 					     lvt_val | APIC_LVT_MASKED);
1275 			}
1276 			atomic_set(&apic->lapic_timer.pending, 0);
1277 
1278 		}
1279 		break;
1280 	}
1281 	case APIC_ICR:
1282 		/* No delay here, so we always clear the pending bit */
1283 		apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1284 		apic_send_ipi(apic);
1285 		break;
1286 
1287 	case APIC_ICR2:
1288 		if (!apic_x2apic_mode(apic))
1289 			val &= 0xff000000;
1290 		apic_set_reg(apic, APIC_ICR2, val);
1291 		break;
1292 
1293 	case APIC_LVT0:
1294 		apic_manage_nmi_watchdog(apic, val);
1295 	case APIC_LVTTHMR:
1296 	case APIC_LVTPC:
1297 	case APIC_LVT1:
1298 	case APIC_LVTERR:
1299 		/* TODO: Check vector */
1300 		if (!kvm_apic_sw_enabled(apic))
1301 			val |= APIC_LVT_MASKED;
1302 
1303 		val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1304 		apic_set_reg(apic, reg, val);
1305 
1306 		break;
1307 
1308 	case APIC_LVTT: {
1309 		u32 timer_mode = val & apic->lapic_timer.timer_mode_mask;
1310 
1311 		if (apic->lapic_timer.timer_mode != timer_mode) {
1312 			apic->lapic_timer.timer_mode = timer_mode;
1313 			hrtimer_cancel(&apic->lapic_timer.timer);
1314 		}
1315 
1316 		if (!kvm_apic_sw_enabled(apic))
1317 			val |= APIC_LVT_MASKED;
1318 		val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1319 		apic_set_reg(apic, APIC_LVTT, val);
1320 		break;
1321 	}
1322 
1323 	case APIC_TMICT:
1324 		if (apic_lvtt_tscdeadline(apic))
1325 			break;
1326 
1327 		hrtimer_cancel(&apic->lapic_timer.timer);
1328 		apic_set_reg(apic, APIC_TMICT, val);
1329 		start_apic_timer(apic);
1330 		break;
1331 
1332 	case APIC_TDCR:
1333 		if (val & 4)
1334 			apic_debug("KVM_WRITE:TDCR %x\n", val);
1335 		apic_set_reg(apic, APIC_TDCR, val);
1336 		update_divide_count(apic);
1337 		break;
1338 
1339 	case APIC_ESR:
1340 		if (apic_x2apic_mode(apic) && val != 0) {
1341 			apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1342 			ret = 1;
1343 		}
1344 		break;
1345 
1346 	case APIC_SELF_IPI:
1347 		if (apic_x2apic_mode(apic)) {
1348 			apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1349 		} else
1350 			ret = 1;
1351 		break;
1352 	default:
1353 		ret = 1;
1354 		break;
1355 	}
1356 	if (ret)
1357 		apic_debug("Local APIC Write to read-only register %x\n", reg);
1358 	return ret;
1359 }
1360 
1361 static int apic_mmio_write(struct kvm_io_device *this,
1362 			    gpa_t address, int len, const void *data)
1363 {
1364 	struct kvm_lapic *apic = to_lapic(this);
1365 	unsigned int offset = address - apic->base_address;
1366 	u32 val;
1367 
1368 	if (!apic_mmio_in_range(apic, address))
1369 		return -EOPNOTSUPP;
1370 
1371 	/*
1372 	 * APIC register must be aligned on 128-bits boundary.
1373 	 * 32/64/128 bits registers must be accessed thru 32 bits.
1374 	 * Refer SDM 8.4.1
1375 	 */
1376 	if (len != 4 || (offset & 0xf)) {
1377 		/* Don't shout loud, $infamous_os would cause only noise. */
1378 		apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1379 		return 0;
1380 	}
1381 
1382 	val = *(u32*)data;
1383 
1384 	/* too common printing */
1385 	if (offset != APIC_EOI)
1386 		apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1387 			   "0x%x\n", __func__, offset, len, val);
1388 
1389 	apic_reg_write(apic, offset & 0xff0, val);
1390 
1391 	return 0;
1392 }
1393 
1394 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1395 {
1396 	if (kvm_vcpu_has_lapic(vcpu))
1397 		apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1398 }
1399 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1400 
1401 /* emulate APIC access in a trap manner */
1402 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1403 {
1404 	u32 val = 0;
1405 
1406 	/* hw has done the conditional check and inst decode */
1407 	offset &= 0xff0;
1408 
1409 	apic_reg_read(vcpu->arch.apic, offset, 4, &val);
1410 
1411 	/* TODO: optimize to just emulate side effect w/o one more write */
1412 	apic_reg_write(vcpu->arch.apic, offset, val);
1413 }
1414 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1415 
1416 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1417 {
1418 	struct kvm_lapic *apic = vcpu->arch.apic;
1419 
1420 	if (!vcpu->arch.apic)
1421 		return;
1422 
1423 	hrtimer_cancel(&apic->lapic_timer.timer);
1424 
1425 	if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1426 		static_key_slow_dec_deferred(&apic_hw_disabled);
1427 
1428 	if (!apic->sw_enabled)
1429 		static_key_slow_dec_deferred(&apic_sw_disabled);
1430 
1431 	if (apic->regs)
1432 		free_page((unsigned long)apic->regs);
1433 
1434 	kfree(apic);
1435 }
1436 
1437 /*
1438  *----------------------------------------------------------------------
1439  * LAPIC interface
1440  *----------------------------------------------------------------------
1441  */
1442 
1443 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1444 {
1445 	struct kvm_lapic *apic = vcpu->arch.apic;
1446 
1447 	if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1448 			apic_lvtt_period(apic))
1449 		return 0;
1450 
1451 	return apic->lapic_timer.tscdeadline;
1452 }
1453 
1454 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1455 {
1456 	struct kvm_lapic *apic = vcpu->arch.apic;
1457 
1458 	if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1459 			apic_lvtt_period(apic))
1460 		return;
1461 
1462 	hrtimer_cancel(&apic->lapic_timer.timer);
1463 	apic->lapic_timer.tscdeadline = data;
1464 	start_apic_timer(apic);
1465 }
1466 
1467 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1468 {
1469 	struct kvm_lapic *apic = vcpu->arch.apic;
1470 
1471 	if (!kvm_vcpu_has_lapic(vcpu))
1472 		return;
1473 
1474 	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1475 		     | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4));
1476 }
1477 
1478 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1479 {
1480 	u64 tpr;
1481 
1482 	if (!kvm_vcpu_has_lapic(vcpu))
1483 		return 0;
1484 
1485 	tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1486 
1487 	return (tpr & 0xf0) >> 4;
1488 }
1489 
1490 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1491 {
1492 	u64 old_value = vcpu->arch.apic_base;
1493 	struct kvm_lapic *apic = vcpu->arch.apic;
1494 
1495 	if (!apic) {
1496 		value |= MSR_IA32_APICBASE_BSP;
1497 		vcpu->arch.apic_base = value;
1498 		return;
1499 	}
1500 
1501 	if (!kvm_vcpu_is_bsp(apic->vcpu))
1502 		value &= ~MSR_IA32_APICBASE_BSP;
1503 	vcpu->arch.apic_base = value;
1504 
1505 	/* update jump label if enable bit changes */
1506 	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1507 		if (value & MSR_IA32_APICBASE_ENABLE)
1508 			static_key_slow_dec_deferred(&apic_hw_disabled);
1509 		else
1510 			static_key_slow_inc(&apic_hw_disabled.key);
1511 		recalculate_apic_map(vcpu->kvm);
1512 	}
1513 
1514 	if ((old_value ^ value) & X2APIC_ENABLE) {
1515 		if (value & X2APIC_ENABLE) {
1516 			u32 id = kvm_apic_id(apic);
1517 			u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
1518 			kvm_apic_set_ldr(apic, ldr);
1519 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1520 		} else
1521 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1522 	}
1523 
1524 	apic->base_address = apic->vcpu->arch.apic_base &
1525 			     MSR_IA32_APICBASE_BASE;
1526 
1527 	if ((value & MSR_IA32_APICBASE_ENABLE) &&
1528 	     apic->base_address != APIC_DEFAULT_PHYS_BASE)
1529 		pr_warn_once("APIC base relocation is unsupported by KVM");
1530 
1531 	/* with FSB delivery interrupt, we can restart APIC functionality */
1532 	apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1533 		   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1534 
1535 }
1536 
1537 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1538 {
1539 	struct kvm_lapic *apic;
1540 	int i;
1541 
1542 	apic_debug("%s\n", __func__);
1543 
1544 	ASSERT(vcpu);
1545 	apic = vcpu->arch.apic;
1546 	ASSERT(apic != NULL);
1547 
1548 	/* Stop the timer in case it's a reset to an active apic */
1549 	hrtimer_cancel(&apic->lapic_timer.timer);
1550 
1551 	kvm_apic_set_id(apic, vcpu->vcpu_id);
1552 	kvm_apic_set_version(apic->vcpu);
1553 
1554 	for (i = 0; i < APIC_LVT_NUM; i++)
1555 		apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1556 	apic->lapic_timer.timer_mode = 0;
1557 	apic_set_reg(apic, APIC_LVT0,
1558 		     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1559 
1560 	apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1561 	apic_set_spiv(apic, 0xff);
1562 	apic_set_reg(apic, APIC_TASKPRI, 0);
1563 	kvm_apic_set_ldr(apic, 0);
1564 	apic_set_reg(apic, APIC_ESR, 0);
1565 	apic_set_reg(apic, APIC_ICR, 0);
1566 	apic_set_reg(apic, APIC_ICR2, 0);
1567 	apic_set_reg(apic, APIC_TDCR, 0);
1568 	apic_set_reg(apic, APIC_TMICT, 0);
1569 	for (i = 0; i < 8; i++) {
1570 		apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1571 		apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1572 		apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1573 	}
1574 	apic->irr_pending = kvm_apic_vid_enabled(vcpu->kvm);
1575 	apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm);
1576 	apic->highest_isr_cache = -1;
1577 	update_divide_count(apic);
1578 	atomic_set(&apic->lapic_timer.pending, 0);
1579 	if (kvm_vcpu_is_bsp(vcpu))
1580 		kvm_lapic_set_base(vcpu,
1581 				vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1582 	vcpu->arch.pv_eoi.msr_val = 0;
1583 	apic_update_ppr(apic);
1584 
1585 	vcpu->arch.apic_arb_prio = 0;
1586 	vcpu->arch.apic_attention = 0;
1587 
1588 	apic_debug("%s: vcpu=%p, id=%d, base_msr="
1589 		   "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1590 		   vcpu, kvm_apic_id(apic),
1591 		   vcpu->arch.apic_base, apic->base_address);
1592 }
1593 
1594 /*
1595  *----------------------------------------------------------------------
1596  * timer interface
1597  *----------------------------------------------------------------------
1598  */
1599 
1600 static bool lapic_is_periodic(struct kvm_lapic *apic)
1601 {
1602 	return apic_lvtt_period(apic);
1603 }
1604 
1605 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1606 {
1607 	struct kvm_lapic *apic = vcpu->arch.apic;
1608 
1609 	if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) &&
1610 			apic_lvt_enabled(apic, APIC_LVTT))
1611 		return atomic_read(&apic->lapic_timer.pending);
1612 
1613 	return 0;
1614 }
1615 
1616 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1617 {
1618 	u32 reg = kvm_apic_get_reg(apic, lvt_type);
1619 	int vector, mode, trig_mode;
1620 
1621 	if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1622 		vector = reg & APIC_VECTOR_MASK;
1623 		mode = reg & APIC_MODE_MASK;
1624 		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1625 		return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1626 					NULL);
1627 	}
1628 	return 0;
1629 }
1630 
1631 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1632 {
1633 	struct kvm_lapic *apic = vcpu->arch.apic;
1634 
1635 	if (apic)
1636 		kvm_apic_local_deliver(apic, APIC_LVT0);
1637 }
1638 
1639 static const struct kvm_io_device_ops apic_mmio_ops = {
1640 	.read     = apic_mmio_read,
1641 	.write    = apic_mmio_write,
1642 };
1643 
1644 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1645 {
1646 	struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1647 	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1648 
1649 	apic_timer_expired(apic);
1650 
1651 	if (lapic_is_periodic(apic)) {
1652 		hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1653 		return HRTIMER_RESTART;
1654 	} else
1655 		return HRTIMER_NORESTART;
1656 }
1657 
1658 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1659 {
1660 	struct kvm_lapic *apic;
1661 
1662 	ASSERT(vcpu != NULL);
1663 	apic_debug("apic_init %d\n", vcpu->vcpu_id);
1664 
1665 	apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1666 	if (!apic)
1667 		goto nomem;
1668 
1669 	vcpu->arch.apic = apic;
1670 
1671 	apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1672 	if (!apic->regs) {
1673 		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1674 		       vcpu->vcpu_id);
1675 		goto nomem_free_apic;
1676 	}
1677 	apic->vcpu = vcpu;
1678 
1679 	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1680 		     HRTIMER_MODE_ABS);
1681 	apic->lapic_timer.timer.function = apic_timer_fn;
1682 
1683 	/*
1684 	 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1685 	 * thinking that APIC satet has changed.
1686 	 */
1687 	vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1688 	kvm_lapic_set_base(vcpu,
1689 			APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE);
1690 
1691 	static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1692 	kvm_lapic_reset(vcpu);
1693 	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1694 
1695 	return 0;
1696 nomem_free_apic:
1697 	kfree(apic);
1698 nomem:
1699 	return -ENOMEM;
1700 }
1701 
1702 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1703 {
1704 	struct kvm_lapic *apic = vcpu->arch.apic;
1705 	int highest_irr;
1706 
1707 	if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic))
1708 		return -1;
1709 
1710 	apic_update_ppr(apic);
1711 	highest_irr = apic_find_highest_irr(apic);
1712 	if ((highest_irr == -1) ||
1713 	    ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI)))
1714 		return -1;
1715 	return highest_irr;
1716 }
1717 
1718 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1719 {
1720 	u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1721 	int r = 0;
1722 
1723 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1724 		r = 1;
1725 	if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1726 	    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1727 		r = 1;
1728 	return r;
1729 }
1730 
1731 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1732 {
1733 	struct kvm_lapic *apic = vcpu->arch.apic;
1734 
1735 	if (!kvm_vcpu_has_lapic(vcpu))
1736 		return;
1737 
1738 	if (atomic_read(&apic->lapic_timer.pending) > 0) {
1739 		kvm_apic_local_deliver(apic, APIC_LVTT);
1740 		if (apic_lvtt_tscdeadline(apic))
1741 			apic->lapic_timer.tscdeadline = 0;
1742 		atomic_set(&apic->lapic_timer.pending, 0);
1743 	}
1744 }
1745 
1746 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1747 {
1748 	int vector = kvm_apic_has_interrupt(vcpu);
1749 	struct kvm_lapic *apic = vcpu->arch.apic;
1750 
1751 	if (vector == -1)
1752 		return -1;
1753 
1754 	/*
1755 	 * We get here even with APIC virtualization enabled, if doing
1756 	 * nested virtualization and L1 runs with the "acknowledge interrupt
1757 	 * on exit" mode.  Then we cannot inject the interrupt via RVI,
1758 	 * because the process would deliver it through the IDT.
1759 	 */
1760 
1761 	apic_set_isr(vector, apic);
1762 	apic_update_ppr(apic);
1763 	apic_clear_irr(vector, apic);
1764 	return vector;
1765 }
1766 
1767 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
1768 		struct kvm_lapic_state *s)
1769 {
1770 	struct kvm_lapic *apic = vcpu->arch.apic;
1771 
1772 	kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
1773 	/* set SPIV separately to get count of SW disabled APICs right */
1774 	apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
1775 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1776 	/* call kvm_apic_set_id() to put apic into apic_map */
1777 	kvm_apic_set_id(apic, kvm_apic_id(apic));
1778 	kvm_apic_set_version(vcpu);
1779 
1780 	apic_update_ppr(apic);
1781 	hrtimer_cancel(&apic->lapic_timer.timer);
1782 	update_divide_count(apic);
1783 	start_apic_timer(apic);
1784 	apic->irr_pending = true;
1785 	apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm) ?
1786 				1 : count_vectors(apic->regs + APIC_ISR);
1787 	apic->highest_isr_cache = -1;
1788 	if (kvm_x86_ops->hwapic_irr_update)
1789 		kvm_x86_ops->hwapic_irr_update(vcpu,
1790 				apic_find_highest_irr(apic));
1791 	if (unlikely(kvm_x86_ops->hwapic_isr_update))
1792 		kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
1793 				apic_find_highest_isr(apic));
1794 	kvm_make_request(KVM_REQ_EVENT, vcpu);
1795 	kvm_rtc_eoi_tracking_restore_one(vcpu);
1796 }
1797 
1798 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1799 {
1800 	struct hrtimer *timer;
1801 
1802 	if (!kvm_vcpu_has_lapic(vcpu))
1803 		return;
1804 
1805 	timer = &vcpu->arch.apic->lapic_timer.timer;
1806 	if (hrtimer_cancel(timer))
1807 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1808 }
1809 
1810 /*
1811  * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1812  *
1813  * Detect whether guest triggered PV EOI since the
1814  * last entry. If yes, set EOI on guests's behalf.
1815  * Clear PV EOI in guest memory in any case.
1816  */
1817 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1818 					struct kvm_lapic *apic)
1819 {
1820 	bool pending;
1821 	int vector;
1822 	/*
1823 	 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1824 	 * and KVM_PV_EOI_ENABLED in guest memory as follows:
1825 	 *
1826 	 * KVM_APIC_PV_EOI_PENDING is unset:
1827 	 * 	-> host disabled PV EOI.
1828 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1829 	 * 	-> host enabled PV EOI, guest did not execute EOI yet.
1830 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1831 	 * 	-> host enabled PV EOI, guest executed EOI.
1832 	 */
1833 	BUG_ON(!pv_eoi_enabled(vcpu));
1834 	pending = pv_eoi_get_pending(vcpu);
1835 	/*
1836 	 * Clear pending bit in any case: it will be set again on vmentry.
1837 	 * While this might not be ideal from performance point of view,
1838 	 * this makes sure pv eoi is only enabled when we know it's safe.
1839 	 */
1840 	pv_eoi_clr_pending(vcpu);
1841 	if (pending)
1842 		return;
1843 	vector = apic_set_eoi(apic);
1844 	trace_kvm_pv_eoi(apic, vector);
1845 }
1846 
1847 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1848 {
1849 	u32 data;
1850 
1851 	if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1852 		apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1853 
1854 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1855 		return;
1856 
1857 	kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1858 				sizeof(u32));
1859 
1860 	apic_set_tpr(vcpu->arch.apic, data & 0xff);
1861 }
1862 
1863 /*
1864  * apic_sync_pv_eoi_to_guest - called before vmentry
1865  *
1866  * Detect whether it's safe to enable PV EOI and
1867  * if yes do so.
1868  */
1869 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
1870 					struct kvm_lapic *apic)
1871 {
1872 	if (!pv_eoi_enabled(vcpu) ||
1873 	    /* IRR set or many bits in ISR: could be nested. */
1874 	    apic->irr_pending ||
1875 	    /* Cache not set: could be safe but we don't bother. */
1876 	    apic->highest_isr_cache == -1 ||
1877 	    /* Need EOI to update ioapic. */
1878 	    kvm_ioapic_handles_vector(vcpu->kvm, apic->highest_isr_cache)) {
1879 		/*
1880 		 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1881 		 * so we need not do anything here.
1882 		 */
1883 		return;
1884 	}
1885 
1886 	pv_eoi_set_pending(apic->vcpu);
1887 }
1888 
1889 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1890 {
1891 	u32 data, tpr;
1892 	int max_irr, max_isr;
1893 	struct kvm_lapic *apic = vcpu->arch.apic;
1894 
1895 	apic_sync_pv_eoi_to_guest(vcpu, apic);
1896 
1897 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1898 		return;
1899 
1900 	tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1901 	max_irr = apic_find_highest_irr(apic);
1902 	if (max_irr < 0)
1903 		max_irr = 0;
1904 	max_isr = apic_find_highest_isr(apic);
1905 	if (max_isr < 0)
1906 		max_isr = 0;
1907 	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1908 
1909 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1910 				sizeof(u32));
1911 }
1912 
1913 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1914 {
1915 	if (vapic_addr) {
1916 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1917 					&vcpu->arch.apic->vapic_cache,
1918 					vapic_addr, sizeof(u32)))
1919 			return -EINVAL;
1920 		__set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1921 	} else {
1922 		__clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1923 	}
1924 
1925 	vcpu->arch.apic->vapic_addr = vapic_addr;
1926 	return 0;
1927 }
1928 
1929 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1930 {
1931 	struct kvm_lapic *apic = vcpu->arch.apic;
1932 	u32 reg = (msr - APIC_BASE_MSR) << 4;
1933 
1934 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1935 		return 1;
1936 
1937 	if (reg == APIC_ICR2)
1938 		return 1;
1939 
1940 	/* if this is ICR write vector before command */
1941 	if (reg == APIC_ICR)
1942 		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1943 	return apic_reg_write(apic, reg, (u32)data);
1944 }
1945 
1946 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1947 {
1948 	struct kvm_lapic *apic = vcpu->arch.apic;
1949 	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1950 
1951 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1952 		return 1;
1953 
1954 	if (reg == APIC_DFR || reg == APIC_ICR2) {
1955 		apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
1956 			   reg);
1957 		return 1;
1958 	}
1959 
1960 	if (apic_reg_read(apic, reg, 4, &low))
1961 		return 1;
1962 	if (reg == APIC_ICR)
1963 		apic_reg_read(apic, APIC_ICR2, 4, &high);
1964 
1965 	*data = (((u64)high) << 32) | low;
1966 
1967 	return 0;
1968 }
1969 
1970 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1971 {
1972 	struct kvm_lapic *apic = vcpu->arch.apic;
1973 
1974 	if (!kvm_vcpu_has_lapic(vcpu))
1975 		return 1;
1976 
1977 	/* if this is ICR write vector before command */
1978 	if (reg == APIC_ICR)
1979 		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1980 	return apic_reg_write(apic, reg, (u32)data);
1981 }
1982 
1983 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
1984 {
1985 	struct kvm_lapic *apic = vcpu->arch.apic;
1986 	u32 low, high = 0;
1987 
1988 	if (!kvm_vcpu_has_lapic(vcpu))
1989 		return 1;
1990 
1991 	if (apic_reg_read(apic, reg, 4, &low))
1992 		return 1;
1993 	if (reg == APIC_ICR)
1994 		apic_reg_read(apic, APIC_ICR2, 4, &high);
1995 
1996 	*data = (((u64)high) << 32) | low;
1997 
1998 	return 0;
1999 }
2000 
2001 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2002 {
2003 	u64 addr = data & ~KVM_MSR_ENABLED;
2004 	if (!IS_ALIGNED(addr, 4))
2005 		return 1;
2006 
2007 	vcpu->arch.pv_eoi.msr_val = data;
2008 	if (!pv_eoi_enabled(vcpu))
2009 		return 0;
2010 	return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
2011 					 addr, sizeof(u8));
2012 }
2013 
2014 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2015 {
2016 	struct kvm_lapic *apic = vcpu->arch.apic;
2017 	u8 sipi_vector;
2018 	unsigned long pe;
2019 
2020 	if (!kvm_vcpu_has_lapic(vcpu) || !apic->pending_events)
2021 		return;
2022 
2023 	pe = xchg(&apic->pending_events, 0);
2024 
2025 	if (test_bit(KVM_APIC_INIT, &pe)) {
2026 		kvm_lapic_reset(vcpu);
2027 		kvm_vcpu_reset(vcpu);
2028 		if (kvm_vcpu_is_bsp(apic->vcpu))
2029 			vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2030 		else
2031 			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2032 	}
2033 	if (test_bit(KVM_APIC_SIPI, &pe) &&
2034 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2035 		/* evaluate pending_events before reading the vector */
2036 		smp_rmb();
2037 		sipi_vector = apic->sipi_vector;
2038 		apic_debug("vcpu %d received sipi with vector # %x\n",
2039 			 vcpu->vcpu_id, sipi_vector);
2040 		kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2041 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2042 	}
2043 }
2044 
2045 void kvm_lapic_init(void)
2046 {
2047 	/* do not patch jump label more than once per second */
2048 	jump_label_rate_limit(&apic_hw_disabled, HZ);
2049 	jump_label_rate_limit(&apic_sw_disabled, HZ);
2050 }
2051