xref: /openbmc/linux/arch/x86/kvm/lapic.c (revision e3b9f1e8)
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 	feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
325 	if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
326 		v |= APIC_LVR_DIRECTED_EOI;
327 	kvm_lapic_set_reg(apic, APIC_LVR, v);
328 }
329 
330 static const unsigned int apic_lvt_mask[KVM_APIC_LVT_NUM] = {
331 	LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
332 	LVT_MASK | APIC_MODE_MASK,	/* LVTTHMR */
333 	LVT_MASK | APIC_MODE_MASK,	/* LVTPC */
334 	LINT_MASK, LINT_MASK,	/* LVT0-1 */
335 	LVT_MASK		/* LVTERR */
336 };
337 
338 static int find_highest_vector(void *bitmap)
339 {
340 	int vec;
341 	u32 *reg;
342 
343 	for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
344 	     vec >= 0; vec -= APIC_VECTORS_PER_REG) {
345 		reg = bitmap + REG_POS(vec);
346 		if (*reg)
347 			return __fls(*reg) + vec;
348 	}
349 
350 	return -1;
351 }
352 
353 static u8 count_vectors(void *bitmap)
354 {
355 	int vec;
356 	u32 *reg;
357 	u8 count = 0;
358 
359 	for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
360 		reg = bitmap + REG_POS(vec);
361 		count += hweight32(*reg);
362 	}
363 
364 	return count;
365 }
366 
367 bool __kvm_apic_update_irr(u32 *pir, void *regs, int *max_irr)
368 {
369 	u32 i, vec;
370 	u32 pir_val, irr_val, prev_irr_val;
371 	int max_updated_irr;
372 
373 	max_updated_irr = -1;
374 	*max_irr = -1;
375 
376 	for (i = vec = 0; i <= 7; i++, vec += 32) {
377 		pir_val = READ_ONCE(pir[i]);
378 		irr_val = *((u32 *)(regs + APIC_IRR + i * 0x10));
379 		if (pir_val) {
380 			prev_irr_val = irr_val;
381 			irr_val |= xchg(&pir[i], 0);
382 			*((u32 *)(regs + APIC_IRR + i * 0x10)) = irr_val;
383 			if (prev_irr_val != irr_val) {
384 				max_updated_irr =
385 					__fls(irr_val ^ prev_irr_val) + vec;
386 			}
387 		}
388 		if (irr_val)
389 			*max_irr = __fls(irr_val) + vec;
390 	}
391 
392 	return ((max_updated_irr != -1) &&
393 		(max_updated_irr == *max_irr));
394 }
395 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
396 
397 bool kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir, int *max_irr)
398 {
399 	struct kvm_lapic *apic = vcpu->arch.apic;
400 
401 	return __kvm_apic_update_irr(pir, apic->regs, max_irr);
402 }
403 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
404 
405 static inline int apic_search_irr(struct kvm_lapic *apic)
406 {
407 	return find_highest_vector(apic->regs + APIC_IRR);
408 }
409 
410 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
411 {
412 	int result;
413 
414 	/*
415 	 * Note that irr_pending is just a hint. It will be always
416 	 * true with virtual interrupt delivery enabled.
417 	 */
418 	if (!apic->irr_pending)
419 		return -1;
420 
421 	result = apic_search_irr(apic);
422 	ASSERT(result == -1 || result >= 16);
423 
424 	return result;
425 }
426 
427 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
428 {
429 	struct kvm_vcpu *vcpu;
430 
431 	vcpu = apic->vcpu;
432 
433 	if (unlikely(vcpu->arch.apicv_active)) {
434 		/* need to update RVI */
435 		apic_clear_vector(vec, apic->regs + APIC_IRR);
436 		kvm_x86_ops->hwapic_irr_update(vcpu,
437 				apic_find_highest_irr(apic));
438 	} else {
439 		apic->irr_pending = false;
440 		apic_clear_vector(vec, apic->regs + APIC_IRR);
441 		if (apic_search_irr(apic) != -1)
442 			apic->irr_pending = true;
443 	}
444 }
445 
446 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
447 {
448 	struct kvm_vcpu *vcpu;
449 
450 	if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
451 		return;
452 
453 	vcpu = apic->vcpu;
454 
455 	/*
456 	 * With APIC virtualization enabled, all caching is disabled
457 	 * because the processor can modify ISR under the hood.  Instead
458 	 * just set SVI.
459 	 */
460 	if (unlikely(vcpu->arch.apicv_active))
461 		kvm_x86_ops->hwapic_isr_update(vcpu, vec);
462 	else {
463 		++apic->isr_count;
464 		BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
465 		/*
466 		 * ISR (in service register) bit is set when injecting an interrupt.
467 		 * The highest vector is injected. Thus the latest bit set matches
468 		 * the highest bit in ISR.
469 		 */
470 		apic->highest_isr_cache = vec;
471 	}
472 }
473 
474 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
475 {
476 	int result;
477 
478 	/*
479 	 * Note that isr_count is always 1, and highest_isr_cache
480 	 * is always -1, with APIC virtualization enabled.
481 	 */
482 	if (!apic->isr_count)
483 		return -1;
484 	if (likely(apic->highest_isr_cache != -1))
485 		return apic->highest_isr_cache;
486 
487 	result = find_highest_vector(apic->regs + APIC_ISR);
488 	ASSERT(result == -1 || result >= 16);
489 
490 	return result;
491 }
492 
493 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
494 {
495 	struct kvm_vcpu *vcpu;
496 	if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
497 		return;
498 
499 	vcpu = apic->vcpu;
500 
501 	/*
502 	 * We do get here for APIC virtualization enabled if the guest
503 	 * uses the Hyper-V APIC enlightenment.  In this case we may need
504 	 * to trigger a new interrupt delivery by writing the SVI field;
505 	 * on the other hand isr_count and highest_isr_cache are unused
506 	 * and must be left alone.
507 	 */
508 	if (unlikely(vcpu->arch.apicv_active))
509 		kvm_x86_ops->hwapic_isr_update(vcpu,
510 					       apic_find_highest_isr(apic));
511 	else {
512 		--apic->isr_count;
513 		BUG_ON(apic->isr_count < 0);
514 		apic->highest_isr_cache = -1;
515 	}
516 }
517 
518 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
519 {
520 	/* This may race with setting of irr in __apic_accept_irq() and
521 	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
522 	 * will cause vmexit immediately and the value will be recalculated
523 	 * on the next vmentry.
524 	 */
525 	return apic_find_highest_irr(vcpu->arch.apic);
526 }
527 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
528 
529 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
530 			     int vector, int level, int trig_mode,
531 			     struct dest_map *dest_map);
532 
533 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
534 		     struct dest_map *dest_map)
535 {
536 	struct kvm_lapic *apic = vcpu->arch.apic;
537 
538 	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
539 			irq->level, irq->trig_mode, dest_map);
540 }
541 
542 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
543 {
544 
545 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
546 				      sizeof(val));
547 }
548 
549 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
550 {
551 
552 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
553 				      sizeof(*val));
554 }
555 
556 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
557 {
558 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
559 }
560 
561 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
562 {
563 	u8 val;
564 	if (pv_eoi_get_user(vcpu, &val) < 0)
565 		apic_debug("Can't read EOI MSR value: 0x%llx\n",
566 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
567 	return val & 0x1;
568 }
569 
570 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
571 {
572 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
573 		apic_debug("Can't set EOI MSR value: 0x%llx\n",
574 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
575 		return;
576 	}
577 	__set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
578 }
579 
580 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
581 {
582 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
583 		apic_debug("Can't clear EOI MSR value: 0x%llx\n",
584 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
585 		return;
586 	}
587 	__clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
588 }
589 
590 static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
591 {
592 	int highest_irr;
593 	if (apic->vcpu->arch.apicv_active)
594 		highest_irr = kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
595 	else
596 		highest_irr = apic_find_highest_irr(apic);
597 	if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
598 		return -1;
599 	return highest_irr;
600 }
601 
602 static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
603 {
604 	u32 tpr, isrv, ppr, old_ppr;
605 	int isr;
606 
607 	old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
608 	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
609 	isr = apic_find_highest_isr(apic);
610 	isrv = (isr != -1) ? isr : 0;
611 
612 	if ((tpr & 0xf0) >= (isrv & 0xf0))
613 		ppr = tpr & 0xff;
614 	else
615 		ppr = isrv & 0xf0;
616 
617 	apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
618 		   apic, ppr, isr, isrv);
619 
620 	*new_ppr = ppr;
621 	if (old_ppr != ppr)
622 		kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
623 
624 	return ppr < old_ppr;
625 }
626 
627 static void apic_update_ppr(struct kvm_lapic *apic)
628 {
629 	u32 ppr;
630 
631 	if (__apic_update_ppr(apic, &ppr) &&
632 	    apic_has_interrupt_for_ppr(apic, ppr) != -1)
633 		kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
634 }
635 
636 void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
637 {
638 	apic_update_ppr(vcpu->arch.apic);
639 }
640 EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
641 
642 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
643 {
644 	kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
645 	apic_update_ppr(apic);
646 }
647 
648 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
649 {
650 	return mda == (apic_x2apic_mode(apic) ?
651 			X2APIC_BROADCAST : APIC_BROADCAST);
652 }
653 
654 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
655 {
656 	if (kvm_apic_broadcast(apic, mda))
657 		return true;
658 
659 	if (apic_x2apic_mode(apic))
660 		return mda == kvm_x2apic_id(apic);
661 
662 	/*
663 	 * Hotplug hack: Make LAPIC in xAPIC mode also accept interrupts as if
664 	 * it were in x2APIC mode.  Hotplugged VCPUs start in xAPIC mode and
665 	 * this allows unique addressing of VCPUs with APIC ID over 0xff.
666 	 * The 0xff condition is needed because writeable xAPIC ID.
667 	 */
668 	if (kvm_x2apic_id(apic) > 0xff && mda == kvm_x2apic_id(apic))
669 		return true;
670 
671 	return mda == kvm_xapic_id(apic);
672 }
673 
674 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
675 {
676 	u32 logical_id;
677 
678 	if (kvm_apic_broadcast(apic, mda))
679 		return true;
680 
681 	logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
682 
683 	if (apic_x2apic_mode(apic))
684 		return ((logical_id >> 16) == (mda >> 16))
685 		       && (logical_id & mda & 0xffff) != 0;
686 
687 	logical_id = GET_APIC_LOGICAL_ID(logical_id);
688 
689 	switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
690 	case APIC_DFR_FLAT:
691 		return (logical_id & mda) != 0;
692 	case APIC_DFR_CLUSTER:
693 		return ((logical_id >> 4) == (mda >> 4))
694 		       && (logical_id & mda & 0xf) != 0;
695 	default:
696 		apic_debug("Bad DFR vcpu %d: %08x\n",
697 			   apic->vcpu->vcpu_id, kvm_lapic_get_reg(apic, APIC_DFR));
698 		return false;
699 	}
700 }
701 
702 /* The KVM local APIC implementation has two quirks:
703  *
704  *  - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
705  *    in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
706  *    KVM doesn't do that aliasing.
707  *
708  *  - in-kernel IOAPIC messages have to be delivered directly to
709  *    x2APIC, because the kernel does not support interrupt remapping.
710  *    In order to support broadcast without interrupt remapping, x2APIC
711  *    rewrites the destination of non-IPI messages from APIC_BROADCAST
712  *    to X2APIC_BROADCAST.
713  *
714  * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API.  This is
715  * important when userspace wants to use x2APIC-format MSIs, because
716  * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
717  */
718 static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
719 		struct kvm_lapic *source, struct kvm_lapic *target)
720 {
721 	bool ipi = source != NULL;
722 
723 	if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
724 	    !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
725 		return X2APIC_BROADCAST;
726 
727 	return dest_id;
728 }
729 
730 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
731 			   int short_hand, unsigned int dest, int dest_mode)
732 {
733 	struct kvm_lapic *target = vcpu->arch.apic;
734 	u32 mda = kvm_apic_mda(vcpu, dest, source, target);
735 
736 	apic_debug("target %p, source %p, dest 0x%x, "
737 		   "dest_mode 0x%x, short_hand 0x%x\n",
738 		   target, source, dest, dest_mode, short_hand);
739 
740 	ASSERT(target);
741 	switch (short_hand) {
742 	case APIC_DEST_NOSHORT:
743 		if (dest_mode == APIC_DEST_PHYSICAL)
744 			return kvm_apic_match_physical_addr(target, mda);
745 		else
746 			return kvm_apic_match_logical_addr(target, mda);
747 	case APIC_DEST_SELF:
748 		return target == source;
749 	case APIC_DEST_ALLINC:
750 		return true;
751 	case APIC_DEST_ALLBUT:
752 		return target != source;
753 	default:
754 		apic_debug("kvm: apic: Bad dest shorthand value %x\n",
755 			   short_hand);
756 		return false;
757 	}
758 }
759 EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
760 
761 int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
762 		       const unsigned long *bitmap, u32 bitmap_size)
763 {
764 	u32 mod;
765 	int i, idx = -1;
766 
767 	mod = vector % dest_vcpus;
768 
769 	for (i = 0; i <= mod; i++) {
770 		idx = find_next_bit(bitmap, bitmap_size, idx + 1);
771 		BUG_ON(idx == bitmap_size);
772 	}
773 
774 	return idx;
775 }
776 
777 static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
778 {
779 	if (!kvm->arch.disabled_lapic_found) {
780 		kvm->arch.disabled_lapic_found = true;
781 		printk(KERN_INFO
782 		       "Disabled LAPIC found during irq injection\n");
783 	}
784 }
785 
786 static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
787 		struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
788 {
789 	if (kvm->arch.x2apic_broadcast_quirk_disabled) {
790 		if ((irq->dest_id == APIC_BROADCAST &&
791 				map->mode != KVM_APIC_MODE_X2APIC))
792 			return true;
793 		if (irq->dest_id == X2APIC_BROADCAST)
794 			return true;
795 	} else {
796 		bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
797 		if (irq->dest_id == (x2apic_ipi ?
798 		                     X2APIC_BROADCAST : APIC_BROADCAST))
799 			return true;
800 	}
801 
802 	return false;
803 }
804 
805 /* Return true if the interrupt can be handled by using *bitmap as index mask
806  * for valid destinations in *dst array.
807  * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
808  * Note: we may have zero kvm_lapic destinations when we return true, which
809  * means that the interrupt should be dropped.  In this case, *bitmap would be
810  * zero and *dst undefined.
811  */
812 static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
813 		struct kvm_lapic **src, struct kvm_lapic_irq *irq,
814 		struct kvm_apic_map *map, struct kvm_lapic ***dst,
815 		unsigned long *bitmap)
816 {
817 	int i, lowest;
818 
819 	if (irq->shorthand == APIC_DEST_SELF && src) {
820 		*dst = src;
821 		*bitmap = 1;
822 		return true;
823 	} else if (irq->shorthand)
824 		return false;
825 
826 	if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
827 		return false;
828 
829 	if (irq->dest_mode == APIC_DEST_PHYSICAL) {
830 		if (irq->dest_id > map->max_apic_id) {
831 			*bitmap = 0;
832 		} else {
833 			*dst = &map->phys_map[irq->dest_id];
834 			*bitmap = 1;
835 		}
836 		return true;
837 	}
838 
839 	*bitmap = 0;
840 	if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
841 				(u16 *)bitmap))
842 		return false;
843 
844 	if (!kvm_lowest_prio_delivery(irq))
845 		return true;
846 
847 	if (!kvm_vector_hashing_enabled()) {
848 		lowest = -1;
849 		for_each_set_bit(i, bitmap, 16) {
850 			if (!(*dst)[i])
851 				continue;
852 			if (lowest < 0)
853 				lowest = i;
854 			else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
855 						(*dst)[lowest]->vcpu) < 0)
856 				lowest = i;
857 		}
858 	} else {
859 		if (!*bitmap)
860 			return true;
861 
862 		lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
863 				bitmap, 16);
864 
865 		if (!(*dst)[lowest]) {
866 			kvm_apic_disabled_lapic_found(kvm);
867 			*bitmap = 0;
868 			return true;
869 		}
870 	}
871 
872 	*bitmap = (lowest >= 0) ? 1 << lowest : 0;
873 
874 	return true;
875 }
876 
877 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
878 		struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
879 {
880 	struct kvm_apic_map *map;
881 	unsigned long bitmap;
882 	struct kvm_lapic **dst = NULL;
883 	int i;
884 	bool ret;
885 
886 	*r = -1;
887 
888 	if (irq->shorthand == APIC_DEST_SELF) {
889 		*r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
890 		return true;
891 	}
892 
893 	rcu_read_lock();
894 	map = rcu_dereference(kvm->arch.apic_map);
895 
896 	ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
897 	if (ret)
898 		for_each_set_bit(i, &bitmap, 16) {
899 			if (!dst[i])
900 				continue;
901 			if (*r < 0)
902 				*r = 0;
903 			*r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
904 		}
905 
906 	rcu_read_unlock();
907 	return ret;
908 }
909 
910 /*
911  * This routine tries to handler interrupts in posted mode, here is how
912  * it deals with different cases:
913  * - For single-destination interrupts, handle it in posted mode
914  * - Else if vector hashing is enabled and it is a lowest-priority
915  *   interrupt, handle it in posted mode and use the following mechanism
916  *   to find the destinaiton vCPU.
917  *	1. For lowest-priority interrupts, store all the possible
918  *	   destination vCPUs in an array.
919  *	2. Use "guest vector % max number of destination vCPUs" to find
920  *	   the right destination vCPU in the array for the lowest-priority
921  *	   interrupt.
922  * - Otherwise, use remapped mode to inject the interrupt.
923  */
924 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
925 			struct kvm_vcpu **dest_vcpu)
926 {
927 	struct kvm_apic_map *map;
928 	unsigned long bitmap;
929 	struct kvm_lapic **dst = NULL;
930 	bool ret = false;
931 
932 	if (irq->shorthand)
933 		return false;
934 
935 	rcu_read_lock();
936 	map = rcu_dereference(kvm->arch.apic_map);
937 
938 	if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
939 			hweight16(bitmap) == 1) {
940 		unsigned long i = find_first_bit(&bitmap, 16);
941 
942 		if (dst[i]) {
943 			*dest_vcpu = dst[i]->vcpu;
944 			ret = true;
945 		}
946 	}
947 
948 	rcu_read_unlock();
949 	return ret;
950 }
951 
952 /*
953  * Add a pending IRQ into lapic.
954  * Return 1 if successfully added and 0 if discarded.
955  */
956 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
957 			     int vector, int level, int trig_mode,
958 			     struct dest_map *dest_map)
959 {
960 	int result = 0;
961 	struct kvm_vcpu *vcpu = apic->vcpu;
962 
963 	trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
964 				  trig_mode, vector);
965 	switch (delivery_mode) {
966 	case APIC_DM_LOWEST:
967 		vcpu->arch.apic_arb_prio++;
968 	case APIC_DM_FIXED:
969 		if (unlikely(trig_mode && !level))
970 			break;
971 
972 		/* FIXME add logic for vcpu on reset */
973 		if (unlikely(!apic_enabled(apic)))
974 			break;
975 
976 		result = 1;
977 
978 		if (dest_map) {
979 			__set_bit(vcpu->vcpu_id, dest_map->map);
980 			dest_map->vectors[vcpu->vcpu_id] = vector;
981 		}
982 
983 		if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
984 			if (trig_mode)
985 				kvm_lapic_set_vector(vector, apic->regs + APIC_TMR);
986 			else
987 				apic_clear_vector(vector, apic->regs + APIC_TMR);
988 		}
989 
990 		if (vcpu->arch.apicv_active)
991 			kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
992 		else {
993 			kvm_lapic_set_irr(vector, apic);
994 
995 			kvm_make_request(KVM_REQ_EVENT, vcpu);
996 			kvm_vcpu_kick(vcpu);
997 		}
998 		break;
999 
1000 	case APIC_DM_REMRD:
1001 		result = 1;
1002 		vcpu->arch.pv.pv_unhalted = 1;
1003 		kvm_make_request(KVM_REQ_EVENT, vcpu);
1004 		kvm_vcpu_kick(vcpu);
1005 		break;
1006 
1007 	case APIC_DM_SMI:
1008 		result = 1;
1009 		kvm_make_request(KVM_REQ_SMI, vcpu);
1010 		kvm_vcpu_kick(vcpu);
1011 		break;
1012 
1013 	case APIC_DM_NMI:
1014 		result = 1;
1015 		kvm_inject_nmi(vcpu);
1016 		kvm_vcpu_kick(vcpu);
1017 		break;
1018 
1019 	case APIC_DM_INIT:
1020 		if (!trig_mode || level) {
1021 			result = 1;
1022 			/* assumes that there are only KVM_APIC_INIT/SIPI */
1023 			apic->pending_events = (1UL << KVM_APIC_INIT);
1024 			/* make sure pending_events is visible before sending
1025 			 * the request */
1026 			smp_wmb();
1027 			kvm_make_request(KVM_REQ_EVENT, vcpu);
1028 			kvm_vcpu_kick(vcpu);
1029 		} else {
1030 			apic_debug("Ignoring de-assert INIT to vcpu %d\n",
1031 				   vcpu->vcpu_id);
1032 		}
1033 		break;
1034 
1035 	case APIC_DM_STARTUP:
1036 		apic_debug("SIPI to vcpu %d vector 0x%02x\n",
1037 			   vcpu->vcpu_id, vector);
1038 		result = 1;
1039 		apic->sipi_vector = vector;
1040 		/* make sure sipi_vector is visible for the receiver */
1041 		smp_wmb();
1042 		set_bit(KVM_APIC_SIPI, &apic->pending_events);
1043 		kvm_make_request(KVM_REQ_EVENT, vcpu);
1044 		kvm_vcpu_kick(vcpu);
1045 		break;
1046 
1047 	case APIC_DM_EXTINT:
1048 		/*
1049 		 * Should only be called by kvm_apic_local_deliver() with LVT0,
1050 		 * before NMI watchdog was enabled. Already handled by
1051 		 * kvm_apic_accept_pic_intr().
1052 		 */
1053 		break;
1054 
1055 	default:
1056 		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1057 		       delivery_mode);
1058 		break;
1059 	}
1060 	return result;
1061 }
1062 
1063 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1064 {
1065 	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1066 }
1067 
1068 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1069 {
1070 	return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1071 }
1072 
1073 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1074 {
1075 	int trigger_mode;
1076 
1077 	/* Eoi the ioapic only if the ioapic doesn't own the vector. */
1078 	if (!kvm_ioapic_handles_vector(apic, vector))
1079 		return;
1080 
1081 	/* Request a KVM exit to inform the userspace IOAPIC. */
1082 	if (irqchip_split(apic->vcpu->kvm)) {
1083 		apic->vcpu->arch.pending_ioapic_eoi = vector;
1084 		kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1085 		return;
1086 	}
1087 
1088 	if (apic_test_vector(vector, apic->regs + APIC_TMR))
1089 		trigger_mode = IOAPIC_LEVEL_TRIG;
1090 	else
1091 		trigger_mode = IOAPIC_EDGE_TRIG;
1092 
1093 	kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1094 }
1095 
1096 static int apic_set_eoi(struct kvm_lapic *apic)
1097 {
1098 	int vector = apic_find_highest_isr(apic);
1099 
1100 	trace_kvm_eoi(apic, vector);
1101 
1102 	/*
1103 	 * Not every write EOI will has corresponding ISR,
1104 	 * one example is when Kernel check timer on setup_IO_APIC
1105 	 */
1106 	if (vector == -1)
1107 		return vector;
1108 
1109 	apic_clear_isr(vector, apic);
1110 	apic_update_ppr(apic);
1111 
1112 	if (test_bit(vector, vcpu_to_synic(apic->vcpu)->vec_bitmap))
1113 		kvm_hv_synic_send_eoi(apic->vcpu, vector);
1114 
1115 	kvm_ioapic_send_eoi(apic, vector);
1116 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1117 	return vector;
1118 }
1119 
1120 /*
1121  * this interface assumes a trap-like exit, which has already finished
1122  * desired side effect including vISR and vPPR update.
1123  */
1124 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1125 {
1126 	struct kvm_lapic *apic = vcpu->arch.apic;
1127 
1128 	trace_kvm_eoi(apic, vector);
1129 
1130 	kvm_ioapic_send_eoi(apic, vector);
1131 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1132 }
1133 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1134 
1135 static void apic_send_ipi(struct kvm_lapic *apic)
1136 {
1137 	u32 icr_low = kvm_lapic_get_reg(apic, APIC_ICR);
1138 	u32 icr_high = kvm_lapic_get_reg(apic, APIC_ICR2);
1139 	struct kvm_lapic_irq irq;
1140 
1141 	irq.vector = icr_low & APIC_VECTOR_MASK;
1142 	irq.delivery_mode = icr_low & APIC_MODE_MASK;
1143 	irq.dest_mode = icr_low & APIC_DEST_MASK;
1144 	irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1145 	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1146 	irq.shorthand = icr_low & APIC_SHORT_MASK;
1147 	irq.msi_redir_hint = false;
1148 	if (apic_x2apic_mode(apic))
1149 		irq.dest_id = icr_high;
1150 	else
1151 		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1152 
1153 	trace_kvm_apic_ipi(icr_low, irq.dest_id);
1154 
1155 	apic_debug("icr_high 0x%x, icr_low 0x%x, "
1156 		   "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
1157 		   "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x, "
1158 		   "msi_redir_hint 0x%x\n",
1159 		   icr_high, icr_low, irq.shorthand, irq.dest_id,
1160 		   irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
1161 		   irq.vector, irq.msi_redir_hint);
1162 
1163 	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1164 }
1165 
1166 static u32 apic_get_tmcct(struct kvm_lapic *apic)
1167 {
1168 	ktime_t remaining, now;
1169 	s64 ns;
1170 	u32 tmcct;
1171 
1172 	ASSERT(apic != NULL);
1173 
1174 	/* if initial count is 0, current count should also be 0 */
1175 	if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1176 		apic->lapic_timer.period == 0)
1177 		return 0;
1178 
1179 	now = ktime_get();
1180 	remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1181 	if (ktime_to_ns(remaining) < 0)
1182 		remaining = 0;
1183 
1184 	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1185 	tmcct = div64_u64(ns,
1186 			 (APIC_BUS_CYCLE_NS * apic->divide_count));
1187 
1188 	return tmcct;
1189 }
1190 
1191 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1192 {
1193 	struct kvm_vcpu *vcpu = apic->vcpu;
1194 	struct kvm_run *run = vcpu->run;
1195 
1196 	kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1197 	run->tpr_access.rip = kvm_rip_read(vcpu);
1198 	run->tpr_access.is_write = write;
1199 }
1200 
1201 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1202 {
1203 	if (apic->vcpu->arch.tpr_access_reporting)
1204 		__report_tpr_access(apic, write);
1205 }
1206 
1207 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1208 {
1209 	u32 val = 0;
1210 
1211 	if (offset >= LAPIC_MMIO_LENGTH)
1212 		return 0;
1213 
1214 	switch (offset) {
1215 	case APIC_ARBPRI:
1216 		apic_debug("Access APIC ARBPRI register which is for P6\n");
1217 		break;
1218 
1219 	case APIC_TMCCT:	/* Timer CCR */
1220 		if (apic_lvtt_tscdeadline(apic))
1221 			return 0;
1222 
1223 		val = apic_get_tmcct(apic);
1224 		break;
1225 	case APIC_PROCPRI:
1226 		apic_update_ppr(apic);
1227 		val = kvm_lapic_get_reg(apic, offset);
1228 		break;
1229 	case APIC_TASKPRI:
1230 		report_tpr_access(apic, false);
1231 		/* fall thru */
1232 	default:
1233 		val = kvm_lapic_get_reg(apic, offset);
1234 		break;
1235 	}
1236 
1237 	return val;
1238 }
1239 
1240 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1241 {
1242 	return container_of(dev, struct kvm_lapic, dev);
1243 }
1244 
1245 int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1246 		void *data)
1247 {
1248 	unsigned char alignment = offset & 0xf;
1249 	u32 result;
1250 	/* this bitmask has a bit cleared for each reserved register */
1251 	static const u64 rmask = 0x43ff01ffffffe70cULL;
1252 
1253 	if ((alignment + len) > 4) {
1254 		apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1255 			   offset, len);
1256 		return 1;
1257 	}
1258 
1259 	if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1260 		apic_debug("KVM_APIC_READ: read reserved register %x\n",
1261 			   offset);
1262 		return 1;
1263 	}
1264 
1265 	result = __apic_read(apic, offset & ~0xf);
1266 
1267 	trace_kvm_apic_read(offset, result);
1268 
1269 	switch (len) {
1270 	case 1:
1271 	case 2:
1272 	case 4:
1273 		memcpy(data, (char *)&result + alignment, len);
1274 		break;
1275 	default:
1276 		printk(KERN_ERR "Local APIC read with len = %x, "
1277 		       "should be 1,2, or 4 instead\n", len);
1278 		break;
1279 	}
1280 	return 0;
1281 }
1282 EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1283 
1284 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1285 {
1286 	return kvm_apic_hw_enabled(apic) &&
1287 	    addr >= apic->base_address &&
1288 	    addr < apic->base_address + LAPIC_MMIO_LENGTH;
1289 }
1290 
1291 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1292 			   gpa_t address, int len, void *data)
1293 {
1294 	struct kvm_lapic *apic = to_lapic(this);
1295 	u32 offset = address - apic->base_address;
1296 
1297 	if (!apic_mmio_in_range(apic, address))
1298 		return -EOPNOTSUPP;
1299 
1300 	kvm_lapic_reg_read(apic, offset, len, data);
1301 
1302 	return 0;
1303 }
1304 
1305 static void update_divide_count(struct kvm_lapic *apic)
1306 {
1307 	u32 tmp1, tmp2, tdcr;
1308 
1309 	tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1310 	tmp1 = tdcr & 0xf;
1311 	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1312 	apic->divide_count = 0x1 << (tmp2 & 0x7);
1313 
1314 	apic_debug("timer divide count is 0x%x\n",
1315 				   apic->divide_count);
1316 }
1317 
1318 static void limit_periodic_timer_frequency(struct kvm_lapic *apic)
1319 {
1320 	/*
1321 	 * Do not allow the guest to program periodic timers with small
1322 	 * interval, since the hrtimers are not throttled by the host
1323 	 * scheduler.
1324 	 */
1325 	if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1326 		s64 min_period = min_timer_period_us * 1000LL;
1327 
1328 		if (apic->lapic_timer.period < min_period) {
1329 			pr_info_ratelimited(
1330 			    "kvm: vcpu %i: requested %lld ns "
1331 			    "lapic timer period limited to %lld ns\n",
1332 			    apic->vcpu->vcpu_id,
1333 			    apic->lapic_timer.period, min_period);
1334 			apic->lapic_timer.period = min_period;
1335 		}
1336 	}
1337 }
1338 
1339 static void apic_update_lvtt(struct kvm_lapic *apic)
1340 {
1341 	u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1342 			apic->lapic_timer.timer_mode_mask;
1343 
1344 	if (apic->lapic_timer.timer_mode != timer_mode) {
1345 		if (apic_lvtt_tscdeadline(apic) != (timer_mode ==
1346 				APIC_LVT_TIMER_TSCDEADLINE)) {
1347 			hrtimer_cancel(&apic->lapic_timer.timer);
1348 			kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1349 			apic->lapic_timer.period = 0;
1350 			apic->lapic_timer.tscdeadline = 0;
1351 		}
1352 		apic->lapic_timer.timer_mode = timer_mode;
1353 		limit_periodic_timer_frequency(apic);
1354 	}
1355 }
1356 
1357 static void apic_timer_expired(struct kvm_lapic *apic)
1358 {
1359 	struct kvm_vcpu *vcpu = apic->vcpu;
1360 	struct swait_queue_head *q = &vcpu->wq;
1361 	struct kvm_timer *ktimer = &apic->lapic_timer;
1362 
1363 	if (atomic_read(&apic->lapic_timer.pending))
1364 		return;
1365 
1366 	atomic_inc(&apic->lapic_timer.pending);
1367 	kvm_set_pending_timer(vcpu);
1368 
1369 	/*
1370 	 * For x86, the atomic_inc() is serialized, thus
1371 	 * using swait_active() is safe.
1372 	 */
1373 	if (swait_active(q))
1374 		swake_up(q);
1375 
1376 	if (apic_lvtt_tscdeadline(apic))
1377 		ktimer->expired_tscdeadline = ktimer->tscdeadline;
1378 }
1379 
1380 /*
1381  * On APICv, this test will cause a busy wait
1382  * during a higher-priority task.
1383  */
1384 
1385 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1386 {
1387 	struct kvm_lapic *apic = vcpu->arch.apic;
1388 	u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1389 
1390 	if (kvm_apic_hw_enabled(apic)) {
1391 		int vec = reg & APIC_VECTOR_MASK;
1392 		void *bitmap = apic->regs + APIC_ISR;
1393 
1394 		if (vcpu->arch.apicv_active)
1395 			bitmap = apic->regs + APIC_IRR;
1396 
1397 		if (apic_test_vector(vec, bitmap))
1398 			return true;
1399 	}
1400 	return false;
1401 }
1402 
1403 void wait_lapic_expire(struct kvm_vcpu *vcpu)
1404 {
1405 	struct kvm_lapic *apic = vcpu->arch.apic;
1406 	u64 guest_tsc, tsc_deadline;
1407 
1408 	if (!lapic_in_kernel(vcpu))
1409 		return;
1410 
1411 	if (apic->lapic_timer.expired_tscdeadline == 0)
1412 		return;
1413 
1414 	if (!lapic_timer_int_injected(vcpu))
1415 		return;
1416 
1417 	tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1418 	apic->lapic_timer.expired_tscdeadline = 0;
1419 	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1420 	trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1421 
1422 	/* __delay is delay_tsc whenever the hardware has TSC, thus always.  */
1423 	if (guest_tsc < tsc_deadline)
1424 		__delay(min(tsc_deadline - guest_tsc,
1425 			nsec_to_cycles(vcpu, lapic_timer_advance_ns)));
1426 }
1427 
1428 static void start_sw_tscdeadline(struct kvm_lapic *apic)
1429 {
1430 	u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1431 	u64 ns = 0;
1432 	ktime_t expire;
1433 	struct kvm_vcpu *vcpu = apic->vcpu;
1434 	unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1435 	unsigned long flags;
1436 	ktime_t now;
1437 
1438 	if (unlikely(!tscdeadline || !this_tsc_khz))
1439 		return;
1440 
1441 	local_irq_save(flags);
1442 
1443 	now = ktime_get();
1444 	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1445 	if (likely(tscdeadline > guest_tsc)) {
1446 		ns = (tscdeadline - guest_tsc) * 1000000ULL;
1447 		do_div(ns, this_tsc_khz);
1448 		expire = ktime_add_ns(now, ns);
1449 		expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1450 		hrtimer_start(&apic->lapic_timer.timer,
1451 				expire, HRTIMER_MODE_ABS_PINNED);
1452 	} else
1453 		apic_timer_expired(apic);
1454 
1455 	local_irq_restore(flags);
1456 }
1457 
1458 static void start_sw_period(struct kvm_lapic *apic)
1459 {
1460 	if (!apic->lapic_timer.period)
1461 		return;
1462 
1463 	if (apic_lvtt_oneshot(apic) &&
1464 	    ktime_after(ktime_get(),
1465 			apic->lapic_timer.target_expiration)) {
1466 		apic_timer_expired(apic);
1467 		return;
1468 	}
1469 
1470 	hrtimer_start(&apic->lapic_timer.timer,
1471 		apic->lapic_timer.target_expiration,
1472 		HRTIMER_MODE_ABS_PINNED);
1473 }
1474 
1475 static void update_target_expiration(struct kvm_lapic *apic, uint32_t old_divisor)
1476 {
1477 	ktime_t now, remaining;
1478 	u64 ns_remaining_old, ns_remaining_new;
1479 
1480 	apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1481 		* APIC_BUS_CYCLE_NS * apic->divide_count;
1482 	limit_periodic_timer_frequency(apic);
1483 
1484 	now = ktime_get();
1485 	remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1486 	if (ktime_to_ns(remaining) < 0)
1487 		remaining = 0;
1488 
1489 	ns_remaining_old = ktime_to_ns(remaining);
1490 	ns_remaining_new = mul_u64_u32_div(ns_remaining_old,
1491 	                                   apic->divide_count, old_divisor);
1492 
1493 	apic->lapic_timer.tscdeadline +=
1494 		nsec_to_cycles(apic->vcpu, ns_remaining_new) -
1495 		nsec_to_cycles(apic->vcpu, ns_remaining_old);
1496 	apic->lapic_timer.target_expiration = ktime_add_ns(now, ns_remaining_new);
1497 }
1498 
1499 static bool set_target_expiration(struct kvm_lapic *apic)
1500 {
1501 	ktime_t now;
1502 	u64 tscl = rdtsc();
1503 
1504 	now = ktime_get();
1505 	apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1506 		* APIC_BUS_CYCLE_NS * apic->divide_count;
1507 
1508 	if (!apic->lapic_timer.period) {
1509 		apic->lapic_timer.tscdeadline = 0;
1510 		return false;
1511 	}
1512 
1513 	limit_periodic_timer_frequency(apic);
1514 
1515 	apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1516 		   PRIx64 ", "
1517 		   "timer initial count 0x%x, period %lldns, "
1518 		   "expire @ 0x%016" PRIx64 ".\n", __func__,
1519 		   APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1520 		   kvm_lapic_get_reg(apic, APIC_TMICT),
1521 		   apic->lapic_timer.period,
1522 		   ktime_to_ns(ktime_add_ns(now,
1523 				apic->lapic_timer.period)));
1524 
1525 	apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1526 		nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1527 	apic->lapic_timer.target_expiration = ktime_add_ns(now, apic->lapic_timer.period);
1528 
1529 	return true;
1530 }
1531 
1532 static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1533 {
1534 	apic->lapic_timer.tscdeadline +=
1535 		nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1536 	apic->lapic_timer.target_expiration =
1537 		ktime_add_ns(apic->lapic_timer.target_expiration,
1538 				apic->lapic_timer.period);
1539 }
1540 
1541 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1542 {
1543 	if (!lapic_in_kernel(vcpu))
1544 		return false;
1545 
1546 	return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1547 }
1548 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1549 
1550 static void cancel_hv_timer(struct kvm_lapic *apic)
1551 {
1552 	WARN_ON(preemptible());
1553 	WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1554 	kvm_x86_ops->cancel_hv_timer(apic->vcpu);
1555 	apic->lapic_timer.hv_timer_in_use = false;
1556 }
1557 
1558 static bool start_hv_timer(struct kvm_lapic *apic)
1559 {
1560 	struct kvm_timer *ktimer = &apic->lapic_timer;
1561 	int r;
1562 
1563 	WARN_ON(preemptible());
1564 	if (!kvm_x86_ops->set_hv_timer)
1565 		return false;
1566 
1567 	if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1568 		return false;
1569 
1570 	if (!ktimer->tscdeadline)
1571 		return false;
1572 
1573 	r = kvm_x86_ops->set_hv_timer(apic->vcpu, ktimer->tscdeadline);
1574 	if (r < 0)
1575 		return false;
1576 
1577 	ktimer->hv_timer_in_use = true;
1578 	hrtimer_cancel(&ktimer->timer);
1579 
1580 	/*
1581 	 * Also recheck ktimer->pending, in case the sw timer triggered in
1582 	 * the window.  For periodic timer, leave the hv timer running for
1583 	 * simplicity, and the deadline will be recomputed on the next vmexit.
1584 	 */
1585 	if (!apic_lvtt_period(apic) && (r || atomic_read(&ktimer->pending))) {
1586 		if (r)
1587 			apic_timer_expired(apic);
1588 		return false;
1589 	}
1590 
1591 	trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, true);
1592 	return true;
1593 }
1594 
1595 static void start_sw_timer(struct kvm_lapic *apic)
1596 {
1597 	struct kvm_timer *ktimer = &apic->lapic_timer;
1598 
1599 	WARN_ON(preemptible());
1600 	if (apic->lapic_timer.hv_timer_in_use)
1601 		cancel_hv_timer(apic);
1602 	if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1603 		return;
1604 
1605 	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1606 		start_sw_period(apic);
1607 	else if (apic_lvtt_tscdeadline(apic))
1608 		start_sw_tscdeadline(apic);
1609 	trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, false);
1610 }
1611 
1612 static void restart_apic_timer(struct kvm_lapic *apic)
1613 {
1614 	preempt_disable();
1615 	if (!start_hv_timer(apic))
1616 		start_sw_timer(apic);
1617 	preempt_enable();
1618 }
1619 
1620 void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1621 {
1622 	struct kvm_lapic *apic = vcpu->arch.apic;
1623 
1624 	preempt_disable();
1625 	/* If the preempt notifier has already run, it also called apic_timer_expired */
1626 	if (!apic->lapic_timer.hv_timer_in_use)
1627 		goto out;
1628 	WARN_ON(swait_active(&vcpu->wq));
1629 	cancel_hv_timer(apic);
1630 	apic_timer_expired(apic);
1631 
1632 	if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1633 		advance_periodic_target_expiration(apic);
1634 		restart_apic_timer(apic);
1635 	}
1636 out:
1637 	preempt_enable();
1638 }
1639 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1640 
1641 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1642 {
1643 	restart_apic_timer(vcpu->arch.apic);
1644 }
1645 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1646 
1647 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1648 {
1649 	struct kvm_lapic *apic = vcpu->arch.apic;
1650 
1651 	preempt_disable();
1652 	/* Possibly the TSC deadline timer is not enabled yet */
1653 	if (apic->lapic_timer.hv_timer_in_use)
1654 		start_sw_timer(apic);
1655 	preempt_enable();
1656 }
1657 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1658 
1659 void kvm_lapic_restart_hv_timer(struct kvm_vcpu *vcpu)
1660 {
1661 	struct kvm_lapic *apic = vcpu->arch.apic;
1662 
1663 	WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1664 	restart_apic_timer(apic);
1665 }
1666 
1667 static void start_apic_timer(struct kvm_lapic *apic)
1668 {
1669 	atomic_set(&apic->lapic_timer.pending, 0);
1670 
1671 	if ((apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1672 	    && !set_target_expiration(apic))
1673 		return;
1674 
1675 	restart_apic_timer(apic);
1676 }
1677 
1678 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1679 {
1680 	bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1681 
1682 	if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1683 		apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1684 		if (lvt0_in_nmi_mode) {
1685 			apic_debug("Receive NMI setting on APIC_LVT0 "
1686 				   "for cpu %d\n", apic->vcpu->vcpu_id);
1687 			atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1688 		} else
1689 			atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1690 	}
1691 }
1692 
1693 int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1694 {
1695 	int ret = 0;
1696 
1697 	trace_kvm_apic_write(reg, val);
1698 
1699 	switch (reg) {
1700 	case APIC_ID:		/* Local APIC ID */
1701 		if (!apic_x2apic_mode(apic))
1702 			kvm_apic_set_xapic_id(apic, val >> 24);
1703 		else
1704 			ret = 1;
1705 		break;
1706 
1707 	case APIC_TASKPRI:
1708 		report_tpr_access(apic, true);
1709 		apic_set_tpr(apic, val & 0xff);
1710 		break;
1711 
1712 	case APIC_EOI:
1713 		apic_set_eoi(apic);
1714 		break;
1715 
1716 	case APIC_LDR:
1717 		if (!apic_x2apic_mode(apic))
1718 			kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1719 		else
1720 			ret = 1;
1721 		break;
1722 
1723 	case APIC_DFR:
1724 		if (!apic_x2apic_mode(apic)) {
1725 			kvm_lapic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1726 			recalculate_apic_map(apic->vcpu->kvm);
1727 		} else
1728 			ret = 1;
1729 		break;
1730 
1731 	case APIC_SPIV: {
1732 		u32 mask = 0x3ff;
1733 		if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1734 			mask |= APIC_SPIV_DIRECTED_EOI;
1735 		apic_set_spiv(apic, val & mask);
1736 		if (!(val & APIC_SPIV_APIC_ENABLED)) {
1737 			int i;
1738 			u32 lvt_val;
1739 
1740 			for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
1741 				lvt_val = kvm_lapic_get_reg(apic,
1742 						       APIC_LVTT + 0x10 * i);
1743 				kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
1744 					     lvt_val | APIC_LVT_MASKED);
1745 			}
1746 			apic_update_lvtt(apic);
1747 			atomic_set(&apic->lapic_timer.pending, 0);
1748 
1749 		}
1750 		break;
1751 	}
1752 	case APIC_ICR:
1753 		/* No delay here, so we always clear the pending bit */
1754 		kvm_lapic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1755 		apic_send_ipi(apic);
1756 		break;
1757 
1758 	case APIC_ICR2:
1759 		if (!apic_x2apic_mode(apic))
1760 			val &= 0xff000000;
1761 		kvm_lapic_set_reg(apic, APIC_ICR2, val);
1762 		break;
1763 
1764 	case APIC_LVT0:
1765 		apic_manage_nmi_watchdog(apic, val);
1766 	case APIC_LVTTHMR:
1767 	case APIC_LVTPC:
1768 	case APIC_LVT1:
1769 	case APIC_LVTERR:
1770 		/* TODO: Check vector */
1771 		if (!kvm_apic_sw_enabled(apic))
1772 			val |= APIC_LVT_MASKED;
1773 
1774 		val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1775 		kvm_lapic_set_reg(apic, reg, val);
1776 
1777 		break;
1778 
1779 	case APIC_LVTT:
1780 		if (!kvm_apic_sw_enabled(apic))
1781 			val |= APIC_LVT_MASKED;
1782 		val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1783 		kvm_lapic_set_reg(apic, APIC_LVTT, val);
1784 		apic_update_lvtt(apic);
1785 		break;
1786 
1787 	case APIC_TMICT:
1788 		if (apic_lvtt_tscdeadline(apic))
1789 			break;
1790 
1791 		hrtimer_cancel(&apic->lapic_timer.timer);
1792 		kvm_lapic_set_reg(apic, APIC_TMICT, val);
1793 		start_apic_timer(apic);
1794 		break;
1795 
1796 	case APIC_TDCR: {
1797 		uint32_t old_divisor = apic->divide_count;
1798 
1799 		if (val & 4)
1800 			apic_debug("KVM_WRITE:TDCR %x\n", val);
1801 		kvm_lapic_set_reg(apic, APIC_TDCR, val);
1802 		update_divide_count(apic);
1803 		if (apic->divide_count != old_divisor &&
1804 				apic->lapic_timer.period) {
1805 			hrtimer_cancel(&apic->lapic_timer.timer);
1806 			update_target_expiration(apic, old_divisor);
1807 			restart_apic_timer(apic);
1808 		}
1809 		break;
1810 	}
1811 	case APIC_ESR:
1812 		if (apic_x2apic_mode(apic) && val != 0) {
1813 			apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1814 			ret = 1;
1815 		}
1816 		break;
1817 
1818 	case APIC_SELF_IPI:
1819 		if (apic_x2apic_mode(apic)) {
1820 			kvm_lapic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1821 		} else
1822 			ret = 1;
1823 		break;
1824 	default:
1825 		ret = 1;
1826 		break;
1827 	}
1828 	if (ret)
1829 		apic_debug("Local APIC Write to read-only register %x\n", reg);
1830 	return ret;
1831 }
1832 EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
1833 
1834 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1835 			    gpa_t address, int len, const void *data)
1836 {
1837 	struct kvm_lapic *apic = to_lapic(this);
1838 	unsigned int offset = address - apic->base_address;
1839 	u32 val;
1840 
1841 	if (!apic_mmio_in_range(apic, address))
1842 		return -EOPNOTSUPP;
1843 
1844 	/*
1845 	 * APIC register must be aligned on 128-bits boundary.
1846 	 * 32/64/128 bits registers must be accessed thru 32 bits.
1847 	 * Refer SDM 8.4.1
1848 	 */
1849 	if (len != 4 || (offset & 0xf)) {
1850 		/* Don't shout loud, $infamous_os would cause only noise. */
1851 		apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1852 		return 0;
1853 	}
1854 
1855 	val = *(u32*)data;
1856 
1857 	/* too common printing */
1858 	if (offset != APIC_EOI)
1859 		apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1860 			   "0x%x\n", __func__, offset, len, val);
1861 
1862 	kvm_lapic_reg_write(apic, offset & 0xff0, val);
1863 
1864 	return 0;
1865 }
1866 
1867 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1868 {
1869 	kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1870 }
1871 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1872 
1873 /* emulate APIC access in a trap manner */
1874 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1875 {
1876 	u32 val = 0;
1877 
1878 	/* hw has done the conditional check and inst decode */
1879 	offset &= 0xff0;
1880 
1881 	kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
1882 
1883 	/* TODO: optimize to just emulate side effect w/o one more write */
1884 	kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
1885 }
1886 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1887 
1888 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1889 {
1890 	struct kvm_lapic *apic = vcpu->arch.apic;
1891 
1892 	if (!vcpu->arch.apic)
1893 		return;
1894 
1895 	hrtimer_cancel(&apic->lapic_timer.timer);
1896 
1897 	if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1898 		static_key_slow_dec_deferred(&apic_hw_disabled);
1899 
1900 	if (!apic->sw_enabled)
1901 		static_key_slow_dec_deferred(&apic_sw_disabled);
1902 
1903 	if (apic->regs)
1904 		free_page((unsigned long)apic->regs);
1905 
1906 	kfree(apic);
1907 }
1908 
1909 /*
1910  *----------------------------------------------------------------------
1911  * LAPIC interface
1912  *----------------------------------------------------------------------
1913  */
1914 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1915 {
1916 	struct kvm_lapic *apic = vcpu->arch.apic;
1917 
1918 	if (!lapic_in_kernel(vcpu) ||
1919 		!apic_lvtt_tscdeadline(apic))
1920 		return 0;
1921 
1922 	return apic->lapic_timer.tscdeadline;
1923 }
1924 
1925 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1926 {
1927 	struct kvm_lapic *apic = vcpu->arch.apic;
1928 
1929 	if (!lapic_in_kernel(vcpu) || apic_lvtt_oneshot(apic) ||
1930 			apic_lvtt_period(apic))
1931 		return;
1932 
1933 	hrtimer_cancel(&apic->lapic_timer.timer);
1934 	apic->lapic_timer.tscdeadline = data;
1935 	start_apic_timer(apic);
1936 }
1937 
1938 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1939 {
1940 	struct kvm_lapic *apic = vcpu->arch.apic;
1941 
1942 	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1943 		     | (kvm_lapic_get_reg(apic, APIC_TASKPRI) & 4));
1944 }
1945 
1946 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1947 {
1948 	u64 tpr;
1949 
1950 	tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1951 
1952 	return (tpr & 0xf0) >> 4;
1953 }
1954 
1955 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1956 {
1957 	u64 old_value = vcpu->arch.apic_base;
1958 	struct kvm_lapic *apic = vcpu->arch.apic;
1959 
1960 	if (!apic)
1961 		value |= MSR_IA32_APICBASE_BSP;
1962 
1963 	vcpu->arch.apic_base = value;
1964 
1965 	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
1966 		kvm_update_cpuid(vcpu);
1967 
1968 	if (!apic)
1969 		return;
1970 
1971 	/* update jump label if enable bit changes */
1972 	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1973 		if (value & MSR_IA32_APICBASE_ENABLE) {
1974 			kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
1975 			static_key_slow_dec_deferred(&apic_hw_disabled);
1976 		} else {
1977 			static_key_slow_inc(&apic_hw_disabled.key);
1978 			recalculate_apic_map(vcpu->kvm);
1979 		}
1980 	}
1981 
1982 	if ((old_value ^ value) & X2APIC_ENABLE) {
1983 		if (value & X2APIC_ENABLE) {
1984 			kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
1985 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1986 		} else
1987 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1988 	}
1989 
1990 	apic->base_address = apic->vcpu->arch.apic_base &
1991 			     MSR_IA32_APICBASE_BASE;
1992 
1993 	if ((value & MSR_IA32_APICBASE_ENABLE) &&
1994 	     apic->base_address != APIC_DEFAULT_PHYS_BASE)
1995 		pr_warn_once("APIC base relocation is unsupported by KVM");
1996 
1997 	/* with FSB delivery interrupt, we can restart APIC functionality */
1998 	apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1999 		   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
2000 
2001 }
2002 
2003 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
2004 {
2005 	struct kvm_lapic *apic = vcpu->arch.apic;
2006 	int i;
2007 
2008 	if (!apic)
2009 		return;
2010 
2011 	apic_debug("%s\n", __func__);
2012 
2013 	/* Stop the timer in case it's a reset to an active apic */
2014 	hrtimer_cancel(&apic->lapic_timer.timer);
2015 
2016 	if (!init_event) {
2017 		kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
2018 		                         MSR_IA32_APICBASE_ENABLE);
2019 		kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2020 	}
2021 	kvm_apic_set_version(apic->vcpu);
2022 
2023 	for (i = 0; i < KVM_APIC_LVT_NUM; i++)
2024 		kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
2025 	apic_update_lvtt(apic);
2026 	if (kvm_vcpu_is_reset_bsp(vcpu) &&
2027 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
2028 		kvm_lapic_set_reg(apic, APIC_LVT0,
2029 			     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
2030 	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2031 
2032 	kvm_lapic_set_reg(apic, APIC_DFR, 0xffffffffU);
2033 	apic_set_spiv(apic, 0xff);
2034 	kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
2035 	if (!apic_x2apic_mode(apic))
2036 		kvm_apic_set_ldr(apic, 0);
2037 	kvm_lapic_set_reg(apic, APIC_ESR, 0);
2038 	kvm_lapic_set_reg(apic, APIC_ICR, 0);
2039 	kvm_lapic_set_reg(apic, APIC_ICR2, 0);
2040 	kvm_lapic_set_reg(apic, APIC_TDCR, 0);
2041 	kvm_lapic_set_reg(apic, APIC_TMICT, 0);
2042 	for (i = 0; i < 8; i++) {
2043 		kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
2044 		kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
2045 		kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
2046 	}
2047 	apic->irr_pending = vcpu->arch.apicv_active;
2048 	apic->isr_count = vcpu->arch.apicv_active ? 1 : 0;
2049 	apic->highest_isr_cache = -1;
2050 	update_divide_count(apic);
2051 	atomic_set(&apic->lapic_timer.pending, 0);
2052 	if (kvm_vcpu_is_bsp(vcpu))
2053 		kvm_lapic_set_base(vcpu,
2054 				vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
2055 	vcpu->arch.pv_eoi.msr_val = 0;
2056 	apic_update_ppr(apic);
2057 	if (vcpu->arch.apicv_active) {
2058 		kvm_x86_ops->apicv_post_state_restore(vcpu);
2059 		kvm_x86_ops->hwapic_irr_update(vcpu, -1);
2060 		kvm_x86_ops->hwapic_isr_update(vcpu, -1);
2061 	}
2062 
2063 	vcpu->arch.apic_arb_prio = 0;
2064 	vcpu->arch.apic_attention = 0;
2065 
2066 	apic_debug("%s: vcpu=%p, id=0x%x, base_msr="
2067 		   "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
2068 		   vcpu, kvm_lapic_get_reg(apic, APIC_ID),
2069 		   vcpu->arch.apic_base, apic->base_address);
2070 }
2071 
2072 /*
2073  *----------------------------------------------------------------------
2074  * timer interface
2075  *----------------------------------------------------------------------
2076  */
2077 
2078 static bool lapic_is_periodic(struct kvm_lapic *apic)
2079 {
2080 	return apic_lvtt_period(apic);
2081 }
2082 
2083 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
2084 {
2085 	struct kvm_lapic *apic = vcpu->arch.apic;
2086 
2087 	if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
2088 		return atomic_read(&apic->lapic_timer.pending);
2089 
2090 	return 0;
2091 }
2092 
2093 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
2094 {
2095 	u32 reg = kvm_lapic_get_reg(apic, lvt_type);
2096 	int vector, mode, trig_mode;
2097 
2098 	if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2099 		vector = reg & APIC_VECTOR_MASK;
2100 		mode = reg & APIC_MODE_MASK;
2101 		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2102 		return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2103 					NULL);
2104 	}
2105 	return 0;
2106 }
2107 
2108 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2109 {
2110 	struct kvm_lapic *apic = vcpu->arch.apic;
2111 
2112 	if (apic)
2113 		kvm_apic_local_deliver(apic, APIC_LVT0);
2114 }
2115 
2116 static const struct kvm_io_device_ops apic_mmio_ops = {
2117 	.read     = apic_mmio_read,
2118 	.write    = apic_mmio_write,
2119 };
2120 
2121 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2122 {
2123 	struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2124 	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2125 
2126 	apic_timer_expired(apic);
2127 
2128 	if (lapic_is_periodic(apic)) {
2129 		advance_periodic_target_expiration(apic);
2130 		hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2131 		return HRTIMER_RESTART;
2132 	} else
2133 		return HRTIMER_NORESTART;
2134 }
2135 
2136 int kvm_create_lapic(struct kvm_vcpu *vcpu)
2137 {
2138 	struct kvm_lapic *apic;
2139 
2140 	ASSERT(vcpu != NULL);
2141 	apic_debug("apic_init %d\n", vcpu->vcpu_id);
2142 
2143 	apic = kzalloc(sizeof(*apic), GFP_KERNEL);
2144 	if (!apic)
2145 		goto nomem;
2146 
2147 	vcpu->arch.apic = apic;
2148 
2149 	apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
2150 	if (!apic->regs) {
2151 		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2152 		       vcpu->vcpu_id);
2153 		goto nomem_free_apic;
2154 	}
2155 	apic->vcpu = vcpu;
2156 
2157 	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2158 		     HRTIMER_MODE_ABS_PINNED);
2159 	apic->lapic_timer.timer.function = apic_timer_fn;
2160 
2161 	/*
2162 	 * APIC is created enabled. This will prevent kvm_lapic_set_base from
2163 	 * thinking that APIC satet has changed.
2164 	 */
2165 	vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2166 	static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2167 	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2168 
2169 	return 0;
2170 nomem_free_apic:
2171 	kfree(apic);
2172 nomem:
2173 	return -ENOMEM;
2174 }
2175 
2176 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2177 {
2178 	struct kvm_lapic *apic = vcpu->arch.apic;
2179 	u32 ppr;
2180 
2181 	if (!apic_enabled(apic))
2182 		return -1;
2183 
2184 	__apic_update_ppr(apic, &ppr);
2185 	return apic_has_interrupt_for_ppr(apic, ppr);
2186 }
2187 
2188 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2189 {
2190 	u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2191 	int r = 0;
2192 
2193 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2194 		r = 1;
2195 	if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2196 	    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2197 		r = 1;
2198 	return r;
2199 }
2200 
2201 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2202 {
2203 	struct kvm_lapic *apic = vcpu->arch.apic;
2204 
2205 	if (atomic_read(&apic->lapic_timer.pending) > 0) {
2206 		kvm_apic_local_deliver(apic, APIC_LVTT);
2207 		if (apic_lvtt_tscdeadline(apic))
2208 			apic->lapic_timer.tscdeadline = 0;
2209 		if (apic_lvtt_oneshot(apic)) {
2210 			apic->lapic_timer.tscdeadline = 0;
2211 			apic->lapic_timer.target_expiration = 0;
2212 		}
2213 		atomic_set(&apic->lapic_timer.pending, 0);
2214 	}
2215 }
2216 
2217 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2218 {
2219 	int vector = kvm_apic_has_interrupt(vcpu);
2220 	struct kvm_lapic *apic = vcpu->arch.apic;
2221 	u32 ppr;
2222 
2223 	if (vector == -1)
2224 		return -1;
2225 
2226 	/*
2227 	 * We get here even with APIC virtualization enabled, if doing
2228 	 * nested virtualization and L1 runs with the "acknowledge interrupt
2229 	 * on exit" mode.  Then we cannot inject the interrupt via RVI,
2230 	 * because the process would deliver it through the IDT.
2231 	 */
2232 
2233 	apic_clear_irr(vector, apic);
2234 	if (test_bit(vector, vcpu_to_synic(vcpu)->auto_eoi_bitmap)) {
2235 		/*
2236 		 * For auto-EOI interrupts, there might be another pending
2237 		 * interrupt above PPR, so check whether to raise another
2238 		 * KVM_REQ_EVENT.
2239 		 */
2240 		apic_update_ppr(apic);
2241 	} else {
2242 		/*
2243 		 * For normal interrupts, PPR has been raised and there cannot
2244 		 * be a higher-priority pending interrupt---except if there was
2245 		 * a concurrent interrupt injection, but that would have
2246 		 * triggered KVM_REQ_EVENT already.
2247 		 */
2248 		apic_set_isr(vector, apic);
2249 		__apic_update_ppr(apic, &ppr);
2250 	}
2251 
2252 	return vector;
2253 }
2254 
2255 static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2256 		struct kvm_lapic_state *s, bool set)
2257 {
2258 	if (apic_x2apic_mode(vcpu->arch.apic)) {
2259 		u32 *id = (u32 *)(s->regs + APIC_ID);
2260 		u32 *ldr = (u32 *)(s->regs + APIC_LDR);
2261 
2262 		if (vcpu->kvm->arch.x2apic_format) {
2263 			if (*id != vcpu->vcpu_id)
2264 				return -EINVAL;
2265 		} else {
2266 			if (set)
2267 				*id >>= 24;
2268 			else
2269 				*id <<= 24;
2270 		}
2271 
2272 		/* In x2APIC mode, the LDR is fixed and based on the id */
2273 		if (set)
2274 			*ldr = kvm_apic_calc_x2apic_ldr(*id);
2275 	}
2276 
2277 	return 0;
2278 }
2279 
2280 int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2281 {
2282 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2283 	return kvm_apic_state_fixup(vcpu, s, false);
2284 }
2285 
2286 int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2287 {
2288 	struct kvm_lapic *apic = vcpu->arch.apic;
2289 	int r;
2290 
2291 
2292 	kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2293 	/* set SPIV separately to get count of SW disabled APICs right */
2294 	apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2295 
2296 	r = kvm_apic_state_fixup(vcpu, s, true);
2297 	if (r)
2298 		return r;
2299 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2300 
2301 	recalculate_apic_map(vcpu->kvm);
2302 	kvm_apic_set_version(vcpu);
2303 
2304 	apic_update_ppr(apic);
2305 	hrtimer_cancel(&apic->lapic_timer.timer);
2306 	apic_update_lvtt(apic);
2307 	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2308 	update_divide_count(apic);
2309 	start_apic_timer(apic);
2310 	apic->irr_pending = true;
2311 	apic->isr_count = vcpu->arch.apicv_active ?
2312 				1 : count_vectors(apic->regs + APIC_ISR);
2313 	apic->highest_isr_cache = -1;
2314 	if (vcpu->arch.apicv_active) {
2315 		kvm_x86_ops->apicv_post_state_restore(vcpu);
2316 		kvm_x86_ops->hwapic_irr_update(vcpu,
2317 				apic_find_highest_irr(apic));
2318 		kvm_x86_ops->hwapic_isr_update(vcpu,
2319 				apic_find_highest_isr(apic));
2320 	}
2321 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2322 	if (ioapic_in_kernel(vcpu->kvm))
2323 		kvm_rtc_eoi_tracking_restore_one(vcpu);
2324 
2325 	vcpu->arch.apic_arb_prio = 0;
2326 
2327 	return 0;
2328 }
2329 
2330 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2331 {
2332 	struct hrtimer *timer;
2333 
2334 	if (!lapic_in_kernel(vcpu))
2335 		return;
2336 
2337 	timer = &vcpu->arch.apic->lapic_timer.timer;
2338 	if (hrtimer_cancel(timer))
2339 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
2340 }
2341 
2342 /*
2343  * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2344  *
2345  * Detect whether guest triggered PV EOI since the
2346  * last entry. If yes, set EOI on guests's behalf.
2347  * Clear PV EOI in guest memory in any case.
2348  */
2349 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2350 					struct kvm_lapic *apic)
2351 {
2352 	bool pending;
2353 	int vector;
2354 	/*
2355 	 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2356 	 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2357 	 *
2358 	 * KVM_APIC_PV_EOI_PENDING is unset:
2359 	 * 	-> host disabled PV EOI.
2360 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2361 	 * 	-> host enabled PV EOI, guest did not execute EOI yet.
2362 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2363 	 * 	-> host enabled PV EOI, guest executed EOI.
2364 	 */
2365 	BUG_ON(!pv_eoi_enabled(vcpu));
2366 	pending = pv_eoi_get_pending(vcpu);
2367 	/*
2368 	 * Clear pending bit in any case: it will be set again on vmentry.
2369 	 * While this might not be ideal from performance point of view,
2370 	 * this makes sure pv eoi is only enabled when we know it's safe.
2371 	 */
2372 	pv_eoi_clr_pending(vcpu);
2373 	if (pending)
2374 		return;
2375 	vector = apic_set_eoi(apic);
2376 	trace_kvm_pv_eoi(apic, vector);
2377 }
2378 
2379 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2380 {
2381 	u32 data;
2382 
2383 	if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2384 		apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2385 
2386 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2387 		return;
2388 
2389 	if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2390 				  sizeof(u32)))
2391 		return;
2392 
2393 	apic_set_tpr(vcpu->arch.apic, data & 0xff);
2394 }
2395 
2396 /*
2397  * apic_sync_pv_eoi_to_guest - called before vmentry
2398  *
2399  * Detect whether it's safe to enable PV EOI and
2400  * if yes do so.
2401  */
2402 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2403 					struct kvm_lapic *apic)
2404 {
2405 	if (!pv_eoi_enabled(vcpu) ||
2406 	    /* IRR set or many bits in ISR: could be nested. */
2407 	    apic->irr_pending ||
2408 	    /* Cache not set: could be safe but we don't bother. */
2409 	    apic->highest_isr_cache == -1 ||
2410 	    /* Need EOI to update ioapic. */
2411 	    kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2412 		/*
2413 		 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2414 		 * so we need not do anything here.
2415 		 */
2416 		return;
2417 	}
2418 
2419 	pv_eoi_set_pending(apic->vcpu);
2420 }
2421 
2422 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2423 {
2424 	u32 data, tpr;
2425 	int max_irr, max_isr;
2426 	struct kvm_lapic *apic = vcpu->arch.apic;
2427 
2428 	apic_sync_pv_eoi_to_guest(vcpu, apic);
2429 
2430 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2431 		return;
2432 
2433 	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2434 	max_irr = apic_find_highest_irr(apic);
2435 	if (max_irr < 0)
2436 		max_irr = 0;
2437 	max_isr = apic_find_highest_isr(apic);
2438 	if (max_isr < 0)
2439 		max_isr = 0;
2440 	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2441 
2442 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2443 				sizeof(u32));
2444 }
2445 
2446 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2447 {
2448 	if (vapic_addr) {
2449 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2450 					&vcpu->arch.apic->vapic_cache,
2451 					vapic_addr, sizeof(u32)))
2452 			return -EINVAL;
2453 		__set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2454 	} else {
2455 		__clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2456 	}
2457 
2458 	vcpu->arch.apic->vapic_addr = vapic_addr;
2459 	return 0;
2460 }
2461 
2462 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2463 {
2464 	struct kvm_lapic *apic = vcpu->arch.apic;
2465 	u32 reg = (msr - APIC_BASE_MSR) << 4;
2466 
2467 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2468 		return 1;
2469 
2470 	if (reg == APIC_ICR2)
2471 		return 1;
2472 
2473 	/* if this is ICR write vector before command */
2474 	if (reg == APIC_ICR)
2475 		kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2476 	return kvm_lapic_reg_write(apic, reg, (u32)data);
2477 }
2478 
2479 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2480 {
2481 	struct kvm_lapic *apic = vcpu->arch.apic;
2482 	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2483 
2484 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2485 		return 1;
2486 
2487 	if (reg == APIC_DFR || reg == APIC_ICR2) {
2488 		apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
2489 			   reg);
2490 		return 1;
2491 	}
2492 
2493 	if (kvm_lapic_reg_read(apic, reg, 4, &low))
2494 		return 1;
2495 	if (reg == APIC_ICR)
2496 		kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2497 
2498 	*data = (((u64)high) << 32) | low;
2499 
2500 	return 0;
2501 }
2502 
2503 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2504 {
2505 	struct kvm_lapic *apic = vcpu->arch.apic;
2506 
2507 	if (!lapic_in_kernel(vcpu))
2508 		return 1;
2509 
2510 	/* if this is ICR write vector before command */
2511 	if (reg == APIC_ICR)
2512 		kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2513 	return kvm_lapic_reg_write(apic, reg, (u32)data);
2514 }
2515 
2516 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2517 {
2518 	struct kvm_lapic *apic = vcpu->arch.apic;
2519 	u32 low, high = 0;
2520 
2521 	if (!lapic_in_kernel(vcpu))
2522 		return 1;
2523 
2524 	if (kvm_lapic_reg_read(apic, reg, 4, &low))
2525 		return 1;
2526 	if (reg == APIC_ICR)
2527 		kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2528 
2529 	*data = (((u64)high) << 32) | low;
2530 
2531 	return 0;
2532 }
2533 
2534 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2535 {
2536 	u64 addr = data & ~KVM_MSR_ENABLED;
2537 	if (!IS_ALIGNED(addr, 4))
2538 		return 1;
2539 
2540 	vcpu->arch.pv_eoi.msr_val = data;
2541 	if (!pv_eoi_enabled(vcpu))
2542 		return 0;
2543 	return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
2544 					 addr, sizeof(u8));
2545 }
2546 
2547 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2548 {
2549 	struct kvm_lapic *apic = vcpu->arch.apic;
2550 	u8 sipi_vector;
2551 	unsigned long pe;
2552 
2553 	if (!lapic_in_kernel(vcpu) || !apic->pending_events)
2554 		return;
2555 
2556 	/*
2557 	 * INITs are latched while in SMM.  Because an SMM CPU cannot
2558 	 * be in KVM_MP_STATE_INIT_RECEIVED state, just eat SIPIs
2559 	 * and delay processing of INIT until the next RSM.
2560 	 */
2561 	if (is_smm(vcpu)) {
2562 		WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2563 		if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2564 			clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2565 		return;
2566 	}
2567 
2568 	pe = xchg(&apic->pending_events, 0);
2569 	if (test_bit(KVM_APIC_INIT, &pe)) {
2570 		kvm_vcpu_reset(vcpu, true);
2571 		if (kvm_vcpu_is_bsp(apic->vcpu))
2572 			vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2573 		else
2574 			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2575 	}
2576 	if (test_bit(KVM_APIC_SIPI, &pe) &&
2577 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2578 		/* evaluate pending_events before reading the vector */
2579 		smp_rmb();
2580 		sipi_vector = apic->sipi_vector;
2581 		apic_debug("vcpu %d received sipi with vector # %x\n",
2582 			 vcpu->vcpu_id, sipi_vector);
2583 		kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2584 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2585 	}
2586 }
2587 
2588 void kvm_lapic_init(void)
2589 {
2590 	/* do not patch jump label more than once per second */
2591 	jump_label_rate_limit(&apic_hw_disabled, HZ);
2592 	jump_label_rate_limit(&apic_sw_disabled, HZ);
2593 }
2594 
2595 void kvm_lapic_exit(void)
2596 {
2597 	static_key_deferred_flush(&apic_hw_disabled);
2598 	static_key_deferred_flush(&apic_sw_disabled);
2599 }
2600