xref: /openbmc/linux/arch/x86/kvm/lapic.c (revision d0659d946be05e098883b6955d2764595997f6a4)
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 		aid = kvm_apic_id(apic);
197 		ldr = kvm_apic_get_reg(apic, APIC_LDR);
198 		cid = apic_cluster_id(new, ldr);
199 		lid = apic_logical_id(new, ldr);
200 
201 		if (aid < ARRAY_SIZE(new->phys_map))
202 			new->phys_map[aid] = apic;
203 		if (lid && cid < ARRAY_SIZE(new->logical_map))
204 			new->logical_map[cid][ffs(lid) - 1] = apic;
205 	}
206 out:
207 	old = rcu_dereference_protected(kvm->arch.apic_map,
208 			lockdep_is_held(&kvm->arch.apic_map_lock));
209 	rcu_assign_pointer(kvm->arch.apic_map, new);
210 	mutex_unlock(&kvm->arch.apic_map_lock);
211 
212 	if (old)
213 		kfree_rcu(old, rcu);
214 
215 	kvm_vcpu_request_scan_ioapic(kvm);
216 }
217 
218 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
219 {
220 	bool enabled = val & APIC_SPIV_APIC_ENABLED;
221 
222 	apic_set_reg(apic, APIC_SPIV, val);
223 
224 	if (enabled != apic->sw_enabled) {
225 		apic->sw_enabled = enabled;
226 		if (enabled) {
227 			static_key_slow_dec_deferred(&apic_sw_disabled);
228 			recalculate_apic_map(apic->vcpu->kvm);
229 		} else
230 			static_key_slow_inc(&apic_sw_disabled.key);
231 	}
232 }
233 
234 static inline void kvm_apic_set_id(struct kvm_lapic *apic, u8 id)
235 {
236 	apic_set_reg(apic, APIC_ID, id << 24);
237 	recalculate_apic_map(apic->vcpu->kvm);
238 }
239 
240 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
241 {
242 	apic_set_reg(apic, APIC_LDR, id);
243 	recalculate_apic_map(apic->vcpu->kvm);
244 }
245 
246 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
247 {
248 	return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
249 }
250 
251 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
252 {
253 	return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
254 }
255 
256 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
257 {
258 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
259 }
260 
261 static inline int apic_lvtt_period(struct kvm_lapic *apic)
262 {
263 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
264 }
265 
266 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
267 {
268 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
269 }
270 
271 static inline int apic_lvt_nmi_mode(u32 lvt_val)
272 {
273 	return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
274 }
275 
276 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
277 {
278 	struct kvm_lapic *apic = vcpu->arch.apic;
279 	struct kvm_cpuid_entry2 *feat;
280 	u32 v = APIC_VERSION;
281 
282 	if (!kvm_vcpu_has_lapic(vcpu))
283 		return;
284 
285 	feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
286 	if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
287 		v |= APIC_LVR_DIRECTED_EOI;
288 	apic_set_reg(apic, APIC_LVR, v);
289 }
290 
291 static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
292 	LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
293 	LVT_MASK | APIC_MODE_MASK,	/* LVTTHMR */
294 	LVT_MASK | APIC_MODE_MASK,	/* LVTPC */
295 	LINT_MASK, LINT_MASK,	/* LVT0-1 */
296 	LVT_MASK		/* LVTERR */
297 };
298 
299 static int find_highest_vector(void *bitmap)
300 {
301 	int vec;
302 	u32 *reg;
303 
304 	for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
305 	     vec >= 0; vec -= APIC_VECTORS_PER_REG) {
306 		reg = bitmap + REG_POS(vec);
307 		if (*reg)
308 			return fls(*reg) - 1 + vec;
309 	}
310 
311 	return -1;
312 }
313 
314 static u8 count_vectors(void *bitmap)
315 {
316 	int vec;
317 	u32 *reg;
318 	u8 count = 0;
319 
320 	for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
321 		reg = bitmap + REG_POS(vec);
322 		count += hweight32(*reg);
323 	}
324 
325 	return count;
326 }
327 
328 void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
329 {
330 	u32 i, pir_val;
331 	struct kvm_lapic *apic = vcpu->arch.apic;
332 
333 	for (i = 0; i <= 7; i++) {
334 		pir_val = xchg(&pir[i], 0);
335 		if (pir_val)
336 			*((u32 *)(apic->regs + APIC_IRR + i * 0x10)) |= pir_val;
337 	}
338 }
339 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
340 
341 static inline void apic_set_irr(int vec, struct kvm_lapic *apic)
342 {
343 	apic_set_vector(vec, apic->regs + APIC_IRR);
344 	/*
345 	 * irr_pending must be true if any interrupt is pending; set it after
346 	 * APIC_IRR to avoid race with apic_clear_irr
347 	 */
348 	apic->irr_pending = true;
349 }
350 
351 static inline int apic_search_irr(struct kvm_lapic *apic)
352 {
353 	return find_highest_vector(apic->regs + APIC_IRR);
354 }
355 
356 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
357 {
358 	int result;
359 
360 	/*
361 	 * Note that irr_pending is just a hint. It will be always
362 	 * true with virtual interrupt delivery enabled.
363 	 */
364 	if (!apic->irr_pending)
365 		return -1;
366 
367 	kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
368 	result = apic_search_irr(apic);
369 	ASSERT(result == -1 || result >= 16);
370 
371 	return result;
372 }
373 
374 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
375 {
376 	struct kvm_vcpu *vcpu;
377 
378 	vcpu = apic->vcpu;
379 
380 	if (unlikely(kvm_apic_vid_enabled(vcpu->kvm))) {
381 		/* try to update RVI */
382 		apic_clear_vector(vec, apic->regs + APIC_IRR);
383 		kvm_make_request(KVM_REQ_EVENT, vcpu);
384 	} else {
385 		apic->irr_pending = false;
386 		apic_clear_vector(vec, apic->regs + APIC_IRR);
387 		if (apic_search_irr(apic) != -1)
388 			apic->irr_pending = true;
389 	}
390 }
391 
392 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
393 {
394 	struct kvm_vcpu *vcpu;
395 
396 	if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
397 		return;
398 
399 	vcpu = apic->vcpu;
400 
401 	/*
402 	 * With APIC virtualization enabled, all caching is disabled
403 	 * because the processor can modify ISR under the hood.  Instead
404 	 * just set SVI.
405 	 */
406 	if (unlikely(kvm_x86_ops->hwapic_isr_update))
407 		kvm_x86_ops->hwapic_isr_update(vcpu->kvm, vec);
408 	else {
409 		++apic->isr_count;
410 		BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
411 		/*
412 		 * ISR (in service register) bit is set when injecting an interrupt.
413 		 * The highest vector is injected. Thus the latest bit set matches
414 		 * the highest bit in ISR.
415 		 */
416 		apic->highest_isr_cache = vec;
417 	}
418 }
419 
420 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
421 {
422 	int result;
423 
424 	/*
425 	 * Note that isr_count is always 1, and highest_isr_cache
426 	 * is always -1, with APIC virtualization enabled.
427 	 */
428 	if (!apic->isr_count)
429 		return -1;
430 	if (likely(apic->highest_isr_cache != -1))
431 		return apic->highest_isr_cache;
432 
433 	result = find_highest_vector(apic->regs + APIC_ISR);
434 	ASSERT(result == -1 || result >= 16);
435 
436 	return result;
437 }
438 
439 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
440 {
441 	struct kvm_vcpu *vcpu;
442 	if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
443 		return;
444 
445 	vcpu = apic->vcpu;
446 
447 	/*
448 	 * We do get here for APIC virtualization enabled if the guest
449 	 * uses the Hyper-V APIC enlightenment.  In this case we may need
450 	 * to trigger a new interrupt delivery by writing the SVI field;
451 	 * on the other hand isr_count and highest_isr_cache are unused
452 	 * and must be left alone.
453 	 */
454 	if (unlikely(kvm_x86_ops->hwapic_isr_update))
455 		kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
456 					       apic_find_highest_isr(apic));
457 	else {
458 		--apic->isr_count;
459 		BUG_ON(apic->isr_count < 0);
460 		apic->highest_isr_cache = -1;
461 	}
462 }
463 
464 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
465 {
466 	int highest_irr;
467 
468 	/* This may race with setting of irr in __apic_accept_irq() and
469 	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
470 	 * will cause vmexit immediately and the value will be recalculated
471 	 * on the next vmentry.
472 	 */
473 	if (!kvm_vcpu_has_lapic(vcpu))
474 		return 0;
475 	highest_irr = apic_find_highest_irr(vcpu->arch.apic);
476 
477 	return highest_irr;
478 }
479 
480 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
481 			     int vector, int level, int trig_mode,
482 			     unsigned long *dest_map);
483 
484 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
485 		unsigned long *dest_map)
486 {
487 	struct kvm_lapic *apic = vcpu->arch.apic;
488 
489 	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
490 			irq->level, irq->trig_mode, dest_map);
491 }
492 
493 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
494 {
495 
496 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
497 				      sizeof(val));
498 }
499 
500 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
501 {
502 
503 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
504 				      sizeof(*val));
505 }
506 
507 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
508 {
509 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
510 }
511 
512 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
513 {
514 	u8 val;
515 	if (pv_eoi_get_user(vcpu, &val) < 0)
516 		apic_debug("Can't read EOI MSR value: 0x%llx\n",
517 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
518 	return val & 0x1;
519 }
520 
521 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
522 {
523 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
524 		apic_debug("Can't set EOI MSR value: 0x%llx\n",
525 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
526 		return;
527 	}
528 	__set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
529 }
530 
531 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
532 {
533 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
534 		apic_debug("Can't clear EOI MSR value: 0x%llx\n",
535 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
536 		return;
537 	}
538 	__clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
539 }
540 
541 void kvm_apic_update_tmr(struct kvm_vcpu *vcpu, u32 *tmr)
542 {
543 	struct kvm_lapic *apic = vcpu->arch.apic;
544 	int i;
545 
546 	for (i = 0; i < 8; i++)
547 		apic_set_reg(apic, APIC_TMR + 0x10 * i, tmr[i]);
548 }
549 
550 static void apic_update_ppr(struct kvm_lapic *apic)
551 {
552 	u32 tpr, isrv, ppr, old_ppr;
553 	int isr;
554 
555 	old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI);
556 	tpr = kvm_apic_get_reg(apic, APIC_TASKPRI);
557 	isr = apic_find_highest_isr(apic);
558 	isrv = (isr != -1) ? isr : 0;
559 
560 	if ((tpr & 0xf0) >= (isrv & 0xf0))
561 		ppr = tpr & 0xff;
562 	else
563 		ppr = isrv & 0xf0;
564 
565 	apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
566 		   apic, ppr, isr, isrv);
567 
568 	if (old_ppr != ppr) {
569 		apic_set_reg(apic, APIC_PROCPRI, ppr);
570 		if (ppr < old_ppr)
571 			kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
572 	}
573 }
574 
575 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
576 {
577 	apic_set_reg(apic, APIC_TASKPRI, tpr);
578 	apic_update_ppr(apic);
579 }
580 
581 static int kvm_apic_broadcast(struct kvm_lapic *apic, u32 dest)
582 {
583 	return dest == (apic_x2apic_mode(apic) ?
584 			X2APIC_BROADCAST : APIC_BROADCAST);
585 }
586 
587 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 dest)
588 {
589 	return kvm_apic_id(apic) == dest || kvm_apic_broadcast(apic, dest);
590 }
591 
592 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
593 {
594 	int result = 0;
595 	u32 logical_id;
596 
597 	if (kvm_apic_broadcast(apic, mda))
598 		return 1;
599 
600 	if (apic_x2apic_mode(apic)) {
601 		logical_id = kvm_apic_get_reg(apic, APIC_LDR);
602 		return logical_id & mda;
603 	}
604 
605 	logical_id = GET_APIC_LOGICAL_ID(kvm_apic_get_reg(apic, APIC_LDR));
606 
607 	switch (kvm_apic_get_reg(apic, APIC_DFR)) {
608 	case APIC_DFR_FLAT:
609 		if (logical_id & mda)
610 			result = 1;
611 		break;
612 	case APIC_DFR_CLUSTER:
613 		if (((logical_id >> 4) == (mda >> 0x4))
614 		    && (logical_id & mda & 0xf))
615 			result = 1;
616 		break;
617 	default:
618 		apic_debug("Bad DFR vcpu %d: %08x\n",
619 			   apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR));
620 		break;
621 	}
622 
623 	return result;
624 }
625 
626 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
627 			   int short_hand, unsigned int dest, int dest_mode)
628 {
629 	int result = 0;
630 	struct kvm_lapic *target = vcpu->arch.apic;
631 
632 	apic_debug("target %p, source %p, dest 0x%x, "
633 		   "dest_mode 0x%x, short_hand 0x%x\n",
634 		   target, source, dest, dest_mode, short_hand);
635 
636 	ASSERT(target);
637 	switch (short_hand) {
638 	case APIC_DEST_NOSHORT:
639 		if (dest_mode == 0)
640 			/* Physical mode. */
641 			result = kvm_apic_match_physical_addr(target, dest);
642 		else
643 			/* Logical mode. */
644 			result = kvm_apic_match_logical_addr(target, dest);
645 		break;
646 	case APIC_DEST_SELF:
647 		result = (target == source);
648 		break;
649 	case APIC_DEST_ALLINC:
650 		result = 1;
651 		break;
652 	case APIC_DEST_ALLBUT:
653 		result = (target != source);
654 		break;
655 	default:
656 		apic_debug("kvm: apic: Bad dest shorthand value %x\n",
657 			   short_hand);
658 		break;
659 	}
660 
661 	return result;
662 }
663 
664 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
665 		struct kvm_lapic_irq *irq, int *r, unsigned long *dest_map)
666 {
667 	struct kvm_apic_map *map;
668 	unsigned long bitmap = 1;
669 	struct kvm_lapic **dst;
670 	int i;
671 	bool ret = false;
672 
673 	*r = -1;
674 
675 	if (irq->shorthand == APIC_DEST_SELF) {
676 		*r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
677 		return true;
678 	}
679 
680 	if (irq->shorthand)
681 		return false;
682 
683 	rcu_read_lock();
684 	map = rcu_dereference(kvm->arch.apic_map);
685 
686 	if (!map)
687 		goto out;
688 
689 	if (irq->dest_id == map->broadcast)
690 		goto out;
691 
692 	ret = true;
693 
694 	if (irq->dest_mode == 0) { /* physical mode */
695 		if (irq->dest_id >= ARRAY_SIZE(map->phys_map))
696 			goto out;
697 
698 		dst = &map->phys_map[irq->dest_id];
699 	} else {
700 		u32 mda = irq->dest_id << (32 - map->ldr_bits);
701 		u16 cid = apic_cluster_id(map, mda);
702 
703 		if (cid >= ARRAY_SIZE(map->logical_map))
704 			goto out;
705 
706 		dst = map->logical_map[cid];
707 
708 		bitmap = apic_logical_id(map, mda);
709 
710 		if (irq->delivery_mode == APIC_DM_LOWEST) {
711 			int l = -1;
712 			for_each_set_bit(i, &bitmap, 16) {
713 				if (!dst[i])
714 					continue;
715 				if (l < 0)
716 					l = i;
717 				else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0)
718 					l = i;
719 			}
720 
721 			bitmap = (l >= 0) ? 1 << l : 0;
722 		}
723 	}
724 
725 	for_each_set_bit(i, &bitmap, 16) {
726 		if (!dst[i])
727 			continue;
728 		if (*r < 0)
729 			*r = 0;
730 		*r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
731 	}
732 out:
733 	rcu_read_unlock();
734 	return ret;
735 }
736 
737 /*
738  * Add a pending IRQ into lapic.
739  * Return 1 if successfully added and 0 if discarded.
740  */
741 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
742 			     int vector, int level, int trig_mode,
743 			     unsigned long *dest_map)
744 {
745 	int result = 0;
746 	struct kvm_vcpu *vcpu = apic->vcpu;
747 
748 	trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
749 				  trig_mode, vector);
750 	switch (delivery_mode) {
751 	case APIC_DM_LOWEST:
752 		vcpu->arch.apic_arb_prio++;
753 	case APIC_DM_FIXED:
754 		/* FIXME add logic for vcpu on reset */
755 		if (unlikely(!apic_enabled(apic)))
756 			break;
757 
758 		result = 1;
759 
760 		if (dest_map)
761 			__set_bit(vcpu->vcpu_id, dest_map);
762 
763 		if (kvm_x86_ops->deliver_posted_interrupt)
764 			kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
765 		else {
766 			apic_set_irr(vector, apic);
767 
768 			kvm_make_request(KVM_REQ_EVENT, vcpu);
769 			kvm_vcpu_kick(vcpu);
770 		}
771 		break;
772 
773 	case APIC_DM_REMRD:
774 		result = 1;
775 		vcpu->arch.pv.pv_unhalted = 1;
776 		kvm_make_request(KVM_REQ_EVENT, vcpu);
777 		kvm_vcpu_kick(vcpu);
778 		break;
779 
780 	case APIC_DM_SMI:
781 		apic_debug("Ignoring guest SMI\n");
782 		break;
783 
784 	case APIC_DM_NMI:
785 		result = 1;
786 		kvm_inject_nmi(vcpu);
787 		kvm_vcpu_kick(vcpu);
788 		break;
789 
790 	case APIC_DM_INIT:
791 		if (!trig_mode || level) {
792 			result = 1;
793 			/* assumes that there are only KVM_APIC_INIT/SIPI */
794 			apic->pending_events = (1UL << KVM_APIC_INIT);
795 			/* make sure pending_events is visible before sending
796 			 * the request */
797 			smp_wmb();
798 			kvm_make_request(KVM_REQ_EVENT, vcpu);
799 			kvm_vcpu_kick(vcpu);
800 		} else {
801 			apic_debug("Ignoring de-assert INIT to vcpu %d\n",
802 				   vcpu->vcpu_id);
803 		}
804 		break;
805 
806 	case APIC_DM_STARTUP:
807 		apic_debug("SIPI to vcpu %d vector 0x%02x\n",
808 			   vcpu->vcpu_id, vector);
809 		result = 1;
810 		apic->sipi_vector = vector;
811 		/* make sure sipi_vector is visible for the receiver */
812 		smp_wmb();
813 		set_bit(KVM_APIC_SIPI, &apic->pending_events);
814 		kvm_make_request(KVM_REQ_EVENT, vcpu);
815 		kvm_vcpu_kick(vcpu);
816 		break;
817 
818 	case APIC_DM_EXTINT:
819 		/*
820 		 * Should only be called by kvm_apic_local_deliver() with LVT0,
821 		 * before NMI watchdog was enabled. Already handled by
822 		 * kvm_apic_accept_pic_intr().
823 		 */
824 		break;
825 
826 	default:
827 		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
828 		       delivery_mode);
829 		break;
830 	}
831 	return result;
832 }
833 
834 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
835 {
836 	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
837 }
838 
839 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
840 {
841 	if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI) &&
842 	    kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
843 		int trigger_mode;
844 		if (apic_test_vector(vector, apic->regs + APIC_TMR))
845 			trigger_mode = IOAPIC_LEVEL_TRIG;
846 		else
847 			trigger_mode = IOAPIC_EDGE_TRIG;
848 		kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
849 	}
850 }
851 
852 static int apic_set_eoi(struct kvm_lapic *apic)
853 {
854 	int vector = apic_find_highest_isr(apic);
855 
856 	trace_kvm_eoi(apic, vector);
857 
858 	/*
859 	 * Not every write EOI will has corresponding ISR,
860 	 * one example is when Kernel check timer on setup_IO_APIC
861 	 */
862 	if (vector == -1)
863 		return vector;
864 
865 	apic_clear_isr(vector, apic);
866 	apic_update_ppr(apic);
867 
868 	kvm_ioapic_send_eoi(apic, vector);
869 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
870 	return vector;
871 }
872 
873 /*
874  * this interface assumes a trap-like exit, which has already finished
875  * desired side effect including vISR and vPPR update.
876  */
877 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
878 {
879 	struct kvm_lapic *apic = vcpu->arch.apic;
880 
881 	trace_kvm_eoi(apic, vector);
882 
883 	kvm_ioapic_send_eoi(apic, vector);
884 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
885 }
886 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
887 
888 static void apic_send_ipi(struct kvm_lapic *apic)
889 {
890 	u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR);
891 	u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2);
892 	struct kvm_lapic_irq irq;
893 
894 	irq.vector = icr_low & APIC_VECTOR_MASK;
895 	irq.delivery_mode = icr_low & APIC_MODE_MASK;
896 	irq.dest_mode = icr_low & APIC_DEST_MASK;
897 	irq.level = icr_low & APIC_INT_ASSERT;
898 	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
899 	irq.shorthand = icr_low & APIC_SHORT_MASK;
900 	if (apic_x2apic_mode(apic))
901 		irq.dest_id = icr_high;
902 	else
903 		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
904 
905 	trace_kvm_apic_ipi(icr_low, irq.dest_id);
906 
907 	apic_debug("icr_high 0x%x, icr_low 0x%x, "
908 		   "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
909 		   "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
910 		   icr_high, icr_low, irq.shorthand, irq.dest_id,
911 		   irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
912 		   irq.vector);
913 
914 	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
915 }
916 
917 static u32 apic_get_tmcct(struct kvm_lapic *apic)
918 {
919 	ktime_t remaining;
920 	s64 ns;
921 	u32 tmcct;
922 
923 	ASSERT(apic != NULL);
924 
925 	/* if initial count is 0, current count should also be 0 */
926 	if (kvm_apic_get_reg(apic, APIC_TMICT) == 0 ||
927 		apic->lapic_timer.period == 0)
928 		return 0;
929 
930 	remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
931 	if (ktime_to_ns(remaining) < 0)
932 		remaining = ktime_set(0, 0);
933 
934 	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
935 	tmcct = div64_u64(ns,
936 			 (APIC_BUS_CYCLE_NS * apic->divide_count));
937 
938 	return tmcct;
939 }
940 
941 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
942 {
943 	struct kvm_vcpu *vcpu = apic->vcpu;
944 	struct kvm_run *run = vcpu->run;
945 
946 	kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
947 	run->tpr_access.rip = kvm_rip_read(vcpu);
948 	run->tpr_access.is_write = write;
949 }
950 
951 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
952 {
953 	if (apic->vcpu->arch.tpr_access_reporting)
954 		__report_tpr_access(apic, write);
955 }
956 
957 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
958 {
959 	u32 val = 0;
960 
961 	if (offset >= LAPIC_MMIO_LENGTH)
962 		return 0;
963 
964 	switch (offset) {
965 	case APIC_ID:
966 		if (apic_x2apic_mode(apic))
967 			val = kvm_apic_id(apic);
968 		else
969 			val = kvm_apic_id(apic) << 24;
970 		break;
971 	case APIC_ARBPRI:
972 		apic_debug("Access APIC ARBPRI register which is for P6\n");
973 		break;
974 
975 	case APIC_TMCCT:	/* Timer CCR */
976 		if (apic_lvtt_tscdeadline(apic))
977 			return 0;
978 
979 		val = apic_get_tmcct(apic);
980 		break;
981 	case APIC_PROCPRI:
982 		apic_update_ppr(apic);
983 		val = kvm_apic_get_reg(apic, offset);
984 		break;
985 	case APIC_TASKPRI:
986 		report_tpr_access(apic, false);
987 		/* fall thru */
988 	default:
989 		val = kvm_apic_get_reg(apic, offset);
990 		break;
991 	}
992 
993 	return val;
994 }
995 
996 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
997 {
998 	return container_of(dev, struct kvm_lapic, dev);
999 }
1000 
1001 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1002 		void *data)
1003 {
1004 	unsigned char alignment = offset & 0xf;
1005 	u32 result;
1006 	/* this bitmask has a bit cleared for each reserved register */
1007 	static const u64 rmask = 0x43ff01ffffffe70cULL;
1008 
1009 	if ((alignment + len) > 4) {
1010 		apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1011 			   offset, len);
1012 		return 1;
1013 	}
1014 
1015 	if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1016 		apic_debug("KVM_APIC_READ: read reserved register %x\n",
1017 			   offset);
1018 		return 1;
1019 	}
1020 
1021 	result = __apic_read(apic, offset & ~0xf);
1022 
1023 	trace_kvm_apic_read(offset, result);
1024 
1025 	switch (len) {
1026 	case 1:
1027 	case 2:
1028 	case 4:
1029 		memcpy(data, (char *)&result + alignment, len);
1030 		break;
1031 	default:
1032 		printk(KERN_ERR "Local APIC read with len = %x, "
1033 		       "should be 1,2, or 4 instead\n", len);
1034 		break;
1035 	}
1036 	return 0;
1037 }
1038 
1039 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1040 {
1041 	return kvm_apic_hw_enabled(apic) &&
1042 	    addr >= apic->base_address &&
1043 	    addr < apic->base_address + LAPIC_MMIO_LENGTH;
1044 }
1045 
1046 static int apic_mmio_read(struct kvm_io_device *this,
1047 			   gpa_t address, int len, void *data)
1048 {
1049 	struct kvm_lapic *apic = to_lapic(this);
1050 	u32 offset = address - apic->base_address;
1051 
1052 	if (!apic_mmio_in_range(apic, address))
1053 		return -EOPNOTSUPP;
1054 
1055 	apic_reg_read(apic, offset, len, data);
1056 
1057 	return 0;
1058 }
1059 
1060 static void update_divide_count(struct kvm_lapic *apic)
1061 {
1062 	u32 tmp1, tmp2, tdcr;
1063 
1064 	tdcr = kvm_apic_get_reg(apic, APIC_TDCR);
1065 	tmp1 = tdcr & 0xf;
1066 	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1067 	apic->divide_count = 0x1 << (tmp2 & 0x7);
1068 
1069 	apic_debug("timer divide count is 0x%x\n",
1070 				   apic->divide_count);
1071 }
1072 
1073 static void apic_timer_expired(struct kvm_lapic *apic)
1074 {
1075 	struct kvm_vcpu *vcpu = apic->vcpu;
1076 	wait_queue_head_t *q = &vcpu->wq;
1077 	struct kvm_timer *ktimer = &apic->lapic_timer;
1078 
1079 	/*
1080 	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1081 	 * vcpu_enter_guest.
1082 	 */
1083 	if (atomic_read(&apic->lapic_timer.pending))
1084 		return;
1085 
1086 	atomic_inc(&apic->lapic_timer.pending);
1087 	/* FIXME: this code should not know anything about vcpus */
1088 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1089 
1090 	if (waitqueue_active(q))
1091 		wake_up_interruptible(q);
1092 
1093 	if (apic_lvtt_tscdeadline(apic))
1094 		ktimer->expired_tscdeadline = ktimer->tscdeadline;
1095 }
1096 
1097 /*
1098  * On APICv, this test will cause a busy wait
1099  * during a higher-priority task.
1100  */
1101 
1102 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1103 {
1104 	struct kvm_lapic *apic = vcpu->arch.apic;
1105 	u32 reg = kvm_apic_get_reg(apic, APIC_LVTT);
1106 
1107 	if (kvm_apic_hw_enabled(apic)) {
1108 		int vec = reg & APIC_VECTOR_MASK;
1109 
1110 		if (kvm_x86_ops->test_posted_interrupt)
1111 			return kvm_x86_ops->test_posted_interrupt(vcpu, vec);
1112 		else {
1113 			if (apic_test_vector(vec, apic->regs + APIC_ISR))
1114 				return true;
1115 		}
1116 	}
1117 	return false;
1118 }
1119 
1120 void wait_lapic_expire(struct kvm_vcpu *vcpu)
1121 {
1122 	struct kvm_lapic *apic = vcpu->arch.apic;
1123 	u64 guest_tsc, tsc_deadline;
1124 
1125 	if (!kvm_vcpu_has_lapic(vcpu))
1126 		return;
1127 
1128 	if (apic->lapic_timer.expired_tscdeadline == 0)
1129 		return;
1130 
1131 	if (!lapic_timer_int_injected(vcpu))
1132 		return;
1133 
1134 	tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1135 	apic->lapic_timer.expired_tscdeadline = 0;
1136 	guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1137 
1138 	/* __delay is delay_tsc whenever the hardware has TSC, thus always.  */
1139 	if (guest_tsc < tsc_deadline)
1140 		__delay(tsc_deadline - guest_tsc);
1141 }
1142 
1143 static void start_apic_timer(struct kvm_lapic *apic)
1144 {
1145 	ktime_t now;
1146 
1147 	atomic_set(&apic->lapic_timer.pending, 0);
1148 
1149 	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1150 		/* lapic timer in oneshot or periodic mode */
1151 		now = apic->lapic_timer.timer.base->get_time();
1152 		apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT)
1153 			    * APIC_BUS_CYCLE_NS * apic->divide_count;
1154 
1155 		if (!apic->lapic_timer.period)
1156 			return;
1157 		/*
1158 		 * Do not allow the guest to program periodic timers with small
1159 		 * interval, since the hrtimers are not throttled by the host
1160 		 * scheduler.
1161 		 */
1162 		if (apic_lvtt_period(apic)) {
1163 			s64 min_period = min_timer_period_us * 1000LL;
1164 
1165 			if (apic->lapic_timer.period < min_period) {
1166 				pr_info_ratelimited(
1167 				    "kvm: vcpu %i: requested %lld ns "
1168 				    "lapic timer period limited to %lld ns\n",
1169 				    apic->vcpu->vcpu_id,
1170 				    apic->lapic_timer.period, min_period);
1171 				apic->lapic_timer.period = min_period;
1172 			}
1173 		}
1174 
1175 		hrtimer_start(&apic->lapic_timer.timer,
1176 			      ktime_add_ns(now, apic->lapic_timer.period),
1177 			      HRTIMER_MODE_ABS);
1178 
1179 		apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1180 			   PRIx64 ", "
1181 			   "timer initial count 0x%x, period %lldns, "
1182 			   "expire @ 0x%016" PRIx64 ".\n", __func__,
1183 			   APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1184 			   kvm_apic_get_reg(apic, APIC_TMICT),
1185 			   apic->lapic_timer.period,
1186 			   ktime_to_ns(ktime_add_ns(now,
1187 					apic->lapic_timer.period)));
1188 	} else if (apic_lvtt_tscdeadline(apic)) {
1189 		/* lapic timer in tsc deadline mode */
1190 		u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1191 		u64 ns = 0;
1192 		ktime_t expire;
1193 		struct kvm_vcpu *vcpu = apic->vcpu;
1194 		unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1195 		unsigned long flags;
1196 
1197 		if (unlikely(!tscdeadline || !this_tsc_khz))
1198 			return;
1199 
1200 		local_irq_save(flags);
1201 
1202 		now = apic->lapic_timer.timer.base->get_time();
1203 		guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1204 		if (likely(tscdeadline > guest_tsc)) {
1205 			ns = (tscdeadline - guest_tsc) * 1000000ULL;
1206 			do_div(ns, this_tsc_khz);
1207 			expire = ktime_add_ns(now, ns);
1208 			expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1209 			hrtimer_start(&apic->lapic_timer.timer,
1210 				      expire, HRTIMER_MODE_ABS);
1211 		} else
1212 			apic_timer_expired(apic);
1213 
1214 		local_irq_restore(flags);
1215 	}
1216 }
1217 
1218 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1219 {
1220 	int nmi_wd_enabled = apic_lvt_nmi_mode(kvm_apic_get_reg(apic, APIC_LVT0));
1221 
1222 	if (apic_lvt_nmi_mode(lvt0_val)) {
1223 		if (!nmi_wd_enabled) {
1224 			apic_debug("Receive NMI setting on APIC_LVT0 "
1225 				   "for cpu %d\n", apic->vcpu->vcpu_id);
1226 			apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
1227 		}
1228 	} else if (nmi_wd_enabled)
1229 		apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
1230 }
1231 
1232 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1233 {
1234 	int ret = 0;
1235 
1236 	trace_kvm_apic_write(reg, val);
1237 
1238 	switch (reg) {
1239 	case APIC_ID:		/* Local APIC ID */
1240 		if (!apic_x2apic_mode(apic))
1241 			kvm_apic_set_id(apic, val >> 24);
1242 		else
1243 			ret = 1;
1244 		break;
1245 
1246 	case APIC_TASKPRI:
1247 		report_tpr_access(apic, true);
1248 		apic_set_tpr(apic, val & 0xff);
1249 		break;
1250 
1251 	case APIC_EOI:
1252 		apic_set_eoi(apic);
1253 		break;
1254 
1255 	case APIC_LDR:
1256 		if (!apic_x2apic_mode(apic))
1257 			kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1258 		else
1259 			ret = 1;
1260 		break;
1261 
1262 	case APIC_DFR:
1263 		if (!apic_x2apic_mode(apic)) {
1264 			apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1265 			recalculate_apic_map(apic->vcpu->kvm);
1266 		} else
1267 			ret = 1;
1268 		break;
1269 
1270 	case APIC_SPIV: {
1271 		u32 mask = 0x3ff;
1272 		if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1273 			mask |= APIC_SPIV_DIRECTED_EOI;
1274 		apic_set_spiv(apic, val & mask);
1275 		if (!(val & APIC_SPIV_APIC_ENABLED)) {
1276 			int i;
1277 			u32 lvt_val;
1278 
1279 			for (i = 0; i < APIC_LVT_NUM; i++) {
1280 				lvt_val = kvm_apic_get_reg(apic,
1281 						       APIC_LVTT + 0x10 * i);
1282 				apic_set_reg(apic, APIC_LVTT + 0x10 * i,
1283 					     lvt_val | APIC_LVT_MASKED);
1284 			}
1285 			atomic_set(&apic->lapic_timer.pending, 0);
1286 
1287 		}
1288 		break;
1289 	}
1290 	case APIC_ICR:
1291 		/* No delay here, so we always clear the pending bit */
1292 		apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1293 		apic_send_ipi(apic);
1294 		break;
1295 
1296 	case APIC_ICR2:
1297 		if (!apic_x2apic_mode(apic))
1298 			val &= 0xff000000;
1299 		apic_set_reg(apic, APIC_ICR2, val);
1300 		break;
1301 
1302 	case APIC_LVT0:
1303 		apic_manage_nmi_watchdog(apic, val);
1304 	case APIC_LVTTHMR:
1305 	case APIC_LVTPC:
1306 	case APIC_LVT1:
1307 	case APIC_LVTERR:
1308 		/* TODO: Check vector */
1309 		if (!kvm_apic_sw_enabled(apic))
1310 			val |= APIC_LVT_MASKED;
1311 
1312 		val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1313 		apic_set_reg(apic, reg, val);
1314 
1315 		break;
1316 
1317 	case APIC_LVTT: {
1318 		u32 timer_mode = val & apic->lapic_timer.timer_mode_mask;
1319 
1320 		if (apic->lapic_timer.timer_mode != timer_mode) {
1321 			apic->lapic_timer.timer_mode = timer_mode;
1322 			hrtimer_cancel(&apic->lapic_timer.timer);
1323 		}
1324 
1325 		if (!kvm_apic_sw_enabled(apic))
1326 			val |= APIC_LVT_MASKED;
1327 		val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1328 		apic_set_reg(apic, APIC_LVTT, val);
1329 		break;
1330 	}
1331 
1332 	case APIC_TMICT:
1333 		if (apic_lvtt_tscdeadline(apic))
1334 			break;
1335 
1336 		hrtimer_cancel(&apic->lapic_timer.timer);
1337 		apic_set_reg(apic, APIC_TMICT, val);
1338 		start_apic_timer(apic);
1339 		break;
1340 
1341 	case APIC_TDCR:
1342 		if (val & 4)
1343 			apic_debug("KVM_WRITE:TDCR %x\n", val);
1344 		apic_set_reg(apic, APIC_TDCR, val);
1345 		update_divide_count(apic);
1346 		break;
1347 
1348 	case APIC_ESR:
1349 		if (apic_x2apic_mode(apic) && val != 0) {
1350 			apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1351 			ret = 1;
1352 		}
1353 		break;
1354 
1355 	case APIC_SELF_IPI:
1356 		if (apic_x2apic_mode(apic)) {
1357 			apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1358 		} else
1359 			ret = 1;
1360 		break;
1361 	default:
1362 		ret = 1;
1363 		break;
1364 	}
1365 	if (ret)
1366 		apic_debug("Local APIC Write to read-only register %x\n", reg);
1367 	return ret;
1368 }
1369 
1370 static int apic_mmio_write(struct kvm_io_device *this,
1371 			    gpa_t address, int len, const void *data)
1372 {
1373 	struct kvm_lapic *apic = to_lapic(this);
1374 	unsigned int offset = address - apic->base_address;
1375 	u32 val;
1376 
1377 	if (!apic_mmio_in_range(apic, address))
1378 		return -EOPNOTSUPP;
1379 
1380 	/*
1381 	 * APIC register must be aligned on 128-bits boundary.
1382 	 * 32/64/128 bits registers must be accessed thru 32 bits.
1383 	 * Refer SDM 8.4.1
1384 	 */
1385 	if (len != 4 || (offset & 0xf)) {
1386 		/* Don't shout loud, $infamous_os would cause only noise. */
1387 		apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1388 		return 0;
1389 	}
1390 
1391 	val = *(u32*)data;
1392 
1393 	/* too common printing */
1394 	if (offset != APIC_EOI)
1395 		apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1396 			   "0x%x\n", __func__, offset, len, val);
1397 
1398 	apic_reg_write(apic, offset & 0xff0, val);
1399 
1400 	return 0;
1401 }
1402 
1403 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1404 {
1405 	if (kvm_vcpu_has_lapic(vcpu))
1406 		apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1407 }
1408 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1409 
1410 /* emulate APIC access in a trap manner */
1411 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1412 {
1413 	u32 val = 0;
1414 
1415 	/* hw has done the conditional check and inst decode */
1416 	offset &= 0xff0;
1417 
1418 	apic_reg_read(vcpu->arch.apic, offset, 4, &val);
1419 
1420 	/* TODO: optimize to just emulate side effect w/o one more write */
1421 	apic_reg_write(vcpu->arch.apic, offset, val);
1422 }
1423 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1424 
1425 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1426 {
1427 	struct kvm_lapic *apic = vcpu->arch.apic;
1428 
1429 	if (!vcpu->arch.apic)
1430 		return;
1431 
1432 	hrtimer_cancel(&apic->lapic_timer.timer);
1433 
1434 	if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1435 		static_key_slow_dec_deferred(&apic_hw_disabled);
1436 
1437 	if (!apic->sw_enabled)
1438 		static_key_slow_dec_deferred(&apic_sw_disabled);
1439 
1440 	if (apic->regs)
1441 		free_page((unsigned long)apic->regs);
1442 
1443 	kfree(apic);
1444 }
1445 
1446 /*
1447  *----------------------------------------------------------------------
1448  * LAPIC interface
1449  *----------------------------------------------------------------------
1450  */
1451 
1452 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1453 {
1454 	struct kvm_lapic *apic = vcpu->arch.apic;
1455 
1456 	if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1457 			apic_lvtt_period(apic))
1458 		return 0;
1459 
1460 	return apic->lapic_timer.tscdeadline;
1461 }
1462 
1463 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1464 {
1465 	struct kvm_lapic *apic = vcpu->arch.apic;
1466 
1467 	if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1468 			apic_lvtt_period(apic))
1469 		return;
1470 
1471 	hrtimer_cancel(&apic->lapic_timer.timer);
1472 	apic->lapic_timer.tscdeadline = data;
1473 	start_apic_timer(apic);
1474 }
1475 
1476 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1477 {
1478 	struct kvm_lapic *apic = vcpu->arch.apic;
1479 
1480 	if (!kvm_vcpu_has_lapic(vcpu))
1481 		return;
1482 
1483 	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1484 		     | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4));
1485 }
1486 
1487 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1488 {
1489 	u64 tpr;
1490 
1491 	if (!kvm_vcpu_has_lapic(vcpu))
1492 		return 0;
1493 
1494 	tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1495 
1496 	return (tpr & 0xf0) >> 4;
1497 }
1498 
1499 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1500 {
1501 	u64 old_value = vcpu->arch.apic_base;
1502 	struct kvm_lapic *apic = vcpu->arch.apic;
1503 
1504 	if (!apic) {
1505 		value |= MSR_IA32_APICBASE_BSP;
1506 		vcpu->arch.apic_base = value;
1507 		return;
1508 	}
1509 
1510 	if (!kvm_vcpu_is_bsp(apic->vcpu))
1511 		value &= ~MSR_IA32_APICBASE_BSP;
1512 	vcpu->arch.apic_base = value;
1513 
1514 	/* update jump label if enable bit changes */
1515 	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1516 		if (value & MSR_IA32_APICBASE_ENABLE)
1517 			static_key_slow_dec_deferred(&apic_hw_disabled);
1518 		else
1519 			static_key_slow_inc(&apic_hw_disabled.key);
1520 		recalculate_apic_map(vcpu->kvm);
1521 	}
1522 
1523 	if ((old_value ^ value) & X2APIC_ENABLE) {
1524 		if (value & X2APIC_ENABLE) {
1525 			u32 id = kvm_apic_id(apic);
1526 			u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
1527 			kvm_apic_set_ldr(apic, ldr);
1528 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1529 		} else
1530 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1531 	}
1532 
1533 	apic->base_address = apic->vcpu->arch.apic_base &
1534 			     MSR_IA32_APICBASE_BASE;
1535 
1536 	if ((value & MSR_IA32_APICBASE_ENABLE) &&
1537 	     apic->base_address != APIC_DEFAULT_PHYS_BASE)
1538 		pr_warn_once("APIC base relocation is unsupported by KVM");
1539 
1540 	/* with FSB delivery interrupt, we can restart APIC functionality */
1541 	apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1542 		   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1543 
1544 }
1545 
1546 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1547 {
1548 	struct kvm_lapic *apic;
1549 	int i;
1550 
1551 	apic_debug("%s\n", __func__);
1552 
1553 	ASSERT(vcpu);
1554 	apic = vcpu->arch.apic;
1555 	ASSERT(apic != NULL);
1556 
1557 	/* Stop the timer in case it's a reset to an active apic */
1558 	hrtimer_cancel(&apic->lapic_timer.timer);
1559 
1560 	kvm_apic_set_id(apic, vcpu->vcpu_id);
1561 	kvm_apic_set_version(apic->vcpu);
1562 
1563 	for (i = 0; i < APIC_LVT_NUM; i++)
1564 		apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1565 	apic->lapic_timer.timer_mode = 0;
1566 	apic_set_reg(apic, APIC_LVT0,
1567 		     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1568 
1569 	apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1570 	apic_set_spiv(apic, 0xff);
1571 	apic_set_reg(apic, APIC_TASKPRI, 0);
1572 	kvm_apic_set_ldr(apic, 0);
1573 	apic_set_reg(apic, APIC_ESR, 0);
1574 	apic_set_reg(apic, APIC_ICR, 0);
1575 	apic_set_reg(apic, APIC_ICR2, 0);
1576 	apic_set_reg(apic, APIC_TDCR, 0);
1577 	apic_set_reg(apic, APIC_TMICT, 0);
1578 	for (i = 0; i < 8; i++) {
1579 		apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1580 		apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1581 		apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1582 	}
1583 	apic->irr_pending = kvm_apic_vid_enabled(vcpu->kvm);
1584 	apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm);
1585 	apic->highest_isr_cache = -1;
1586 	update_divide_count(apic);
1587 	atomic_set(&apic->lapic_timer.pending, 0);
1588 	if (kvm_vcpu_is_bsp(vcpu))
1589 		kvm_lapic_set_base(vcpu,
1590 				vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1591 	vcpu->arch.pv_eoi.msr_val = 0;
1592 	apic_update_ppr(apic);
1593 
1594 	vcpu->arch.apic_arb_prio = 0;
1595 	vcpu->arch.apic_attention = 0;
1596 
1597 	apic_debug("%s: vcpu=%p, id=%d, base_msr="
1598 		   "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1599 		   vcpu, kvm_apic_id(apic),
1600 		   vcpu->arch.apic_base, apic->base_address);
1601 }
1602 
1603 /*
1604  *----------------------------------------------------------------------
1605  * timer interface
1606  *----------------------------------------------------------------------
1607  */
1608 
1609 static bool lapic_is_periodic(struct kvm_lapic *apic)
1610 {
1611 	return apic_lvtt_period(apic);
1612 }
1613 
1614 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1615 {
1616 	struct kvm_lapic *apic = vcpu->arch.apic;
1617 
1618 	if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) &&
1619 			apic_lvt_enabled(apic, APIC_LVTT))
1620 		return atomic_read(&apic->lapic_timer.pending);
1621 
1622 	return 0;
1623 }
1624 
1625 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1626 {
1627 	u32 reg = kvm_apic_get_reg(apic, lvt_type);
1628 	int vector, mode, trig_mode;
1629 
1630 	if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1631 		vector = reg & APIC_VECTOR_MASK;
1632 		mode = reg & APIC_MODE_MASK;
1633 		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1634 		return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1635 					NULL);
1636 	}
1637 	return 0;
1638 }
1639 
1640 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1641 {
1642 	struct kvm_lapic *apic = vcpu->arch.apic;
1643 
1644 	if (apic)
1645 		kvm_apic_local_deliver(apic, APIC_LVT0);
1646 }
1647 
1648 static const struct kvm_io_device_ops apic_mmio_ops = {
1649 	.read     = apic_mmio_read,
1650 	.write    = apic_mmio_write,
1651 };
1652 
1653 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1654 {
1655 	struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1656 	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1657 
1658 	apic_timer_expired(apic);
1659 
1660 	if (lapic_is_periodic(apic)) {
1661 		hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1662 		return HRTIMER_RESTART;
1663 	} else
1664 		return HRTIMER_NORESTART;
1665 }
1666 
1667 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1668 {
1669 	struct kvm_lapic *apic;
1670 
1671 	ASSERT(vcpu != NULL);
1672 	apic_debug("apic_init %d\n", vcpu->vcpu_id);
1673 
1674 	apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1675 	if (!apic)
1676 		goto nomem;
1677 
1678 	vcpu->arch.apic = apic;
1679 
1680 	apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1681 	if (!apic->regs) {
1682 		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1683 		       vcpu->vcpu_id);
1684 		goto nomem_free_apic;
1685 	}
1686 	apic->vcpu = vcpu;
1687 
1688 	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1689 		     HRTIMER_MODE_ABS);
1690 	apic->lapic_timer.timer.function = apic_timer_fn;
1691 
1692 	/*
1693 	 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1694 	 * thinking that APIC satet has changed.
1695 	 */
1696 	vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1697 	kvm_lapic_set_base(vcpu,
1698 			APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE);
1699 
1700 	static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1701 	kvm_lapic_reset(vcpu);
1702 	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1703 
1704 	return 0;
1705 nomem_free_apic:
1706 	kfree(apic);
1707 nomem:
1708 	return -ENOMEM;
1709 }
1710 
1711 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1712 {
1713 	struct kvm_lapic *apic = vcpu->arch.apic;
1714 	int highest_irr;
1715 
1716 	if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic))
1717 		return -1;
1718 
1719 	apic_update_ppr(apic);
1720 	highest_irr = apic_find_highest_irr(apic);
1721 	if ((highest_irr == -1) ||
1722 	    ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI)))
1723 		return -1;
1724 	return highest_irr;
1725 }
1726 
1727 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1728 {
1729 	u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1730 	int r = 0;
1731 
1732 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1733 		r = 1;
1734 	if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1735 	    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1736 		r = 1;
1737 	return r;
1738 }
1739 
1740 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1741 {
1742 	struct kvm_lapic *apic = vcpu->arch.apic;
1743 
1744 	if (!kvm_vcpu_has_lapic(vcpu))
1745 		return;
1746 
1747 	if (atomic_read(&apic->lapic_timer.pending) > 0) {
1748 		kvm_apic_local_deliver(apic, APIC_LVTT);
1749 		if (apic_lvtt_tscdeadline(apic))
1750 			apic->lapic_timer.tscdeadline = 0;
1751 		atomic_set(&apic->lapic_timer.pending, 0);
1752 	}
1753 }
1754 
1755 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1756 {
1757 	int vector = kvm_apic_has_interrupt(vcpu);
1758 	struct kvm_lapic *apic = vcpu->arch.apic;
1759 
1760 	if (vector == -1)
1761 		return -1;
1762 
1763 	/*
1764 	 * We get here even with APIC virtualization enabled, if doing
1765 	 * nested virtualization and L1 runs with the "acknowledge interrupt
1766 	 * on exit" mode.  Then we cannot inject the interrupt via RVI,
1767 	 * because the process would deliver it through the IDT.
1768 	 */
1769 
1770 	apic_set_isr(vector, apic);
1771 	apic_update_ppr(apic);
1772 	apic_clear_irr(vector, apic);
1773 	return vector;
1774 }
1775 
1776 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
1777 		struct kvm_lapic_state *s)
1778 {
1779 	struct kvm_lapic *apic = vcpu->arch.apic;
1780 
1781 	kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
1782 	/* set SPIV separately to get count of SW disabled APICs right */
1783 	apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
1784 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1785 	/* call kvm_apic_set_id() to put apic into apic_map */
1786 	kvm_apic_set_id(apic, kvm_apic_id(apic));
1787 	kvm_apic_set_version(vcpu);
1788 
1789 	apic_update_ppr(apic);
1790 	hrtimer_cancel(&apic->lapic_timer.timer);
1791 	update_divide_count(apic);
1792 	start_apic_timer(apic);
1793 	apic->irr_pending = true;
1794 	apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm) ?
1795 				1 : count_vectors(apic->regs + APIC_ISR);
1796 	apic->highest_isr_cache = -1;
1797 	if (kvm_x86_ops->hwapic_irr_update)
1798 		kvm_x86_ops->hwapic_irr_update(vcpu,
1799 				apic_find_highest_irr(apic));
1800 	if (unlikely(kvm_x86_ops->hwapic_isr_update))
1801 		kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
1802 				apic_find_highest_isr(apic));
1803 	kvm_make_request(KVM_REQ_EVENT, vcpu);
1804 	kvm_rtc_eoi_tracking_restore_one(vcpu);
1805 }
1806 
1807 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1808 {
1809 	struct hrtimer *timer;
1810 
1811 	if (!kvm_vcpu_has_lapic(vcpu))
1812 		return;
1813 
1814 	timer = &vcpu->arch.apic->lapic_timer.timer;
1815 	if (hrtimer_cancel(timer))
1816 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1817 }
1818 
1819 /*
1820  * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1821  *
1822  * Detect whether guest triggered PV EOI since the
1823  * last entry. If yes, set EOI on guests's behalf.
1824  * Clear PV EOI in guest memory in any case.
1825  */
1826 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1827 					struct kvm_lapic *apic)
1828 {
1829 	bool pending;
1830 	int vector;
1831 	/*
1832 	 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1833 	 * and KVM_PV_EOI_ENABLED in guest memory as follows:
1834 	 *
1835 	 * KVM_APIC_PV_EOI_PENDING is unset:
1836 	 * 	-> host disabled PV EOI.
1837 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1838 	 * 	-> host enabled PV EOI, guest did not execute EOI yet.
1839 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1840 	 * 	-> host enabled PV EOI, guest executed EOI.
1841 	 */
1842 	BUG_ON(!pv_eoi_enabled(vcpu));
1843 	pending = pv_eoi_get_pending(vcpu);
1844 	/*
1845 	 * Clear pending bit in any case: it will be set again on vmentry.
1846 	 * While this might not be ideal from performance point of view,
1847 	 * this makes sure pv eoi is only enabled when we know it's safe.
1848 	 */
1849 	pv_eoi_clr_pending(vcpu);
1850 	if (pending)
1851 		return;
1852 	vector = apic_set_eoi(apic);
1853 	trace_kvm_pv_eoi(apic, vector);
1854 }
1855 
1856 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1857 {
1858 	u32 data;
1859 
1860 	if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1861 		apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1862 
1863 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1864 		return;
1865 
1866 	kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1867 				sizeof(u32));
1868 
1869 	apic_set_tpr(vcpu->arch.apic, data & 0xff);
1870 }
1871 
1872 /*
1873  * apic_sync_pv_eoi_to_guest - called before vmentry
1874  *
1875  * Detect whether it's safe to enable PV EOI and
1876  * if yes do so.
1877  */
1878 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
1879 					struct kvm_lapic *apic)
1880 {
1881 	if (!pv_eoi_enabled(vcpu) ||
1882 	    /* IRR set or many bits in ISR: could be nested. */
1883 	    apic->irr_pending ||
1884 	    /* Cache not set: could be safe but we don't bother. */
1885 	    apic->highest_isr_cache == -1 ||
1886 	    /* Need EOI to update ioapic. */
1887 	    kvm_ioapic_handles_vector(vcpu->kvm, apic->highest_isr_cache)) {
1888 		/*
1889 		 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1890 		 * so we need not do anything here.
1891 		 */
1892 		return;
1893 	}
1894 
1895 	pv_eoi_set_pending(apic->vcpu);
1896 }
1897 
1898 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1899 {
1900 	u32 data, tpr;
1901 	int max_irr, max_isr;
1902 	struct kvm_lapic *apic = vcpu->arch.apic;
1903 
1904 	apic_sync_pv_eoi_to_guest(vcpu, apic);
1905 
1906 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1907 		return;
1908 
1909 	tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1910 	max_irr = apic_find_highest_irr(apic);
1911 	if (max_irr < 0)
1912 		max_irr = 0;
1913 	max_isr = apic_find_highest_isr(apic);
1914 	if (max_isr < 0)
1915 		max_isr = 0;
1916 	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1917 
1918 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1919 				sizeof(u32));
1920 }
1921 
1922 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1923 {
1924 	if (vapic_addr) {
1925 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1926 					&vcpu->arch.apic->vapic_cache,
1927 					vapic_addr, sizeof(u32)))
1928 			return -EINVAL;
1929 		__set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1930 	} else {
1931 		__clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1932 	}
1933 
1934 	vcpu->arch.apic->vapic_addr = vapic_addr;
1935 	return 0;
1936 }
1937 
1938 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1939 {
1940 	struct kvm_lapic *apic = vcpu->arch.apic;
1941 	u32 reg = (msr - APIC_BASE_MSR) << 4;
1942 
1943 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1944 		return 1;
1945 
1946 	if (reg == APIC_ICR2)
1947 		return 1;
1948 
1949 	/* if this is ICR write vector before command */
1950 	if (reg == APIC_ICR)
1951 		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1952 	return apic_reg_write(apic, reg, (u32)data);
1953 }
1954 
1955 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1956 {
1957 	struct kvm_lapic *apic = vcpu->arch.apic;
1958 	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1959 
1960 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1961 		return 1;
1962 
1963 	if (reg == APIC_DFR || reg == APIC_ICR2) {
1964 		apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
1965 			   reg);
1966 		return 1;
1967 	}
1968 
1969 	if (apic_reg_read(apic, reg, 4, &low))
1970 		return 1;
1971 	if (reg == APIC_ICR)
1972 		apic_reg_read(apic, APIC_ICR2, 4, &high);
1973 
1974 	*data = (((u64)high) << 32) | low;
1975 
1976 	return 0;
1977 }
1978 
1979 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1980 {
1981 	struct kvm_lapic *apic = vcpu->arch.apic;
1982 
1983 	if (!kvm_vcpu_has_lapic(vcpu))
1984 		return 1;
1985 
1986 	/* if this is ICR write vector before command */
1987 	if (reg == APIC_ICR)
1988 		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1989 	return apic_reg_write(apic, reg, (u32)data);
1990 }
1991 
1992 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
1993 {
1994 	struct kvm_lapic *apic = vcpu->arch.apic;
1995 	u32 low, high = 0;
1996 
1997 	if (!kvm_vcpu_has_lapic(vcpu))
1998 		return 1;
1999 
2000 	if (apic_reg_read(apic, reg, 4, &low))
2001 		return 1;
2002 	if (reg == APIC_ICR)
2003 		apic_reg_read(apic, APIC_ICR2, 4, &high);
2004 
2005 	*data = (((u64)high) << 32) | low;
2006 
2007 	return 0;
2008 }
2009 
2010 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2011 {
2012 	u64 addr = data & ~KVM_MSR_ENABLED;
2013 	if (!IS_ALIGNED(addr, 4))
2014 		return 1;
2015 
2016 	vcpu->arch.pv_eoi.msr_val = data;
2017 	if (!pv_eoi_enabled(vcpu))
2018 		return 0;
2019 	return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
2020 					 addr, sizeof(u8));
2021 }
2022 
2023 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2024 {
2025 	struct kvm_lapic *apic = vcpu->arch.apic;
2026 	u8 sipi_vector;
2027 	unsigned long pe;
2028 
2029 	if (!kvm_vcpu_has_lapic(vcpu) || !apic->pending_events)
2030 		return;
2031 
2032 	pe = xchg(&apic->pending_events, 0);
2033 
2034 	if (test_bit(KVM_APIC_INIT, &pe)) {
2035 		kvm_lapic_reset(vcpu);
2036 		kvm_vcpu_reset(vcpu);
2037 		if (kvm_vcpu_is_bsp(apic->vcpu))
2038 			vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2039 		else
2040 			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2041 	}
2042 	if (test_bit(KVM_APIC_SIPI, &pe) &&
2043 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2044 		/* evaluate pending_events before reading the vector */
2045 		smp_rmb();
2046 		sipi_vector = apic->sipi_vector;
2047 		apic_debug("vcpu %d received sipi with vector # %x\n",
2048 			 vcpu->vcpu_id, sipi_vector);
2049 		kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2050 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2051 	}
2052 }
2053 
2054 void kvm_lapic_init(void)
2055 {
2056 	/* do not patch jump label more than once per second */
2057 	jump_label_rate_limit(&apic_hw_disabled, HZ);
2058 	jump_label_rate_limit(&apic_sw_disabled, HZ);
2059 }
2060