xref: /openbmc/linux/arch/x86/kvm/lapic.c (revision 1e0ad70cc1957b9050368e433b1061a2cd1ce543)
1 
2 /*
3  * Local APIC virtualization
4  *
5  * Copyright (C) 2006 Qumranet, Inc.
6  * Copyright (C) 2007 Novell
7  * Copyright (C) 2007 Intel
8  * Copyright 2009 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Dor Laor <dor.laor@qumranet.com>
12  *   Gregory Haskins <ghaskins@novell.com>
13  *   Yaozu (Eddie) Dong <eddie.dong@intel.com>
14  *
15  * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
16  *
17  * This work is licensed under the terms of the GNU GPL, version 2.  See
18  * the COPYING file in the top-level directory.
19  */
20 
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/mm.h>
24 #include <linux/highmem.h>
25 #include <linux/smp.h>
26 #include <linux/hrtimer.h>
27 #include <linux/io.h>
28 #include <linux/module.h>
29 #include <linux/math64.h>
30 #include <linux/slab.h>
31 #include <asm/processor.h>
32 #include <asm/msr.h>
33 #include <asm/page.h>
34 #include <asm/current.h>
35 #include <asm/apicdef.h>
36 #include <linux/atomic.h>
37 #include <linux/jump_label.h>
38 #include "kvm_cache_regs.h"
39 #include "irq.h"
40 #include "trace.h"
41 #include "x86.h"
42 #include "cpuid.h"
43 
44 #ifndef CONFIG_X86_64
45 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
46 #else
47 #define mod_64(x, y) ((x) % (y))
48 #endif
49 
50 #define PRId64 "d"
51 #define PRIx64 "llx"
52 #define PRIu64 "u"
53 #define PRIo64 "o"
54 
55 #define APIC_BUS_CYCLE_NS 1
56 
57 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
58 #define apic_debug(fmt, arg...)
59 
60 #define APIC_LVT_NUM			6
61 /* 14 is the version for Xeon and Pentium 8.4.8*/
62 #define APIC_VERSION			(0x14UL | ((APIC_LVT_NUM - 1) << 16))
63 #define LAPIC_MMIO_LENGTH		(1 << 12)
64 /* followed define is not in apicdef.h */
65 #define APIC_SHORT_MASK			0xc0000
66 #define APIC_DEST_NOSHORT		0x0
67 #define APIC_DEST_MASK			0x800
68 #define MAX_APIC_VECTOR			256
69 #define APIC_VECTORS_PER_REG		32
70 
71 #define APIC_BROADCAST			0xFF
72 #define X2APIC_BROADCAST		0xFFFFFFFFul
73 
74 #define VEC_POS(v) ((v) & (32 - 1))
75 #define REG_POS(v) (((v) >> 5) << 4)
76 
77 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
78 {
79 	*((u32 *) (apic->regs + reg_off)) = val;
80 }
81 
82 static inline int apic_test_vector(int vec, void *bitmap)
83 {
84 	return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
85 }
86 
87 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
88 {
89 	struct kvm_lapic *apic = vcpu->arch.apic;
90 
91 	return apic_test_vector(vector, apic->regs + APIC_ISR) ||
92 		apic_test_vector(vector, apic->regs + APIC_IRR);
93 }
94 
95 static inline void apic_set_vector(int vec, void *bitmap)
96 {
97 	set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
98 }
99 
100 static inline void apic_clear_vector(int vec, void *bitmap)
101 {
102 	clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
103 }
104 
105 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
106 {
107 	return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
108 }
109 
110 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
111 {
112 	return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
113 }
114 
115 struct static_key_deferred apic_hw_disabled __read_mostly;
116 struct static_key_deferred apic_sw_disabled __read_mostly;
117 
118 static inline int apic_enabled(struct kvm_lapic *apic)
119 {
120 	return kvm_apic_sw_enabled(apic) &&	kvm_apic_hw_enabled(apic);
121 }
122 
123 #define LVT_MASK	\
124 	(APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
125 
126 #define LINT_MASK	\
127 	(LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
128 	 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
129 
130 static inline int kvm_apic_id(struct kvm_lapic *apic)
131 {
132 	return (kvm_apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
133 }
134 
135 #define KVM_X2APIC_CID_BITS 0
136 
137 static void recalculate_apic_map(struct kvm *kvm)
138 {
139 	struct kvm_apic_map *new, *old = NULL;
140 	struct kvm_vcpu *vcpu;
141 	int i;
142 
143 	new = kzalloc(sizeof(struct kvm_apic_map), GFP_KERNEL);
144 
145 	mutex_lock(&kvm->arch.apic_map_lock);
146 
147 	if (!new)
148 		goto out;
149 
150 	new->ldr_bits = 8;
151 	/* flat mode is default */
152 	new->cid_shift = 8;
153 	new->cid_mask = 0;
154 	new->lid_mask = 0xff;
155 	new->broadcast = APIC_BROADCAST;
156 
157 	kvm_for_each_vcpu(i, vcpu, kvm) {
158 		struct kvm_lapic *apic = vcpu->arch.apic;
159 		u16 cid, lid;
160 		u32 ldr;
161 
162 		if (!kvm_apic_present(vcpu))
163 			continue;
164 
165 		/*
166 		 * All APICs have to be configured in the same mode by an OS.
167 		 * We take advatage of this while building logical id loockup
168 		 * table. After reset APICs are in xapic/flat mode, so if we
169 		 * find apic with different setting we assume this is the mode
170 		 * OS wants all apics to be in; build lookup table accordingly.
171 		 */
172 		if (apic_x2apic_mode(apic)) {
173 			new->ldr_bits = 32;
174 			new->cid_shift = 16;
175 			new->cid_mask = (1 << KVM_X2APIC_CID_BITS) - 1;
176 			new->lid_mask = 0xffff;
177 			new->broadcast = X2APIC_BROADCAST;
178 		} else if (kvm_apic_sw_enabled(apic) &&
179 				!new->cid_mask /* flat mode */ &&
180 				kvm_apic_get_reg(apic, APIC_DFR) == APIC_DFR_CLUSTER) {
181 			new->cid_shift = 4;
182 			new->cid_mask = 0xf;
183 			new->lid_mask = 0xf;
184 		}
185 
186 		new->phys_map[kvm_apic_id(apic)] = apic;
187 
188 		ldr = kvm_apic_get_reg(apic, APIC_LDR);
189 		cid = apic_cluster_id(new, ldr);
190 		lid = apic_logical_id(new, ldr);
191 
192 		if (lid)
193 			new->logical_map[cid][ffs(lid) - 1] = apic;
194 	}
195 out:
196 	old = rcu_dereference_protected(kvm->arch.apic_map,
197 			lockdep_is_held(&kvm->arch.apic_map_lock));
198 	rcu_assign_pointer(kvm->arch.apic_map, new);
199 	mutex_unlock(&kvm->arch.apic_map_lock);
200 
201 	if (old)
202 		kfree_rcu(old, rcu);
203 
204 	kvm_vcpu_request_scan_ioapic(kvm);
205 }
206 
207 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
208 {
209 	u32 prev = kvm_apic_get_reg(apic, APIC_SPIV);
210 
211 	apic_set_reg(apic, APIC_SPIV, val);
212 	if ((prev ^ val) & APIC_SPIV_APIC_ENABLED) {
213 		if (val & APIC_SPIV_APIC_ENABLED) {
214 			static_key_slow_dec_deferred(&apic_sw_disabled);
215 			recalculate_apic_map(apic->vcpu->kvm);
216 		} else
217 			static_key_slow_inc(&apic_sw_disabled.key);
218 	}
219 }
220 
221 static inline void kvm_apic_set_id(struct kvm_lapic *apic, u8 id)
222 {
223 	apic_set_reg(apic, APIC_ID, id << 24);
224 	recalculate_apic_map(apic->vcpu->kvm);
225 }
226 
227 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
228 {
229 	apic_set_reg(apic, APIC_LDR, id);
230 	recalculate_apic_map(apic->vcpu->kvm);
231 }
232 
233 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
234 {
235 	return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
236 }
237 
238 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
239 {
240 	return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
241 }
242 
243 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
244 {
245 	return ((kvm_apic_get_reg(apic, APIC_LVTT) &
246 		apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_ONESHOT);
247 }
248 
249 static inline int apic_lvtt_period(struct kvm_lapic *apic)
250 {
251 	return ((kvm_apic_get_reg(apic, APIC_LVTT) &
252 		apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_PERIODIC);
253 }
254 
255 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
256 {
257 	return ((kvm_apic_get_reg(apic, APIC_LVTT) &
258 		apic->lapic_timer.timer_mode_mask) ==
259 			APIC_LVT_TIMER_TSCDEADLINE);
260 }
261 
262 static inline int apic_lvt_nmi_mode(u32 lvt_val)
263 {
264 	return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
265 }
266 
267 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
268 {
269 	struct kvm_lapic *apic = vcpu->arch.apic;
270 	struct kvm_cpuid_entry2 *feat;
271 	u32 v = APIC_VERSION;
272 
273 	if (!kvm_vcpu_has_lapic(vcpu))
274 		return;
275 
276 	feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
277 	if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
278 		v |= APIC_LVR_DIRECTED_EOI;
279 	apic_set_reg(apic, APIC_LVR, v);
280 }
281 
282 static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
283 	LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
284 	LVT_MASK | APIC_MODE_MASK,	/* LVTTHMR */
285 	LVT_MASK | APIC_MODE_MASK,	/* LVTPC */
286 	LINT_MASK, LINT_MASK,	/* LVT0-1 */
287 	LVT_MASK		/* LVTERR */
288 };
289 
290 static int find_highest_vector(void *bitmap)
291 {
292 	int vec;
293 	u32 *reg;
294 
295 	for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
296 	     vec >= 0; vec -= APIC_VECTORS_PER_REG) {
297 		reg = bitmap + REG_POS(vec);
298 		if (*reg)
299 			return fls(*reg) - 1 + vec;
300 	}
301 
302 	return -1;
303 }
304 
305 static u8 count_vectors(void *bitmap)
306 {
307 	int vec;
308 	u32 *reg;
309 	u8 count = 0;
310 
311 	for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
312 		reg = bitmap + REG_POS(vec);
313 		count += hweight32(*reg);
314 	}
315 
316 	return count;
317 }
318 
319 void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
320 {
321 	u32 i, pir_val;
322 	struct kvm_lapic *apic = vcpu->arch.apic;
323 
324 	for (i = 0; i <= 7; i++) {
325 		pir_val = xchg(&pir[i], 0);
326 		if (pir_val)
327 			*((u32 *)(apic->regs + APIC_IRR + i * 0x10)) |= pir_val;
328 	}
329 }
330 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
331 
332 static inline void apic_set_irr(int vec, struct kvm_lapic *apic)
333 {
334 	apic->irr_pending = true;
335 	apic_set_vector(vec, apic->regs + APIC_IRR);
336 }
337 
338 static inline int apic_search_irr(struct kvm_lapic *apic)
339 {
340 	return find_highest_vector(apic->regs + APIC_IRR);
341 }
342 
343 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
344 {
345 	int result;
346 
347 	/*
348 	 * Note that irr_pending is just a hint. It will be always
349 	 * true with virtual interrupt delivery enabled.
350 	 */
351 	if (!apic->irr_pending)
352 		return -1;
353 
354 	kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
355 	result = apic_search_irr(apic);
356 	ASSERT(result == -1 || result >= 16);
357 
358 	return result;
359 }
360 
361 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
362 {
363 	struct kvm_vcpu *vcpu;
364 
365 	vcpu = apic->vcpu;
366 
367 	apic_clear_vector(vec, apic->regs + APIC_IRR);
368 	if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
369 		/* try to update RVI */
370 		kvm_make_request(KVM_REQ_EVENT, vcpu);
371 	else {
372 		vec = apic_search_irr(apic);
373 		apic->irr_pending = (vec != -1);
374 	}
375 }
376 
377 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
378 {
379 	struct kvm_vcpu *vcpu;
380 
381 	if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
382 		return;
383 
384 	vcpu = apic->vcpu;
385 
386 	/*
387 	 * With APIC virtualization enabled, all caching is disabled
388 	 * because the processor can modify ISR under the hood.  Instead
389 	 * just set SVI.
390 	 */
391 	if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
392 		kvm_x86_ops->hwapic_isr_update(vcpu->kvm, vec);
393 	else {
394 		++apic->isr_count;
395 		BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
396 		/*
397 		 * ISR (in service register) bit is set when injecting an interrupt.
398 		 * The highest vector is injected. Thus the latest bit set matches
399 		 * the highest bit in ISR.
400 		 */
401 		apic->highest_isr_cache = vec;
402 	}
403 }
404 
405 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
406 {
407 	int result;
408 
409 	/*
410 	 * Note that isr_count is always 1, and highest_isr_cache
411 	 * is always -1, with APIC virtualization enabled.
412 	 */
413 	if (!apic->isr_count)
414 		return -1;
415 	if (likely(apic->highest_isr_cache != -1))
416 		return apic->highest_isr_cache;
417 
418 	result = find_highest_vector(apic->regs + APIC_ISR);
419 	ASSERT(result == -1 || result >= 16);
420 
421 	return result;
422 }
423 
424 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
425 {
426 	struct kvm_vcpu *vcpu;
427 	if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
428 		return;
429 
430 	vcpu = apic->vcpu;
431 
432 	/*
433 	 * We do get here for APIC virtualization enabled if the guest
434 	 * uses the Hyper-V APIC enlightenment.  In this case we may need
435 	 * to trigger a new interrupt delivery by writing the SVI field;
436 	 * on the other hand isr_count and highest_isr_cache are unused
437 	 * and must be left alone.
438 	 */
439 	if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
440 		kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
441 					       apic_find_highest_isr(apic));
442 	else {
443 		--apic->isr_count;
444 		BUG_ON(apic->isr_count < 0);
445 		apic->highest_isr_cache = -1;
446 	}
447 }
448 
449 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
450 {
451 	int highest_irr;
452 
453 	/* This may race with setting of irr in __apic_accept_irq() and
454 	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
455 	 * will cause vmexit immediately and the value will be recalculated
456 	 * on the next vmentry.
457 	 */
458 	if (!kvm_vcpu_has_lapic(vcpu))
459 		return 0;
460 	highest_irr = apic_find_highest_irr(vcpu->arch.apic);
461 
462 	return highest_irr;
463 }
464 
465 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
466 			     int vector, int level, int trig_mode,
467 			     unsigned long *dest_map);
468 
469 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
470 		unsigned long *dest_map)
471 {
472 	struct kvm_lapic *apic = vcpu->arch.apic;
473 
474 	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
475 			irq->level, irq->trig_mode, dest_map);
476 }
477 
478 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
479 {
480 
481 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
482 				      sizeof(val));
483 }
484 
485 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
486 {
487 
488 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
489 				      sizeof(*val));
490 }
491 
492 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
493 {
494 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
495 }
496 
497 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
498 {
499 	u8 val;
500 	if (pv_eoi_get_user(vcpu, &val) < 0)
501 		apic_debug("Can't read EOI MSR value: 0x%llx\n",
502 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
503 	return val & 0x1;
504 }
505 
506 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
507 {
508 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
509 		apic_debug("Can't set EOI MSR value: 0x%llx\n",
510 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
511 		return;
512 	}
513 	__set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
514 }
515 
516 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
517 {
518 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
519 		apic_debug("Can't clear EOI MSR value: 0x%llx\n",
520 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
521 		return;
522 	}
523 	__clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
524 }
525 
526 void kvm_apic_update_tmr(struct kvm_vcpu *vcpu, u32 *tmr)
527 {
528 	struct kvm_lapic *apic = vcpu->arch.apic;
529 	int i;
530 
531 	for (i = 0; i < 8; i++)
532 		apic_set_reg(apic, APIC_TMR + 0x10 * i, tmr[i]);
533 }
534 
535 static void apic_update_ppr(struct kvm_lapic *apic)
536 {
537 	u32 tpr, isrv, ppr, old_ppr;
538 	int isr;
539 
540 	old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI);
541 	tpr = kvm_apic_get_reg(apic, APIC_TASKPRI);
542 	isr = apic_find_highest_isr(apic);
543 	isrv = (isr != -1) ? isr : 0;
544 
545 	if ((tpr & 0xf0) >= (isrv & 0xf0))
546 		ppr = tpr & 0xff;
547 	else
548 		ppr = isrv & 0xf0;
549 
550 	apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
551 		   apic, ppr, isr, isrv);
552 
553 	if (old_ppr != ppr) {
554 		apic_set_reg(apic, APIC_PROCPRI, ppr);
555 		if (ppr < old_ppr)
556 			kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
557 	}
558 }
559 
560 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
561 {
562 	apic_set_reg(apic, APIC_TASKPRI, tpr);
563 	apic_update_ppr(apic);
564 }
565 
566 static int kvm_apic_broadcast(struct kvm_lapic *apic, u32 dest)
567 {
568 	return dest == (apic_x2apic_mode(apic) ?
569 			X2APIC_BROADCAST : APIC_BROADCAST);
570 }
571 
572 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 dest)
573 {
574 	return kvm_apic_id(apic) == dest || kvm_apic_broadcast(apic, dest);
575 }
576 
577 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
578 {
579 	int result = 0;
580 	u32 logical_id;
581 
582 	if (kvm_apic_broadcast(apic, mda))
583 		return 1;
584 
585 	if (apic_x2apic_mode(apic)) {
586 		logical_id = kvm_apic_get_reg(apic, APIC_LDR);
587 		return logical_id & mda;
588 	}
589 
590 	logical_id = GET_APIC_LOGICAL_ID(kvm_apic_get_reg(apic, APIC_LDR));
591 
592 	switch (kvm_apic_get_reg(apic, APIC_DFR)) {
593 	case APIC_DFR_FLAT:
594 		if (logical_id & mda)
595 			result = 1;
596 		break;
597 	case APIC_DFR_CLUSTER:
598 		if (((logical_id >> 4) == (mda >> 0x4))
599 		    && (logical_id & mda & 0xf))
600 			result = 1;
601 		break;
602 	default:
603 		apic_debug("Bad DFR vcpu %d: %08x\n",
604 			   apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR));
605 		break;
606 	}
607 
608 	return result;
609 }
610 
611 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
612 			   int short_hand, unsigned int dest, int dest_mode)
613 {
614 	int result = 0;
615 	struct kvm_lapic *target = vcpu->arch.apic;
616 
617 	apic_debug("target %p, source %p, dest 0x%x, "
618 		   "dest_mode 0x%x, short_hand 0x%x\n",
619 		   target, source, dest, dest_mode, short_hand);
620 
621 	ASSERT(target);
622 	switch (short_hand) {
623 	case APIC_DEST_NOSHORT:
624 		if (dest_mode == 0)
625 			/* Physical mode. */
626 			result = kvm_apic_match_physical_addr(target, dest);
627 		else
628 			/* Logical mode. */
629 			result = kvm_apic_match_logical_addr(target, dest);
630 		break;
631 	case APIC_DEST_SELF:
632 		result = (target == source);
633 		break;
634 	case APIC_DEST_ALLINC:
635 		result = 1;
636 		break;
637 	case APIC_DEST_ALLBUT:
638 		result = (target != source);
639 		break;
640 	default:
641 		apic_debug("kvm: apic: Bad dest shorthand value %x\n",
642 			   short_hand);
643 		break;
644 	}
645 
646 	return result;
647 }
648 
649 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
650 		struct kvm_lapic_irq *irq, int *r, unsigned long *dest_map)
651 {
652 	struct kvm_apic_map *map;
653 	unsigned long bitmap = 1;
654 	struct kvm_lapic **dst;
655 	int i;
656 	bool ret = false;
657 
658 	*r = -1;
659 
660 	if (irq->shorthand == APIC_DEST_SELF) {
661 		*r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
662 		return true;
663 	}
664 
665 	if (irq->shorthand)
666 		return false;
667 
668 	rcu_read_lock();
669 	map = rcu_dereference(kvm->arch.apic_map);
670 
671 	if (!map)
672 		goto out;
673 
674 	if (irq->dest_id == map->broadcast)
675 		goto out;
676 
677 	if (irq->dest_mode == 0) { /* physical mode */
678 		if (irq->delivery_mode == APIC_DM_LOWEST)
679 			goto out;
680 		dst = &map->phys_map[irq->dest_id & 0xff];
681 	} else {
682 		u32 mda = irq->dest_id << (32 - map->ldr_bits);
683 
684 		dst = map->logical_map[apic_cluster_id(map, mda)];
685 
686 		bitmap = apic_logical_id(map, mda);
687 
688 		if (irq->delivery_mode == APIC_DM_LOWEST) {
689 			int l = -1;
690 			for_each_set_bit(i, &bitmap, 16) {
691 				if (!dst[i])
692 					continue;
693 				if (l < 0)
694 					l = i;
695 				else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0)
696 					l = i;
697 			}
698 
699 			bitmap = (l >= 0) ? 1 << l : 0;
700 		}
701 	}
702 
703 	for_each_set_bit(i, &bitmap, 16) {
704 		if (!dst[i])
705 			continue;
706 		if (*r < 0)
707 			*r = 0;
708 		*r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
709 	}
710 
711 	ret = true;
712 out:
713 	rcu_read_unlock();
714 	return ret;
715 }
716 
717 /*
718  * Add a pending IRQ into lapic.
719  * Return 1 if successfully added and 0 if discarded.
720  */
721 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
722 			     int vector, int level, int trig_mode,
723 			     unsigned long *dest_map)
724 {
725 	int result = 0;
726 	struct kvm_vcpu *vcpu = apic->vcpu;
727 
728 	trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
729 				  trig_mode, vector);
730 	switch (delivery_mode) {
731 	case APIC_DM_LOWEST:
732 		vcpu->arch.apic_arb_prio++;
733 	case APIC_DM_FIXED:
734 		/* FIXME add logic for vcpu on reset */
735 		if (unlikely(!apic_enabled(apic)))
736 			break;
737 
738 		result = 1;
739 
740 		if (dest_map)
741 			__set_bit(vcpu->vcpu_id, dest_map);
742 
743 		if (kvm_x86_ops->deliver_posted_interrupt)
744 			kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
745 		else {
746 			apic_set_irr(vector, apic);
747 
748 			kvm_make_request(KVM_REQ_EVENT, vcpu);
749 			kvm_vcpu_kick(vcpu);
750 		}
751 		break;
752 
753 	case APIC_DM_REMRD:
754 		result = 1;
755 		vcpu->arch.pv.pv_unhalted = 1;
756 		kvm_make_request(KVM_REQ_EVENT, vcpu);
757 		kvm_vcpu_kick(vcpu);
758 		break;
759 
760 	case APIC_DM_SMI:
761 		apic_debug("Ignoring guest SMI\n");
762 		break;
763 
764 	case APIC_DM_NMI:
765 		result = 1;
766 		kvm_inject_nmi(vcpu);
767 		kvm_vcpu_kick(vcpu);
768 		break;
769 
770 	case APIC_DM_INIT:
771 		if (!trig_mode || level) {
772 			result = 1;
773 			/* assumes that there are only KVM_APIC_INIT/SIPI */
774 			apic->pending_events = (1UL << KVM_APIC_INIT);
775 			/* make sure pending_events is visible before sending
776 			 * the request */
777 			smp_wmb();
778 			kvm_make_request(KVM_REQ_EVENT, vcpu);
779 			kvm_vcpu_kick(vcpu);
780 		} else {
781 			apic_debug("Ignoring de-assert INIT to vcpu %d\n",
782 				   vcpu->vcpu_id);
783 		}
784 		break;
785 
786 	case APIC_DM_STARTUP:
787 		apic_debug("SIPI to vcpu %d vector 0x%02x\n",
788 			   vcpu->vcpu_id, vector);
789 		result = 1;
790 		apic->sipi_vector = vector;
791 		/* make sure sipi_vector is visible for the receiver */
792 		smp_wmb();
793 		set_bit(KVM_APIC_SIPI, &apic->pending_events);
794 		kvm_make_request(KVM_REQ_EVENT, vcpu);
795 		kvm_vcpu_kick(vcpu);
796 		break;
797 
798 	case APIC_DM_EXTINT:
799 		/*
800 		 * Should only be called by kvm_apic_local_deliver() with LVT0,
801 		 * before NMI watchdog was enabled. Already handled by
802 		 * kvm_apic_accept_pic_intr().
803 		 */
804 		break;
805 
806 	default:
807 		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
808 		       delivery_mode);
809 		break;
810 	}
811 	return result;
812 }
813 
814 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
815 {
816 	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
817 }
818 
819 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
820 {
821 	if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI) &&
822 	    kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
823 		int trigger_mode;
824 		if (apic_test_vector(vector, apic->regs + APIC_TMR))
825 			trigger_mode = IOAPIC_LEVEL_TRIG;
826 		else
827 			trigger_mode = IOAPIC_EDGE_TRIG;
828 		kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
829 	}
830 }
831 
832 static int apic_set_eoi(struct kvm_lapic *apic)
833 {
834 	int vector = apic_find_highest_isr(apic);
835 
836 	trace_kvm_eoi(apic, vector);
837 
838 	/*
839 	 * Not every write EOI will has corresponding ISR,
840 	 * one example is when Kernel check timer on setup_IO_APIC
841 	 */
842 	if (vector == -1)
843 		return vector;
844 
845 	apic_clear_isr(vector, apic);
846 	apic_update_ppr(apic);
847 
848 	kvm_ioapic_send_eoi(apic, vector);
849 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
850 	return vector;
851 }
852 
853 /*
854  * this interface assumes a trap-like exit, which has already finished
855  * desired side effect including vISR and vPPR update.
856  */
857 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
858 {
859 	struct kvm_lapic *apic = vcpu->arch.apic;
860 
861 	trace_kvm_eoi(apic, vector);
862 
863 	kvm_ioapic_send_eoi(apic, vector);
864 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
865 }
866 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
867 
868 static void apic_send_ipi(struct kvm_lapic *apic)
869 {
870 	u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR);
871 	u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2);
872 	struct kvm_lapic_irq irq;
873 
874 	irq.vector = icr_low & APIC_VECTOR_MASK;
875 	irq.delivery_mode = icr_low & APIC_MODE_MASK;
876 	irq.dest_mode = icr_low & APIC_DEST_MASK;
877 	irq.level = icr_low & APIC_INT_ASSERT;
878 	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
879 	irq.shorthand = icr_low & APIC_SHORT_MASK;
880 	if (apic_x2apic_mode(apic))
881 		irq.dest_id = icr_high;
882 	else
883 		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
884 
885 	trace_kvm_apic_ipi(icr_low, irq.dest_id);
886 
887 	apic_debug("icr_high 0x%x, icr_low 0x%x, "
888 		   "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
889 		   "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
890 		   icr_high, icr_low, irq.shorthand, irq.dest_id,
891 		   irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
892 		   irq.vector);
893 
894 	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
895 }
896 
897 static u32 apic_get_tmcct(struct kvm_lapic *apic)
898 {
899 	ktime_t remaining;
900 	s64 ns;
901 	u32 tmcct;
902 
903 	ASSERT(apic != NULL);
904 
905 	/* if initial count is 0, current count should also be 0 */
906 	if (kvm_apic_get_reg(apic, APIC_TMICT) == 0 ||
907 		apic->lapic_timer.period == 0)
908 		return 0;
909 
910 	remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
911 	if (ktime_to_ns(remaining) < 0)
912 		remaining = ktime_set(0, 0);
913 
914 	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
915 	tmcct = div64_u64(ns,
916 			 (APIC_BUS_CYCLE_NS * apic->divide_count));
917 
918 	return tmcct;
919 }
920 
921 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
922 {
923 	struct kvm_vcpu *vcpu = apic->vcpu;
924 	struct kvm_run *run = vcpu->run;
925 
926 	kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
927 	run->tpr_access.rip = kvm_rip_read(vcpu);
928 	run->tpr_access.is_write = write;
929 }
930 
931 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
932 {
933 	if (apic->vcpu->arch.tpr_access_reporting)
934 		__report_tpr_access(apic, write);
935 }
936 
937 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
938 {
939 	u32 val = 0;
940 
941 	if (offset >= LAPIC_MMIO_LENGTH)
942 		return 0;
943 
944 	switch (offset) {
945 	case APIC_ID:
946 		if (apic_x2apic_mode(apic))
947 			val = kvm_apic_id(apic);
948 		else
949 			val = kvm_apic_id(apic) << 24;
950 		break;
951 	case APIC_ARBPRI:
952 		apic_debug("Access APIC ARBPRI register which is for P6\n");
953 		break;
954 
955 	case APIC_TMCCT:	/* Timer CCR */
956 		if (apic_lvtt_tscdeadline(apic))
957 			return 0;
958 
959 		val = apic_get_tmcct(apic);
960 		break;
961 	case APIC_PROCPRI:
962 		apic_update_ppr(apic);
963 		val = kvm_apic_get_reg(apic, offset);
964 		break;
965 	case APIC_TASKPRI:
966 		report_tpr_access(apic, false);
967 		/* fall thru */
968 	default:
969 		val = kvm_apic_get_reg(apic, offset);
970 		break;
971 	}
972 
973 	return val;
974 }
975 
976 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
977 {
978 	return container_of(dev, struct kvm_lapic, dev);
979 }
980 
981 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
982 		void *data)
983 {
984 	unsigned char alignment = offset & 0xf;
985 	u32 result;
986 	/* this bitmask has a bit cleared for each reserved register */
987 	static const u64 rmask = 0x43ff01ffffffe70cULL;
988 
989 	if ((alignment + len) > 4) {
990 		apic_debug("KVM_APIC_READ: alignment error %x %d\n",
991 			   offset, len);
992 		return 1;
993 	}
994 
995 	if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
996 		apic_debug("KVM_APIC_READ: read reserved register %x\n",
997 			   offset);
998 		return 1;
999 	}
1000 
1001 	result = __apic_read(apic, offset & ~0xf);
1002 
1003 	trace_kvm_apic_read(offset, result);
1004 
1005 	switch (len) {
1006 	case 1:
1007 	case 2:
1008 	case 4:
1009 		memcpy(data, (char *)&result + alignment, len);
1010 		break;
1011 	default:
1012 		printk(KERN_ERR "Local APIC read with len = %x, "
1013 		       "should be 1,2, or 4 instead\n", len);
1014 		break;
1015 	}
1016 	return 0;
1017 }
1018 
1019 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1020 {
1021 	return kvm_apic_hw_enabled(apic) &&
1022 	    addr >= apic->base_address &&
1023 	    addr < apic->base_address + LAPIC_MMIO_LENGTH;
1024 }
1025 
1026 static int apic_mmio_read(struct kvm_io_device *this,
1027 			   gpa_t address, int len, void *data)
1028 {
1029 	struct kvm_lapic *apic = to_lapic(this);
1030 	u32 offset = address - apic->base_address;
1031 
1032 	if (!apic_mmio_in_range(apic, address))
1033 		return -EOPNOTSUPP;
1034 
1035 	apic_reg_read(apic, offset, len, data);
1036 
1037 	return 0;
1038 }
1039 
1040 static void update_divide_count(struct kvm_lapic *apic)
1041 {
1042 	u32 tmp1, tmp2, tdcr;
1043 
1044 	tdcr = kvm_apic_get_reg(apic, APIC_TDCR);
1045 	tmp1 = tdcr & 0xf;
1046 	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1047 	apic->divide_count = 0x1 << (tmp2 & 0x7);
1048 
1049 	apic_debug("timer divide count is 0x%x\n",
1050 				   apic->divide_count);
1051 }
1052 
1053 static void apic_timer_expired(struct kvm_lapic *apic)
1054 {
1055 	struct kvm_vcpu *vcpu = apic->vcpu;
1056 	wait_queue_head_t *q = &vcpu->wq;
1057 
1058 	/*
1059 	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1060 	 * vcpu_enter_guest.
1061 	 */
1062 	if (atomic_read(&apic->lapic_timer.pending))
1063 		return;
1064 
1065 	atomic_inc(&apic->lapic_timer.pending);
1066 	/* FIXME: this code should not know anything about vcpus */
1067 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1068 
1069 	if (waitqueue_active(q))
1070 		wake_up_interruptible(q);
1071 }
1072 
1073 static void start_apic_timer(struct kvm_lapic *apic)
1074 {
1075 	ktime_t now;
1076 	atomic_set(&apic->lapic_timer.pending, 0);
1077 
1078 	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1079 		/* lapic timer in oneshot or periodic mode */
1080 		now = apic->lapic_timer.timer.base->get_time();
1081 		apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT)
1082 			    * APIC_BUS_CYCLE_NS * apic->divide_count;
1083 
1084 		if (!apic->lapic_timer.period)
1085 			return;
1086 		/*
1087 		 * Do not allow the guest to program periodic timers with small
1088 		 * interval, since the hrtimers are not throttled by the host
1089 		 * scheduler.
1090 		 */
1091 		if (apic_lvtt_period(apic)) {
1092 			s64 min_period = min_timer_period_us * 1000LL;
1093 
1094 			if (apic->lapic_timer.period < min_period) {
1095 				pr_info_ratelimited(
1096 				    "kvm: vcpu %i: requested %lld ns "
1097 				    "lapic timer period limited to %lld ns\n",
1098 				    apic->vcpu->vcpu_id,
1099 				    apic->lapic_timer.period, min_period);
1100 				apic->lapic_timer.period = min_period;
1101 			}
1102 		}
1103 
1104 		hrtimer_start(&apic->lapic_timer.timer,
1105 			      ktime_add_ns(now, apic->lapic_timer.period),
1106 			      HRTIMER_MODE_ABS);
1107 
1108 		apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1109 			   PRIx64 ", "
1110 			   "timer initial count 0x%x, period %lldns, "
1111 			   "expire @ 0x%016" PRIx64 ".\n", __func__,
1112 			   APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1113 			   kvm_apic_get_reg(apic, APIC_TMICT),
1114 			   apic->lapic_timer.period,
1115 			   ktime_to_ns(ktime_add_ns(now,
1116 					apic->lapic_timer.period)));
1117 	} else if (apic_lvtt_tscdeadline(apic)) {
1118 		/* lapic timer in tsc deadline mode */
1119 		u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1120 		u64 ns = 0;
1121 		struct kvm_vcpu *vcpu = apic->vcpu;
1122 		unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1123 		unsigned long flags;
1124 
1125 		if (unlikely(!tscdeadline || !this_tsc_khz))
1126 			return;
1127 
1128 		local_irq_save(flags);
1129 
1130 		now = apic->lapic_timer.timer.base->get_time();
1131 		guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1132 		if (likely(tscdeadline > guest_tsc)) {
1133 			ns = (tscdeadline - guest_tsc) * 1000000ULL;
1134 			do_div(ns, this_tsc_khz);
1135 			hrtimer_start(&apic->lapic_timer.timer,
1136 				ktime_add_ns(now, ns), HRTIMER_MODE_ABS);
1137 		} else
1138 			apic_timer_expired(apic);
1139 
1140 		local_irq_restore(flags);
1141 	}
1142 }
1143 
1144 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1145 {
1146 	int nmi_wd_enabled = apic_lvt_nmi_mode(kvm_apic_get_reg(apic, APIC_LVT0));
1147 
1148 	if (apic_lvt_nmi_mode(lvt0_val)) {
1149 		if (!nmi_wd_enabled) {
1150 			apic_debug("Receive NMI setting on APIC_LVT0 "
1151 				   "for cpu %d\n", apic->vcpu->vcpu_id);
1152 			apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
1153 		}
1154 	} else if (nmi_wd_enabled)
1155 		apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
1156 }
1157 
1158 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1159 {
1160 	int ret = 0;
1161 
1162 	trace_kvm_apic_write(reg, val);
1163 
1164 	switch (reg) {
1165 	case APIC_ID:		/* Local APIC ID */
1166 		if (!apic_x2apic_mode(apic))
1167 			kvm_apic_set_id(apic, val >> 24);
1168 		else
1169 			ret = 1;
1170 		break;
1171 
1172 	case APIC_TASKPRI:
1173 		report_tpr_access(apic, true);
1174 		apic_set_tpr(apic, val & 0xff);
1175 		break;
1176 
1177 	case APIC_EOI:
1178 		apic_set_eoi(apic);
1179 		break;
1180 
1181 	case APIC_LDR:
1182 		if (!apic_x2apic_mode(apic))
1183 			kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1184 		else
1185 			ret = 1;
1186 		break;
1187 
1188 	case APIC_DFR:
1189 		if (!apic_x2apic_mode(apic)) {
1190 			apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1191 			recalculate_apic_map(apic->vcpu->kvm);
1192 		} else
1193 			ret = 1;
1194 		break;
1195 
1196 	case APIC_SPIV: {
1197 		u32 mask = 0x3ff;
1198 		if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1199 			mask |= APIC_SPIV_DIRECTED_EOI;
1200 		apic_set_spiv(apic, val & mask);
1201 		if (!(val & APIC_SPIV_APIC_ENABLED)) {
1202 			int i;
1203 			u32 lvt_val;
1204 
1205 			for (i = 0; i < APIC_LVT_NUM; i++) {
1206 				lvt_val = kvm_apic_get_reg(apic,
1207 						       APIC_LVTT + 0x10 * i);
1208 				apic_set_reg(apic, APIC_LVTT + 0x10 * i,
1209 					     lvt_val | APIC_LVT_MASKED);
1210 			}
1211 			atomic_set(&apic->lapic_timer.pending, 0);
1212 
1213 		}
1214 		break;
1215 	}
1216 	case APIC_ICR:
1217 		/* No delay here, so we always clear the pending bit */
1218 		apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1219 		apic_send_ipi(apic);
1220 		break;
1221 
1222 	case APIC_ICR2:
1223 		if (!apic_x2apic_mode(apic))
1224 			val &= 0xff000000;
1225 		apic_set_reg(apic, APIC_ICR2, val);
1226 		break;
1227 
1228 	case APIC_LVT0:
1229 		apic_manage_nmi_watchdog(apic, val);
1230 	case APIC_LVTTHMR:
1231 	case APIC_LVTPC:
1232 	case APIC_LVT1:
1233 	case APIC_LVTERR:
1234 		/* TODO: Check vector */
1235 		if (!kvm_apic_sw_enabled(apic))
1236 			val |= APIC_LVT_MASKED;
1237 
1238 		val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1239 		apic_set_reg(apic, reg, val);
1240 
1241 		break;
1242 
1243 	case APIC_LVTT:
1244 		if ((kvm_apic_get_reg(apic, APIC_LVTT) &
1245 		    apic->lapic_timer.timer_mode_mask) !=
1246 		   (val & apic->lapic_timer.timer_mode_mask))
1247 			hrtimer_cancel(&apic->lapic_timer.timer);
1248 
1249 		if (!kvm_apic_sw_enabled(apic))
1250 			val |= APIC_LVT_MASKED;
1251 		val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1252 		apic_set_reg(apic, APIC_LVTT, val);
1253 		break;
1254 
1255 	case APIC_TMICT:
1256 		if (apic_lvtt_tscdeadline(apic))
1257 			break;
1258 
1259 		hrtimer_cancel(&apic->lapic_timer.timer);
1260 		apic_set_reg(apic, APIC_TMICT, val);
1261 		start_apic_timer(apic);
1262 		break;
1263 
1264 	case APIC_TDCR:
1265 		if (val & 4)
1266 			apic_debug("KVM_WRITE:TDCR %x\n", val);
1267 		apic_set_reg(apic, APIC_TDCR, val);
1268 		update_divide_count(apic);
1269 		break;
1270 
1271 	case APIC_ESR:
1272 		if (apic_x2apic_mode(apic) && val != 0) {
1273 			apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1274 			ret = 1;
1275 		}
1276 		break;
1277 
1278 	case APIC_SELF_IPI:
1279 		if (apic_x2apic_mode(apic)) {
1280 			apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1281 		} else
1282 			ret = 1;
1283 		break;
1284 	default:
1285 		ret = 1;
1286 		break;
1287 	}
1288 	if (ret)
1289 		apic_debug("Local APIC Write to read-only register %x\n", reg);
1290 	return ret;
1291 }
1292 
1293 static int apic_mmio_write(struct kvm_io_device *this,
1294 			    gpa_t address, int len, const void *data)
1295 {
1296 	struct kvm_lapic *apic = to_lapic(this);
1297 	unsigned int offset = address - apic->base_address;
1298 	u32 val;
1299 
1300 	if (!apic_mmio_in_range(apic, address))
1301 		return -EOPNOTSUPP;
1302 
1303 	/*
1304 	 * APIC register must be aligned on 128-bits boundary.
1305 	 * 32/64/128 bits registers must be accessed thru 32 bits.
1306 	 * Refer SDM 8.4.1
1307 	 */
1308 	if (len != 4 || (offset & 0xf)) {
1309 		/* Don't shout loud, $infamous_os would cause only noise. */
1310 		apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1311 		return 0;
1312 	}
1313 
1314 	val = *(u32*)data;
1315 
1316 	/* too common printing */
1317 	if (offset != APIC_EOI)
1318 		apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1319 			   "0x%x\n", __func__, offset, len, val);
1320 
1321 	apic_reg_write(apic, offset & 0xff0, val);
1322 
1323 	return 0;
1324 }
1325 
1326 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1327 {
1328 	if (kvm_vcpu_has_lapic(vcpu))
1329 		apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1330 }
1331 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1332 
1333 /* emulate APIC access in a trap manner */
1334 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1335 {
1336 	u32 val = 0;
1337 
1338 	/* hw has done the conditional check and inst decode */
1339 	offset &= 0xff0;
1340 
1341 	apic_reg_read(vcpu->arch.apic, offset, 4, &val);
1342 
1343 	/* TODO: optimize to just emulate side effect w/o one more write */
1344 	apic_reg_write(vcpu->arch.apic, offset, val);
1345 }
1346 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1347 
1348 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1349 {
1350 	struct kvm_lapic *apic = vcpu->arch.apic;
1351 
1352 	if (!vcpu->arch.apic)
1353 		return;
1354 
1355 	hrtimer_cancel(&apic->lapic_timer.timer);
1356 
1357 	if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1358 		static_key_slow_dec_deferred(&apic_hw_disabled);
1359 
1360 	if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED))
1361 		static_key_slow_dec_deferred(&apic_sw_disabled);
1362 
1363 	if (apic->regs)
1364 		free_page((unsigned long)apic->regs);
1365 
1366 	kfree(apic);
1367 }
1368 
1369 /*
1370  *----------------------------------------------------------------------
1371  * LAPIC interface
1372  *----------------------------------------------------------------------
1373  */
1374 
1375 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1376 {
1377 	struct kvm_lapic *apic = vcpu->arch.apic;
1378 
1379 	if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1380 			apic_lvtt_period(apic))
1381 		return 0;
1382 
1383 	return apic->lapic_timer.tscdeadline;
1384 }
1385 
1386 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1387 {
1388 	struct kvm_lapic *apic = vcpu->arch.apic;
1389 
1390 	if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1391 			apic_lvtt_period(apic))
1392 		return;
1393 
1394 	hrtimer_cancel(&apic->lapic_timer.timer);
1395 	apic->lapic_timer.tscdeadline = data;
1396 	start_apic_timer(apic);
1397 }
1398 
1399 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1400 {
1401 	struct kvm_lapic *apic = vcpu->arch.apic;
1402 
1403 	if (!kvm_vcpu_has_lapic(vcpu))
1404 		return;
1405 
1406 	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1407 		     | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4));
1408 }
1409 
1410 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1411 {
1412 	u64 tpr;
1413 
1414 	if (!kvm_vcpu_has_lapic(vcpu))
1415 		return 0;
1416 
1417 	tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1418 
1419 	return (tpr & 0xf0) >> 4;
1420 }
1421 
1422 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1423 {
1424 	u64 old_value = vcpu->arch.apic_base;
1425 	struct kvm_lapic *apic = vcpu->arch.apic;
1426 
1427 	if (!apic) {
1428 		value |= MSR_IA32_APICBASE_BSP;
1429 		vcpu->arch.apic_base = value;
1430 		return;
1431 	}
1432 
1433 	if (!kvm_vcpu_is_bsp(apic->vcpu))
1434 		value &= ~MSR_IA32_APICBASE_BSP;
1435 	vcpu->arch.apic_base = value;
1436 
1437 	/* update jump label if enable bit changes */
1438 	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1439 		if (value & MSR_IA32_APICBASE_ENABLE)
1440 			static_key_slow_dec_deferred(&apic_hw_disabled);
1441 		else
1442 			static_key_slow_inc(&apic_hw_disabled.key);
1443 		recalculate_apic_map(vcpu->kvm);
1444 	}
1445 
1446 	if ((old_value ^ value) & X2APIC_ENABLE) {
1447 		if (value & X2APIC_ENABLE) {
1448 			u32 id = kvm_apic_id(apic);
1449 			u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
1450 			kvm_apic_set_ldr(apic, ldr);
1451 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1452 		} else
1453 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1454 	}
1455 
1456 	apic->base_address = apic->vcpu->arch.apic_base &
1457 			     MSR_IA32_APICBASE_BASE;
1458 
1459 	/* with FSB delivery interrupt, we can restart APIC functionality */
1460 	apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1461 		   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1462 
1463 }
1464 
1465 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1466 {
1467 	struct kvm_lapic *apic;
1468 	int i;
1469 
1470 	apic_debug("%s\n", __func__);
1471 
1472 	ASSERT(vcpu);
1473 	apic = vcpu->arch.apic;
1474 	ASSERT(apic != NULL);
1475 
1476 	/* Stop the timer in case it's a reset to an active apic */
1477 	hrtimer_cancel(&apic->lapic_timer.timer);
1478 
1479 	kvm_apic_set_id(apic, vcpu->vcpu_id);
1480 	kvm_apic_set_version(apic->vcpu);
1481 
1482 	for (i = 0; i < APIC_LVT_NUM; i++)
1483 		apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1484 	apic_set_reg(apic, APIC_LVT0,
1485 		     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1486 
1487 	apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1488 	apic_set_spiv(apic, 0xff);
1489 	apic_set_reg(apic, APIC_TASKPRI, 0);
1490 	kvm_apic_set_ldr(apic, 0);
1491 	apic_set_reg(apic, APIC_ESR, 0);
1492 	apic_set_reg(apic, APIC_ICR, 0);
1493 	apic_set_reg(apic, APIC_ICR2, 0);
1494 	apic_set_reg(apic, APIC_TDCR, 0);
1495 	apic_set_reg(apic, APIC_TMICT, 0);
1496 	for (i = 0; i < 8; i++) {
1497 		apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1498 		apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1499 		apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1500 	}
1501 	apic->irr_pending = kvm_apic_vid_enabled(vcpu->kvm);
1502 	apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm);
1503 	apic->highest_isr_cache = -1;
1504 	update_divide_count(apic);
1505 	atomic_set(&apic->lapic_timer.pending, 0);
1506 	if (kvm_vcpu_is_bsp(vcpu))
1507 		kvm_lapic_set_base(vcpu,
1508 				vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1509 	vcpu->arch.pv_eoi.msr_val = 0;
1510 	apic_update_ppr(apic);
1511 
1512 	vcpu->arch.apic_arb_prio = 0;
1513 	vcpu->arch.apic_attention = 0;
1514 
1515 	apic_debug("%s: vcpu=%p, id=%d, base_msr="
1516 		   "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1517 		   vcpu, kvm_apic_id(apic),
1518 		   vcpu->arch.apic_base, apic->base_address);
1519 }
1520 
1521 /*
1522  *----------------------------------------------------------------------
1523  * timer interface
1524  *----------------------------------------------------------------------
1525  */
1526 
1527 static bool lapic_is_periodic(struct kvm_lapic *apic)
1528 {
1529 	return apic_lvtt_period(apic);
1530 }
1531 
1532 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1533 {
1534 	struct kvm_lapic *apic = vcpu->arch.apic;
1535 
1536 	if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) &&
1537 			apic_lvt_enabled(apic, APIC_LVTT))
1538 		return atomic_read(&apic->lapic_timer.pending);
1539 
1540 	return 0;
1541 }
1542 
1543 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1544 {
1545 	u32 reg = kvm_apic_get_reg(apic, lvt_type);
1546 	int vector, mode, trig_mode;
1547 
1548 	if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1549 		vector = reg & APIC_VECTOR_MASK;
1550 		mode = reg & APIC_MODE_MASK;
1551 		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1552 		return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1553 					NULL);
1554 	}
1555 	return 0;
1556 }
1557 
1558 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1559 {
1560 	struct kvm_lapic *apic = vcpu->arch.apic;
1561 
1562 	if (apic)
1563 		kvm_apic_local_deliver(apic, APIC_LVT0);
1564 }
1565 
1566 static const struct kvm_io_device_ops apic_mmio_ops = {
1567 	.read     = apic_mmio_read,
1568 	.write    = apic_mmio_write,
1569 };
1570 
1571 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1572 {
1573 	struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1574 	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1575 
1576 	apic_timer_expired(apic);
1577 
1578 	if (lapic_is_periodic(apic)) {
1579 		hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1580 		return HRTIMER_RESTART;
1581 	} else
1582 		return HRTIMER_NORESTART;
1583 }
1584 
1585 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1586 {
1587 	struct kvm_lapic *apic;
1588 
1589 	ASSERT(vcpu != NULL);
1590 	apic_debug("apic_init %d\n", vcpu->vcpu_id);
1591 
1592 	apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1593 	if (!apic)
1594 		goto nomem;
1595 
1596 	vcpu->arch.apic = apic;
1597 
1598 	apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1599 	if (!apic->regs) {
1600 		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1601 		       vcpu->vcpu_id);
1602 		goto nomem_free_apic;
1603 	}
1604 	apic->vcpu = vcpu;
1605 
1606 	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1607 		     HRTIMER_MODE_ABS);
1608 	apic->lapic_timer.timer.function = apic_timer_fn;
1609 
1610 	/*
1611 	 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1612 	 * thinking that APIC satet has changed.
1613 	 */
1614 	vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1615 	kvm_lapic_set_base(vcpu,
1616 			APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE);
1617 
1618 	static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1619 	kvm_lapic_reset(vcpu);
1620 	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1621 
1622 	return 0;
1623 nomem_free_apic:
1624 	kfree(apic);
1625 nomem:
1626 	return -ENOMEM;
1627 }
1628 
1629 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1630 {
1631 	struct kvm_lapic *apic = vcpu->arch.apic;
1632 	int highest_irr;
1633 
1634 	if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic))
1635 		return -1;
1636 
1637 	apic_update_ppr(apic);
1638 	highest_irr = apic_find_highest_irr(apic);
1639 	if ((highest_irr == -1) ||
1640 	    ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI)))
1641 		return -1;
1642 	return highest_irr;
1643 }
1644 
1645 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1646 {
1647 	u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1648 	int r = 0;
1649 
1650 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1651 		r = 1;
1652 	if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1653 	    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1654 		r = 1;
1655 	return r;
1656 }
1657 
1658 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1659 {
1660 	struct kvm_lapic *apic = vcpu->arch.apic;
1661 
1662 	if (!kvm_vcpu_has_lapic(vcpu))
1663 		return;
1664 
1665 	if (atomic_read(&apic->lapic_timer.pending) > 0) {
1666 		kvm_apic_local_deliver(apic, APIC_LVTT);
1667 		if (apic_lvtt_tscdeadline(apic))
1668 			apic->lapic_timer.tscdeadline = 0;
1669 		atomic_set(&apic->lapic_timer.pending, 0);
1670 	}
1671 }
1672 
1673 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1674 {
1675 	int vector = kvm_apic_has_interrupt(vcpu);
1676 	struct kvm_lapic *apic = vcpu->arch.apic;
1677 
1678 	if (vector == -1)
1679 		return -1;
1680 
1681 	/*
1682 	 * We get here even with APIC virtualization enabled, if doing
1683 	 * nested virtualization and L1 runs with the "acknowledge interrupt
1684 	 * on exit" mode.  Then we cannot inject the interrupt via RVI,
1685 	 * because the process would deliver it through the IDT.
1686 	 */
1687 
1688 	apic_set_isr(vector, apic);
1689 	apic_update_ppr(apic);
1690 	apic_clear_irr(vector, apic);
1691 	return vector;
1692 }
1693 
1694 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
1695 		struct kvm_lapic_state *s)
1696 {
1697 	struct kvm_lapic *apic = vcpu->arch.apic;
1698 
1699 	kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
1700 	/* set SPIV separately to get count of SW disabled APICs right */
1701 	apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
1702 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1703 	/* call kvm_apic_set_id() to put apic into apic_map */
1704 	kvm_apic_set_id(apic, kvm_apic_id(apic));
1705 	kvm_apic_set_version(vcpu);
1706 
1707 	apic_update_ppr(apic);
1708 	hrtimer_cancel(&apic->lapic_timer.timer);
1709 	update_divide_count(apic);
1710 	start_apic_timer(apic);
1711 	apic->irr_pending = true;
1712 	apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm) ?
1713 				1 : count_vectors(apic->regs + APIC_ISR);
1714 	apic->highest_isr_cache = -1;
1715 	kvm_x86_ops->hwapic_isr_update(vcpu->kvm, apic_find_highest_isr(apic));
1716 	kvm_make_request(KVM_REQ_EVENT, vcpu);
1717 	kvm_rtc_eoi_tracking_restore_one(vcpu);
1718 }
1719 
1720 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1721 {
1722 	struct hrtimer *timer;
1723 
1724 	if (!kvm_vcpu_has_lapic(vcpu))
1725 		return;
1726 
1727 	timer = &vcpu->arch.apic->lapic_timer.timer;
1728 	if (hrtimer_cancel(timer))
1729 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1730 }
1731 
1732 /*
1733  * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1734  *
1735  * Detect whether guest triggered PV EOI since the
1736  * last entry. If yes, set EOI on guests's behalf.
1737  * Clear PV EOI in guest memory in any case.
1738  */
1739 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1740 					struct kvm_lapic *apic)
1741 {
1742 	bool pending;
1743 	int vector;
1744 	/*
1745 	 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1746 	 * and KVM_PV_EOI_ENABLED in guest memory as follows:
1747 	 *
1748 	 * KVM_APIC_PV_EOI_PENDING is unset:
1749 	 * 	-> host disabled PV EOI.
1750 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1751 	 * 	-> host enabled PV EOI, guest did not execute EOI yet.
1752 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1753 	 * 	-> host enabled PV EOI, guest executed EOI.
1754 	 */
1755 	BUG_ON(!pv_eoi_enabled(vcpu));
1756 	pending = pv_eoi_get_pending(vcpu);
1757 	/*
1758 	 * Clear pending bit in any case: it will be set again on vmentry.
1759 	 * While this might not be ideal from performance point of view,
1760 	 * this makes sure pv eoi is only enabled when we know it's safe.
1761 	 */
1762 	pv_eoi_clr_pending(vcpu);
1763 	if (pending)
1764 		return;
1765 	vector = apic_set_eoi(apic);
1766 	trace_kvm_pv_eoi(apic, vector);
1767 }
1768 
1769 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1770 {
1771 	u32 data;
1772 
1773 	if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1774 		apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1775 
1776 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1777 		return;
1778 
1779 	kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1780 				sizeof(u32));
1781 
1782 	apic_set_tpr(vcpu->arch.apic, data & 0xff);
1783 }
1784 
1785 /*
1786  * apic_sync_pv_eoi_to_guest - called before vmentry
1787  *
1788  * Detect whether it's safe to enable PV EOI and
1789  * if yes do so.
1790  */
1791 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
1792 					struct kvm_lapic *apic)
1793 {
1794 	if (!pv_eoi_enabled(vcpu) ||
1795 	    /* IRR set or many bits in ISR: could be nested. */
1796 	    apic->irr_pending ||
1797 	    /* Cache not set: could be safe but we don't bother. */
1798 	    apic->highest_isr_cache == -1 ||
1799 	    /* Need EOI to update ioapic. */
1800 	    kvm_ioapic_handles_vector(vcpu->kvm, apic->highest_isr_cache)) {
1801 		/*
1802 		 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1803 		 * so we need not do anything here.
1804 		 */
1805 		return;
1806 	}
1807 
1808 	pv_eoi_set_pending(apic->vcpu);
1809 }
1810 
1811 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1812 {
1813 	u32 data, tpr;
1814 	int max_irr, max_isr;
1815 	struct kvm_lapic *apic = vcpu->arch.apic;
1816 
1817 	apic_sync_pv_eoi_to_guest(vcpu, apic);
1818 
1819 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1820 		return;
1821 
1822 	tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1823 	max_irr = apic_find_highest_irr(apic);
1824 	if (max_irr < 0)
1825 		max_irr = 0;
1826 	max_isr = apic_find_highest_isr(apic);
1827 	if (max_isr < 0)
1828 		max_isr = 0;
1829 	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1830 
1831 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1832 				sizeof(u32));
1833 }
1834 
1835 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1836 {
1837 	if (vapic_addr) {
1838 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1839 					&vcpu->arch.apic->vapic_cache,
1840 					vapic_addr, sizeof(u32)))
1841 			return -EINVAL;
1842 		__set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1843 	} else {
1844 		__clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1845 	}
1846 
1847 	vcpu->arch.apic->vapic_addr = vapic_addr;
1848 	return 0;
1849 }
1850 
1851 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1852 {
1853 	struct kvm_lapic *apic = vcpu->arch.apic;
1854 	u32 reg = (msr - APIC_BASE_MSR) << 4;
1855 
1856 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1857 		return 1;
1858 
1859 	/* if this is ICR write vector before command */
1860 	if (msr == 0x830)
1861 		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1862 	return apic_reg_write(apic, reg, (u32)data);
1863 }
1864 
1865 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1866 {
1867 	struct kvm_lapic *apic = vcpu->arch.apic;
1868 	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1869 
1870 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1871 		return 1;
1872 
1873 	if (apic_reg_read(apic, reg, 4, &low))
1874 		return 1;
1875 	if (msr == 0x830)
1876 		apic_reg_read(apic, APIC_ICR2, 4, &high);
1877 
1878 	*data = (((u64)high) << 32) | low;
1879 
1880 	return 0;
1881 }
1882 
1883 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1884 {
1885 	struct kvm_lapic *apic = vcpu->arch.apic;
1886 
1887 	if (!kvm_vcpu_has_lapic(vcpu))
1888 		return 1;
1889 
1890 	/* if this is ICR write vector before command */
1891 	if (reg == APIC_ICR)
1892 		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1893 	return apic_reg_write(apic, reg, (u32)data);
1894 }
1895 
1896 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
1897 {
1898 	struct kvm_lapic *apic = vcpu->arch.apic;
1899 	u32 low, high = 0;
1900 
1901 	if (!kvm_vcpu_has_lapic(vcpu))
1902 		return 1;
1903 
1904 	if (apic_reg_read(apic, reg, 4, &low))
1905 		return 1;
1906 	if (reg == APIC_ICR)
1907 		apic_reg_read(apic, APIC_ICR2, 4, &high);
1908 
1909 	*data = (((u64)high) << 32) | low;
1910 
1911 	return 0;
1912 }
1913 
1914 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
1915 {
1916 	u64 addr = data & ~KVM_MSR_ENABLED;
1917 	if (!IS_ALIGNED(addr, 4))
1918 		return 1;
1919 
1920 	vcpu->arch.pv_eoi.msr_val = data;
1921 	if (!pv_eoi_enabled(vcpu))
1922 		return 0;
1923 	return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
1924 					 addr, sizeof(u8));
1925 }
1926 
1927 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
1928 {
1929 	struct kvm_lapic *apic = vcpu->arch.apic;
1930 	unsigned int sipi_vector;
1931 	unsigned long pe;
1932 
1933 	if (!kvm_vcpu_has_lapic(vcpu) || !apic->pending_events)
1934 		return;
1935 
1936 	pe = xchg(&apic->pending_events, 0);
1937 
1938 	if (test_bit(KVM_APIC_INIT, &pe)) {
1939 		kvm_lapic_reset(vcpu);
1940 		kvm_vcpu_reset(vcpu);
1941 		if (kvm_vcpu_is_bsp(apic->vcpu))
1942 			vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1943 		else
1944 			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
1945 	}
1946 	if (test_bit(KVM_APIC_SIPI, &pe) &&
1947 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
1948 		/* evaluate pending_events before reading the vector */
1949 		smp_rmb();
1950 		sipi_vector = apic->sipi_vector;
1951 		apic_debug("vcpu %d received sipi with vector # %x\n",
1952 			 vcpu->vcpu_id, sipi_vector);
1953 		kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
1954 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1955 	}
1956 }
1957 
1958 void kvm_lapic_init(void)
1959 {
1960 	/* do not patch jump label more than once per second */
1961 	jump_label_rate_limit(&apic_hw_disabled, HZ);
1962 	jump_label_rate_limit(&apic_sw_disabled, HZ);
1963 }
1964