xref: /openbmc/linux/arch/x86/kvm/lapic.c (revision 5c73cc4b6c83e88863a5de869cc5df3b913aef4a)
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 /* The logical map is definitely wrong if we have multiple
137  * modes at the same time.  (Physical map is always right.)
138  */
139 static inline bool kvm_apic_logical_map_valid(struct kvm_apic_map *map)
140 {
141 	return !(map->mode & (map->mode - 1));
142 }
143 
144 static inline void
145 apic_logical_id(struct kvm_apic_map *map, u32 dest_id, u16 *cid, u16 *lid)
146 {
147 	unsigned lid_bits;
148 
149 	BUILD_BUG_ON(KVM_APIC_MODE_XAPIC_CLUSTER !=  4);
150 	BUILD_BUG_ON(KVM_APIC_MODE_XAPIC_FLAT    !=  8);
151 	BUILD_BUG_ON(KVM_APIC_MODE_X2APIC        != 16);
152 	lid_bits = map->mode;
153 
154 	*cid = dest_id >> lid_bits;
155 	*lid = dest_id & ((1 << lid_bits) - 1);
156 }
157 
158 static void recalculate_apic_map(struct kvm *kvm)
159 {
160 	struct kvm_apic_map *new, *old = NULL;
161 	struct kvm_vcpu *vcpu;
162 	int i;
163 
164 	new = kzalloc(sizeof(struct kvm_apic_map), GFP_KERNEL);
165 
166 	mutex_lock(&kvm->arch.apic_map_lock);
167 
168 	if (!new)
169 		goto out;
170 
171 	kvm_for_each_vcpu(i, vcpu, kvm) {
172 		struct kvm_lapic *apic = vcpu->arch.apic;
173 		u16 cid, lid;
174 		u32 ldr, aid;
175 
176 		if (!kvm_apic_present(vcpu))
177 			continue;
178 
179 		aid = kvm_apic_id(apic);
180 		ldr = kvm_apic_get_reg(apic, APIC_LDR);
181 
182 		if (aid < ARRAY_SIZE(new->phys_map))
183 			new->phys_map[aid] = apic;
184 
185 		if (apic_x2apic_mode(apic)) {
186 			new->mode |= KVM_APIC_MODE_X2APIC;
187 		} else if (ldr) {
188 			ldr = GET_APIC_LOGICAL_ID(ldr);
189 			if (kvm_apic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
190 				new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
191 			else
192 				new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
193 		}
194 
195 		if (!kvm_apic_logical_map_valid(new))
196 			continue;
197 
198 		apic_logical_id(new, ldr, &cid, &lid);
199 
200 		if (lid && cid < ARRAY_SIZE(new->logical_map))
201 			new->logical_map[cid][ffs(lid) - 1] = apic;
202 	}
203 out:
204 	old = rcu_dereference_protected(kvm->arch.apic_map,
205 			lockdep_is_held(&kvm->arch.apic_map_lock));
206 	rcu_assign_pointer(kvm->arch.apic_map, new);
207 	mutex_unlock(&kvm->arch.apic_map_lock);
208 
209 	if (old)
210 		kfree_rcu(old, rcu);
211 
212 	kvm_vcpu_request_scan_ioapic(kvm);
213 }
214 
215 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
216 {
217 	bool enabled = val & APIC_SPIV_APIC_ENABLED;
218 
219 	apic_set_reg(apic, APIC_SPIV, val);
220 
221 	if (enabled != apic->sw_enabled) {
222 		apic->sw_enabled = enabled;
223 		if (enabled) {
224 			static_key_slow_dec_deferred(&apic_sw_disabled);
225 			recalculate_apic_map(apic->vcpu->kvm);
226 		} else
227 			static_key_slow_inc(&apic_sw_disabled.key);
228 	}
229 }
230 
231 static inline void kvm_apic_set_id(struct kvm_lapic *apic, u8 id)
232 {
233 	apic_set_reg(apic, APIC_ID, id << 24);
234 	recalculate_apic_map(apic->vcpu->kvm);
235 }
236 
237 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
238 {
239 	apic_set_reg(apic, APIC_LDR, id);
240 	recalculate_apic_map(apic->vcpu->kvm);
241 }
242 
243 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
244 {
245 	return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
246 }
247 
248 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
249 {
250 	return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
251 }
252 
253 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
254 {
255 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
256 }
257 
258 static inline int apic_lvtt_period(struct kvm_lapic *apic)
259 {
260 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
261 }
262 
263 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
264 {
265 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
266 }
267 
268 static inline int apic_lvt_nmi_mode(u32 lvt_val)
269 {
270 	return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
271 }
272 
273 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
274 {
275 	struct kvm_lapic *apic = vcpu->arch.apic;
276 	struct kvm_cpuid_entry2 *feat;
277 	u32 v = APIC_VERSION;
278 
279 	if (!kvm_vcpu_has_lapic(vcpu))
280 		return;
281 
282 	feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
283 	if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
284 		v |= APIC_LVR_DIRECTED_EOI;
285 	apic_set_reg(apic, APIC_LVR, v);
286 }
287 
288 static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
289 	LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
290 	LVT_MASK | APIC_MODE_MASK,	/* LVTTHMR */
291 	LVT_MASK | APIC_MODE_MASK,	/* LVTPC */
292 	LINT_MASK, LINT_MASK,	/* LVT0-1 */
293 	LVT_MASK		/* LVTERR */
294 };
295 
296 static int find_highest_vector(void *bitmap)
297 {
298 	int vec;
299 	u32 *reg;
300 
301 	for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
302 	     vec >= 0; vec -= APIC_VECTORS_PER_REG) {
303 		reg = bitmap + REG_POS(vec);
304 		if (*reg)
305 			return fls(*reg) - 1 + vec;
306 	}
307 
308 	return -1;
309 }
310 
311 static u8 count_vectors(void *bitmap)
312 {
313 	int vec;
314 	u32 *reg;
315 	u8 count = 0;
316 
317 	for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
318 		reg = bitmap + REG_POS(vec);
319 		count += hweight32(*reg);
320 	}
321 
322 	return count;
323 }
324 
325 void __kvm_apic_update_irr(u32 *pir, void *regs)
326 {
327 	u32 i, pir_val;
328 
329 	for (i = 0; i <= 7; i++) {
330 		pir_val = xchg(&pir[i], 0);
331 		if (pir_val)
332 			*((u32 *)(regs + APIC_IRR + i * 0x10)) |= pir_val;
333 	}
334 }
335 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
336 
337 void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
338 {
339 	struct kvm_lapic *apic = vcpu->arch.apic;
340 
341 	__kvm_apic_update_irr(pir, apic->regs);
342 }
343 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
344 
345 static inline void apic_set_irr(int vec, struct kvm_lapic *apic)
346 {
347 	apic_set_vector(vec, apic->regs + APIC_IRR);
348 	/*
349 	 * irr_pending must be true if any interrupt is pending; set it after
350 	 * APIC_IRR to avoid race with apic_clear_irr
351 	 */
352 	apic->irr_pending = true;
353 }
354 
355 static inline int apic_search_irr(struct kvm_lapic *apic)
356 {
357 	return find_highest_vector(apic->regs + APIC_IRR);
358 }
359 
360 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
361 {
362 	int result;
363 
364 	/*
365 	 * Note that irr_pending is just a hint. It will be always
366 	 * true with virtual interrupt delivery enabled.
367 	 */
368 	if (!apic->irr_pending)
369 		return -1;
370 
371 	kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
372 	result = apic_search_irr(apic);
373 	ASSERT(result == -1 || result >= 16);
374 
375 	return result;
376 }
377 
378 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
379 {
380 	struct kvm_vcpu *vcpu;
381 
382 	vcpu = apic->vcpu;
383 
384 	if (unlikely(kvm_apic_vid_enabled(vcpu->kvm))) {
385 		/* try to update RVI */
386 		apic_clear_vector(vec, apic->regs + APIC_IRR);
387 		kvm_make_request(KVM_REQ_EVENT, vcpu);
388 	} else {
389 		apic->irr_pending = false;
390 		apic_clear_vector(vec, apic->regs + APIC_IRR);
391 		if (apic_search_irr(apic) != -1)
392 			apic->irr_pending = true;
393 	}
394 }
395 
396 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
397 {
398 	struct kvm_vcpu *vcpu;
399 
400 	if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
401 		return;
402 
403 	vcpu = apic->vcpu;
404 
405 	/*
406 	 * With APIC virtualization enabled, all caching is disabled
407 	 * because the processor can modify ISR under the hood.  Instead
408 	 * just set SVI.
409 	 */
410 	if (unlikely(kvm_x86_ops->hwapic_isr_update))
411 		kvm_x86_ops->hwapic_isr_update(vcpu->kvm, vec);
412 	else {
413 		++apic->isr_count;
414 		BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
415 		/*
416 		 * ISR (in service register) bit is set when injecting an interrupt.
417 		 * The highest vector is injected. Thus the latest bit set matches
418 		 * the highest bit in ISR.
419 		 */
420 		apic->highest_isr_cache = vec;
421 	}
422 }
423 
424 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
425 {
426 	int result;
427 
428 	/*
429 	 * Note that isr_count is always 1, and highest_isr_cache
430 	 * is always -1, with APIC virtualization enabled.
431 	 */
432 	if (!apic->isr_count)
433 		return -1;
434 	if (likely(apic->highest_isr_cache != -1))
435 		return apic->highest_isr_cache;
436 
437 	result = find_highest_vector(apic->regs + APIC_ISR);
438 	ASSERT(result == -1 || result >= 16);
439 
440 	return result;
441 }
442 
443 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
444 {
445 	struct kvm_vcpu *vcpu;
446 	if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
447 		return;
448 
449 	vcpu = apic->vcpu;
450 
451 	/*
452 	 * We do get here for APIC virtualization enabled if the guest
453 	 * uses the Hyper-V APIC enlightenment.  In this case we may need
454 	 * to trigger a new interrupt delivery by writing the SVI field;
455 	 * on the other hand isr_count and highest_isr_cache are unused
456 	 * and must be left alone.
457 	 */
458 	if (unlikely(kvm_x86_ops->hwapic_isr_update))
459 		kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
460 					       apic_find_highest_isr(apic));
461 	else {
462 		--apic->isr_count;
463 		BUG_ON(apic->isr_count < 0);
464 		apic->highest_isr_cache = -1;
465 	}
466 }
467 
468 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
469 {
470 	int highest_irr;
471 
472 	/* This may race with setting of irr in __apic_accept_irq() and
473 	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
474 	 * will cause vmexit immediately and the value will be recalculated
475 	 * on the next vmentry.
476 	 */
477 	if (!kvm_vcpu_has_lapic(vcpu))
478 		return 0;
479 	highest_irr = apic_find_highest_irr(vcpu->arch.apic);
480 
481 	return highest_irr;
482 }
483 
484 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
485 			     int vector, int level, int trig_mode,
486 			     unsigned long *dest_map);
487 
488 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
489 		unsigned long *dest_map)
490 {
491 	struct kvm_lapic *apic = vcpu->arch.apic;
492 
493 	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
494 			irq->level, irq->trig_mode, dest_map);
495 }
496 
497 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
498 {
499 
500 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
501 				      sizeof(val));
502 }
503 
504 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
505 {
506 
507 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
508 				      sizeof(*val));
509 }
510 
511 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
512 {
513 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
514 }
515 
516 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
517 {
518 	u8 val;
519 	if (pv_eoi_get_user(vcpu, &val) < 0)
520 		apic_debug("Can't read EOI MSR value: 0x%llx\n",
521 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
522 	return val & 0x1;
523 }
524 
525 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
526 {
527 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
528 		apic_debug("Can't set EOI MSR value: 0x%llx\n",
529 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
530 		return;
531 	}
532 	__set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
533 }
534 
535 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
536 {
537 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
538 		apic_debug("Can't clear EOI MSR value: 0x%llx\n",
539 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
540 		return;
541 	}
542 	__clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
543 }
544 
545 void kvm_apic_update_tmr(struct kvm_vcpu *vcpu, u32 *tmr)
546 {
547 	struct kvm_lapic *apic = vcpu->arch.apic;
548 	int i;
549 
550 	for (i = 0; i < 8; i++)
551 		apic_set_reg(apic, APIC_TMR + 0x10 * i, tmr[i]);
552 }
553 
554 static void apic_update_ppr(struct kvm_lapic *apic)
555 {
556 	u32 tpr, isrv, ppr, old_ppr;
557 	int isr;
558 
559 	old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI);
560 	tpr = kvm_apic_get_reg(apic, APIC_TASKPRI);
561 	isr = apic_find_highest_isr(apic);
562 	isrv = (isr != -1) ? isr : 0;
563 
564 	if ((tpr & 0xf0) >= (isrv & 0xf0))
565 		ppr = tpr & 0xff;
566 	else
567 		ppr = isrv & 0xf0;
568 
569 	apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
570 		   apic, ppr, isr, isrv);
571 
572 	if (old_ppr != ppr) {
573 		apic_set_reg(apic, APIC_PROCPRI, ppr);
574 		if (ppr < old_ppr)
575 			kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
576 	}
577 }
578 
579 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
580 {
581 	apic_set_reg(apic, APIC_TASKPRI, tpr);
582 	apic_update_ppr(apic);
583 }
584 
585 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
586 {
587 	if (apic_x2apic_mode(apic))
588 		return mda == X2APIC_BROADCAST;
589 
590 	return GET_APIC_DEST_FIELD(mda) == APIC_BROADCAST;
591 }
592 
593 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
594 {
595 	if (kvm_apic_broadcast(apic, mda))
596 		return true;
597 
598 	if (apic_x2apic_mode(apic))
599 		return mda == kvm_apic_id(apic);
600 
601 	return mda == SET_APIC_DEST_FIELD(kvm_apic_id(apic));
602 }
603 
604 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
605 {
606 	u32 logical_id;
607 
608 	if (kvm_apic_broadcast(apic, mda))
609 		return true;
610 
611 	logical_id = kvm_apic_get_reg(apic, APIC_LDR);
612 
613 	if (apic_x2apic_mode(apic))
614 		return ((logical_id >> 16) == (mda >> 16))
615 		       && (logical_id & mda & 0xffff) != 0;
616 
617 	logical_id = GET_APIC_LOGICAL_ID(logical_id);
618 	mda = GET_APIC_DEST_FIELD(mda);
619 
620 	switch (kvm_apic_get_reg(apic, APIC_DFR)) {
621 	case APIC_DFR_FLAT:
622 		return (logical_id & mda) != 0;
623 	case APIC_DFR_CLUSTER:
624 		return ((logical_id >> 4) == (mda >> 4))
625 		       && (logical_id & mda & 0xf) != 0;
626 	default:
627 		apic_debug("Bad DFR vcpu %d: %08x\n",
628 			   apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR));
629 		return false;
630 	}
631 }
632 
633 /* KVM APIC implementation has two quirks
634  *  - dest always begins at 0 while xAPIC MDA has offset 24,
635  *  - IOxAPIC messages have to be delivered (directly) to x2APIC.
636  */
637 static u32 kvm_apic_mda(unsigned int dest_id, struct kvm_lapic *source,
638                                               struct kvm_lapic *target)
639 {
640 	bool ipi = source != NULL;
641 	bool x2apic_mda = apic_x2apic_mode(ipi ? source : target);
642 
643 	if (!ipi && dest_id == APIC_BROADCAST && x2apic_mda)
644 		return X2APIC_BROADCAST;
645 
646 	return x2apic_mda ? dest_id : SET_APIC_DEST_FIELD(dest_id);
647 }
648 
649 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
650 			   int short_hand, unsigned int dest, int dest_mode)
651 {
652 	struct kvm_lapic *target = vcpu->arch.apic;
653 	u32 mda = kvm_apic_mda(dest, source, target);
654 
655 	apic_debug("target %p, source %p, dest 0x%x, "
656 		   "dest_mode 0x%x, short_hand 0x%x\n",
657 		   target, source, dest, dest_mode, short_hand);
658 
659 	ASSERT(target);
660 	switch (short_hand) {
661 	case APIC_DEST_NOSHORT:
662 		if (dest_mode == APIC_DEST_PHYSICAL)
663 			return kvm_apic_match_physical_addr(target, mda);
664 		else
665 			return kvm_apic_match_logical_addr(target, mda);
666 	case APIC_DEST_SELF:
667 		return target == source;
668 	case APIC_DEST_ALLINC:
669 		return true;
670 	case APIC_DEST_ALLBUT:
671 		return target != source;
672 	default:
673 		apic_debug("kvm: apic: Bad dest shorthand value %x\n",
674 			   short_hand);
675 		return false;
676 	}
677 }
678 
679 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
680 		struct kvm_lapic_irq *irq, int *r, unsigned long *dest_map)
681 {
682 	struct kvm_apic_map *map;
683 	unsigned long bitmap = 1;
684 	struct kvm_lapic **dst;
685 	int i;
686 	bool ret = false;
687 	bool x2apic_ipi = src && apic_x2apic_mode(src);
688 
689 	*r = -1;
690 
691 	if (irq->shorthand == APIC_DEST_SELF) {
692 		*r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
693 		return true;
694 	}
695 
696 	if (irq->shorthand)
697 		return false;
698 
699 	if (irq->dest_id == (x2apic_ipi ? X2APIC_BROADCAST : APIC_BROADCAST))
700 		return false;
701 
702 	rcu_read_lock();
703 	map = rcu_dereference(kvm->arch.apic_map);
704 
705 	if (!map)
706 		goto out;
707 
708 	ret = true;
709 
710 	if (irq->dest_mode == APIC_DEST_PHYSICAL) {
711 		if (irq->dest_id >= ARRAY_SIZE(map->phys_map))
712 			goto out;
713 
714 		dst = &map->phys_map[irq->dest_id];
715 	} else {
716 		u16 cid;
717 
718 		if (!kvm_apic_logical_map_valid(map)) {
719 			ret = false;
720 			goto out;
721 		}
722 
723 		apic_logical_id(map, irq->dest_id, &cid, (u16 *)&bitmap);
724 
725 		if (cid >= ARRAY_SIZE(map->logical_map))
726 			goto out;
727 
728 		dst = map->logical_map[cid];
729 
730 		if (irq->delivery_mode == APIC_DM_LOWEST) {
731 			int l = -1;
732 			for_each_set_bit(i, &bitmap, 16) {
733 				if (!dst[i])
734 					continue;
735 				if (l < 0)
736 					l = i;
737 				else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0)
738 					l = i;
739 			}
740 
741 			bitmap = (l >= 0) ? 1 << l : 0;
742 		}
743 	}
744 
745 	for_each_set_bit(i, &bitmap, 16) {
746 		if (!dst[i])
747 			continue;
748 		if (*r < 0)
749 			*r = 0;
750 		*r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
751 	}
752 out:
753 	rcu_read_unlock();
754 	return ret;
755 }
756 
757 /*
758  * Add a pending IRQ into lapic.
759  * Return 1 if successfully added and 0 if discarded.
760  */
761 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
762 			     int vector, int level, int trig_mode,
763 			     unsigned long *dest_map)
764 {
765 	int result = 0;
766 	struct kvm_vcpu *vcpu = apic->vcpu;
767 
768 	trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
769 				  trig_mode, vector);
770 	switch (delivery_mode) {
771 	case APIC_DM_LOWEST:
772 		vcpu->arch.apic_arb_prio++;
773 	case APIC_DM_FIXED:
774 		/* FIXME add logic for vcpu on reset */
775 		if (unlikely(!apic_enabled(apic)))
776 			break;
777 
778 		result = 1;
779 
780 		if (dest_map)
781 			__set_bit(vcpu->vcpu_id, dest_map);
782 
783 		if (kvm_x86_ops->deliver_posted_interrupt)
784 			kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
785 		else {
786 			apic_set_irr(vector, apic);
787 
788 			kvm_make_request(KVM_REQ_EVENT, vcpu);
789 			kvm_vcpu_kick(vcpu);
790 		}
791 		break;
792 
793 	case APIC_DM_REMRD:
794 		result = 1;
795 		vcpu->arch.pv.pv_unhalted = 1;
796 		kvm_make_request(KVM_REQ_EVENT, vcpu);
797 		kvm_vcpu_kick(vcpu);
798 		break;
799 
800 	case APIC_DM_SMI:
801 		apic_debug("Ignoring guest SMI\n");
802 		break;
803 
804 	case APIC_DM_NMI:
805 		result = 1;
806 		kvm_inject_nmi(vcpu);
807 		kvm_vcpu_kick(vcpu);
808 		break;
809 
810 	case APIC_DM_INIT:
811 		if (!trig_mode || level) {
812 			result = 1;
813 			/* assumes that there are only KVM_APIC_INIT/SIPI */
814 			apic->pending_events = (1UL << KVM_APIC_INIT);
815 			/* make sure pending_events is visible before sending
816 			 * the request */
817 			smp_wmb();
818 			kvm_make_request(KVM_REQ_EVENT, vcpu);
819 			kvm_vcpu_kick(vcpu);
820 		} else {
821 			apic_debug("Ignoring de-assert INIT to vcpu %d\n",
822 				   vcpu->vcpu_id);
823 		}
824 		break;
825 
826 	case APIC_DM_STARTUP:
827 		apic_debug("SIPI to vcpu %d vector 0x%02x\n",
828 			   vcpu->vcpu_id, vector);
829 		result = 1;
830 		apic->sipi_vector = vector;
831 		/* make sure sipi_vector is visible for the receiver */
832 		smp_wmb();
833 		set_bit(KVM_APIC_SIPI, &apic->pending_events);
834 		kvm_make_request(KVM_REQ_EVENT, vcpu);
835 		kvm_vcpu_kick(vcpu);
836 		break;
837 
838 	case APIC_DM_EXTINT:
839 		/*
840 		 * Should only be called by kvm_apic_local_deliver() with LVT0,
841 		 * before NMI watchdog was enabled. Already handled by
842 		 * kvm_apic_accept_pic_intr().
843 		 */
844 		break;
845 
846 	default:
847 		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
848 		       delivery_mode);
849 		break;
850 	}
851 	return result;
852 }
853 
854 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
855 {
856 	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
857 }
858 
859 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
860 {
861 	if (kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
862 		int trigger_mode;
863 		if (apic_test_vector(vector, apic->regs + APIC_TMR))
864 			trigger_mode = IOAPIC_LEVEL_TRIG;
865 		else
866 			trigger_mode = IOAPIC_EDGE_TRIG;
867 		kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
868 	}
869 }
870 
871 static int apic_set_eoi(struct kvm_lapic *apic)
872 {
873 	int vector = apic_find_highest_isr(apic);
874 
875 	trace_kvm_eoi(apic, vector);
876 
877 	/*
878 	 * Not every write EOI will has corresponding ISR,
879 	 * one example is when Kernel check timer on setup_IO_APIC
880 	 */
881 	if (vector == -1)
882 		return vector;
883 
884 	apic_clear_isr(vector, apic);
885 	apic_update_ppr(apic);
886 
887 	kvm_ioapic_send_eoi(apic, vector);
888 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
889 	return vector;
890 }
891 
892 /*
893  * this interface assumes a trap-like exit, which has already finished
894  * desired side effect including vISR and vPPR update.
895  */
896 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
897 {
898 	struct kvm_lapic *apic = vcpu->arch.apic;
899 
900 	trace_kvm_eoi(apic, vector);
901 
902 	kvm_ioapic_send_eoi(apic, vector);
903 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
904 }
905 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
906 
907 static void apic_send_ipi(struct kvm_lapic *apic)
908 {
909 	u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR);
910 	u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2);
911 	struct kvm_lapic_irq irq;
912 
913 	irq.vector = icr_low & APIC_VECTOR_MASK;
914 	irq.delivery_mode = icr_low & APIC_MODE_MASK;
915 	irq.dest_mode = icr_low & APIC_DEST_MASK;
916 	irq.level = icr_low & APIC_INT_ASSERT;
917 	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
918 	irq.shorthand = icr_low & APIC_SHORT_MASK;
919 	if (apic_x2apic_mode(apic))
920 		irq.dest_id = icr_high;
921 	else
922 		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
923 
924 	trace_kvm_apic_ipi(icr_low, irq.dest_id);
925 
926 	apic_debug("icr_high 0x%x, icr_low 0x%x, "
927 		   "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
928 		   "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
929 		   icr_high, icr_low, irq.shorthand, irq.dest_id,
930 		   irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
931 		   irq.vector);
932 
933 	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
934 }
935 
936 static u32 apic_get_tmcct(struct kvm_lapic *apic)
937 {
938 	ktime_t remaining;
939 	s64 ns;
940 	u32 tmcct;
941 
942 	ASSERT(apic != NULL);
943 
944 	/* if initial count is 0, current count should also be 0 */
945 	if (kvm_apic_get_reg(apic, APIC_TMICT) == 0 ||
946 		apic->lapic_timer.period == 0)
947 		return 0;
948 
949 	remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
950 	if (ktime_to_ns(remaining) < 0)
951 		remaining = ktime_set(0, 0);
952 
953 	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
954 	tmcct = div64_u64(ns,
955 			 (APIC_BUS_CYCLE_NS * apic->divide_count));
956 
957 	return tmcct;
958 }
959 
960 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
961 {
962 	struct kvm_vcpu *vcpu = apic->vcpu;
963 	struct kvm_run *run = vcpu->run;
964 
965 	kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
966 	run->tpr_access.rip = kvm_rip_read(vcpu);
967 	run->tpr_access.is_write = write;
968 }
969 
970 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
971 {
972 	if (apic->vcpu->arch.tpr_access_reporting)
973 		__report_tpr_access(apic, write);
974 }
975 
976 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
977 {
978 	u32 val = 0;
979 
980 	if (offset >= LAPIC_MMIO_LENGTH)
981 		return 0;
982 
983 	switch (offset) {
984 	case APIC_ID:
985 		if (apic_x2apic_mode(apic))
986 			val = kvm_apic_id(apic);
987 		else
988 			val = kvm_apic_id(apic) << 24;
989 		break;
990 	case APIC_ARBPRI:
991 		apic_debug("Access APIC ARBPRI register which is for P6\n");
992 		break;
993 
994 	case APIC_TMCCT:	/* Timer CCR */
995 		if (apic_lvtt_tscdeadline(apic))
996 			return 0;
997 
998 		val = apic_get_tmcct(apic);
999 		break;
1000 	case APIC_PROCPRI:
1001 		apic_update_ppr(apic);
1002 		val = kvm_apic_get_reg(apic, offset);
1003 		break;
1004 	case APIC_TASKPRI:
1005 		report_tpr_access(apic, false);
1006 		/* fall thru */
1007 	default:
1008 		val = kvm_apic_get_reg(apic, offset);
1009 		break;
1010 	}
1011 
1012 	return val;
1013 }
1014 
1015 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1016 {
1017 	return container_of(dev, struct kvm_lapic, dev);
1018 }
1019 
1020 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1021 		void *data)
1022 {
1023 	unsigned char alignment = offset & 0xf;
1024 	u32 result;
1025 	/* this bitmask has a bit cleared for each reserved register */
1026 	static const u64 rmask = 0x43ff01ffffffe70cULL;
1027 
1028 	if ((alignment + len) > 4) {
1029 		apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1030 			   offset, len);
1031 		return 1;
1032 	}
1033 
1034 	if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1035 		apic_debug("KVM_APIC_READ: read reserved register %x\n",
1036 			   offset);
1037 		return 1;
1038 	}
1039 
1040 	result = __apic_read(apic, offset & ~0xf);
1041 
1042 	trace_kvm_apic_read(offset, result);
1043 
1044 	switch (len) {
1045 	case 1:
1046 	case 2:
1047 	case 4:
1048 		memcpy(data, (char *)&result + alignment, len);
1049 		break;
1050 	default:
1051 		printk(KERN_ERR "Local APIC read with len = %x, "
1052 		       "should be 1,2, or 4 instead\n", len);
1053 		break;
1054 	}
1055 	return 0;
1056 }
1057 
1058 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1059 {
1060 	return kvm_apic_hw_enabled(apic) &&
1061 	    addr >= apic->base_address &&
1062 	    addr < apic->base_address + LAPIC_MMIO_LENGTH;
1063 }
1064 
1065 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1066 			   gpa_t address, int len, void *data)
1067 {
1068 	struct kvm_lapic *apic = to_lapic(this);
1069 	u32 offset = address - apic->base_address;
1070 
1071 	if (!apic_mmio_in_range(apic, address))
1072 		return -EOPNOTSUPP;
1073 
1074 	apic_reg_read(apic, offset, len, data);
1075 
1076 	return 0;
1077 }
1078 
1079 static void update_divide_count(struct kvm_lapic *apic)
1080 {
1081 	u32 tmp1, tmp2, tdcr;
1082 
1083 	tdcr = kvm_apic_get_reg(apic, APIC_TDCR);
1084 	tmp1 = tdcr & 0xf;
1085 	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1086 	apic->divide_count = 0x1 << (tmp2 & 0x7);
1087 
1088 	apic_debug("timer divide count is 0x%x\n",
1089 				   apic->divide_count);
1090 }
1091 
1092 static void apic_timer_expired(struct kvm_lapic *apic)
1093 {
1094 	struct kvm_vcpu *vcpu = apic->vcpu;
1095 	wait_queue_head_t *q = &vcpu->wq;
1096 	struct kvm_timer *ktimer = &apic->lapic_timer;
1097 
1098 	if (atomic_read(&apic->lapic_timer.pending))
1099 		return;
1100 
1101 	atomic_inc(&apic->lapic_timer.pending);
1102 	kvm_set_pending_timer(vcpu);
1103 
1104 	if (waitqueue_active(q))
1105 		wake_up_interruptible(q);
1106 
1107 	if (apic_lvtt_tscdeadline(apic))
1108 		ktimer->expired_tscdeadline = ktimer->tscdeadline;
1109 }
1110 
1111 /*
1112  * On APICv, this test will cause a busy wait
1113  * during a higher-priority task.
1114  */
1115 
1116 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1117 {
1118 	struct kvm_lapic *apic = vcpu->arch.apic;
1119 	u32 reg = kvm_apic_get_reg(apic, APIC_LVTT);
1120 
1121 	if (kvm_apic_hw_enabled(apic)) {
1122 		int vec = reg & APIC_VECTOR_MASK;
1123 		void *bitmap = apic->regs + APIC_ISR;
1124 
1125 		if (kvm_x86_ops->deliver_posted_interrupt)
1126 			bitmap = apic->regs + APIC_IRR;
1127 
1128 		if (apic_test_vector(vec, bitmap))
1129 			return true;
1130 	}
1131 	return false;
1132 }
1133 
1134 void wait_lapic_expire(struct kvm_vcpu *vcpu)
1135 {
1136 	struct kvm_lapic *apic = vcpu->arch.apic;
1137 	u64 guest_tsc, tsc_deadline;
1138 
1139 	if (!kvm_vcpu_has_lapic(vcpu))
1140 		return;
1141 
1142 	if (apic->lapic_timer.expired_tscdeadline == 0)
1143 		return;
1144 
1145 	if (!lapic_timer_int_injected(vcpu))
1146 		return;
1147 
1148 	tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1149 	apic->lapic_timer.expired_tscdeadline = 0;
1150 	guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1151 	trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1152 
1153 	/* __delay is delay_tsc whenever the hardware has TSC, thus always.  */
1154 	if (guest_tsc < tsc_deadline)
1155 		__delay(tsc_deadline - guest_tsc);
1156 }
1157 
1158 static void start_apic_timer(struct kvm_lapic *apic)
1159 {
1160 	ktime_t now;
1161 
1162 	atomic_set(&apic->lapic_timer.pending, 0);
1163 
1164 	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1165 		/* lapic timer in oneshot or periodic mode */
1166 		now = apic->lapic_timer.timer.base->get_time();
1167 		apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT)
1168 			    * APIC_BUS_CYCLE_NS * apic->divide_count;
1169 
1170 		if (!apic->lapic_timer.period)
1171 			return;
1172 		/*
1173 		 * Do not allow the guest to program periodic timers with small
1174 		 * interval, since the hrtimers are not throttled by the host
1175 		 * scheduler.
1176 		 */
1177 		if (apic_lvtt_period(apic)) {
1178 			s64 min_period = min_timer_period_us * 1000LL;
1179 
1180 			if (apic->lapic_timer.period < min_period) {
1181 				pr_info_ratelimited(
1182 				    "kvm: vcpu %i: requested %lld ns "
1183 				    "lapic timer period limited to %lld ns\n",
1184 				    apic->vcpu->vcpu_id,
1185 				    apic->lapic_timer.period, min_period);
1186 				apic->lapic_timer.period = min_period;
1187 			}
1188 		}
1189 
1190 		hrtimer_start(&apic->lapic_timer.timer,
1191 			      ktime_add_ns(now, apic->lapic_timer.period),
1192 			      HRTIMER_MODE_ABS);
1193 
1194 		apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1195 			   PRIx64 ", "
1196 			   "timer initial count 0x%x, period %lldns, "
1197 			   "expire @ 0x%016" PRIx64 ".\n", __func__,
1198 			   APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1199 			   kvm_apic_get_reg(apic, APIC_TMICT),
1200 			   apic->lapic_timer.period,
1201 			   ktime_to_ns(ktime_add_ns(now,
1202 					apic->lapic_timer.period)));
1203 	} else if (apic_lvtt_tscdeadline(apic)) {
1204 		/* lapic timer in tsc deadline mode */
1205 		u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1206 		u64 ns = 0;
1207 		ktime_t expire;
1208 		struct kvm_vcpu *vcpu = apic->vcpu;
1209 		unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1210 		unsigned long flags;
1211 
1212 		if (unlikely(!tscdeadline || !this_tsc_khz))
1213 			return;
1214 
1215 		local_irq_save(flags);
1216 
1217 		now = apic->lapic_timer.timer.base->get_time();
1218 		guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1219 		if (likely(tscdeadline > guest_tsc)) {
1220 			ns = (tscdeadline - guest_tsc) * 1000000ULL;
1221 			do_div(ns, this_tsc_khz);
1222 			expire = ktime_add_ns(now, ns);
1223 			expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1224 			hrtimer_start(&apic->lapic_timer.timer,
1225 				      expire, HRTIMER_MODE_ABS);
1226 		} else
1227 			apic_timer_expired(apic);
1228 
1229 		local_irq_restore(flags);
1230 	}
1231 }
1232 
1233 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1234 {
1235 	int nmi_wd_enabled = apic_lvt_nmi_mode(kvm_apic_get_reg(apic, APIC_LVT0));
1236 
1237 	if (apic_lvt_nmi_mode(lvt0_val)) {
1238 		if (!nmi_wd_enabled) {
1239 			apic_debug("Receive NMI setting on APIC_LVT0 "
1240 				   "for cpu %d\n", apic->vcpu->vcpu_id);
1241 			apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
1242 		}
1243 	} else if (nmi_wd_enabled)
1244 		apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
1245 }
1246 
1247 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1248 {
1249 	int ret = 0;
1250 
1251 	trace_kvm_apic_write(reg, val);
1252 
1253 	switch (reg) {
1254 	case APIC_ID:		/* Local APIC ID */
1255 		if (!apic_x2apic_mode(apic))
1256 			kvm_apic_set_id(apic, val >> 24);
1257 		else
1258 			ret = 1;
1259 		break;
1260 
1261 	case APIC_TASKPRI:
1262 		report_tpr_access(apic, true);
1263 		apic_set_tpr(apic, val & 0xff);
1264 		break;
1265 
1266 	case APIC_EOI:
1267 		apic_set_eoi(apic);
1268 		break;
1269 
1270 	case APIC_LDR:
1271 		if (!apic_x2apic_mode(apic))
1272 			kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1273 		else
1274 			ret = 1;
1275 		break;
1276 
1277 	case APIC_DFR:
1278 		if (!apic_x2apic_mode(apic)) {
1279 			apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1280 			recalculate_apic_map(apic->vcpu->kvm);
1281 		} else
1282 			ret = 1;
1283 		break;
1284 
1285 	case APIC_SPIV: {
1286 		u32 mask = 0x3ff;
1287 		if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1288 			mask |= APIC_SPIV_DIRECTED_EOI;
1289 		apic_set_spiv(apic, val & mask);
1290 		if (!(val & APIC_SPIV_APIC_ENABLED)) {
1291 			int i;
1292 			u32 lvt_val;
1293 
1294 			for (i = 0; i < APIC_LVT_NUM; i++) {
1295 				lvt_val = kvm_apic_get_reg(apic,
1296 						       APIC_LVTT + 0x10 * i);
1297 				apic_set_reg(apic, APIC_LVTT + 0x10 * i,
1298 					     lvt_val | APIC_LVT_MASKED);
1299 			}
1300 			atomic_set(&apic->lapic_timer.pending, 0);
1301 
1302 		}
1303 		break;
1304 	}
1305 	case APIC_ICR:
1306 		/* No delay here, so we always clear the pending bit */
1307 		apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1308 		apic_send_ipi(apic);
1309 		break;
1310 
1311 	case APIC_ICR2:
1312 		if (!apic_x2apic_mode(apic))
1313 			val &= 0xff000000;
1314 		apic_set_reg(apic, APIC_ICR2, val);
1315 		break;
1316 
1317 	case APIC_LVT0:
1318 		apic_manage_nmi_watchdog(apic, val);
1319 	case APIC_LVTTHMR:
1320 	case APIC_LVTPC:
1321 	case APIC_LVT1:
1322 	case APIC_LVTERR:
1323 		/* TODO: Check vector */
1324 		if (!kvm_apic_sw_enabled(apic))
1325 			val |= APIC_LVT_MASKED;
1326 
1327 		val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1328 		apic_set_reg(apic, reg, val);
1329 
1330 		break;
1331 
1332 	case APIC_LVTT: {
1333 		u32 timer_mode = val & apic->lapic_timer.timer_mode_mask;
1334 
1335 		if (apic->lapic_timer.timer_mode != timer_mode) {
1336 			apic->lapic_timer.timer_mode = timer_mode;
1337 			hrtimer_cancel(&apic->lapic_timer.timer);
1338 		}
1339 
1340 		if (!kvm_apic_sw_enabled(apic))
1341 			val |= APIC_LVT_MASKED;
1342 		val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1343 		apic_set_reg(apic, APIC_LVTT, val);
1344 		break;
1345 	}
1346 
1347 	case APIC_TMICT:
1348 		if (apic_lvtt_tscdeadline(apic))
1349 			break;
1350 
1351 		hrtimer_cancel(&apic->lapic_timer.timer);
1352 		apic_set_reg(apic, APIC_TMICT, val);
1353 		start_apic_timer(apic);
1354 		break;
1355 
1356 	case APIC_TDCR:
1357 		if (val & 4)
1358 			apic_debug("KVM_WRITE:TDCR %x\n", val);
1359 		apic_set_reg(apic, APIC_TDCR, val);
1360 		update_divide_count(apic);
1361 		break;
1362 
1363 	case APIC_ESR:
1364 		if (apic_x2apic_mode(apic) && val != 0) {
1365 			apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1366 			ret = 1;
1367 		}
1368 		break;
1369 
1370 	case APIC_SELF_IPI:
1371 		if (apic_x2apic_mode(apic)) {
1372 			apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1373 		} else
1374 			ret = 1;
1375 		break;
1376 	default:
1377 		ret = 1;
1378 		break;
1379 	}
1380 	if (ret)
1381 		apic_debug("Local APIC Write to read-only register %x\n", reg);
1382 	return ret;
1383 }
1384 
1385 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1386 			    gpa_t address, int len, const void *data)
1387 {
1388 	struct kvm_lapic *apic = to_lapic(this);
1389 	unsigned int offset = address - apic->base_address;
1390 	u32 val;
1391 
1392 	if (!apic_mmio_in_range(apic, address))
1393 		return -EOPNOTSUPP;
1394 
1395 	/*
1396 	 * APIC register must be aligned on 128-bits boundary.
1397 	 * 32/64/128 bits registers must be accessed thru 32 bits.
1398 	 * Refer SDM 8.4.1
1399 	 */
1400 	if (len != 4 || (offset & 0xf)) {
1401 		/* Don't shout loud, $infamous_os would cause only noise. */
1402 		apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1403 		return 0;
1404 	}
1405 
1406 	val = *(u32*)data;
1407 
1408 	/* too common printing */
1409 	if (offset != APIC_EOI)
1410 		apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1411 			   "0x%x\n", __func__, offset, len, val);
1412 
1413 	apic_reg_write(apic, offset & 0xff0, val);
1414 
1415 	return 0;
1416 }
1417 
1418 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1419 {
1420 	if (kvm_vcpu_has_lapic(vcpu))
1421 		apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1422 }
1423 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1424 
1425 /* emulate APIC access in a trap manner */
1426 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1427 {
1428 	u32 val = 0;
1429 
1430 	/* hw has done the conditional check and inst decode */
1431 	offset &= 0xff0;
1432 
1433 	apic_reg_read(vcpu->arch.apic, offset, 4, &val);
1434 
1435 	/* TODO: optimize to just emulate side effect w/o one more write */
1436 	apic_reg_write(vcpu->arch.apic, offset, val);
1437 }
1438 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1439 
1440 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1441 {
1442 	struct kvm_lapic *apic = vcpu->arch.apic;
1443 
1444 	if (!vcpu->arch.apic)
1445 		return;
1446 
1447 	hrtimer_cancel(&apic->lapic_timer.timer);
1448 
1449 	if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1450 		static_key_slow_dec_deferred(&apic_hw_disabled);
1451 
1452 	if (!apic->sw_enabled)
1453 		static_key_slow_dec_deferred(&apic_sw_disabled);
1454 
1455 	if (apic->regs)
1456 		free_page((unsigned long)apic->regs);
1457 
1458 	kfree(apic);
1459 }
1460 
1461 /*
1462  *----------------------------------------------------------------------
1463  * LAPIC interface
1464  *----------------------------------------------------------------------
1465  */
1466 
1467 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1468 {
1469 	struct kvm_lapic *apic = vcpu->arch.apic;
1470 
1471 	if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1472 			apic_lvtt_period(apic))
1473 		return 0;
1474 
1475 	return apic->lapic_timer.tscdeadline;
1476 }
1477 
1478 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1479 {
1480 	struct kvm_lapic *apic = vcpu->arch.apic;
1481 
1482 	if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1483 			apic_lvtt_period(apic))
1484 		return;
1485 
1486 	hrtimer_cancel(&apic->lapic_timer.timer);
1487 	apic->lapic_timer.tscdeadline = data;
1488 	start_apic_timer(apic);
1489 }
1490 
1491 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1492 {
1493 	struct kvm_lapic *apic = vcpu->arch.apic;
1494 
1495 	if (!kvm_vcpu_has_lapic(vcpu))
1496 		return;
1497 
1498 	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1499 		     | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4));
1500 }
1501 
1502 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1503 {
1504 	u64 tpr;
1505 
1506 	if (!kvm_vcpu_has_lapic(vcpu))
1507 		return 0;
1508 
1509 	tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1510 
1511 	return (tpr & 0xf0) >> 4;
1512 }
1513 
1514 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1515 {
1516 	u64 old_value = vcpu->arch.apic_base;
1517 	struct kvm_lapic *apic = vcpu->arch.apic;
1518 
1519 	if (!apic) {
1520 		value |= MSR_IA32_APICBASE_BSP;
1521 		vcpu->arch.apic_base = value;
1522 		return;
1523 	}
1524 
1525 	vcpu->arch.apic_base = value;
1526 
1527 	/* update jump label if enable bit changes */
1528 	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1529 		if (value & MSR_IA32_APICBASE_ENABLE)
1530 			static_key_slow_dec_deferred(&apic_hw_disabled);
1531 		else
1532 			static_key_slow_inc(&apic_hw_disabled.key);
1533 		recalculate_apic_map(vcpu->kvm);
1534 	}
1535 
1536 	if ((old_value ^ value) & X2APIC_ENABLE) {
1537 		if (value & X2APIC_ENABLE) {
1538 			u32 id = kvm_apic_id(apic);
1539 			u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
1540 			kvm_apic_set_ldr(apic, ldr);
1541 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1542 		} else
1543 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1544 	}
1545 
1546 	apic->base_address = apic->vcpu->arch.apic_base &
1547 			     MSR_IA32_APICBASE_BASE;
1548 
1549 	if ((value & MSR_IA32_APICBASE_ENABLE) &&
1550 	     apic->base_address != APIC_DEFAULT_PHYS_BASE)
1551 		pr_warn_once("APIC base relocation is unsupported by KVM");
1552 
1553 	/* with FSB delivery interrupt, we can restart APIC functionality */
1554 	apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1555 		   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1556 
1557 }
1558 
1559 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1560 {
1561 	struct kvm_lapic *apic;
1562 	int i;
1563 
1564 	apic_debug("%s\n", __func__);
1565 
1566 	ASSERT(vcpu);
1567 	apic = vcpu->arch.apic;
1568 	ASSERT(apic != NULL);
1569 
1570 	/* Stop the timer in case it's a reset to an active apic */
1571 	hrtimer_cancel(&apic->lapic_timer.timer);
1572 
1573 	kvm_apic_set_id(apic, vcpu->vcpu_id);
1574 	kvm_apic_set_version(apic->vcpu);
1575 
1576 	for (i = 0; i < APIC_LVT_NUM; i++)
1577 		apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1578 	apic->lapic_timer.timer_mode = 0;
1579 	apic_set_reg(apic, APIC_LVT0,
1580 		     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1581 
1582 	apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1583 	apic_set_spiv(apic, 0xff);
1584 	apic_set_reg(apic, APIC_TASKPRI, 0);
1585 	kvm_apic_set_ldr(apic, 0);
1586 	apic_set_reg(apic, APIC_ESR, 0);
1587 	apic_set_reg(apic, APIC_ICR, 0);
1588 	apic_set_reg(apic, APIC_ICR2, 0);
1589 	apic_set_reg(apic, APIC_TDCR, 0);
1590 	apic_set_reg(apic, APIC_TMICT, 0);
1591 	for (i = 0; i < 8; i++) {
1592 		apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1593 		apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1594 		apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1595 	}
1596 	apic->irr_pending = kvm_apic_vid_enabled(vcpu->kvm);
1597 	apic->isr_count = kvm_x86_ops->hwapic_isr_update ? 1 : 0;
1598 	apic->highest_isr_cache = -1;
1599 	update_divide_count(apic);
1600 	atomic_set(&apic->lapic_timer.pending, 0);
1601 	if (kvm_vcpu_is_bsp(vcpu))
1602 		kvm_lapic_set_base(vcpu,
1603 				vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1604 	vcpu->arch.pv_eoi.msr_val = 0;
1605 	apic_update_ppr(apic);
1606 
1607 	vcpu->arch.apic_arb_prio = 0;
1608 	vcpu->arch.apic_attention = 0;
1609 
1610 	apic_debug("%s: vcpu=%p, id=%d, base_msr="
1611 		   "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1612 		   vcpu, kvm_apic_id(apic),
1613 		   vcpu->arch.apic_base, apic->base_address);
1614 }
1615 
1616 /*
1617  *----------------------------------------------------------------------
1618  * timer interface
1619  *----------------------------------------------------------------------
1620  */
1621 
1622 static bool lapic_is_periodic(struct kvm_lapic *apic)
1623 {
1624 	return apic_lvtt_period(apic);
1625 }
1626 
1627 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1628 {
1629 	struct kvm_lapic *apic = vcpu->arch.apic;
1630 
1631 	if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) &&
1632 			apic_lvt_enabled(apic, APIC_LVTT))
1633 		return atomic_read(&apic->lapic_timer.pending);
1634 
1635 	return 0;
1636 }
1637 
1638 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1639 {
1640 	u32 reg = kvm_apic_get_reg(apic, lvt_type);
1641 	int vector, mode, trig_mode;
1642 
1643 	if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1644 		vector = reg & APIC_VECTOR_MASK;
1645 		mode = reg & APIC_MODE_MASK;
1646 		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1647 		return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1648 					NULL);
1649 	}
1650 	return 0;
1651 }
1652 
1653 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1654 {
1655 	struct kvm_lapic *apic = vcpu->arch.apic;
1656 
1657 	if (apic)
1658 		kvm_apic_local_deliver(apic, APIC_LVT0);
1659 }
1660 
1661 static const struct kvm_io_device_ops apic_mmio_ops = {
1662 	.read     = apic_mmio_read,
1663 	.write    = apic_mmio_write,
1664 };
1665 
1666 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1667 {
1668 	struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1669 	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1670 
1671 	apic_timer_expired(apic);
1672 
1673 	if (lapic_is_periodic(apic)) {
1674 		hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1675 		return HRTIMER_RESTART;
1676 	} else
1677 		return HRTIMER_NORESTART;
1678 }
1679 
1680 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1681 {
1682 	struct kvm_lapic *apic;
1683 
1684 	ASSERT(vcpu != NULL);
1685 	apic_debug("apic_init %d\n", vcpu->vcpu_id);
1686 
1687 	apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1688 	if (!apic)
1689 		goto nomem;
1690 
1691 	vcpu->arch.apic = apic;
1692 
1693 	apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1694 	if (!apic->regs) {
1695 		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1696 		       vcpu->vcpu_id);
1697 		goto nomem_free_apic;
1698 	}
1699 	apic->vcpu = vcpu;
1700 
1701 	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1702 		     HRTIMER_MODE_ABS);
1703 	apic->lapic_timer.timer.function = apic_timer_fn;
1704 
1705 	/*
1706 	 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1707 	 * thinking that APIC satet has changed.
1708 	 */
1709 	vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1710 	kvm_lapic_set_base(vcpu,
1711 			APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE);
1712 
1713 	static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1714 	kvm_lapic_reset(vcpu);
1715 	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1716 
1717 	return 0;
1718 nomem_free_apic:
1719 	kfree(apic);
1720 nomem:
1721 	return -ENOMEM;
1722 }
1723 
1724 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1725 {
1726 	struct kvm_lapic *apic = vcpu->arch.apic;
1727 	int highest_irr;
1728 
1729 	if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic))
1730 		return -1;
1731 
1732 	apic_update_ppr(apic);
1733 	highest_irr = apic_find_highest_irr(apic);
1734 	if ((highest_irr == -1) ||
1735 	    ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI)))
1736 		return -1;
1737 	return highest_irr;
1738 }
1739 
1740 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1741 {
1742 	u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1743 	int r = 0;
1744 
1745 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1746 		r = 1;
1747 	if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1748 	    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1749 		r = 1;
1750 	return r;
1751 }
1752 
1753 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1754 {
1755 	struct kvm_lapic *apic = vcpu->arch.apic;
1756 
1757 	if (!kvm_vcpu_has_lapic(vcpu))
1758 		return;
1759 
1760 	if (atomic_read(&apic->lapic_timer.pending) > 0) {
1761 		kvm_apic_local_deliver(apic, APIC_LVTT);
1762 		if (apic_lvtt_tscdeadline(apic))
1763 			apic->lapic_timer.tscdeadline = 0;
1764 		atomic_set(&apic->lapic_timer.pending, 0);
1765 	}
1766 }
1767 
1768 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1769 {
1770 	int vector = kvm_apic_has_interrupt(vcpu);
1771 	struct kvm_lapic *apic = vcpu->arch.apic;
1772 
1773 	if (vector == -1)
1774 		return -1;
1775 
1776 	/*
1777 	 * We get here even with APIC virtualization enabled, if doing
1778 	 * nested virtualization and L1 runs with the "acknowledge interrupt
1779 	 * on exit" mode.  Then we cannot inject the interrupt via RVI,
1780 	 * because the process would deliver it through the IDT.
1781 	 */
1782 
1783 	apic_set_isr(vector, apic);
1784 	apic_update_ppr(apic);
1785 	apic_clear_irr(vector, apic);
1786 	return vector;
1787 }
1788 
1789 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
1790 		struct kvm_lapic_state *s)
1791 {
1792 	struct kvm_lapic *apic = vcpu->arch.apic;
1793 
1794 	kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
1795 	/* set SPIV separately to get count of SW disabled APICs right */
1796 	apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
1797 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1798 	/* call kvm_apic_set_id() to put apic into apic_map */
1799 	kvm_apic_set_id(apic, kvm_apic_id(apic));
1800 	kvm_apic_set_version(vcpu);
1801 
1802 	apic_update_ppr(apic);
1803 	hrtimer_cancel(&apic->lapic_timer.timer);
1804 	update_divide_count(apic);
1805 	start_apic_timer(apic);
1806 	apic->irr_pending = true;
1807 	apic->isr_count = kvm_x86_ops->hwapic_isr_update ?
1808 				1 : count_vectors(apic->regs + APIC_ISR);
1809 	apic->highest_isr_cache = -1;
1810 	if (kvm_x86_ops->hwapic_irr_update)
1811 		kvm_x86_ops->hwapic_irr_update(vcpu,
1812 				apic_find_highest_irr(apic));
1813 	if (unlikely(kvm_x86_ops->hwapic_isr_update))
1814 		kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
1815 				apic_find_highest_isr(apic));
1816 	kvm_make_request(KVM_REQ_EVENT, vcpu);
1817 	kvm_rtc_eoi_tracking_restore_one(vcpu);
1818 }
1819 
1820 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1821 {
1822 	struct hrtimer *timer;
1823 
1824 	if (!kvm_vcpu_has_lapic(vcpu))
1825 		return;
1826 
1827 	timer = &vcpu->arch.apic->lapic_timer.timer;
1828 	if (hrtimer_cancel(timer))
1829 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1830 }
1831 
1832 /*
1833  * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1834  *
1835  * Detect whether guest triggered PV EOI since the
1836  * last entry. If yes, set EOI on guests's behalf.
1837  * Clear PV EOI in guest memory in any case.
1838  */
1839 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1840 					struct kvm_lapic *apic)
1841 {
1842 	bool pending;
1843 	int vector;
1844 	/*
1845 	 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1846 	 * and KVM_PV_EOI_ENABLED in guest memory as follows:
1847 	 *
1848 	 * KVM_APIC_PV_EOI_PENDING is unset:
1849 	 * 	-> host disabled PV EOI.
1850 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1851 	 * 	-> host enabled PV EOI, guest did not execute EOI yet.
1852 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1853 	 * 	-> host enabled PV EOI, guest executed EOI.
1854 	 */
1855 	BUG_ON(!pv_eoi_enabled(vcpu));
1856 	pending = pv_eoi_get_pending(vcpu);
1857 	/*
1858 	 * Clear pending bit in any case: it will be set again on vmentry.
1859 	 * While this might not be ideal from performance point of view,
1860 	 * this makes sure pv eoi is only enabled when we know it's safe.
1861 	 */
1862 	pv_eoi_clr_pending(vcpu);
1863 	if (pending)
1864 		return;
1865 	vector = apic_set_eoi(apic);
1866 	trace_kvm_pv_eoi(apic, vector);
1867 }
1868 
1869 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1870 {
1871 	u32 data;
1872 
1873 	if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1874 		apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1875 
1876 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1877 		return;
1878 
1879 	kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1880 				sizeof(u32));
1881 
1882 	apic_set_tpr(vcpu->arch.apic, data & 0xff);
1883 }
1884 
1885 /*
1886  * apic_sync_pv_eoi_to_guest - called before vmentry
1887  *
1888  * Detect whether it's safe to enable PV EOI and
1889  * if yes do so.
1890  */
1891 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
1892 					struct kvm_lapic *apic)
1893 {
1894 	if (!pv_eoi_enabled(vcpu) ||
1895 	    /* IRR set or many bits in ISR: could be nested. */
1896 	    apic->irr_pending ||
1897 	    /* Cache not set: could be safe but we don't bother. */
1898 	    apic->highest_isr_cache == -1 ||
1899 	    /* Need EOI to update ioapic. */
1900 	    kvm_ioapic_handles_vector(vcpu->kvm, apic->highest_isr_cache)) {
1901 		/*
1902 		 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1903 		 * so we need not do anything here.
1904 		 */
1905 		return;
1906 	}
1907 
1908 	pv_eoi_set_pending(apic->vcpu);
1909 }
1910 
1911 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1912 {
1913 	u32 data, tpr;
1914 	int max_irr, max_isr;
1915 	struct kvm_lapic *apic = vcpu->arch.apic;
1916 
1917 	apic_sync_pv_eoi_to_guest(vcpu, apic);
1918 
1919 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1920 		return;
1921 
1922 	tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1923 	max_irr = apic_find_highest_irr(apic);
1924 	if (max_irr < 0)
1925 		max_irr = 0;
1926 	max_isr = apic_find_highest_isr(apic);
1927 	if (max_isr < 0)
1928 		max_isr = 0;
1929 	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1930 
1931 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1932 				sizeof(u32));
1933 }
1934 
1935 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1936 {
1937 	if (vapic_addr) {
1938 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1939 					&vcpu->arch.apic->vapic_cache,
1940 					vapic_addr, sizeof(u32)))
1941 			return -EINVAL;
1942 		__set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1943 	} else {
1944 		__clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1945 	}
1946 
1947 	vcpu->arch.apic->vapic_addr = vapic_addr;
1948 	return 0;
1949 }
1950 
1951 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1952 {
1953 	struct kvm_lapic *apic = vcpu->arch.apic;
1954 	u32 reg = (msr - APIC_BASE_MSR) << 4;
1955 
1956 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1957 		return 1;
1958 
1959 	if (reg == APIC_ICR2)
1960 		return 1;
1961 
1962 	/* if this is ICR write vector before command */
1963 	if (reg == APIC_ICR)
1964 		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1965 	return apic_reg_write(apic, reg, (u32)data);
1966 }
1967 
1968 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1969 {
1970 	struct kvm_lapic *apic = vcpu->arch.apic;
1971 	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1972 
1973 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1974 		return 1;
1975 
1976 	if (reg == APIC_DFR || reg == APIC_ICR2) {
1977 		apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
1978 			   reg);
1979 		return 1;
1980 	}
1981 
1982 	if (apic_reg_read(apic, reg, 4, &low))
1983 		return 1;
1984 	if (reg == APIC_ICR)
1985 		apic_reg_read(apic, APIC_ICR2, 4, &high);
1986 
1987 	*data = (((u64)high) << 32) | low;
1988 
1989 	return 0;
1990 }
1991 
1992 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1993 {
1994 	struct kvm_lapic *apic = vcpu->arch.apic;
1995 
1996 	if (!kvm_vcpu_has_lapic(vcpu))
1997 		return 1;
1998 
1999 	/* if this is ICR write vector before command */
2000 	if (reg == APIC_ICR)
2001 		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2002 	return apic_reg_write(apic, reg, (u32)data);
2003 }
2004 
2005 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2006 {
2007 	struct kvm_lapic *apic = vcpu->arch.apic;
2008 	u32 low, high = 0;
2009 
2010 	if (!kvm_vcpu_has_lapic(vcpu))
2011 		return 1;
2012 
2013 	if (apic_reg_read(apic, reg, 4, &low))
2014 		return 1;
2015 	if (reg == APIC_ICR)
2016 		apic_reg_read(apic, APIC_ICR2, 4, &high);
2017 
2018 	*data = (((u64)high) << 32) | low;
2019 
2020 	return 0;
2021 }
2022 
2023 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2024 {
2025 	u64 addr = data & ~KVM_MSR_ENABLED;
2026 	if (!IS_ALIGNED(addr, 4))
2027 		return 1;
2028 
2029 	vcpu->arch.pv_eoi.msr_val = data;
2030 	if (!pv_eoi_enabled(vcpu))
2031 		return 0;
2032 	return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
2033 					 addr, sizeof(u8));
2034 }
2035 
2036 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2037 {
2038 	struct kvm_lapic *apic = vcpu->arch.apic;
2039 	u8 sipi_vector;
2040 	unsigned long pe;
2041 
2042 	if (!kvm_vcpu_has_lapic(vcpu) || !apic->pending_events)
2043 		return;
2044 
2045 	pe = xchg(&apic->pending_events, 0);
2046 
2047 	if (test_bit(KVM_APIC_INIT, &pe)) {
2048 		kvm_lapic_reset(vcpu);
2049 		kvm_vcpu_reset(vcpu);
2050 		if (kvm_vcpu_is_bsp(apic->vcpu))
2051 			vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2052 		else
2053 			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2054 	}
2055 	if (test_bit(KVM_APIC_SIPI, &pe) &&
2056 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2057 		/* evaluate pending_events before reading the vector */
2058 		smp_rmb();
2059 		sipi_vector = apic->sipi_vector;
2060 		apic_debug("vcpu %d received sipi with vector # %x\n",
2061 			 vcpu->vcpu_id, sipi_vector);
2062 		kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2063 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2064 	}
2065 }
2066 
2067 void kvm_lapic_init(void)
2068 {
2069 	/* do not patch jump label more than once per second */
2070 	jump_label_rate_limit(&apic_hw_disabled, HZ);
2071 	jump_label_rate_limit(&apic_sw_disabled, HZ);
2072 }
2073