xref: /openbmc/linux/arch/x86/kvm/lapic.c (revision 9c6d26df1fae6ad4718d51c48e6517913304ed27)
1 
2 /*
3  * Local APIC virtualization
4  *
5  * Copyright (C) 2006 Qumranet, Inc.
6  * Copyright (C) 2007 Novell
7  * Copyright (C) 2007 Intel
8  * Copyright 2009 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Dor Laor <dor.laor@qumranet.com>
12  *   Gregory Haskins <ghaskins@novell.com>
13  *   Yaozu (Eddie) Dong <eddie.dong@intel.com>
14  *
15  * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
16  *
17  * This work is licensed under the terms of the GNU GPL, version 2.  See
18  * the COPYING file in the top-level directory.
19  */
20 
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/mm.h>
24 #include <linux/highmem.h>
25 #include <linux/smp.h>
26 #include <linux/hrtimer.h>
27 #include <linux/io.h>
28 #include <linux/export.h>
29 #include <linux/math64.h>
30 #include <linux/slab.h>
31 #include <asm/processor.h>
32 #include <asm/msr.h>
33 #include <asm/page.h>
34 #include <asm/current.h>
35 #include <asm/apicdef.h>
36 #include <asm/delay.h>
37 #include <linux/atomic.h>
38 #include <linux/jump_label.h>
39 #include "kvm_cache_regs.h"
40 #include "irq.h"
41 #include "trace.h"
42 #include "x86.h"
43 #include "cpuid.h"
44 #include "hyperv.h"
45 
46 #ifndef CONFIG_X86_64
47 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
48 #else
49 #define mod_64(x, y) ((x) % (y))
50 #endif
51 
52 #define PRId64 "d"
53 #define PRIx64 "llx"
54 #define PRIu64 "u"
55 #define PRIo64 "o"
56 
57 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
58 #define apic_debug(fmt, arg...)
59 
60 /* 14 is the version for Xeon and Pentium 8.4.8*/
61 #define APIC_VERSION			(0x14UL | ((KVM_APIC_LVT_NUM - 1) << 16))
62 #define LAPIC_MMIO_LENGTH		(1 << 12)
63 /* followed define is not in apicdef.h */
64 #define APIC_SHORT_MASK			0xc0000
65 #define APIC_DEST_NOSHORT		0x0
66 #define APIC_DEST_MASK			0x800
67 #define MAX_APIC_VECTOR			256
68 #define APIC_VECTORS_PER_REG		32
69 
70 #define APIC_BROADCAST			0xFF
71 #define X2APIC_BROADCAST		0xFFFFFFFFul
72 
73 static inline int apic_test_vector(int vec, void *bitmap)
74 {
75 	return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
76 }
77 
78 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
79 {
80 	struct kvm_lapic *apic = vcpu->arch.apic;
81 
82 	return apic_test_vector(vector, apic->regs + APIC_ISR) ||
83 		apic_test_vector(vector, apic->regs + APIC_IRR);
84 }
85 
86 static inline void apic_clear_vector(int vec, void *bitmap)
87 {
88 	clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
89 }
90 
91 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
92 {
93 	return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
94 }
95 
96 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
97 {
98 	return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
99 }
100 
101 struct static_key_deferred apic_hw_disabled __read_mostly;
102 struct static_key_deferred apic_sw_disabled __read_mostly;
103 
104 static inline int apic_enabled(struct kvm_lapic *apic)
105 {
106 	return kvm_apic_sw_enabled(apic) &&	kvm_apic_hw_enabled(apic);
107 }
108 
109 #define LVT_MASK	\
110 	(APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
111 
112 #define LINT_MASK	\
113 	(LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
114 	 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
115 
116 static inline u8 kvm_xapic_id(struct kvm_lapic *apic)
117 {
118 	return kvm_lapic_get_reg(apic, APIC_ID) >> 24;
119 }
120 
121 static inline u32 kvm_x2apic_id(struct kvm_lapic *apic)
122 {
123 	return apic->vcpu->vcpu_id;
124 }
125 
126 static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
127 		u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
128 	switch (map->mode) {
129 	case KVM_APIC_MODE_X2APIC: {
130 		u32 offset = (dest_id >> 16) * 16;
131 		u32 max_apic_id = map->max_apic_id;
132 
133 		if (offset <= max_apic_id) {
134 			u8 cluster_size = min(max_apic_id - offset + 1, 16U);
135 
136 			*cluster = &map->phys_map[offset];
137 			*mask = dest_id & (0xffff >> (16 - cluster_size));
138 		} else {
139 			*mask = 0;
140 		}
141 
142 		return true;
143 		}
144 	case KVM_APIC_MODE_XAPIC_FLAT:
145 		*cluster = map->xapic_flat_map;
146 		*mask = dest_id & 0xff;
147 		return true;
148 	case KVM_APIC_MODE_XAPIC_CLUSTER:
149 		*cluster = map->xapic_cluster_map[(dest_id >> 4) & 0xf];
150 		*mask = dest_id & 0xf;
151 		return true;
152 	default:
153 		/* Not optimized. */
154 		return false;
155 	}
156 }
157 
158 static void kvm_apic_map_free(struct rcu_head *rcu)
159 {
160 	struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
161 
162 	kvfree(map);
163 }
164 
165 static void recalculate_apic_map(struct kvm *kvm)
166 {
167 	struct kvm_apic_map *new, *old = NULL;
168 	struct kvm_vcpu *vcpu;
169 	int i;
170 	u32 max_id = 255; /* enough space for any xAPIC ID */
171 
172 	mutex_lock(&kvm->arch.apic_map_lock);
173 
174 	kvm_for_each_vcpu(i, vcpu, kvm)
175 		if (kvm_apic_present(vcpu))
176 			max_id = max(max_id, kvm_x2apic_id(vcpu->arch.apic));
177 
178 	new = kvzalloc(sizeof(struct kvm_apic_map) +
179 	                   sizeof(struct kvm_lapic *) * ((u64)max_id + 1), GFP_KERNEL);
180 
181 	if (!new)
182 		goto out;
183 
184 	new->max_apic_id = max_id;
185 
186 	kvm_for_each_vcpu(i, vcpu, kvm) {
187 		struct kvm_lapic *apic = vcpu->arch.apic;
188 		struct kvm_lapic **cluster;
189 		u16 mask;
190 		u32 ldr;
191 		u8 xapic_id;
192 		u32 x2apic_id;
193 
194 		if (!kvm_apic_present(vcpu))
195 			continue;
196 
197 		xapic_id = kvm_xapic_id(apic);
198 		x2apic_id = kvm_x2apic_id(apic);
199 
200 		/* Hotplug hack: see kvm_apic_match_physical_addr(), ... */
201 		if ((apic_x2apic_mode(apic) || x2apic_id > 0xff) &&
202 				x2apic_id <= new->max_apic_id)
203 			new->phys_map[x2apic_id] = apic;
204 		/*
205 		 * ... xAPIC ID of VCPUs with APIC ID > 0xff will wrap-around,
206 		 * prevent them from masking VCPUs with APIC ID <= 0xff.
207 		 */
208 		if (!apic_x2apic_mode(apic) && !new->phys_map[xapic_id])
209 			new->phys_map[xapic_id] = apic;
210 
211 		ldr = kvm_lapic_get_reg(apic, APIC_LDR);
212 
213 		if (apic_x2apic_mode(apic)) {
214 			new->mode |= KVM_APIC_MODE_X2APIC;
215 		} else if (ldr) {
216 			ldr = GET_APIC_LOGICAL_ID(ldr);
217 			if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
218 				new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
219 			else
220 				new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
221 		}
222 
223 		if (!kvm_apic_map_get_logical_dest(new, ldr, &cluster, &mask))
224 			continue;
225 
226 		if (mask)
227 			cluster[ffs(mask) - 1] = apic;
228 	}
229 out:
230 	old = rcu_dereference_protected(kvm->arch.apic_map,
231 			lockdep_is_held(&kvm->arch.apic_map_lock));
232 	rcu_assign_pointer(kvm->arch.apic_map, new);
233 	mutex_unlock(&kvm->arch.apic_map_lock);
234 
235 	if (old)
236 		call_rcu(&old->rcu, kvm_apic_map_free);
237 
238 	kvm_make_scan_ioapic_request(kvm);
239 }
240 
241 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
242 {
243 	bool enabled = val & APIC_SPIV_APIC_ENABLED;
244 
245 	kvm_lapic_set_reg(apic, APIC_SPIV, val);
246 
247 	if (enabled != apic->sw_enabled) {
248 		apic->sw_enabled = enabled;
249 		if (enabled) {
250 			static_key_slow_dec_deferred(&apic_sw_disabled);
251 			recalculate_apic_map(apic->vcpu->kvm);
252 		} else
253 			static_key_slow_inc(&apic_sw_disabled.key);
254 	}
255 }
256 
257 static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
258 {
259 	kvm_lapic_set_reg(apic, APIC_ID, id << 24);
260 	recalculate_apic_map(apic->vcpu->kvm);
261 }
262 
263 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
264 {
265 	kvm_lapic_set_reg(apic, APIC_LDR, id);
266 	recalculate_apic_map(apic->vcpu->kvm);
267 }
268 
269 static inline u32 kvm_apic_calc_x2apic_ldr(u32 id)
270 {
271 	return ((id >> 4) << 16) | (1 << (id & 0xf));
272 }
273 
274 static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
275 {
276 	u32 ldr = kvm_apic_calc_x2apic_ldr(id);
277 
278 	WARN_ON_ONCE(id != apic->vcpu->vcpu_id);
279 
280 	kvm_lapic_set_reg(apic, APIC_ID, id);
281 	kvm_lapic_set_reg(apic, APIC_LDR, ldr);
282 	recalculate_apic_map(apic->vcpu->kvm);
283 }
284 
285 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
286 {
287 	return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
288 }
289 
290 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
291 {
292 	return kvm_lapic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
293 }
294 
295 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
296 {
297 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
298 }
299 
300 static inline int apic_lvtt_period(struct kvm_lapic *apic)
301 {
302 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
303 }
304 
305 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
306 {
307 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
308 }
309 
310 static inline int apic_lvt_nmi_mode(u32 lvt_val)
311 {
312 	return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
313 }
314 
315 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
316 {
317 	struct kvm_lapic *apic = vcpu->arch.apic;
318 	struct kvm_cpuid_entry2 *feat;
319 	u32 v = APIC_VERSION;
320 
321 	if (!lapic_in_kernel(vcpu))
322 		return;
323 
324 	/*
325 	 * KVM emulates 82093AA datasheet (with in-kernel IOAPIC implementation)
326 	 * which doesn't have EOI register; Some buggy OSes (e.g. Windows with
327 	 * Hyper-V role) disable EOI broadcast in lapic not checking for IOAPIC
328 	 * version first and level-triggered interrupts never get EOIed in
329 	 * IOAPIC.
330 	 */
331 	feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
332 	if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))) &&
333 	    !ioapic_in_kernel(vcpu->kvm))
334 		v |= APIC_LVR_DIRECTED_EOI;
335 	kvm_lapic_set_reg(apic, APIC_LVR, v);
336 }
337 
338 static const unsigned int apic_lvt_mask[KVM_APIC_LVT_NUM] = {
339 	LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
340 	LVT_MASK | APIC_MODE_MASK,	/* LVTTHMR */
341 	LVT_MASK | APIC_MODE_MASK,	/* LVTPC */
342 	LINT_MASK, LINT_MASK,	/* LVT0-1 */
343 	LVT_MASK		/* LVTERR */
344 };
345 
346 static int find_highest_vector(void *bitmap)
347 {
348 	int vec;
349 	u32 *reg;
350 
351 	for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
352 	     vec >= 0; vec -= APIC_VECTORS_PER_REG) {
353 		reg = bitmap + REG_POS(vec);
354 		if (*reg)
355 			return __fls(*reg) + vec;
356 	}
357 
358 	return -1;
359 }
360 
361 static u8 count_vectors(void *bitmap)
362 {
363 	int vec;
364 	u32 *reg;
365 	u8 count = 0;
366 
367 	for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
368 		reg = bitmap + REG_POS(vec);
369 		count += hweight32(*reg);
370 	}
371 
372 	return count;
373 }
374 
375 bool __kvm_apic_update_irr(u32 *pir, void *regs, int *max_irr)
376 {
377 	u32 i, vec;
378 	u32 pir_val, irr_val, prev_irr_val;
379 	int max_updated_irr;
380 
381 	max_updated_irr = -1;
382 	*max_irr = -1;
383 
384 	for (i = vec = 0; i <= 7; i++, vec += 32) {
385 		pir_val = READ_ONCE(pir[i]);
386 		irr_val = *((u32 *)(regs + APIC_IRR + i * 0x10));
387 		if (pir_val) {
388 			prev_irr_val = irr_val;
389 			irr_val |= xchg(&pir[i], 0);
390 			*((u32 *)(regs + APIC_IRR + i * 0x10)) = irr_val;
391 			if (prev_irr_val != irr_val) {
392 				max_updated_irr =
393 					__fls(irr_val ^ prev_irr_val) + vec;
394 			}
395 		}
396 		if (irr_val)
397 			*max_irr = __fls(irr_val) + vec;
398 	}
399 
400 	return ((max_updated_irr != -1) &&
401 		(max_updated_irr == *max_irr));
402 }
403 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
404 
405 bool kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir, int *max_irr)
406 {
407 	struct kvm_lapic *apic = vcpu->arch.apic;
408 
409 	return __kvm_apic_update_irr(pir, apic->regs, max_irr);
410 }
411 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
412 
413 static inline int apic_search_irr(struct kvm_lapic *apic)
414 {
415 	return find_highest_vector(apic->regs + APIC_IRR);
416 }
417 
418 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
419 {
420 	int result;
421 
422 	/*
423 	 * Note that irr_pending is just a hint. It will be always
424 	 * true with virtual interrupt delivery enabled.
425 	 */
426 	if (!apic->irr_pending)
427 		return -1;
428 
429 	result = apic_search_irr(apic);
430 	ASSERT(result == -1 || result >= 16);
431 
432 	return result;
433 }
434 
435 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
436 {
437 	struct kvm_vcpu *vcpu;
438 
439 	vcpu = apic->vcpu;
440 
441 	if (unlikely(vcpu->arch.apicv_active)) {
442 		/* need to update RVI */
443 		apic_clear_vector(vec, apic->regs + APIC_IRR);
444 		kvm_x86_ops->hwapic_irr_update(vcpu,
445 				apic_find_highest_irr(apic));
446 	} else {
447 		apic->irr_pending = false;
448 		apic_clear_vector(vec, apic->regs + APIC_IRR);
449 		if (apic_search_irr(apic) != -1)
450 			apic->irr_pending = true;
451 	}
452 }
453 
454 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
455 {
456 	struct kvm_vcpu *vcpu;
457 
458 	if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
459 		return;
460 
461 	vcpu = apic->vcpu;
462 
463 	/*
464 	 * With APIC virtualization enabled, all caching is disabled
465 	 * because the processor can modify ISR under the hood.  Instead
466 	 * just set SVI.
467 	 */
468 	if (unlikely(vcpu->arch.apicv_active))
469 		kvm_x86_ops->hwapic_isr_update(vcpu, vec);
470 	else {
471 		++apic->isr_count;
472 		BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
473 		/*
474 		 * ISR (in service register) bit is set when injecting an interrupt.
475 		 * The highest vector is injected. Thus the latest bit set matches
476 		 * the highest bit in ISR.
477 		 */
478 		apic->highest_isr_cache = vec;
479 	}
480 }
481 
482 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
483 {
484 	int result;
485 
486 	/*
487 	 * Note that isr_count is always 1, and highest_isr_cache
488 	 * is always -1, with APIC virtualization enabled.
489 	 */
490 	if (!apic->isr_count)
491 		return -1;
492 	if (likely(apic->highest_isr_cache != -1))
493 		return apic->highest_isr_cache;
494 
495 	result = find_highest_vector(apic->regs + APIC_ISR);
496 	ASSERT(result == -1 || result >= 16);
497 
498 	return result;
499 }
500 
501 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
502 {
503 	struct kvm_vcpu *vcpu;
504 	if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
505 		return;
506 
507 	vcpu = apic->vcpu;
508 
509 	/*
510 	 * We do get here for APIC virtualization enabled if the guest
511 	 * uses the Hyper-V APIC enlightenment.  In this case we may need
512 	 * to trigger a new interrupt delivery by writing the SVI field;
513 	 * on the other hand isr_count and highest_isr_cache are unused
514 	 * and must be left alone.
515 	 */
516 	if (unlikely(vcpu->arch.apicv_active))
517 		kvm_x86_ops->hwapic_isr_update(vcpu,
518 					       apic_find_highest_isr(apic));
519 	else {
520 		--apic->isr_count;
521 		BUG_ON(apic->isr_count < 0);
522 		apic->highest_isr_cache = -1;
523 	}
524 }
525 
526 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
527 {
528 	/* This may race with setting of irr in __apic_accept_irq() and
529 	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
530 	 * will cause vmexit immediately and the value will be recalculated
531 	 * on the next vmentry.
532 	 */
533 	return apic_find_highest_irr(vcpu->arch.apic);
534 }
535 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
536 
537 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
538 			     int vector, int level, int trig_mode,
539 			     struct dest_map *dest_map);
540 
541 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
542 		     struct dest_map *dest_map)
543 {
544 	struct kvm_lapic *apic = vcpu->arch.apic;
545 
546 	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
547 			irq->level, irq->trig_mode, dest_map);
548 }
549 
550 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
551 {
552 
553 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
554 				      sizeof(val));
555 }
556 
557 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
558 {
559 
560 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
561 				      sizeof(*val));
562 }
563 
564 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
565 {
566 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
567 }
568 
569 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
570 {
571 	u8 val;
572 	if (pv_eoi_get_user(vcpu, &val) < 0)
573 		apic_debug("Can't read EOI MSR value: 0x%llx\n",
574 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
575 	return val & 0x1;
576 }
577 
578 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
579 {
580 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
581 		apic_debug("Can't set EOI MSR value: 0x%llx\n",
582 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
583 		return;
584 	}
585 	__set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
586 }
587 
588 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
589 {
590 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
591 		apic_debug("Can't clear EOI MSR value: 0x%llx\n",
592 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
593 		return;
594 	}
595 	__clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
596 }
597 
598 static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
599 {
600 	int highest_irr;
601 	if (apic->vcpu->arch.apicv_active)
602 		highest_irr = kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
603 	else
604 		highest_irr = apic_find_highest_irr(apic);
605 	if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
606 		return -1;
607 	return highest_irr;
608 }
609 
610 static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
611 {
612 	u32 tpr, isrv, ppr, old_ppr;
613 	int isr;
614 
615 	old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
616 	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
617 	isr = apic_find_highest_isr(apic);
618 	isrv = (isr != -1) ? isr : 0;
619 
620 	if ((tpr & 0xf0) >= (isrv & 0xf0))
621 		ppr = tpr & 0xff;
622 	else
623 		ppr = isrv & 0xf0;
624 
625 	apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
626 		   apic, ppr, isr, isrv);
627 
628 	*new_ppr = ppr;
629 	if (old_ppr != ppr)
630 		kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
631 
632 	return ppr < old_ppr;
633 }
634 
635 static void apic_update_ppr(struct kvm_lapic *apic)
636 {
637 	u32 ppr;
638 
639 	if (__apic_update_ppr(apic, &ppr) &&
640 	    apic_has_interrupt_for_ppr(apic, ppr) != -1)
641 		kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
642 }
643 
644 void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
645 {
646 	apic_update_ppr(vcpu->arch.apic);
647 }
648 EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
649 
650 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
651 {
652 	kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
653 	apic_update_ppr(apic);
654 }
655 
656 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
657 {
658 	return mda == (apic_x2apic_mode(apic) ?
659 			X2APIC_BROADCAST : APIC_BROADCAST);
660 }
661 
662 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
663 {
664 	if (kvm_apic_broadcast(apic, mda))
665 		return true;
666 
667 	if (apic_x2apic_mode(apic))
668 		return mda == kvm_x2apic_id(apic);
669 
670 	/*
671 	 * Hotplug hack: Make LAPIC in xAPIC mode also accept interrupts as if
672 	 * it were in x2APIC mode.  Hotplugged VCPUs start in xAPIC mode and
673 	 * this allows unique addressing of VCPUs with APIC ID over 0xff.
674 	 * The 0xff condition is needed because writeable xAPIC ID.
675 	 */
676 	if (kvm_x2apic_id(apic) > 0xff && mda == kvm_x2apic_id(apic))
677 		return true;
678 
679 	return mda == kvm_xapic_id(apic);
680 }
681 
682 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
683 {
684 	u32 logical_id;
685 
686 	if (kvm_apic_broadcast(apic, mda))
687 		return true;
688 
689 	logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
690 
691 	if (apic_x2apic_mode(apic))
692 		return ((logical_id >> 16) == (mda >> 16))
693 		       && (logical_id & mda & 0xffff) != 0;
694 
695 	logical_id = GET_APIC_LOGICAL_ID(logical_id);
696 
697 	switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
698 	case APIC_DFR_FLAT:
699 		return (logical_id & mda) != 0;
700 	case APIC_DFR_CLUSTER:
701 		return ((logical_id >> 4) == (mda >> 4))
702 		       && (logical_id & mda & 0xf) != 0;
703 	default:
704 		apic_debug("Bad DFR vcpu %d: %08x\n",
705 			   apic->vcpu->vcpu_id, kvm_lapic_get_reg(apic, APIC_DFR));
706 		return false;
707 	}
708 }
709 
710 /* The KVM local APIC implementation has two quirks:
711  *
712  *  - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
713  *    in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
714  *    KVM doesn't do that aliasing.
715  *
716  *  - in-kernel IOAPIC messages have to be delivered directly to
717  *    x2APIC, because the kernel does not support interrupt remapping.
718  *    In order to support broadcast without interrupt remapping, x2APIC
719  *    rewrites the destination of non-IPI messages from APIC_BROADCAST
720  *    to X2APIC_BROADCAST.
721  *
722  * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API.  This is
723  * important when userspace wants to use x2APIC-format MSIs, because
724  * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
725  */
726 static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
727 		struct kvm_lapic *source, struct kvm_lapic *target)
728 {
729 	bool ipi = source != NULL;
730 
731 	if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
732 	    !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
733 		return X2APIC_BROADCAST;
734 
735 	return dest_id;
736 }
737 
738 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
739 			   int short_hand, unsigned int dest, int dest_mode)
740 {
741 	struct kvm_lapic *target = vcpu->arch.apic;
742 	u32 mda = kvm_apic_mda(vcpu, dest, source, target);
743 
744 	apic_debug("target %p, source %p, dest 0x%x, "
745 		   "dest_mode 0x%x, short_hand 0x%x\n",
746 		   target, source, dest, dest_mode, short_hand);
747 
748 	ASSERT(target);
749 	switch (short_hand) {
750 	case APIC_DEST_NOSHORT:
751 		if (dest_mode == APIC_DEST_PHYSICAL)
752 			return kvm_apic_match_physical_addr(target, mda);
753 		else
754 			return kvm_apic_match_logical_addr(target, mda);
755 	case APIC_DEST_SELF:
756 		return target == source;
757 	case APIC_DEST_ALLINC:
758 		return true;
759 	case APIC_DEST_ALLBUT:
760 		return target != source;
761 	default:
762 		apic_debug("kvm: apic: Bad dest shorthand value %x\n",
763 			   short_hand);
764 		return false;
765 	}
766 }
767 EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
768 
769 int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
770 		       const unsigned long *bitmap, u32 bitmap_size)
771 {
772 	u32 mod;
773 	int i, idx = -1;
774 
775 	mod = vector % dest_vcpus;
776 
777 	for (i = 0; i <= mod; i++) {
778 		idx = find_next_bit(bitmap, bitmap_size, idx + 1);
779 		BUG_ON(idx == bitmap_size);
780 	}
781 
782 	return idx;
783 }
784 
785 static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
786 {
787 	if (!kvm->arch.disabled_lapic_found) {
788 		kvm->arch.disabled_lapic_found = true;
789 		printk(KERN_INFO
790 		       "Disabled LAPIC found during irq injection\n");
791 	}
792 }
793 
794 static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
795 		struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
796 {
797 	if (kvm->arch.x2apic_broadcast_quirk_disabled) {
798 		if ((irq->dest_id == APIC_BROADCAST &&
799 				map->mode != KVM_APIC_MODE_X2APIC))
800 			return true;
801 		if (irq->dest_id == X2APIC_BROADCAST)
802 			return true;
803 	} else {
804 		bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
805 		if (irq->dest_id == (x2apic_ipi ?
806 		                     X2APIC_BROADCAST : APIC_BROADCAST))
807 			return true;
808 	}
809 
810 	return false;
811 }
812 
813 /* Return true if the interrupt can be handled by using *bitmap as index mask
814  * for valid destinations in *dst array.
815  * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
816  * Note: we may have zero kvm_lapic destinations when we return true, which
817  * means that the interrupt should be dropped.  In this case, *bitmap would be
818  * zero and *dst undefined.
819  */
820 static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
821 		struct kvm_lapic **src, struct kvm_lapic_irq *irq,
822 		struct kvm_apic_map *map, struct kvm_lapic ***dst,
823 		unsigned long *bitmap)
824 {
825 	int i, lowest;
826 
827 	if (irq->shorthand == APIC_DEST_SELF && src) {
828 		*dst = src;
829 		*bitmap = 1;
830 		return true;
831 	} else if (irq->shorthand)
832 		return false;
833 
834 	if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
835 		return false;
836 
837 	if (irq->dest_mode == APIC_DEST_PHYSICAL) {
838 		if (irq->dest_id > map->max_apic_id) {
839 			*bitmap = 0;
840 		} else {
841 			*dst = &map->phys_map[irq->dest_id];
842 			*bitmap = 1;
843 		}
844 		return true;
845 	}
846 
847 	*bitmap = 0;
848 	if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
849 				(u16 *)bitmap))
850 		return false;
851 
852 	if (!kvm_lowest_prio_delivery(irq))
853 		return true;
854 
855 	if (!kvm_vector_hashing_enabled()) {
856 		lowest = -1;
857 		for_each_set_bit(i, bitmap, 16) {
858 			if (!(*dst)[i])
859 				continue;
860 			if (lowest < 0)
861 				lowest = i;
862 			else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
863 						(*dst)[lowest]->vcpu) < 0)
864 				lowest = i;
865 		}
866 	} else {
867 		if (!*bitmap)
868 			return true;
869 
870 		lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
871 				bitmap, 16);
872 
873 		if (!(*dst)[lowest]) {
874 			kvm_apic_disabled_lapic_found(kvm);
875 			*bitmap = 0;
876 			return true;
877 		}
878 	}
879 
880 	*bitmap = (lowest >= 0) ? 1 << lowest : 0;
881 
882 	return true;
883 }
884 
885 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
886 		struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
887 {
888 	struct kvm_apic_map *map;
889 	unsigned long bitmap;
890 	struct kvm_lapic **dst = NULL;
891 	int i;
892 	bool ret;
893 
894 	*r = -1;
895 
896 	if (irq->shorthand == APIC_DEST_SELF) {
897 		*r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
898 		return true;
899 	}
900 
901 	rcu_read_lock();
902 	map = rcu_dereference(kvm->arch.apic_map);
903 
904 	ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
905 	if (ret)
906 		for_each_set_bit(i, &bitmap, 16) {
907 			if (!dst[i])
908 				continue;
909 			if (*r < 0)
910 				*r = 0;
911 			*r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
912 		}
913 
914 	rcu_read_unlock();
915 	return ret;
916 }
917 
918 /*
919  * This routine tries to handler interrupts in posted mode, here is how
920  * it deals with different cases:
921  * - For single-destination interrupts, handle it in posted mode
922  * - Else if vector hashing is enabled and it is a lowest-priority
923  *   interrupt, handle it in posted mode and use the following mechanism
924  *   to find the destinaiton vCPU.
925  *	1. For lowest-priority interrupts, store all the possible
926  *	   destination vCPUs in an array.
927  *	2. Use "guest vector % max number of destination vCPUs" to find
928  *	   the right destination vCPU in the array for the lowest-priority
929  *	   interrupt.
930  * - Otherwise, use remapped mode to inject the interrupt.
931  */
932 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
933 			struct kvm_vcpu **dest_vcpu)
934 {
935 	struct kvm_apic_map *map;
936 	unsigned long bitmap;
937 	struct kvm_lapic **dst = NULL;
938 	bool ret = false;
939 
940 	if (irq->shorthand)
941 		return false;
942 
943 	rcu_read_lock();
944 	map = rcu_dereference(kvm->arch.apic_map);
945 
946 	if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
947 			hweight16(bitmap) == 1) {
948 		unsigned long i = find_first_bit(&bitmap, 16);
949 
950 		if (dst[i]) {
951 			*dest_vcpu = dst[i]->vcpu;
952 			ret = true;
953 		}
954 	}
955 
956 	rcu_read_unlock();
957 	return ret;
958 }
959 
960 /*
961  * Add a pending IRQ into lapic.
962  * Return 1 if successfully added and 0 if discarded.
963  */
964 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
965 			     int vector, int level, int trig_mode,
966 			     struct dest_map *dest_map)
967 {
968 	int result = 0;
969 	struct kvm_vcpu *vcpu = apic->vcpu;
970 
971 	trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
972 				  trig_mode, vector);
973 	switch (delivery_mode) {
974 	case APIC_DM_LOWEST:
975 		vcpu->arch.apic_arb_prio++;
976 	case APIC_DM_FIXED:
977 		if (unlikely(trig_mode && !level))
978 			break;
979 
980 		/* FIXME add logic for vcpu on reset */
981 		if (unlikely(!apic_enabled(apic)))
982 			break;
983 
984 		result = 1;
985 
986 		if (dest_map) {
987 			__set_bit(vcpu->vcpu_id, dest_map->map);
988 			dest_map->vectors[vcpu->vcpu_id] = vector;
989 		}
990 
991 		if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
992 			if (trig_mode)
993 				kvm_lapic_set_vector(vector, apic->regs + APIC_TMR);
994 			else
995 				apic_clear_vector(vector, apic->regs + APIC_TMR);
996 		}
997 
998 		if (vcpu->arch.apicv_active)
999 			kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
1000 		else {
1001 			kvm_lapic_set_irr(vector, apic);
1002 
1003 			kvm_make_request(KVM_REQ_EVENT, vcpu);
1004 			kvm_vcpu_kick(vcpu);
1005 		}
1006 		break;
1007 
1008 	case APIC_DM_REMRD:
1009 		result = 1;
1010 		vcpu->arch.pv.pv_unhalted = 1;
1011 		kvm_make_request(KVM_REQ_EVENT, vcpu);
1012 		kvm_vcpu_kick(vcpu);
1013 		break;
1014 
1015 	case APIC_DM_SMI:
1016 		result = 1;
1017 		kvm_make_request(KVM_REQ_SMI, vcpu);
1018 		kvm_vcpu_kick(vcpu);
1019 		break;
1020 
1021 	case APIC_DM_NMI:
1022 		result = 1;
1023 		kvm_inject_nmi(vcpu);
1024 		kvm_vcpu_kick(vcpu);
1025 		break;
1026 
1027 	case APIC_DM_INIT:
1028 		if (!trig_mode || level) {
1029 			result = 1;
1030 			/* assumes that there are only KVM_APIC_INIT/SIPI */
1031 			apic->pending_events = (1UL << KVM_APIC_INIT);
1032 			/* make sure pending_events is visible before sending
1033 			 * the request */
1034 			smp_wmb();
1035 			kvm_make_request(KVM_REQ_EVENT, vcpu);
1036 			kvm_vcpu_kick(vcpu);
1037 		} else {
1038 			apic_debug("Ignoring de-assert INIT to vcpu %d\n",
1039 				   vcpu->vcpu_id);
1040 		}
1041 		break;
1042 
1043 	case APIC_DM_STARTUP:
1044 		apic_debug("SIPI to vcpu %d vector 0x%02x\n",
1045 			   vcpu->vcpu_id, vector);
1046 		result = 1;
1047 		apic->sipi_vector = vector;
1048 		/* make sure sipi_vector is visible for the receiver */
1049 		smp_wmb();
1050 		set_bit(KVM_APIC_SIPI, &apic->pending_events);
1051 		kvm_make_request(KVM_REQ_EVENT, vcpu);
1052 		kvm_vcpu_kick(vcpu);
1053 		break;
1054 
1055 	case APIC_DM_EXTINT:
1056 		/*
1057 		 * Should only be called by kvm_apic_local_deliver() with LVT0,
1058 		 * before NMI watchdog was enabled. Already handled by
1059 		 * kvm_apic_accept_pic_intr().
1060 		 */
1061 		break;
1062 
1063 	default:
1064 		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1065 		       delivery_mode);
1066 		break;
1067 	}
1068 	return result;
1069 }
1070 
1071 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1072 {
1073 	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1074 }
1075 
1076 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1077 {
1078 	return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1079 }
1080 
1081 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1082 {
1083 	int trigger_mode;
1084 
1085 	/* Eoi the ioapic only if the ioapic doesn't own the vector. */
1086 	if (!kvm_ioapic_handles_vector(apic, vector))
1087 		return;
1088 
1089 	/* Request a KVM exit to inform the userspace IOAPIC. */
1090 	if (irqchip_split(apic->vcpu->kvm)) {
1091 		apic->vcpu->arch.pending_ioapic_eoi = vector;
1092 		kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1093 		return;
1094 	}
1095 
1096 	if (apic_test_vector(vector, apic->regs + APIC_TMR))
1097 		trigger_mode = IOAPIC_LEVEL_TRIG;
1098 	else
1099 		trigger_mode = IOAPIC_EDGE_TRIG;
1100 
1101 	kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1102 }
1103 
1104 static int apic_set_eoi(struct kvm_lapic *apic)
1105 {
1106 	int vector = apic_find_highest_isr(apic);
1107 
1108 	trace_kvm_eoi(apic, vector);
1109 
1110 	/*
1111 	 * Not every write EOI will has corresponding ISR,
1112 	 * one example is when Kernel check timer on setup_IO_APIC
1113 	 */
1114 	if (vector == -1)
1115 		return vector;
1116 
1117 	apic_clear_isr(vector, apic);
1118 	apic_update_ppr(apic);
1119 
1120 	if (test_bit(vector, vcpu_to_synic(apic->vcpu)->vec_bitmap))
1121 		kvm_hv_synic_send_eoi(apic->vcpu, vector);
1122 
1123 	kvm_ioapic_send_eoi(apic, vector);
1124 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1125 	return vector;
1126 }
1127 
1128 /*
1129  * this interface assumes a trap-like exit, which has already finished
1130  * desired side effect including vISR and vPPR update.
1131  */
1132 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1133 {
1134 	struct kvm_lapic *apic = vcpu->arch.apic;
1135 
1136 	trace_kvm_eoi(apic, vector);
1137 
1138 	kvm_ioapic_send_eoi(apic, vector);
1139 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1140 }
1141 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1142 
1143 static void apic_send_ipi(struct kvm_lapic *apic)
1144 {
1145 	u32 icr_low = kvm_lapic_get_reg(apic, APIC_ICR);
1146 	u32 icr_high = kvm_lapic_get_reg(apic, APIC_ICR2);
1147 	struct kvm_lapic_irq irq;
1148 
1149 	irq.vector = icr_low & APIC_VECTOR_MASK;
1150 	irq.delivery_mode = icr_low & APIC_MODE_MASK;
1151 	irq.dest_mode = icr_low & APIC_DEST_MASK;
1152 	irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1153 	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1154 	irq.shorthand = icr_low & APIC_SHORT_MASK;
1155 	irq.msi_redir_hint = false;
1156 	if (apic_x2apic_mode(apic))
1157 		irq.dest_id = icr_high;
1158 	else
1159 		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1160 
1161 	trace_kvm_apic_ipi(icr_low, irq.dest_id);
1162 
1163 	apic_debug("icr_high 0x%x, icr_low 0x%x, "
1164 		   "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
1165 		   "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x, "
1166 		   "msi_redir_hint 0x%x\n",
1167 		   icr_high, icr_low, irq.shorthand, irq.dest_id,
1168 		   irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
1169 		   irq.vector, irq.msi_redir_hint);
1170 
1171 	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1172 }
1173 
1174 static u32 apic_get_tmcct(struct kvm_lapic *apic)
1175 {
1176 	ktime_t remaining, now;
1177 	s64 ns;
1178 	u32 tmcct;
1179 
1180 	ASSERT(apic != NULL);
1181 
1182 	/* if initial count is 0, current count should also be 0 */
1183 	if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1184 		apic->lapic_timer.period == 0)
1185 		return 0;
1186 
1187 	now = ktime_get();
1188 	remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1189 	if (ktime_to_ns(remaining) < 0)
1190 		remaining = 0;
1191 
1192 	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1193 	tmcct = div64_u64(ns,
1194 			 (APIC_BUS_CYCLE_NS * apic->divide_count));
1195 
1196 	return tmcct;
1197 }
1198 
1199 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1200 {
1201 	struct kvm_vcpu *vcpu = apic->vcpu;
1202 	struct kvm_run *run = vcpu->run;
1203 
1204 	kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1205 	run->tpr_access.rip = kvm_rip_read(vcpu);
1206 	run->tpr_access.is_write = write;
1207 }
1208 
1209 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1210 {
1211 	if (apic->vcpu->arch.tpr_access_reporting)
1212 		__report_tpr_access(apic, write);
1213 }
1214 
1215 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1216 {
1217 	u32 val = 0;
1218 
1219 	if (offset >= LAPIC_MMIO_LENGTH)
1220 		return 0;
1221 
1222 	switch (offset) {
1223 	case APIC_ARBPRI:
1224 		apic_debug("Access APIC ARBPRI register which is for P6\n");
1225 		break;
1226 
1227 	case APIC_TMCCT:	/* Timer CCR */
1228 		if (apic_lvtt_tscdeadline(apic))
1229 			return 0;
1230 
1231 		val = apic_get_tmcct(apic);
1232 		break;
1233 	case APIC_PROCPRI:
1234 		apic_update_ppr(apic);
1235 		val = kvm_lapic_get_reg(apic, offset);
1236 		break;
1237 	case APIC_TASKPRI:
1238 		report_tpr_access(apic, false);
1239 		/* fall thru */
1240 	default:
1241 		val = kvm_lapic_get_reg(apic, offset);
1242 		break;
1243 	}
1244 
1245 	return val;
1246 }
1247 
1248 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1249 {
1250 	return container_of(dev, struct kvm_lapic, dev);
1251 }
1252 
1253 int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1254 		void *data)
1255 {
1256 	unsigned char alignment = offset & 0xf;
1257 	u32 result;
1258 	/* this bitmask has a bit cleared for each reserved register */
1259 	static const u64 rmask = 0x43ff01ffffffe70cULL;
1260 
1261 	if ((alignment + len) > 4) {
1262 		apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1263 			   offset, len);
1264 		return 1;
1265 	}
1266 
1267 	if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1268 		apic_debug("KVM_APIC_READ: read reserved register %x\n",
1269 			   offset);
1270 		return 1;
1271 	}
1272 
1273 	result = __apic_read(apic, offset & ~0xf);
1274 
1275 	trace_kvm_apic_read(offset, result);
1276 
1277 	switch (len) {
1278 	case 1:
1279 	case 2:
1280 	case 4:
1281 		memcpy(data, (char *)&result + alignment, len);
1282 		break;
1283 	default:
1284 		printk(KERN_ERR "Local APIC read with len = %x, "
1285 		       "should be 1,2, or 4 instead\n", len);
1286 		break;
1287 	}
1288 	return 0;
1289 }
1290 EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1291 
1292 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1293 {
1294 	return kvm_apic_hw_enabled(apic) &&
1295 	    addr >= apic->base_address &&
1296 	    addr < apic->base_address + LAPIC_MMIO_LENGTH;
1297 }
1298 
1299 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1300 			   gpa_t address, int len, void *data)
1301 {
1302 	struct kvm_lapic *apic = to_lapic(this);
1303 	u32 offset = address - apic->base_address;
1304 
1305 	if (!apic_mmio_in_range(apic, address))
1306 		return -EOPNOTSUPP;
1307 
1308 	kvm_lapic_reg_read(apic, offset, len, data);
1309 
1310 	return 0;
1311 }
1312 
1313 static void update_divide_count(struct kvm_lapic *apic)
1314 {
1315 	u32 tmp1, tmp2, tdcr;
1316 
1317 	tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1318 	tmp1 = tdcr & 0xf;
1319 	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1320 	apic->divide_count = 0x1 << (tmp2 & 0x7);
1321 
1322 	apic_debug("timer divide count is 0x%x\n",
1323 				   apic->divide_count);
1324 }
1325 
1326 static void limit_periodic_timer_frequency(struct kvm_lapic *apic)
1327 {
1328 	/*
1329 	 * Do not allow the guest to program periodic timers with small
1330 	 * interval, since the hrtimers are not throttled by the host
1331 	 * scheduler.
1332 	 */
1333 	if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1334 		s64 min_period = min_timer_period_us * 1000LL;
1335 
1336 		if (apic->lapic_timer.period < min_period) {
1337 			pr_info_ratelimited(
1338 			    "kvm: vcpu %i: requested %lld ns "
1339 			    "lapic timer period limited to %lld ns\n",
1340 			    apic->vcpu->vcpu_id,
1341 			    apic->lapic_timer.period, min_period);
1342 			apic->lapic_timer.period = min_period;
1343 		}
1344 	}
1345 }
1346 
1347 static void apic_update_lvtt(struct kvm_lapic *apic)
1348 {
1349 	u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1350 			apic->lapic_timer.timer_mode_mask;
1351 
1352 	if (apic->lapic_timer.timer_mode != timer_mode) {
1353 		if (apic_lvtt_tscdeadline(apic) != (timer_mode ==
1354 				APIC_LVT_TIMER_TSCDEADLINE)) {
1355 			hrtimer_cancel(&apic->lapic_timer.timer);
1356 			kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1357 			apic->lapic_timer.period = 0;
1358 			apic->lapic_timer.tscdeadline = 0;
1359 		}
1360 		apic->lapic_timer.timer_mode = timer_mode;
1361 		limit_periodic_timer_frequency(apic);
1362 	}
1363 }
1364 
1365 static void apic_timer_expired(struct kvm_lapic *apic)
1366 {
1367 	struct kvm_vcpu *vcpu = apic->vcpu;
1368 	struct swait_queue_head *q = &vcpu->wq;
1369 	struct kvm_timer *ktimer = &apic->lapic_timer;
1370 
1371 	if (atomic_read(&apic->lapic_timer.pending))
1372 		return;
1373 
1374 	atomic_inc(&apic->lapic_timer.pending);
1375 	kvm_set_pending_timer(vcpu);
1376 
1377 	/*
1378 	 * For x86, the atomic_inc() is serialized, thus
1379 	 * using swait_active() is safe.
1380 	 */
1381 	if (swait_active(q))
1382 		swake_up(q);
1383 
1384 	if (apic_lvtt_tscdeadline(apic))
1385 		ktimer->expired_tscdeadline = ktimer->tscdeadline;
1386 }
1387 
1388 /*
1389  * On APICv, this test will cause a busy wait
1390  * during a higher-priority task.
1391  */
1392 
1393 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1394 {
1395 	struct kvm_lapic *apic = vcpu->arch.apic;
1396 	u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1397 
1398 	if (kvm_apic_hw_enabled(apic)) {
1399 		int vec = reg & APIC_VECTOR_MASK;
1400 		void *bitmap = apic->regs + APIC_ISR;
1401 
1402 		if (vcpu->arch.apicv_active)
1403 			bitmap = apic->regs + APIC_IRR;
1404 
1405 		if (apic_test_vector(vec, bitmap))
1406 			return true;
1407 	}
1408 	return false;
1409 }
1410 
1411 void wait_lapic_expire(struct kvm_vcpu *vcpu)
1412 {
1413 	struct kvm_lapic *apic = vcpu->arch.apic;
1414 	u64 guest_tsc, tsc_deadline;
1415 
1416 	if (!lapic_in_kernel(vcpu))
1417 		return;
1418 
1419 	if (apic->lapic_timer.expired_tscdeadline == 0)
1420 		return;
1421 
1422 	if (!lapic_timer_int_injected(vcpu))
1423 		return;
1424 
1425 	tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1426 	apic->lapic_timer.expired_tscdeadline = 0;
1427 	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1428 	trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1429 
1430 	/* __delay is delay_tsc whenever the hardware has TSC, thus always.  */
1431 	if (guest_tsc < tsc_deadline)
1432 		__delay(min(tsc_deadline - guest_tsc,
1433 			nsec_to_cycles(vcpu, lapic_timer_advance_ns)));
1434 }
1435 
1436 static void start_sw_tscdeadline(struct kvm_lapic *apic)
1437 {
1438 	u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1439 	u64 ns = 0;
1440 	ktime_t expire;
1441 	struct kvm_vcpu *vcpu = apic->vcpu;
1442 	unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1443 	unsigned long flags;
1444 	ktime_t now;
1445 
1446 	if (unlikely(!tscdeadline || !this_tsc_khz))
1447 		return;
1448 
1449 	local_irq_save(flags);
1450 
1451 	now = ktime_get();
1452 	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1453 	if (likely(tscdeadline > guest_tsc)) {
1454 		ns = (tscdeadline - guest_tsc) * 1000000ULL;
1455 		do_div(ns, this_tsc_khz);
1456 		expire = ktime_add_ns(now, ns);
1457 		expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1458 		hrtimer_start(&apic->lapic_timer.timer,
1459 				expire, HRTIMER_MODE_ABS_PINNED);
1460 	} else
1461 		apic_timer_expired(apic);
1462 
1463 	local_irq_restore(flags);
1464 }
1465 
1466 static void start_sw_period(struct kvm_lapic *apic)
1467 {
1468 	if (!apic->lapic_timer.period)
1469 		return;
1470 
1471 	if (apic_lvtt_oneshot(apic) &&
1472 	    ktime_after(ktime_get(),
1473 			apic->lapic_timer.target_expiration)) {
1474 		apic_timer_expired(apic);
1475 		return;
1476 	}
1477 
1478 	hrtimer_start(&apic->lapic_timer.timer,
1479 		apic->lapic_timer.target_expiration,
1480 		HRTIMER_MODE_ABS_PINNED);
1481 }
1482 
1483 static void update_target_expiration(struct kvm_lapic *apic, uint32_t old_divisor)
1484 {
1485 	ktime_t now, remaining;
1486 	u64 ns_remaining_old, ns_remaining_new;
1487 
1488 	apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1489 		* APIC_BUS_CYCLE_NS * apic->divide_count;
1490 	limit_periodic_timer_frequency(apic);
1491 
1492 	now = ktime_get();
1493 	remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1494 	if (ktime_to_ns(remaining) < 0)
1495 		remaining = 0;
1496 
1497 	ns_remaining_old = ktime_to_ns(remaining);
1498 	ns_remaining_new = mul_u64_u32_div(ns_remaining_old,
1499 	                                   apic->divide_count, old_divisor);
1500 
1501 	apic->lapic_timer.tscdeadline +=
1502 		nsec_to_cycles(apic->vcpu, ns_remaining_new) -
1503 		nsec_to_cycles(apic->vcpu, ns_remaining_old);
1504 	apic->lapic_timer.target_expiration = ktime_add_ns(now, ns_remaining_new);
1505 }
1506 
1507 static bool set_target_expiration(struct kvm_lapic *apic)
1508 {
1509 	ktime_t now;
1510 	u64 tscl = rdtsc();
1511 
1512 	now = ktime_get();
1513 	apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1514 		* APIC_BUS_CYCLE_NS * apic->divide_count;
1515 
1516 	if (!apic->lapic_timer.period) {
1517 		apic->lapic_timer.tscdeadline = 0;
1518 		return false;
1519 	}
1520 
1521 	limit_periodic_timer_frequency(apic);
1522 
1523 	apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1524 		   PRIx64 ", "
1525 		   "timer initial count 0x%x, period %lldns, "
1526 		   "expire @ 0x%016" PRIx64 ".\n", __func__,
1527 		   APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1528 		   kvm_lapic_get_reg(apic, APIC_TMICT),
1529 		   apic->lapic_timer.period,
1530 		   ktime_to_ns(ktime_add_ns(now,
1531 				apic->lapic_timer.period)));
1532 
1533 	apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1534 		nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1535 	apic->lapic_timer.target_expiration = ktime_add_ns(now, apic->lapic_timer.period);
1536 
1537 	return true;
1538 }
1539 
1540 static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1541 {
1542 	apic->lapic_timer.tscdeadline +=
1543 		nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1544 	apic->lapic_timer.target_expiration =
1545 		ktime_add_ns(apic->lapic_timer.target_expiration,
1546 				apic->lapic_timer.period);
1547 }
1548 
1549 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1550 {
1551 	if (!lapic_in_kernel(vcpu))
1552 		return false;
1553 
1554 	return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1555 }
1556 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1557 
1558 static void cancel_hv_timer(struct kvm_lapic *apic)
1559 {
1560 	WARN_ON(preemptible());
1561 	WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1562 	kvm_x86_ops->cancel_hv_timer(apic->vcpu);
1563 	apic->lapic_timer.hv_timer_in_use = false;
1564 }
1565 
1566 static bool start_hv_timer(struct kvm_lapic *apic)
1567 {
1568 	struct kvm_timer *ktimer = &apic->lapic_timer;
1569 	int r;
1570 
1571 	WARN_ON(preemptible());
1572 	if (!kvm_x86_ops->set_hv_timer)
1573 		return false;
1574 
1575 	if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1576 		return false;
1577 
1578 	if (!ktimer->tscdeadline)
1579 		return false;
1580 
1581 	r = kvm_x86_ops->set_hv_timer(apic->vcpu, ktimer->tscdeadline);
1582 	if (r < 0)
1583 		return false;
1584 
1585 	ktimer->hv_timer_in_use = true;
1586 	hrtimer_cancel(&ktimer->timer);
1587 
1588 	/*
1589 	 * Also recheck ktimer->pending, in case the sw timer triggered in
1590 	 * the window.  For periodic timer, leave the hv timer running for
1591 	 * simplicity, and the deadline will be recomputed on the next vmexit.
1592 	 */
1593 	if (!apic_lvtt_period(apic) && (r || atomic_read(&ktimer->pending))) {
1594 		if (r)
1595 			apic_timer_expired(apic);
1596 		return false;
1597 	}
1598 
1599 	trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, true);
1600 	return true;
1601 }
1602 
1603 static void start_sw_timer(struct kvm_lapic *apic)
1604 {
1605 	struct kvm_timer *ktimer = &apic->lapic_timer;
1606 
1607 	WARN_ON(preemptible());
1608 	if (apic->lapic_timer.hv_timer_in_use)
1609 		cancel_hv_timer(apic);
1610 	if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1611 		return;
1612 
1613 	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1614 		start_sw_period(apic);
1615 	else if (apic_lvtt_tscdeadline(apic))
1616 		start_sw_tscdeadline(apic);
1617 	trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, false);
1618 }
1619 
1620 static void restart_apic_timer(struct kvm_lapic *apic)
1621 {
1622 	preempt_disable();
1623 	if (!start_hv_timer(apic))
1624 		start_sw_timer(apic);
1625 	preempt_enable();
1626 }
1627 
1628 void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1629 {
1630 	struct kvm_lapic *apic = vcpu->arch.apic;
1631 
1632 	preempt_disable();
1633 	/* If the preempt notifier has already run, it also called apic_timer_expired */
1634 	if (!apic->lapic_timer.hv_timer_in_use)
1635 		goto out;
1636 	WARN_ON(swait_active(&vcpu->wq));
1637 	cancel_hv_timer(apic);
1638 	apic_timer_expired(apic);
1639 
1640 	if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1641 		advance_periodic_target_expiration(apic);
1642 		restart_apic_timer(apic);
1643 	}
1644 out:
1645 	preempt_enable();
1646 }
1647 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1648 
1649 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1650 {
1651 	restart_apic_timer(vcpu->arch.apic);
1652 }
1653 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1654 
1655 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1656 {
1657 	struct kvm_lapic *apic = vcpu->arch.apic;
1658 
1659 	preempt_disable();
1660 	/* Possibly the TSC deadline timer is not enabled yet */
1661 	if (apic->lapic_timer.hv_timer_in_use)
1662 		start_sw_timer(apic);
1663 	preempt_enable();
1664 }
1665 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1666 
1667 void kvm_lapic_restart_hv_timer(struct kvm_vcpu *vcpu)
1668 {
1669 	struct kvm_lapic *apic = vcpu->arch.apic;
1670 
1671 	WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1672 	restart_apic_timer(apic);
1673 }
1674 
1675 static void start_apic_timer(struct kvm_lapic *apic)
1676 {
1677 	atomic_set(&apic->lapic_timer.pending, 0);
1678 
1679 	if ((apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1680 	    && !set_target_expiration(apic))
1681 		return;
1682 
1683 	restart_apic_timer(apic);
1684 }
1685 
1686 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1687 {
1688 	bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1689 
1690 	if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1691 		apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1692 		if (lvt0_in_nmi_mode) {
1693 			apic_debug("Receive NMI setting on APIC_LVT0 "
1694 				   "for cpu %d\n", apic->vcpu->vcpu_id);
1695 			atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1696 		} else
1697 			atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1698 	}
1699 }
1700 
1701 int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1702 {
1703 	int ret = 0;
1704 
1705 	trace_kvm_apic_write(reg, val);
1706 
1707 	switch (reg) {
1708 	case APIC_ID:		/* Local APIC ID */
1709 		if (!apic_x2apic_mode(apic))
1710 			kvm_apic_set_xapic_id(apic, val >> 24);
1711 		else
1712 			ret = 1;
1713 		break;
1714 
1715 	case APIC_TASKPRI:
1716 		report_tpr_access(apic, true);
1717 		apic_set_tpr(apic, val & 0xff);
1718 		break;
1719 
1720 	case APIC_EOI:
1721 		apic_set_eoi(apic);
1722 		break;
1723 
1724 	case APIC_LDR:
1725 		if (!apic_x2apic_mode(apic))
1726 			kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1727 		else
1728 			ret = 1;
1729 		break;
1730 
1731 	case APIC_DFR:
1732 		if (!apic_x2apic_mode(apic)) {
1733 			kvm_lapic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1734 			recalculate_apic_map(apic->vcpu->kvm);
1735 		} else
1736 			ret = 1;
1737 		break;
1738 
1739 	case APIC_SPIV: {
1740 		u32 mask = 0x3ff;
1741 		if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1742 			mask |= APIC_SPIV_DIRECTED_EOI;
1743 		apic_set_spiv(apic, val & mask);
1744 		if (!(val & APIC_SPIV_APIC_ENABLED)) {
1745 			int i;
1746 			u32 lvt_val;
1747 
1748 			for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
1749 				lvt_val = kvm_lapic_get_reg(apic,
1750 						       APIC_LVTT + 0x10 * i);
1751 				kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
1752 					     lvt_val | APIC_LVT_MASKED);
1753 			}
1754 			apic_update_lvtt(apic);
1755 			atomic_set(&apic->lapic_timer.pending, 0);
1756 
1757 		}
1758 		break;
1759 	}
1760 	case APIC_ICR:
1761 		/* No delay here, so we always clear the pending bit */
1762 		kvm_lapic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1763 		apic_send_ipi(apic);
1764 		break;
1765 
1766 	case APIC_ICR2:
1767 		if (!apic_x2apic_mode(apic))
1768 			val &= 0xff000000;
1769 		kvm_lapic_set_reg(apic, APIC_ICR2, val);
1770 		break;
1771 
1772 	case APIC_LVT0:
1773 		apic_manage_nmi_watchdog(apic, val);
1774 	case APIC_LVTTHMR:
1775 	case APIC_LVTPC:
1776 	case APIC_LVT1:
1777 	case APIC_LVTERR:
1778 		/* TODO: Check vector */
1779 		if (!kvm_apic_sw_enabled(apic))
1780 			val |= APIC_LVT_MASKED;
1781 
1782 		val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1783 		kvm_lapic_set_reg(apic, reg, val);
1784 
1785 		break;
1786 
1787 	case APIC_LVTT:
1788 		if (!kvm_apic_sw_enabled(apic))
1789 			val |= APIC_LVT_MASKED;
1790 		val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1791 		kvm_lapic_set_reg(apic, APIC_LVTT, val);
1792 		apic_update_lvtt(apic);
1793 		break;
1794 
1795 	case APIC_TMICT:
1796 		if (apic_lvtt_tscdeadline(apic))
1797 			break;
1798 
1799 		hrtimer_cancel(&apic->lapic_timer.timer);
1800 		kvm_lapic_set_reg(apic, APIC_TMICT, val);
1801 		start_apic_timer(apic);
1802 		break;
1803 
1804 	case APIC_TDCR: {
1805 		uint32_t old_divisor = apic->divide_count;
1806 
1807 		if (val & 4)
1808 			apic_debug("KVM_WRITE:TDCR %x\n", val);
1809 		kvm_lapic_set_reg(apic, APIC_TDCR, val);
1810 		update_divide_count(apic);
1811 		if (apic->divide_count != old_divisor &&
1812 				apic->lapic_timer.period) {
1813 			hrtimer_cancel(&apic->lapic_timer.timer);
1814 			update_target_expiration(apic, old_divisor);
1815 			restart_apic_timer(apic);
1816 		}
1817 		break;
1818 	}
1819 	case APIC_ESR:
1820 		if (apic_x2apic_mode(apic) && val != 0) {
1821 			apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1822 			ret = 1;
1823 		}
1824 		break;
1825 
1826 	case APIC_SELF_IPI:
1827 		if (apic_x2apic_mode(apic)) {
1828 			kvm_lapic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1829 		} else
1830 			ret = 1;
1831 		break;
1832 	default:
1833 		ret = 1;
1834 		break;
1835 	}
1836 	if (ret)
1837 		apic_debug("Local APIC Write to read-only register %x\n", reg);
1838 	return ret;
1839 }
1840 EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
1841 
1842 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1843 			    gpa_t address, int len, const void *data)
1844 {
1845 	struct kvm_lapic *apic = to_lapic(this);
1846 	unsigned int offset = address - apic->base_address;
1847 	u32 val;
1848 
1849 	if (!apic_mmio_in_range(apic, address))
1850 		return -EOPNOTSUPP;
1851 
1852 	/*
1853 	 * APIC register must be aligned on 128-bits boundary.
1854 	 * 32/64/128 bits registers must be accessed thru 32 bits.
1855 	 * Refer SDM 8.4.1
1856 	 */
1857 	if (len != 4 || (offset & 0xf)) {
1858 		/* Don't shout loud, $infamous_os would cause only noise. */
1859 		apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1860 		return 0;
1861 	}
1862 
1863 	val = *(u32*)data;
1864 
1865 	/* too common printing */
1866 	if (offset != APIC_EOI)
1867 		apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1868 			   "0x%x\n", __func__, offset, len, val);
1869 
1870 	kvm_lapic_reg_write(apic, offset & 0xff0, val);
1871 
1872 	return 0;
1873 }
1874 
1875 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1876 {
1877 	kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1878 }
1879 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1880 
1881 /* emulate APIC access in a trap manner */
1882 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1883 {
1884 	u32 val = 0;
1885 
1886 	/* hw has done the conditional check and inst decode */
1887 	offset &= 0xff0;
1888 
1889 	kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
1890 
1891 	/* TODO: optimize to just emulate side effect w/o one more write */
1892 	kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
1893 }
1894 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1895 
1896 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1897 {
1898 	struct kvm_lapic *apic = vcpu->arch.apic;
1899 
1900 	if (!vcpu->arch.apic)
1901 		return;
1902 
1903 	hrtimer_cancel(&apic->lapic_timer.timer);
1904 
1905 	if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1906 		static_key_slow_dec_deferred(&apic_hw_disabled);
1907 
1908 	if (!apic->sw_enabled)
1909 		static_key_slow_dec_deferred(&apic_sw_disabled);
1910 
1911 	if (apic->regs)
1912 		free_page((unsigned long)apic->regs);
1913 
1914 	kfree(apic);
1915 }
1916 
1917 /*
1918  *----------------------------------------------------------------------
1919  * LAPIC interface
1920  *----------------------------------------------------------------------
1921  */
1922 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1923 {
1924 	struct kvm_lapic *apic = vcpu->arch.apic;
1925 
1926 	if (!lapic_in_kernel(vcpu) ||
1927 		!apic_lvtt_tscdeadline(apic))
1928 		return 0;
1929 
1930 	return apic->lapic_timer.tscdeadline;
1931 }
1932 
1933 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1934 {
1935 	struct kvm_lapic *apic = vcpu->arch.apic;
1936 
1937 	if (!lapic_in_kernel(vcpu) || apic_lvtt_oneshot(apic) ||
1938 			apic_lvtt_period(apic))
1939 		return;
1940 
1941 	hrtimer_cancel(&apic->lapic_timer.timer);
1942 	apic->lapic_timer.tscdeadline = data;
1943 	start_apic_timer(apic);
1944 }
1945 
1946 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1947 {
1948 	struct kvm_lapic *apic = vcpu->arch.apic;
1949 
1950 	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1951 		     | (kvm_lapic_get_reg(apic, APIC_TASKPRI) & 4));
1952 }
1953 
1954 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1955 {
1956 	u64 tpr;
1957 
1958 	tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1959 
1960 	return (tpr & 0xf0) >> 4;
1961 }
1962 
1963 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1964 {
1965 	u64 old_value = vcpu->arch.apic_base;
1966 	struct kvm_lapic *apic = vcpu->arch.apic;
1967 
1968 	if (!apic)
1969 		value |= MSR_IA32_APICBASE_BSP;
1970 
1971 	vcpu->arch.apic_base = value;
1972 
1973 	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
1974 		kvm_update_cpuid(vcpu);
1975 
1976 	if (!apic)
1977 		return;
1978 
1979 	/* update jump label if enable bit changes */
1980 	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1981 		if (value & MSR_IA32_APICBASE_ENABLE) {
1982 			kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
1983 			static_key_slow_dec_deferred(&apic_hw_disabled);
1984 		} else {
1985 			static_key_slow_inc(&apic_hw_disabled.key);
1986 			recalculate_apic_map(vcpu->kvm);
1987 		}
1988 	}
1989 
1990 	if ((old_value ^ value) & X2APIC_ENABLE) {
1991 		if (value & X2APIC_ENABLE) {
1992 			kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
1993 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1994 		} else
1995 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1996 	}
1997 
1998 	apic->base_address = apic->vcpu->arch.apic_base &
1999 			     MSR_IA32_APICBASE_BASE;
2000 
2001 	if ((value & MSR_IA32_APICBASE_ENABLE) &&
2002 	     apic->base_address != APIC_DEFAULT_PHYS_BASE)
2003 		pr_warn_once("APIC base relocation is unsupported by KVM");
2004 
2005 	/* with FSB delivery interrupt, we can restart APIC functionality */
2006 	apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
2007 		   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
2008 
2009 }
2010 
2011 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
2012 {
2013 	struct kvm_lapic *apic = vcpu->arch.apic;
2014 	int i;
2015 
2016 	if (!apic)
2017 		return;
2018 
2019 	apic_debug("%s\n", __func__);
2020 
2021 	/* Stop the timer in case it's a reset to an active apic */
2022 	hrtimer_cancel(&apic->lapic_timer.timer);
2023 
2024 	if (!init_event) {
2025 		kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
2026 		                         MSR_IA32_APICBASE_ENABLE);
2027 		kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2028 	}
2029 	kvm_apic_set_version(apic->vcpu);
2030 
2031 	for (i = 0; i < KVM_APIC_LVT_NUM; i++)
2032 		kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
2033 	apic_update_lvtt(apic);
2034 	if (kvm_vcpu_is_reset_bsp(vcpu) &&
2035 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
2036 		kvm_lapic_set_reg(apic, APIC_LVT0,
2037 			     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
2038 	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2039 
2040 	kvm_lapic_set_reg(apic, APIC_DFR, 0xffffffffU);
2041 	apic_set_spiv(apic, 0xff);
2042 	kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
2043 	if (!apic_x2apic_mode(apic))
2044 		kvm_apic_set_ldr(apic, 0);
2045 	kvm_lapic_set_reg(apic, APIC_ESR, 0);
2046 	kvm_lapic_set_reg(apic, APIC_ICR, 0);
2047 	kvm_lapic_set_reg(apic, APIC_ICR2, 0);
2048 	kvm_lapic_set_reg(apic, APIC_TDCR, 0);
2049 	kvm_lapic_set_reg(apic, APIC_TMICT, 0);
2050 	for (i = 0; i < 8; i++) {
2051 		kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
2052 		kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
2053 		kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
2054 	}
2055 	apic->irr_pending = vcpu->arch.apicv_active;
2056 	apic->isr_count = vcpu->arch.apicv_active ? 1 : 0;
2057 	apic->highest_isr_cache = -1;
2058 	update_divide_count(apic);
2059 	atomic_set(&apic->lapic_timer.pending, 0);
2060 	if (kvm_vcpu_is_bsp(vcpu))
2061 		kvm_lapic_set_base(vcpu,
2062 				vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
2063 	vcpu->arch.pv_eoi.msr_val = 0;
2064 	apic_update_ppr(apic);
2065 	if (vcpu->arch.apicv_active) {
2066 		kvm_x86_ops->apicv_post_state_restore(vcpu);
2067 		kvm_x86_ops->hwapic_irr_update(vcpu, -1);
2068 		kvm_x86_ops->hwapic_isr_update(vcpu, -1);
2069 	}
2070 
2071 	vcpu->arch.apic_arb_prio = 0;
2072 	vcpu->arch.apic_attention = 0;
2073 
2074 	apic_debug("%s: vcpu=%p, id=0x%x, base_msr="
2075 		   "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
2076 		   vcpu, kvm_lapic_get_reg(apic, APIC_ID),
2077 		   vcpu->arch.apic_base, apic->base_address);
2078 }
2079 
2080 /*
2081  *----------------------------------------------------------------------
2082  * timer interface
2083  *----------------------------------------------------------------------
2084  */
2085 
2086 static bool lapic_is_periodic(struct kvm_lapic *apic)
2087 {
2088 	return apic_lvtt_period(apic);
2089 }
2090 
2091 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
2092 {
2093 	struct kvm_lapic *apic = vcpu->arch.apic;
2094 
2095 	if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
2096 		return atomic_read(&apic->lapic_timer.pending);
2097 
2098 	return 0;
2099 }
2100 
2101 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
2102 {
2103 	u32 reg = kvm_lapic_get_reg(apic, lvt_type);
2104 	int vector, mode, trig_mode;
2105 
2106 	if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2107 		vector = reg & APIC_VECTOR_MASK;
2108 		mode = reg & APIC_MODE_MASK;
2109 		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2110 		return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2111 					NULL);
2112 	}
2113 	return 0;
2114 }
2115 
2116 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2117 {
2118 	struct kvm_lapic *apic = vcpu->arch.apic;
2119 
2120 	if (apic)
2121 		kvm_apic_local_deliver(apic, APIC_LVT0);
2122 }
2123 
2124 static const struct kvm_io_device_ops apic_mmio_ops = {
2125 	.read     = apic_mmio_read,
2126 	.write    = apic_mmio_write,
2127 };
2128 
2129 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2130 {
2131 	struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2132 	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2133 
2134 	apic_timer_expired(apic);
2135 
2136 	if (lapic_is_periodic(apic)) {
2137 		advance_periodic_target_expiration(apic);
2138 		hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2139 		return HRTIMER_RESTART;
2140 	} else
2141 		return HRTIMER_NORESTART;
2142 }
2143 
2144 int kvm_create_lapic(struct kvm_vcpu *vcpu)
2145 {
2146 	struct kvm_lapic *apic;
2147 
2148 	ASSERT(vcpu != NULL);
2149 	apic_debug("apic_init %d\n", vcpu->vcpu_id);
2150 
2151 	apic = kzalloc(sizeof(*apic), GFP_KERNEL);
2152 	if (!apic)
2153 		goto nomem;
2154 
2155 	vcpu->arch.apic = apic;
2156 
2157 	apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
2158 	if (!apic->regs) {
2159 		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2160 		       vcpu->vcpu_id);
2161 		goto nomem_free_apic;
2162 	}
2163 	apic->vcpu = vcpu;
2164 
2165 	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2166 		     HRTIMER_MODE_ABS_PINNED);
2167 	apic->lapic_timer.timer.function = apic_timer_fn;
2168 
2169 	/*
2170 	 * APIC is created enabled. This will prevent kvm_lapic_set_base from
2171 	 * thinking that APIC satet has changed.
2172 	 */
2173 	vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2174 	static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2175 	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2176 
2177 	return 0;
2178 nomem_free_apic:
2179 	kfree(apic);
2180 nomem:
2181 	return -ENOMEM;
2182 }
2183 
2184 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2185 {
2186 	struct kvm_lapic *apic = vcpu->arch.apic;
2187 	u32 ppr;
2188 
2189 	if (!apic_enabled(apic))
2190 		return -1;
2191 
2192 	__apic_update_ppr(apic, &ppr);
2193 	return apic_has_interrupt_for_ppr(apic, ppr);
2194 }
2195 
2196 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2197 {
2198 	u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2199 	int r = 0;
2200 
2201 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2202 		r = 1;
2203 	if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2204 	    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2205 		r = 1;
2206 	return r;
2207 }
2208 
2209 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2210 {
2211 	struct kvm_lapic *apic = vcpu->arch.apic;
2212 
2213 	if (atomic_read(&apic->lapic_timer.pending) > 0) {
2214 		kvm_apic_local_deliver(apic, APIC_LVTT);
2215 		if (apic_lvtt_tscdeadline(apic))
2216 			apic->lapic_timer.tscdeadline = 0;
2217 		if (apic_lvtt_oneshot(apic)) {
2218 			apic->lapic_timer.tscdeadline = 0;
2219 			apic->lapic_timer.target_expiration = 0;
2220 		}
2221 		atomic_set(&apic->lapic_timer.pending, 0);
2222 	}
2223 }
2224 
2225 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2226 {
2227 	int vector = kvm_apic_has_interrupt(vcpu);
2228 	struct kvm_lapic *apic = vcpu->arch.apic;
2229 	u32 ppr;
2230 
2231 	if (vector == -1)
2232 		return -1;
2233 
2234 	/*
2235 	 * We get here even with APIC virtualization enabled, if doing
2236 	 * nested virtualization and L1 runs with the "acknowledge interrupt
2237 	 * on exit" mode.  Then we cannot inject the interrupt via RVI,
2238 	 * because the process would deliver it through the IDT.
2239 	 */
2240 
2241 	apic_clear_irr(vector, apic);
2242 	if (test_bit(vector, vcpu_to_synic(vcpu)->auto_eoi_bitmap)) {
2243 		/*
2244 		 * For auto-EOI interrupts, there might be another pending
2245 		 * interrupt above PPR, so check whether to raise another
2246 		 * KVM_REQ_EVENT.
2247 		 */
2248 		apic_update_ppr(apic);
2249 	} else {
2250 		/*
2251 		 * For normal interrupts, PPR has been raised and there cannot
2252 		 * be a higher-priority pending interrupt---except if there was
2253 		 * a concurrent interrupt injection, but that would have
2254 		 * triggered KVM_REQ_EVENT already.
2255 		 */
2256 		apic_set_isr(vector, apic);
2257 		__apic_update_ppr(apic, &ppr);
2258 	}
2259 
2260 	return vector;
2261 }
2262 
2263 static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2264 		struct kvm_lapic_state *s, bool set)
2265 {
2266 	if (apic_x2apic_mode(vcpu->arch.apic)) {
2267 		u32 *id = (u32 *)(s->regs + APIC_ID);
2268 		u32 *ldr = (u32 *)(s->regs + APIC_LDR);
2269 
2270 		if (vcpu->kvm->arch.x2apic_format) {
2271 			if (*id != vcpu->vcpu_id)
2272 				return -EINVAL;
2273 		} else {
2274 			if (set)
2275 				*id >>= 24;
2276 			else
2277 				*id <<= 24;
2278 		}
2279 
2280 		/* In x2APIC mode, the LDR is fixed and based on the id */
2281 		if (set)
2282 			*ldr = kvm_apic_calc_x2apic_ldr(*id);
2283 	}
2284 
2285 	return 0;
2286 }
2287 
2288 int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2289 {
2290 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2291 	return kvm_apic_state_fixup(vcpu, s, false);
2292 }
2293 
2294 int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2295 {
2296 	struct kvm_lapic *apic = vcpu->arch.apic;
2297 	int r;
2298 
2299 
2300 	kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2301 	/* set SPIV separately to get count of SW disabled APICs right */
2302 	apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2303 
2304 	r = kvm_apic_state_fixup(vcpu, s, true);
2305 	if (r)
2306 		return r;
2307 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2308 
2309 	recalculate_apic_map(vcpu->kvm);
2310 	kvm_apic_set_version(vcpu);
2311 
2312 	apic_update_ppr(apic);
2313 	hrtimer_cancel(&apic->lapic_timer.timer);
2314 	apic_update_lvtt(apic);
2315 	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2316 	update_divide_count(apic);
2317 	start_apic_timer(apic);
2318 	apic->irr_pending = true;
2319 	apic->isr_count = vcpu->arch.apicv_active ?
2320 				1 : count_vectors(apic->regs + APIC_ISR);
2321 	apic->highest_isr_cache = -1;
2322 	if (vcpu->arch.apicv_active) {
2323 		kvm_x86_ops->apicv_post_state_restore(vcpu);
2324 		kvm_x86_ops->hwapic_irr_update(vcpu,
2325 				apic_find_highest_irr(apic));
2326 		kvm_x86_ops->hwapic_isr_update(vcpu,
2327 				apic_find_highest_isr(apic));
2328 	}
2329 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2330 	if (ioapic_in_kernel(vcpu->kvm))
2331 		kvm_rtc_eoi_tracking_restore_one(vcpu);
2332 
2333 	vcpu->arch.apic_arb_prio = 0;
2334 
2335 	return 0;
2336 }
2337 
2338 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2339 {
2340 	struct hrtimer *timer;
2341 
2342 	if (!lapic_in_kernel(vcpu))
2343 		return;
2344 
2345 	timer = &vcpu->arch.apic->lapic_timer.timer;
2346 	if (hrtimer_cancel(timer))
2347 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
2348 }
2349 
2350 /*
2351  * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2352  *
2353  * Detect whether guest triggered PV EOI since the
2354  * last entry. If yes, set EOI on guests's behalf.
2355  * Clear PV EOI in guest memory in any case.
2356  */
2357 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2358 					struct kvm_lapic *apic)
2359 {
2360 	bool pending;
2361 	int vector;
2362 	/*
2363 	 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2364 	 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2365 	 *
2366 	 * KVM_APIC_PV_EOI_PENDING is unset:
2367 	 * 	-> host disabled PV EOI.
2368 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2369 	 * 	-> host enabled PV EOI, guest did not execute EOI yet.
2370 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2371 	 * 	-> host enabled PV EOI, guest executed EOI.
2372 	 */
2373 	BUG_ON(!pv_eoi_enabled(vcpu));
2374 	pending = pv_eoi_get_pending(vcpu);
2375 	/*
2376 	 * Clear pending bit in any case: it will be set again on vmentry.
2377 	 * While this might not be ideal from performance point of view,
2378 	 * this makes sure pv eoi is only enabled when we know it's safe.
2379 	 */
2380 	pv_eoi_clr_pending(vcpu);
2381 	if (pending)
2382 		return;
2383 	vector = apic_set_eoi(apic);
2384 	trace_kvm_pv_eoi(apic, vector);
2385 }
2386 
2387 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2388 {
2389 	u32 data;
2390 
2391 	if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2392 		apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2393 
2394 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2395 		return;
2396 
2397 	if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2398 				  sizeof(u32)))
2399 		return;
2400 
2401 	apic_set_tpr(vcpu->arch.apic, data & 0xff);
2402 }
2403 
2404 /*
2405  * apic_sync_pv_eoi_to_guest - called before vmentry
2406  *
2407  * Detect whether it's safe to enable PV EOI and
2408  * if yes do so.
2409  */
2410 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2411 					struct kvm_lapic *apic)
2412 {
2413 	if (!pv_eoi_enabled(vcpu) ||
2414 	    /* IRR set or many bits in ISR: could be nested. */
2415 	    apic->irr_pending ||
2416 	    /* Cache not set: could be safe but we don't bother. */
2417 	    apic->highest_isr_cache == -1 ||
2418 	    /* Need EOI to update ioapic. */
2419 	    kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2420 		/*
2421 		 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2422 		 * so we need not do anything here.
2423 		 */
2424 		return;
2425 	}
2426 
2427 	pv_eoi_set_pending(apic->vcpu);
2428 }
2429 
2430 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2431 {
2432 	u32 data, tpr;
2433 	int max_irr, max_isr;
2434 	struct kvm_lapic *apic = vcpu->arch.apic;
2435 
2436 	apic_sync_pv_eoi_to_guest(vcpu, apic);
2437 
2438 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2439 		return;
2440 
2441 	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2442 	max_irr = apic_find_highest_irr(apic);
2443 	if (max_irr < 0)
2444 		max_irr = 0;
2445 	max_isr = apic_find_highest_isr(apic);
2446 	if (max_isr < 0)
2447 		max_isr = 0;
2448 	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2449 
2450 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2451 				sizeof(u32));
2452 }
2453 
2454 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2455 {
2456 	if (vapic_addr) {
2457 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2458 					&vcpu->arch.apic->vapic_cache,
2459 					vapic_addr, sizeof(u32)))
2460 			return -EINVAL;
2461 		__set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2462 	} else {
2463 		__clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2464 	}
2465 
2466 	vcpu->arch.apic->vapic_addr = vapic_addr;
2467 	return 0;
2468 }
2469 
2470 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2471 {
2472 	struct kvm_lapic *apic = vcpu->arch.apic;
2473 	u32 reg = (msr - APIC_BASE_MSR) << 4;
2474 
2475 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2476 		return 1;
2477 
2478 	if (reg == APIC_ICR2)
2479 		return 1;
2480 
2481 	/* if this is ICR write vector before command */
2482 	if (reg == APIC_ICR)
2483 		kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2484 	return kvm_lapic_reg_write(apic, reg, (u32)data);
2485 }
2486 
2487 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2488 {
2489 	struct kvm_lapic *apic = vcpu->arch.apic;
2490 	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2491 
2492 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2493 		return 1;
2494 
2495 	if (reg == APIC_DFR || reg == APIC_ICR2) {
2496 		apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
2497 			   reg);
2498 		return 1;
2499 	}
2500 
2501 	if (kvm_lapic_reg_read(apic, reg, 4, &low))
2502 		return 1;
2503 	if (reg == APIC_ICR)
2504 		kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2505 
2506 	*data = (((u64)high) << 32) | low;
2507 
2508 	return 0;
2509 }
2510 
2511 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2512 {
2513 	struct kvm_lapic *apic = vcpu->arch.apic;
2514 
2515 	if (!lapic_in_kernel(vcpu))
2516 		return 1;
2517 
2518 	/* if this is ICR write vector before command */
2519 	if (reg == APIC_ICR)
2520 		kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2521 	return kvm_lapic_reg_write(apic, reg, (u32)data);
2522 }
2523 
2524 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2525 {
2526 	struct kvm_lapic *apic = vcpu->arch.apic;
2527 	u32 low, high = 0;
2528 
2529 	if (!lapic_in_kernel(vcpu))
2530 		return 1;
2531 
2532 	if (kvm_lapic_reg_read(apic, reg, 4, &low))
2533 		return 1;
2534 	if (reg == APIC_ICR)
2535 		kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2536 
2537 	*data = (((u64)high) << 32) | low;
2538 
2539 	return 0;
2540 }
2541 
2542 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2543 {
2544 	u64 addr = data & ~KVM_MSR_ENABLED;
2545 	if (!IS_ALIGNED(addr, 4))
2546 		return 1;
2547 
2548 	vcpu->arch.pv_eoi.msr_val = data;
2549 	if (!pv_eoi_enabled(vcpu))
2550 		return 0;
2551 	return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
2552 					 addr, sizeof(u8));
2553 }
2554 
2555 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2556 {
2557 	struct kvm_lapic *apic = vcpu->arch.apic;
2558 	u8 sipi_vector;
2559 	unsigned long pe;
2560 
2561 	if (!lapic_in_kernel(vcpu) || !apic->pending_events)
2562 		return;
2563 
2564 	/*
2565 	 * INITs are latched while in SMM.  Because an SMM CPU cannot
2566 	 * be in KVM_MP_STATE_INIT_RECEIVED state, just eat SIPIs
2567 	 * and delay processing of INIT until the next RSM.
2568 	 */
2569 	if (is_smm(vcpu)) {
2570 		WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2571 		if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2572 			clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2573 		return;
2574 	}
2575 
2576 	pe = xchg(&apic->pending_events, 0);
2577 	if (test_bit(KVM_APIC_INIT, &pe)) {
2578 		kvm_vcpu_reset(vcpu, true);
2579 		if (kvm_vcpu_is_bsp(apic->vcpu))
2580 			vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2581 		else
2582 			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2583 	}
2584 	if (test_bit(KVM_APIC_SIPI, &pe) &&
2585 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2586 		/* evaluate pending_events before reading the vector */
2587 		smp_rmb();
2588 		sipi_vector = apic->sipi_vector;
2589 		apic_debug("vcpu %d received sipi with vector # %x\n",
2590 			 vcpu->vcpu_id, sipi_vector);
2591 		kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2592 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2593 	}
2594 }
2595 
2596 void kvm_lapic_init(void)
2597 {
2598 	/* do not patch jump label more than once per second */
2599 	jump_label_rate_limit(&apic_hw_disabled, HZ);
2600 	jump_label_rate_limit(&apic_sw_disabled, HZ);
2601 }
2602 
2603 void kvm_lapic_exit(void)
2604 {
2605 	static_key_deferred_flush(&apic_hw_disabled);
2606 	static_key_deferred_flush(&apic_sw_disabled);
2607 }
2608