xref: /openbmc/linux/arch/x86/kvm/lapic.c (revision bcb84fb4)
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_BUS_CYCLE_NS 1
58 
59 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
60 #define apic_debug(fmt, arg...)
61 
62 /* 14 is the version for Xeon and Pentium 8.4.8*/
63 #define APIC_VERSION			(0x14UL | ((KVM_APIC_LVT_NUM - 1) << 16))
64 #define LAPIC_MMIO_LENGTH		(1 << 12)
65 /* followed define is not in apicdef.h */
66 #define APIC_SHORT_MASK			0xc0000
67 #define APIC_DEST_NOSHORT		0x0
68 #define APIC_DEST_MASK			0x800
69 #define MAX_APIC_VECTOR			256
70 #define APIC_VECTORS_PER_REG		32
71 
72 #define APIC_BROADCAST			0xFF
73 #define X2APIC_BROADCAST		0xFFFFFFFFul
74 
75 static inline int apic_test_vector(int vec, void *bitmap)
76 {
77 	return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
78 }
79 
80 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
81 {
82 	struct kvm_lapic *apic = vcpu->arch.apic;
83 
84 	return apic_test_vector(vector, apic->regs + APIC_ISR) ||
85 		apic_test_vector(vector, apic->regs + APIC_IRR);
86 }
87 
88 static inline void apic_clear_vector(int vec, void *bitmap)
89 {
90 	clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
91 }
92 
93 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
94 {
95 	return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
96 }
97 
98 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
99 {
100 	return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
101 }
102 
103 struct static_key_deferred apic_hw_disabled __read_mostly;
104 struct static_key_deferred apic_sw_disabled __read_mostly;
105 
106 static inline int apic_enabled(struct kvm_lapic *apic)
107 {
108 	return kvm_apic_sw_enabled(apic) &&	kvm_apic_hw_enabled(apic);
109 }
110 
111 #define LVT_MASK	\
112 	(APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
113 
114 #define LINT_MASK	\
115 	(LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
116 	 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
117 
118 static inline u8 kvm_xapic_id(struct kvm_lapic *apic)
119 {
120 	return kvm_lapic_get_reg(apic, APIC_ID) >> 24;
121 }
122 
123 static inline u32 kvm_x2apic_id(struct kvm_lapic *apic)
124 {
125 	return apic->vcpu->vcpu_id;
126 }
127 
128 static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
129 		u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
130 	switch (map->mode) {
131 	case KVM_APIC_MODE_X2APIC: {
132 		u32 offset = (dest_id >> 16) * 16;
133 		u32 max_apic_id = map->max_apic_id;
134 
135 		if (offset <= max_apic_id) {
136 			u8 cluster_size = min(max_apic_id - offset + 1, 16U);
137 
138 			*cluster = &map->phys_map[offset];
139 			*mask = dest_id & (0xffff >> (16 - cluster_size));
140 		} else {
141 			*mask = 0;
142 		}
143 
144 		return true;
145 		}
146 	case KVM_APIC_MODE_XAPIC_FLAT:
147 		*cluster = map->xapic_flat_map;
148 		*mask = dest_id & 0xff;
149 		return true;
150 	case KVM_APIC_MODE_XAPIC_CLUSTER:
151 		*cluster = map->xapic_cluster_map[(dest_id >> 4) & 0xf];
152 		*mask = dest_id & 0xf;
153 		return true;
154 	default:
155 		/* Not optimized. */
156 		return false;
157 	}
158 }
159 
160 static void kvm_apic_map_free(struct rcu_head *rcu)
161 {
162 	struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
163 
164 	kvfree(map);
165 }
166 
167 static void recalculate_apic_map(struct kvm *kvm)
168 {
169 	struct kvm_apic_map *new, *old = NULL;
170 	struct kvm_vcpu *vcpu;
171 	int i;
172 	u32 max_id = 255; /* enough space for any xAPIC ID */
173 
174 	mutex_lock(&kvm->arch.apic_map_lock);
175 
176 	kvm_for_each_vcpu(i, vcpu, kvm)
177 		if (kvm_apic_present(vcpu))
178 			max_id = max(max_id, kvm_x2apic_id(vcpu->arch.apic));
179 
180 	new = kvm_kvzalloc(sizeof(struct kvm_apic_map) +
181 	                   sizeof(struct kvm_lapic *) * ((u64)max_id + 1));
182 
183 	if (!new)
184 		goto out;
185 
186 	new->max_apic_id = max_id;
187 
188 	kvm_for_each_vcpu(i, vcpu, kvm) {
189 		struct kvm_lapic *apic = vcpu->arch.apic;
190 		struct kvm_lapic **cluster;
191 		u16 mask;
192 		u32 ldr;
193 		u8 xapic_id;
194 		u32 x2apic_id;
195 
196 		if (!kvm_apic_present(vcpu))
197 			continue;
198 
199 		xapic_id = kvm_xapic_id(apic);
200 		x2apic_id = kvm_x2apic_id(apic);
201 
202 		/* Hotplug hack: see kvm_apic_match_physical_addr(), ... */
203 		if ((apic_x2apic_mode(apic) || x2apic_id > 0xff) &&
204 				x2apic_id <= new->max_apic_id)
205 			new->phys_map[x2apic_id] = apic;
206 		/*
207 		 * ... xAPIC ID of VCPUs with APIC ID > 0xff will wrap-around,
208 		 * prevent them from masking VCPUs with APIC ID <= 0xff.
209 		 */
210 		if (!apic_x2apic_mode(apic) && !new->phys_map[xapic_id])
211 			new->phys_map[xapic_id] = apic;
212 
213 		ldr = kvm_lapic_get_reg(apic, APIC_LDR);
214 
215 		if (apic_x2apic_mode(apic)) {
216 			new->mode |= KVM_APIC_MODE_X2APIC;
217 		} else if (ldr) {
218 			ldr = GET_APIC_LOGICAL_ID(ldr);
219 			if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
220 				new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
221 			else
222 				new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
223 		}
224 
225 		if (!kvm_apic_map_get_logical_dest(new, ldr, &cluster, &mask))
226 			continue;
227 
228 		if (mask)
229 			cluster[ffs(mask) - 1] = apic;
230 	}
231 out:
232 	old = rcu_dereference_protected(kvm->arch.apic_map,
233 			lockdep_is_held(&kvm->arch.apic_map_lock));
234 	rcu_assign_pointer(kvm->arch.apic_map, new);
235 	mutex_unlock(&kvm->arch.apic_map_lock);
236 
237 	if (old)
238 		call_rcu(&old->rcu, kvm_apic_map_free);
239 
240 	kvm_make_scan_ioapic_request(kvm);
241 }
242 
243 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
244 {
245 	bool enabled = val & APIC_SPIV_APIC_ENABLED;
246 
247 	kvm_lapic_set_reg(apic, APIC_SPIV, val);
248 
249 	if (enabled != apic->sw_enabled) {
250 		apic->sw_enabled = enabled;
251 		if (enabled) {
252 			static_key_slow_dec_deferred(&apic_sw_disabled);
253 			recalculate_apic_map(apic->vcpu->kvm);
254 		} else
255 			static_key_slow_inc(&apic_sw_disabled.key);
256 	}
257 }
258 
259 static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
260 {
261 	kvm_lapic_set_reg(apic, APIC_ID, id << 24);
262 	recalculate_apic_map(apic->vcpu->kvm);
263 }
264 
265 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
266 {
267 	kvm_lapic_set_reg(apic, APIC_LDR, id);
268 	recalculate_apic_map(apic->vcpu->kvm);
269 }
270 
271 static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
272 {
273 	u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
274 
275 	WARN_ON_ONCE(id != apic->vcpu->vcpu_id);
276 
277 	kvm_lapic_set_reg(apic, APIC_ID, id);
278 	kvm_lapic_set_reg(apic, APIC_LDR, ldr);
279 	recalculate_apic_map(apic->vcpu->kvm);
280 }
281 
282 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
283 {
284 	return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
285 }
286 
287 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
288 {
289 	return kvm_lapic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
290 }
291 
292 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
293 {
294 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
295 }
296 
297 static inline int apic_lvtt_period(struct kvm_lapic *apic)
298 {
299 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
300 }
301 
302 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
303 {
304 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
305 }
306 
307 static inline int apic_lvt_nmi_mode(u32 lvt_val)
308 {
309 	return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
310 }
311 
312 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
313 {
314 	struct kvm_lapic *apic = vcpu->arch.apic;
315 	struct kvm_cpuid_entry2 *feat;
316 	u32 v = APIC_VERSION;
317 
318 	if (!lapic_in_kernel(vcpu))
319 		return;
320 
321 	feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
322 	if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
323 		v |= APIC_LVR_DIRECTED_EOI;
324 	kvm_lapic_set_reg(apic, APIC_LVR, v);
325 }
326 
327 static const unsigned int apic_lvt_mask[KVM_APIC_LVT_NUM] = {
328 	LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
329 	LVT_MASK | APIC_MODE_MASK,	/* LVTTHMR */
330 	LVT_MASK | APIC_MODE_MASK,	/* LVTPC */
331 	LINT_MASK, LINT_MASK,	/* LVT0-1 */
332 	LVT_MASK		/* LVTERR */
333 };
334 
335 static int find_highest_vector(void *bitmap)
336 {
337 	int vec;
338 	u32 *reg;
339 
340 	for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
341 	     vec >= 0; vec -= APIC_VECTORS_PER_REG) {
342 		reg = bitmap + REG_POS(vec);
343 		if (*reg)
344 			return __fls(*reg) + vec;
345 	}
346 
347 	return -1;
348 }
349 
350 static u8 count_vectors(void *bitmap)
351 {
352 	int vec;
353 	u32 *reg;
354 	u8 count = 0;
355 
356 	for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
357 		reg = bitmap + REG_POS(vec);
358 		count += hweight32(*reg);
359 	}
360 
361 	return count;
362 }
363 
364 int __kvm_apic_update_irr(u32 *pir, void *regs)
365 {
366 	u32 i, vec;
367 	u32 pir_val, irr_val;
368 	int max_irr = -1;
369 
370 	for (i = vec = 0; i <= 7; i++, vec += 32) {
371 		pir_val = READ_ONCE(pir[i]);
372 		irr_val = *((u32 *)(regs + APIC_IRR + i * 0x10));
373 		if (pir_val) {
374 			irr_val |= xchg(&pir[i], 0);
375 			*((u32 *)(regs + APIC_IRR + i * 0x10)) = irr_val;
376 		}
377 		if (irr_val)
378 			max_irr = __fls(irr_val) + vec;
379 	}
380 
381 	return max_irr;
382 }
383 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
384 
385 int kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
386 {
387 	struct kvm_lapic *apic = vcpu->arch.apic;
388 
389 	return __kvm_apic_update_irr(pir, apic->regs);
390 }
391 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
392 
393 static inline int apic_search_irr(struct kvm_lapic *apic)
394 {
395 	return find_highest_vector(apic->regs + APIC_IRR);
396 }
397 
398 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
399 {
400 	int result;
401 
402 	/*
403 	 * Note that irr_pending is just a hint. It will be always
404 	 * true with virtual interrupt delivery enabled.
405 	 */
406 	if (!apic->irr_pending)
407 		return -1;
408 
409 	result = apic_search_irr(apic);
410 	ASSERT(result == -1 || result >= 16);
411 
412 	return result;
413 }
414 
415 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
416 {
417 	struct kvm_vcpu *vcpu;
418 
419 	vcpu = apic->vcpu;
420 
421 	if (unlikely(vcpu->arch.apicv_active)) {
422 		/* need to update RVI */
423 		apic_clear_vector(vec, apic->regs + APIC_IRR);
424 		kvm_x86_ops->hwapic_irr_update(vcpu,
425 				apic_find_highest_irr(apic));
426 	} else {
427 		apic->irr_pending = false;
428 		apic_clear_vector(vec, apic->regs + APIC_IRR);
429 		if (apic_search_irr(apic) != -1)
430 			apic->irr_pending = true;
431 	}
432 }
433 
434 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
435 {
436 	struct kvm_vcpu *vcpu;
437 
438 	if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
439 		return;
440 
441 	vcpu = apic->vcpu;
442 
443 	/*
444 	 * With APIC virtualization enabled, all caching is disabled
445 	 * because the processor can modify ISR under the hood.  Instead
446 	 * just set SVI.
447 	 */
448 	if (unlikely(vcpu->arch.apicv_active))
449 		kvm_x86_ops->hwapic_isr_update(vcpu, vec);
450 	else {
451 		++apic->isr_count;
452 		BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
453 		/*
454 		 * ISR (in service register) bit is set when injecting an interrupt.
455 		 * The highest vector is injected. Thus the latest bit set matches
456 		 * the highest bit in ISR.
457 		 */
458 		apic->highest_isr_cache = vec;
459 	}
460 }
461 
462 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
463 {
464 	int result;
465 
466 	/*
467 	 * Note that isr_count is always 1, and highest_isr_cache
468 	 * is always -1, with APIC virtualization enabled.
469 	 */
470 	if (!apic->isr_count)
471 		return -1;
472 	if (likely(apic->highest_isr_cache != -1))
473 		return apic->highest_isr_cache;
474 
475 	result = find_highest_vector(apic->regs + APIC_ISR);
476 	ASSERT(result == -1 || result >= 16);
477 
478 	return result;
479 }
480 
481 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
482 {
483 	struct kvm_vcpu *vcpu;
484 	if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
485 		return;
486 
487 	vcpu = apic->vcpu;
488 
489 	/*
490 	 * We do get here for APIC virtualization enabled if the guest
491 	 * uses the Hyper-V APIC enlightenment.  In this case we may need
492 	 * to trigger a new interrupt delivery by writing the SVI field;
493 	 * on the other hand isr_count and highest_isr_cache are unused
494 	 * and must be left alone.
495 	 */
496 	if (unlikely(vcpu->arch.apicv_active))
497 		kvm_x86_ops->hwapic_isr_update(vcpu,
498 					       apic_find_highest_isr(apic));
499 	else {
500 		--apic->isr_count;
501 		BUG_ON(apic->isr_count < 0);
502 		apic->highest_isr_cache = -1;
503 	}
504 }
505 
506 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
507 {
508 	/* This may race with setting of irr in __apic_accept_irq() and
509 	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
510 	 * will cause vmexit immediately and the value will be recalculated
511 	 * on the next vmentry.
512 	 */
513 	return apic_find_highest_irr(vcpu->arch.apic);
514 }
515 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
516 
517 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
518 			     int vector, int level, int trig_mode,
519 			     struct dest_map *dest_map);
520 
521 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
522 		     struct dest_map *dest_map)
523 {
524 	struct kvm_lapic *apic = vcpu->arch.apic;
525 
526 	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
527 			irq->level, irq->trig_mode, dest_map);
528 }
529 
530 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
531 {
532 	return kvm_vcpu_write_guest_cached(vcpu, &vcpu->arch.pv_eoi.data, &val,
533 					   sizeof(val));
534 }
535 
536 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
537 {
538 	return kvm_vcpu_read_guest_cached(vcpu, &vcpu->arch.pv_eoi.data, val,
539 					  sizeof(*val));
540 }
541 
542 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
543 {
544 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
545 }
546 
547 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
548 {
549 	u8 val;
550 	if (pv_eoi_get_user(vcpu, &val) < 0)
551 		apic_debug("Can't read EOI MSR value: 0x%llx\n",
552 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
553 	return val & 0x1;
554 }
555 
556 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
557 {
558 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
559 		apic_debug("Can't set EOI MSR value: 0x%llx\n",
560 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
561 		return;
562 	}
563 	__set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
564 }
565 
566 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
567 {
568 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
569 		apic_debug("Can't clear EOI MSR value: 0x%llx\n",
570 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
571 		return;
572 	}
573 	__clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
574 }
575 
576 static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
577 {
578 	int highest_irr;
579 	if (kvm_x86_ops->sync_pir_to_irr && apic->vcpu->arch.apicv_active)
580 		highest_irr = kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
581 	else
582 		highest_irr = apic_find_highest_irr(apic);
583 	if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
584 		return -1;
585 	return highest_irr;
586 }
587 
588 static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
589 {
590 	u32 tpr, isrv, ppr, old_ppr;
591 	int isr;
592 
593 	old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
594 	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
595 	isr = apic_find_highest_isr(apic);
596 	isrv = (isr != -1) ? isr : 0;
597 
598 	if ((tpr & 0xf0) >= (isrv & 0xf0))
599 		ppr = tpr & 0xff;
600 	else
601 		ppr = isrv & 0xf0;
602 
603 	apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
604 		   apic, ppr, isr, isrv);
605 
606 	*new_ppr = ppr;
607 	if (old_ppr != ppr)
608 		kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
609 
610 	return ppr < old_ppr;
611 }
612 
613 static void apic_update_ppr(struct kvm_lapic *apic)
614 {
615 	u32 ppr;
616 
617 	if (__apic_update_ppr(apic, &ppr) &&
618 	    apic_has_interrupt_for_ppr(apic, ppr) != -1)
619 		kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
620 }
621 
622 void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
623 {
624 	apic_update_ppr(vcpu->arch.apic);
625 }
626 EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
627 
628 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
629 {
630 	kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
631 	apic_update_ppr(apic);
632 }
633 
634 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
635 {
636 	return mda == (apic_x2apic_mode(apic) ?
637 			X2APIC_BROADCAST : APIC_BROADCAST);
638 }
639 
640 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
641 {
642 	if (kvm_apic_broadcast(apic, mda))
643 		return true;
644 
645 	if (apic_x2apic_mode(apic))
646 		return mda == kvm_x2apic_id(apic);
647 
648 	/*
649 	 * Hotplug hack: Make LAPIC in xAPIC mode also accept interrupts as if
650 	 * it were in x2APIC mode.  Hotplugged VCPUs start in xAPIC mode and
651 	 * this allows unique addressing of VCPUs with APIC ID over 0xff.
652 	 * The 0xff condition is needed because writeable xAPIC ID.
653 	 */
654 	if (kvm_x2apic_id(apic) > 0xff && mda == kvm_x2apic_id(apic))
655 		return true;
656 
657 	return mda == kvm_xapic_id(apic);
658 }
659 
660 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
661 {
662 	u32 logical_id;
663 
664 	if (kvm_apic_broadcast(apic, mda))
665 		return true;
666 
667 	logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
668 
669 	if (apic_x2apic_mode(apic))
670 		return ((logical_id >> 16) == (mda >> 16))
671 		       && (logical_id & mda & 0xffff) != 0;
672 
673 	logical_id = GET_APIC_LOGICAL_ID(logical_id);
674 
675 	switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
676 	case APIC_DFR_FLAT:
677 		return (logical_id & mda) != 0;
678 	case APIC_DFR_CLUSTER:
679 		return ((logical_id >> 4) == (mda >> 4))
680 		       && (logical_id & mda & 0xf) != 0;
681 	default:
682 		apic_debug("Bad DFR vcpu %d: %08x\n",
683 			   apic->vcpu->vcpu_id, kvm_lapic_get_reg(apic, APIC_DFR));
684 		return false;
685 	}
686 }
687 
688 /* The KVM local APIC implementation has two quirks:
689  *
690  *  - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
691  *    in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
692  *    KVM doesn't do that aliasing.
693  *
694  *  - in-kernel IOAPIC messages have to be delivered directly to
695  *    x2APIC, because the kernel does not support interrupt remapping.
696  *    In order to support broadcast without interrupt remapping, x2APIC
697  *    rewrites the destination of non-IPI messages from APIC_BROADCAST
698  *    to X2APIC_BROADCAST.
699  *
700  * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API.  This is
701  * important when userspace wants to use x2APIC-format MSIs, because
702  * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
703  */
704 static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
705 		struct kvm_lapic *source, struct kvm_lapic *target)
706 {
707 	bool ipi = source != NULL;
708 
709 	if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
710 	    !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
711 		return X2APIC_BROADCAST;
712 
713 	return dest_id;
714 }
715 
716 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
717 			   int short_hand, unsigned int dest, int dest_mode)
718 {
719 	struct kvm_lapic *target = vcpu->arch.apic;
720 	u32 mda = kvm_apic_mda(vcpu, dest, source, target);
721 
722 	apic_debug("target %p, source %p, dest 0x%x, "
723 		   "dest_mode 0x%x, short_hand 0x%x\n",
724 		   target, source, dest, dest_mode, short_hand);
725 
726 	ASSERT(target);
727 	switch (short_hand) {
728 	case APIC_DEST_NOSHORT:
729 		if (dest_mode == APIC_DEST_PHYSICAL)
730 			return kvm_apic_match_physical_addr(target, mda);
731 		else
732 			return kvm_apic_match_logical_addr(target, mda);
733 	case APIC_DEST_SELF:
734 		return target == source;
735 	case APIC_DEST_ALLINC:
736 		return true;
737 	case APIC_DEST_ALLBUT:
738 		return target != source;
739 	default:
740 		apic_debug("kvm: apic: Bad dest shorthand value %x\n",
741 			   short_hand);
742 		return false;
743 	}
744 }
745 EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
746 
747 int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
748 		       const unsigned long *bitmap, u32 bitmap_size)
749 {
750 	u32 mod;
751 	int i, idx = -1;
752 
753 	mod = vector % dest_vcpus;
754 
755 	for (i = 0; i <= mod; i++) {
756 		idx = find_next_bit(bitmap, bitmap_size, idx + 1);
757 		BUG_ON(idx == bitmap_size);
758 	}
759 
760 	return idx;
761 }
762 
763 static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
764 {
765 	if (!kvm->arch.disabled_lapic_found) {
766 		kvm->arch.disabled_lapic_found = true;
767 		printk(KERN_INFO
768 		       "Disabled LAPIC found during irq injection\n");
769 	}
770 }
771 
772 static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
773 		struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
774 {
775 	if (kvm->arch.x2apic_broadcast_quirk_disabled) {
776 		if ((irq->dest_id == APIC_BROADCAST &&
777 				map->mode != KVM_APIC_MODE_X2APIC))
778 			return true;
779 		if (irq->dest_id == X2APIC_BROADCAST)
780 			return true;
781 	} else {
782 		bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
783 		if (irq->dest_id == (x2apic_ipi ?
784 		                     X2APIC_BROADCAST : APIC_BROADCAST))
785 			return true;
786 	}
787 
788 	return false;
789 }
790 
791 /* Return true if the interrupt can be handled by using *bitmap as index mask
792  * for valid destinations in *dst array.
793  * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
794  * Note: we may have zero kvm_lapic destinations when we return true, which
795  * means that the interrupt should be dropped.  In this case, *bitmap would be
796  * zero and *dst undefined.
797  */
798 static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
799 		struct kvm_lapic **src, struct kvm_lapic_irq *irq,
800 		struct kvm_apic_map *map, struct kvm_lapic ***dst,
801 		unsigned long *bitmap)
802 {
803 	int i, lowest;
804 
805 	if (irq->shorthand == APIC_DEST_SELF && src) {
806 		*dst = src;
807 		*bitmap = 1;
808 		return true;
809 	} else if (irq->shorthand)
810 		return false;
811 
812 	if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
813 		return false;
814 
815 	if (irq->dest_mode == APIC_DEST_PHYSICAL) {
816 		if (irq->dest_id > map->max_apic_id) {
817 			*bitmap = 0;
818 		} else {
819 			*dst = &map->phys_map[irq->dest_id];
820 			*bitmap = 1;
821 		}
822 		return true;
823 	}
824 
825 	*bitmap = 0;
826 	if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
827 				(u16 *)bitmap))
828 		return false;
829 
830 	if (!kvm_lowest_prio_delivery(irq))
831 		return true;
832 
833 	if (!kvm_vector_hashing_enabled()) {
834 		lowest = -1;
835 		for_each_set_bit(i, bitmap, 16) {
836 			if (!(*dst)[i])
837 				continue;
838 			if (lowest < 0)
839 				lowest = i;
840 			else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
841 						(*dst)[lowest]->vcpu) < 0)
842 				lowest = i;
843 		}
844 	} else {
845 		if (!*bitmap)
846 			return true;
847 
848 		lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
849 				bitmap, 16);
850 
851 		if (!(*dst)[lowest]) {
852 			kvm_apic_disabled_lapic_found(kvm);
853 			*bitmap = 0;
854 			return true;
855 		}
856 	}
857 
858 	*bitmap = (lowest >= 0) ? 1 << lowest : 0;
859 
860 	return true;
861 }
862 
863 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
864 		struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
865 {
866 	struct kvm_apic_map *map;
867 	unsigned long bitmap;
868 	struct kvm_lapic **dst = NULL;
869 	int i;
870 	bool ret;
871 
872 	*r = -1;
873 
874 	if (irq->shorthand == APIC_DEST_SELF) {
875 		*r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
876 		return true;
877 	}
878 
879 	rcu_read_lock();
880 	map = rcu_dereference(kvm->arch.apic_map);
881 
882 	ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
883 	if (ret)
884 		for_each_set_bit(i, &bitmap, 16) {
885 			if (!dst[i])
886 				continue;
887 			if (*r < 0)
888 				*r = 0;
889 			*r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
890 		}
891 
892 	rcu_read_unlock();
893 	return ret;
894 }
895 
896 /*
897  * This routine tries to handler interrupts in posted mode, here is how
898  * it deals with different cases:
899  * - For single-destination interrupts, handle it in posted mode
900  * - Else if vector hashing is enabled and it is a lowest-priority
901  *   interrupt, handle it in posted mode and use the following mechanism
902  *   to find the destinaiton vCPU.
903  *	1. For lowest-priority interrupts, store all the possible
904  *	   destination vCPUs in an array.
905  *	2. Use "guest vector % max number of destination vCPUs" to find
906  *	   the right destination vCPU in the array for the lowest-priority
907  *	   interrupt.
908  * - Otherwise, use remapped mode to inject the interrupt.
909  */
910 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
911 			struct kvm_vcpu **dest_vcpu)
912 {
913 	struct kvm_apic_map *map;
914 	unsigned long bitmap;
915 	struct kvm_lapic **dst = NULL;
916 	bool ret = false;
917 
918 	if (irq->shorthand)
919 		return false;
920 
921 	rcu_read_lock();
922 	map = rcu_dereference(kvm->arch.apic_map);
923 
924 	if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
925 			hweight16(bitmap) == 1) {
926 		unsigned long i = find_first_bit(&bitmap, 16);
927 
928 		if (dst[i]) {
929 			*dest_vcpu = dst[i]->vcpu;
930 			ret = true;
931 		}
932 	}
933 
934 	rcu_read_unlock();
935 	return ret;
936 }
937 
938 /*
939  * Add a pending IRQ into lapic.
940  * Return 1 if successfully added and 0 if discarded.
941  */
942 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
943 			     int vector, int level, int trig_mode,
944 			     struct dest_map *dest_map)
945 {
946 	int result = 0;
947 	struct kvm_vcpu *vcpu = apic->vcpu;
948 
949 	trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
950 				  trig_mode, vector);
951 	switch (delivery_mode) {
952 	case APIC_DM_LOWEST:
953 		vcpu->arch.apic_arb_prio++;
954 	case APIC_DM_FIXED:
955 		if (unlikely(trig_mode && !level))
956 			break;
957 
958 		/* FIXME add logic for vcpu on reset */
959 		if (unlikely(!apic_enabled(apic)))
960 			break;
961 
962 		result = 1;
963 
964 		if (dest_map) {
965 			__set_bit(vcpu->vcpu_id, dest_map->map);
966 			dest_map->vectors[vcpu->vcpu_id] = vector;
967 		}
968 
969 		if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
970 			if (trig_mode)
971 				kvm_lapic_set_vector(vector, apic->regs + APIC_TMR);
972 			else
973 				apic_clear_vector(vector, apic->regs + APIC_TMR);
974 		}
975 
976 		if (vcpu->arch.apicv_active)
977 			kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
978 		else {
979 			kvm_lapic_set_irr(vector, apic);
980 
981 			kvm_make_request(KVM_REQ_EVENT, vcpu);
982 			kvm_vcpu_kick(vcpu);
983 		}
984 		break;
985 
986 	case APIC_DM_REMRD:
987 		result = 1;
988 		vcpu->arch.pv.pv_unhalted = 1;
989 		kvm_make_request(KVM_REQ_EVENT, vcpu);
990 		kvm_vcpu_kick(vcpu);
991 		break;
992 
993 	case APIC_DM_SMI:
994 		result = 1;
995 		kvm_make_request(KVM_REQ_SMI, vcpu);
996 		kvm_vcpu_kick(vcpu);
997 		break;
998 
999 	case APIC_DM_NMI:
1000 		result = 1;
1001 		kvm_inject_nmi(vcpu);
1002 		kvm_vcpu_kick(vcpu);
1003 		break;
1004 
1005 	case APIC_DM_INIT:
1006 		if (!trig_mode || level) {
1007 			result = 1;
1008 			/* assumes that there are only KVM_APIC_INIT/SIPI */
1009 			apic->pending_events = (1UL << KVM_APIC_INIT);
1010 			/* make sure pending_events is visible before sending
1011 			 * the request */
1012 			smp_wmb();
1013 			kvm_make_request(KVM_REQ_EVENT, vcpu);
1014 			kvm_vcpu_kick(vcpu);
1015 		} else {
1016 			apic_debug("Ignoring de-assert INIT to vcpu %d\n",
1017 				   vcpu->vcpu_id);
1018 		}
1019 		break;
1020 
1021 	case APIC_DM_STARTUP:
1022 		apic_debug("SIPI to vcpu %d vector 0x%02x\n",
1023 			   vcpu->vcpu_id, vector);
1024 		result = 1;
1025 		apic->sipi_vector = vector;
1026 		/* make sure sipi_vector is visible for the receiver */
1027 		smp_wmb();
1028 		set_bit(KVM_APIC_SIPI, &apic->pending_events);
1029 		kvm_make_request(KVM_REQ_EVENT, vcpu);
1030 		kvm_vcpu_kick(vcpu);
1031 		break;
1032 
1033 	case APIC_DM_EXTINT:
1034 		/*
1035 		 * Should only be called by kvm_apic_local_deliver() with LVT0,
1036 		 * before NMI watchdog was enabled. Already handled by
1037 		 * kvm_apic_accept_pic_intr().
1038 		 */
1039 		break;
1040 
1041 	default:
1042 		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1043 		       delivery_mode);
1044 		break;
1045 	}
1046 	return result;
1047 }
1048 
1049 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1050 {
1051 	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1052 }
1053 
1054 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1055 {
1056 	return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1057 }
1058 
1059 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1060 {
1061 	int trigger_mode;
1062 
1063 	/* Eoi the ioapic only if the ioapic doesn't own the vector. */
1064 	if (!kvm_ioapic_handles_vector(apic, vector))
1065 		return;
1066 
1067 	/* Request a KVM exit to inform the userspace IOAPIC. */
1068 	if (irqchip_split(apic->vcpu->kvm)) {
1069 		apic->vcpu->arch.pending_ioapic_eoi = vector;
1070 		kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1071 		return;
1072 	}
1073 
1074 	if (apic_test_vector(vector, apic->regs + APIC_TMR))
1075 		trigger_mode = IOAPIC_LEVEL_TRIG;
1076 	else
1077 		trigger_mode = IOAPIC_EDGE_TRIG;
1078 
1079 	kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1080 }
1081 
1082 static int apic_set_eoi(struct kvm_lapic *apic)
1083 {
1084 	int vector = apic_find_highest_isr(apic);
1085 
1086 	trace_kvm_eoi(apic, vector);
1087 
1088 	/*
1089 	 * Not every write EOI will has corresponding ISR,
1090 	 * one example is when Kernel check timer on setup_IO_APIC
1091 	 */
1092 	if (vector == -1)
1093 		return vector;
1094 
1095 	apic_clear_isr(vector, apic);
1096 	apic_update_ppr(apic);
1097 
1098 	if (test_bit(vector, vcpu_to_synic(apic->vcpu)->vec_bitmap))
1099 		kvm_hv_synic_send_eoi(apic->vcpu, vector);
1100 
1101 	kvm_ioapic_send_eoi(apic, vector);
1102 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1103 	return vector;
1104 }
1105 
1106 /*
1107  * this interface assumes a trap-like exit, which has already finished
1108  * desired side effect including vISR and vPPR update.
1109  */
1110 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1111 {
1112 	struct kvm_lapic *apic = vcpu->arch.apic;
1113 
1114 	trace_kvm_eoi(apic, vector);
1115 
1116 	kvm_ioapic_send_eoi(apic, vector);
1117 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1118 }
1119 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1120 
1121 static void apic_send_ipi(struct kvm_lapic *apic)
1122 {
1123 	u32 icr_low = kvm_lapic_get_reg(apic, APIC_ICR);
1124 	u32 icr_high = kvm_lapic_get_reg(apic, APIC_ICR2);
1125 	struct kvm_lapic_irq irq;
1126 
1127 	irq.vector = icr_low & APIC_VECTOR_MASK;
1128 	irq.delivery_mode = icr_low & APIC_MODE_MASK;
1129 	irq.dest_mode = icr_low & APIC_DEST_MASK;
1130 	irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1131 	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1132 	irq.shorthand = icr_low & APIC_SHORT_MASK;
1133 	irq.msi_redir_hint = false;
1134 	if (apic_x2apic_mode(apic))
1135 		irq.dest_id = icr_high;
1136 	else
1137 		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1138 
1139 	trace_kvm_apic_ipi(icr_low, irq.dest_id);
1140 
1141 	apic_debug("icr_high 0x%x, icr_low 0x%x, "
1142 		   "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
1143 		   "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x, "
1144 		   "msi_redir_hint 0x%x\n",
1145 		   icr_high, icr_low, irq.shorthand, irq.dest_id,
1146 		   irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
1147 		   irq.vector, irq.msi_redir_hint);
1148 
1149 	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1150 }
1151 
1152 static u32 apic_get_tmcct(struct kvm_lapic *apic)
1153 {
1154 	ktime_t remaining, now;
1155 	s64 ns;
1156 	u32 tmcct;
1157 
1158 	ASSERT(apic != NULL);
1159 
1160 	/* if initial count is 0, current count should also be 0 */
1161 	if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1162 		apic->lapic_timer.period == 0)
1163 		return 0;
1164 
1165 	now = ktime_get();
1166 	remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1167 	if (ktime_to_ns(remaining) < 0)
1168 		remaining = 0;
1169 
1170 	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1171 	tmcct = div64_u64(ns,
1172 			 (APIC_BUS_CYCLE_NS * apic->divide_count));
1173 
1174 	return tmcct;
1175 }
1176 
1177 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1178 {
1179 	struct kvm_vcpu *vcpu = apic->vcpu;
1180 	struct kvm_run *run = vcpu->run;
1181 
1182 	kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1183 	run->tpr_access.rip = kvm_rip_read(vcpu);
1184 	run->tpr_access.is_write = write;
1185 }
1186 
1187 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1188 {
1189 	if (apic->vcpu->arch.tpr_access_reporting)
1190 		__report_tpr_access(apic, write);
1191 }
1192 
1193 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1194 {
1195 	u32 val = 0;
1196 
1197 	if (offset >= LAPIC_MMIO_LENGTH)
1198 		return 0;
1199 
1200 	switch (offset) {
1201 	case APIC_ARBPRI:
1202 		apic_debug("Access APIC ARBPRI register which is for P6\n");
1203 		break;
1204 
1205 	case APIC_TMCCT:	/* Timer CCR */
1206 		if (apic_lvtt_tscdeadline(apic))
1207 			return 0;
1208 
1209 		val = apic_get_tmcct(apic);
1210 		break;
1211 	case APIC_PROCPRI:
1212 		apic_update_ppr(apic);
1213 		val = kvm_lapic_get_reg(apic, offset);
1214 		break;
1215 	case APIC_TASKPRI:
1216 		report_tpr_access(apic, false);
1217 		/* fall thru */
1218 	default:
1219 		val = kvm_lapic_get_reg(apic, offset);
1220 		break;
1221 	}
1222 
1223 	return val;
1224 }
1225 
1226 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1227 {
1228 	return container_of(dev, struct kvm_lapic, dev);
1229 }
1230 
1231 int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1232 		void *data)
1233 {
1234 	unsigned char alignment = offset & 0xf;
1235 	u32 result;
1236 	/* this bitmask has a bit cleared for each reserved register */
1237 	static const u64 rmask = 0x43ff01ffffffe70cULL;
1238 
1239 	if ((alignment + len) > 4) {
1240 		apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1241 			   offset, len);
1242 		return 1;
1243 	}
1244 
1245 	if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1246 		apic_debug("KVM_APIC_READ: read reserved register %x\n",
1247 			   offset);
1248 		return 1;
1249 	}
1250 
1251 	result = __apic_read(apic, offset & ~0xf);
1252 
1253 	trace_kvm_apic_read(offset, result);
1254 
1255 	switch (len) {
1256 	case 1:
1257 	case 2:
1258 	case 4:
1259 		memcpy(data, (char *)&result + alignment, len);
1260 		break;
1261 	default:
1262 		printk(KERN_ERR "Local APIC read with len = %x, "
1263 		       "should be 1,2, or 4 instead\n", len);
1264 		break;
1265 	}
1266 	return 0;
1267 }
1268 EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1269 
1270 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1271 {
1272 	return kvm_apic_hw_enabled(apic) &&
1273 	    addr >= apic->base_address &&
1274 	    addr < apic->base_address + LAPIC_MMIO_LENGTH;
1275 }
1276 
1277 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1278 			   gpa_t address, int len, void *data)
1279 {
1280 	struct kvm_lapic *apic = to_lapic(this);
1281 	u32 offset = address - apic->base_address;
1282 
1283 	if (!apic_mmio_in_range(apic, address))
1284 		return -EOPNOTSUPP;
1285 
1286 	kvm_lapic_reg_read(apic, offset, len, data);
1287 
1288 	return 0;
1289 }
1290 
1291 static void update_divide_count(struct kvm_lapic *apic)
1292 {
1293 	u32 tmp1, tmp2, tdcr;
1294 
1295 	tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1296 	tmp1 = tdcr & 0xf;
1297 	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1298 	apic->divide_count = 0x1 << (tmp2 & 0x7);
1299 
1300 	apic_debug("timer divide count is 0x%x\n",
1301 				   apic->divide_count);
1302 }
1303 
1304 static void apic_update_lvtt(struct kvm_lapic *apic)
1305 {
1306 	u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1307 			apic->lapic_timer.timer_mode_mask;
1308 
1309 	if (apic->lapic_timer.timer_mode != timer_mode) {
1310 		apic->lapic_timer.timer_mode = timer_mode;
1311 		hrtimer_cancel(&apic->lapic_timer.timer);
1312 	}
1313 }
1314 
1315 static void apic_timer_expired(struct kvm_lapic *apic)
1316 {
1317 	struct kvm_vcpu *vcpu = apic->vcpu;
1318 	struct swait_queue_head *q = &vcpu->wq;
1319 	struct kvm_timer *ktimer = &apic->lapic_timer;
1320 
1321 	if (atomic_read(&apic->lapic_timer.pending))
1322 		return;
1323 
1324 	atomic_inc(&apic->lapic_timer.pending);
1325 	kvm_set_pending_timer(vcpu);
1326 
1327 	if (swait_active(q))
1328 		swake_up(q);
1329 
1330 	if (apic_lvtt_tscdeadline(apic))
1331 		ktimer->expired_tscdeadline = ktimer->tscdeadline;
1332 }
1333 
1334 /*
1335  * On APICv, this test will cause a busy wait
1336  * during a higher-priority task.
1337  */
1338 
1339 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1340 {
1341 	struct kvm_lapic *apic = vcpu->arch.apic;
1342 	u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1343 
1344 	if (kvm_apic_hw_enabled(apic)) {
1345 		int vec = reg & APIC_VECTOR_MASK;
1346 		void *bitmap = apic->regs + APIC_ISR;
1347 
1348 		if (vcpu->arch.apicv_active)
1349 			bitmap = apic->regs + APIC_IRR;
1350 
1351 		if (apic_test_vector(vec, bitmap))
1352 			return true;
1353 	}
1354 	return false;
1355 }
1356 
1357 void wait_lapic_expire(struct kvm_vcpu *vcpu)
1358 {
1359 	struct kvm_lapic *apic = vcpu->arch.apic;
1360 	u64 guest_tsc, tsc_deadline;
1361 
1362 	if (!lapic_in_kernel(vcpu))
1363 		return;
1364 
1365 	if (apic->lapic_timer.expired_tscdeadline == 0)
1366 		return;
1367 
1368 	if (!lapic_timer_int_injected(vcpu))
1369 		return;
1370 
1371 	tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1372 	apic->lapic_timer.expired_tscdeadline = 0;
1373 	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1374 	trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1375 
1376 	/* __delay is delay_tsc whenever the hardware has TSC, thus always.  */
1377 	if (guest_tsc < tsc_deadline)
1378 		__delay(min(tsc_deadline - guest_tsc,
1379 			nsec_to_cycles(vcpu, lapic_timer_advance_ns)));
1380 }
1381 
1382 static void start_sw_tscdeadline(struct kvm_lapic *apic)
1383 {
1384 	u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1385 	u64 ns = 0;
1386 	ktime_t expire;
1387 	struct kvm_vcpu *vcpu = apic->vcpu;
1388 	unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1389 	unsigned long flags;
1390 	ktime_t now;
1391 
1392 	if (unlikely(!tscdeadline || !this_tsc_khz))
1393 		return;
1394 
1395 	local_irq_save(flags);
1396 
1397 	now = ktime_get();
1398 	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1399 	if (likely(tscdeadline > guest_tsc)) {
1400 		ns = (tscdeadline - guest_tsc) * 1000000ULL;
1401 		do_div(ns, this_tsc_khz);
1402 		expire = ktime_add_ns(now, ns);
1403 		expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1404 		hrtimer_start(&apic->lapic_timer.timer,
1405 				expire, HRTIMER_MODE_ABS_PINNED);
1406 	} else
1407 		apic_timer_expired(apic);
1408 
1409 	local_irq_restore(flags);
1410 }
1411 
1412 static void start_sw_period(struct kvm_lapic *apic)
1413 {
1414 	if (!apic->lapic_timer.period)
1415 		return;
1416 
1417 	if (apic_lvtt_oneshot(apic) &&
1418 	    ktime_after(ktime_get(),
1419 			apic->lapic_timer.target_expiration)) {
1420 		apic_timer_expired(apic);
1421 		return;
1422 	}
1423 
1424 	hrtimer_start(&apic->lapic_timer.timer,
1425 		apic->lapic_timer.target_expiration,
1426 		HRTIMER_MODE_ABS_PINNED);
1427 }
1428 
1429 static bool set_target_expiration(struct kvm_lapic *apic)
1430 {
1431 	ktime_t now;
1432 	u64 tscl = rdtsc();
1433 
1434 	now = ktime_get();
1435 	apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1436 		* APIC_BUS_CYCLE_NS * apic->divide_count;
1437 
1438 	if (!apic->lapic_timer.period)
1439 		return false;
1440 
1441 	/*
1442 	 * Do not allow the guest to program periodic timers with small
1443 	 * interval, since the hrtimers are not throttled by the host
1444 	 * scheduler.
1445 	 */
1446 	if (apic_lvtt_period(apic)) {
1447 		s64 min_period = min_timer_period_us * 1000LL;
1448 
1449 		if (apic->lapic_timer.period < min_period) {
1450 			pr_info_ratelimited(
1451 			    "kvm: vcpu %i: requested %lld ns "
1452 			    "lapic timer period limited to %lld ns\n",
1453 			    apic->vcpu->vcpu_id,
1454 			    apic->lapic_timer.period, min_period);
1455 			apic->lapic_timer.period = min_period;
1456 		}
1457 	}
1458 
1459 	apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1460 		   PRIx64 ", "
1461 		   "timer initial count 0x%x, period %lldns, "
1462 		   "expire @ 0x%016" PRIx64 ".\n", __func__,
1463 		   APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1464 		   kvm_lapic_get_reg(apic, APIC_TMICT),
1465 		   apic->lapic_timer.period,
1466 		   ktime_to_ns(ktime_add_ns(now,
1467 				apic->lapic_timer.period)));
1468 
1469 	apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1470 		nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1471 	apic->lapic_timer.target_expiration = ktime_add_ns(now, apic->lapic_timer.period);
1472 
1473 	return true;
1474 }
1475 
1476 static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1477 {
1478 	apic->lapic_timer.tscdeadline +=
1479 		nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1480 	apic->lapic_timer.target_expiration =
1481 		ktime_add_ns(apic->lapic_timer.target_expiration,
1482 				apic->lapic_timer.period);
1483 }
1484 
1485 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1486 {
1487 	if (!lapic_in_kernel(vcpu))
1488 		return false;
1489 
1490 	return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1491 }
1492 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1493 
1494 static void cancel_hv_timer(struct kvm_lapic *apic)
1495 {
1496 	kvm_x86_ops->cancel_hv_timer(apic->vcpu);
1497 	apic->lapic_timer.hv_timer_in_use = false;
1498 }
1499 
1500 static bool start_hv_timer(struct kvm_lapic *apic)
1501 {
1502 	u64 tscdeadline = apic->lapic_timer.tscdeadline;
1503 
1504 	if ((atomic_read(&apic->lapic_timer.pending) &&
1505 		!apic_lvtt_period(apic)) ||
1506 		kvm_x86_ops->set_hv_timer(apic->vcpu, tscdeadline)) {
1507 		if (apic->lapic_timer.hv_timer_in_use)
1508 			cancel_hv_timer(apic);
1509 	} else {
1510 		apic->lapic_timer.hv_timer_in_use = true;
1511 		hrtimer_cancel(&apic->lapic_timer.timer);
1512 
1513 		/* In case the sw timer triggered in the window */
1514 		if (atomic_read(&apic->lapic_timer.pending) &&
1515 			!apic_lvtt_period(apic))
1516 			cancel_hv_timer(apic);
1517 	}
1518 	trace_kvm_hv_timer_state(apic->vcpu->vcpu_id,
1519 			apic->lapic_timer.hv_timer_in_use);
1520 	return apic->lapic_timer.hv_timer_in_use;
1521 }
1522 
1523 void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1524 {
1525 	struct kvm_lapic *apic = vcpu->arch.apic;
1526 
1527 	WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1528 	WARN_ON(swait_active(&vcpu->wq));
1529 	cancel_hv_timer(apic);
1530 	apic_timer_expired(apic);
1531 
1532 	if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1533 		advance_periodic_target_expiration(apic);
1534 		if (!start_hv_timer(apic))
1535 			start_sw_period(apic);
1536 	}
1537 }
1538 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1539 
1540 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1541 {
1542 	struct kvm_lapic *apic = vcpu->arch.apic;
1543 
1544 	WARN_ON(apic->lapic_timer.hv_timer_in_use);
1545 
1546 	start_hv_timer(apic);
1547 }
1548 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1549 
1550 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1551 {
1552 	struct kvm_lapic *apic = vcpu->arch.apic;
1553 
1554 	/* Possibly the TSC deadline timer is not enabled yet */
1555 	if (!apic->lapic_timer.hv_timer_in_use)
1556 		return;
1557 
1558 	cancel_hv_timer(apic);
1559 
1560 	if (atomic_read(&apic->lapic_timer.pending))
1561 		return;
1562 
1563 	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1564 		start_sw_period(apic);
1565 	else if (apic_lvtt_tscdeadline(apic))
1566 		start_sw_tscdeadline(apic);
1567 }
1568 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1569 
1570 static void start_apic_timer(struct kvm_lapic *apic)
1571 {
1572 	atomic_set(&apic->lapic_timer.pending, 0);
1573 
1574 	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1575 		if (set_target_expiration(apic) &&
1576 			!(kvm_x86_ops->set_hv_timer && start_hv_timer(apic)))
1577 			start_sw_period(apic);
1578 	} else if (apic_lvtt_tscdeadline(apic)) {
1579 		if (!(kvm_x86_ops->set_hv_timer && start_hv_timer(apic)))
1580 			start_sw_tscdeadline(apic);
1581 	}
1582 }
1583 
1584 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1585 {
1586 	bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1587 
1588 	if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1589 		apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1590 		if (lvt0_in_nmi_mode) {
1591 			apic_debug("Receive NMI setting on APIC_LVT0 "
1592 				   "for cpu %d\n", apic->vcpu->vcpu_id);
1593 			atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1594 		} else
1595 			atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1596 	}
1597 }
1598 
1599 int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1600 {
1601 	int ret = 0;
1602 
1603 	trace_kvm_apic_write(reg, val);
1604 
1605 	switch (reg) {
1606 	case APIC_ID:		/* Local APIC ID */
1607 		if (!apic_x2apic_mode(apic))
1608 			kvm_apic_set_xapic_id(apic, val >> 24);
1609 		else
1610 			ret = 1;
1611 		break;
1612 
1613 	case APIC_TASKPRI:
1614 		report_tpr_access(apic, true);
1615 		apic_set_tpr(apic, val & 0xff);
1616 		break;
1617 
1618 	case APIC_EOI:
1619 		apic_set_eoi(apic);
1620 		break;
1621 
1622 	case APIC_LDR:
1623 		if (!apic_x2apic_mode(apic))
1624 			kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1625 		else
1626 			ret = 1;
1627 		break;
1628 
1629 	case APIC_DFR:
1630 		if (!apic_x2apic_mode(apic)) {
1631 			kvm_lapic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1632 			recalculate_apic_map(apic->vcpu->kvm);
1633 		} else
1634 			ret = 1;
1635 		break;
1636 
1637 	case APIC_SPIV: {
1638 		u32 mask = 0x3ff;
1639 		if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1640 			mask |= APIC_SPIV_DIRECTED_EOI;
1641 		apic_set_spiv(apic, val & mask);
1642 		if (!(val & APIC_SPIV_APIC_ENABLED)) {
1643 			int i;
1644 			u32 lvt_val;
1645 
1646 			for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
1647 				lvt_val = kvm_lapic_get_reg(apic,
1648 						       APIC_LVTT + 0x10 * i);
1649 				kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
1650 					     lvt_val | APIC_LVT_MASKED);
1651 			}
1652 			apic_update_lvtt(apic);
1653 			atomic_set(&apic->lapic_timer.pending, 0);
1654 
1655 		}
1656 		break;
1657 	}
1658 	case APIC_ICR:
1659 		/* No delay here, so we always clear the pending bit */
1660 		kvm_lapic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1661 		apic_send_ipi(apic);
1662 		break;
1663 
1664 	case APIC_ICR2:
1665 		if (!apic_x2apic_mode(apic))
1666 			val &= 0xff000000;
1667 		kvm_lapic_set_reg(apic, APIC_ICR2, val);
1668 		break;
1669 
1670 	case APIC_LVT0:
1671 		apic_manage_nmi_watchdog(apic, val);
1672 	case APIC_LVTTHMR:
1673 	case APIC_LVTPC:
1674 	case APIC_LVT1:
1675 	case APIC_LVTERR:
1676 		/* TODO: Check vector */
1677 		if (!kvm_apic_sw_enabled(apic))
1678 			val |= APIC_LVT_MASKED;
1679 
1680 		val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1681 		kvm_lapic_set_reg(apic, reg, val);
1682 
1683 		break;
1684 
1685 	case APIC_LVTT:
1686 		if (!kvm_apic_sw_enabled(apic))
1687 			val |= APIC_LVT_MASKED;
1688 		val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1689 		kvm_lapic_set_reg(apic, APIC_LVTT, val);
1690 		apic_update_lvtt(apic);
1691 		break;
1692 
1693 	case APIC_TMICT:
1694 		if (apic_lvtt_tscdeadline(apic))
1695 			break;
1696 
1697 		hrtimer_cancel(&apic->lapic_timer.timer);
1698 		kvm_lapic_set_reg(apic, APIC_TMICT, val);
1699 		start_apic_timer(apic);
1700 		break;
1701 
1702 	case APIC_TDCR:
1703 		if (val & 4)
1704 			apic_debug("KVM_WRITE:TDCR %x\n", val);
1705 		kvm_lapic_set_reg(apic, APIC_TDCR, val);
1706 		update_divide_count(apic);
1707 		break;
1708 
1709 	case APIC_ESR:
1710 		if (apic_x2apic_mode(apic) && val != 0) {
1711 			apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1712 			ret = 1;
1713 		}
1714 		break;
1715 
1716 	case APIC_SELF_IPI:
1717 		if (apic_x2apic_mode(apic)) {
1718 			kvm_lapic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1719 		} else
1720 			ret = 1;
1721 		break;
1722 	default:
1723 		ret = 1;
1724 		break;
1725 	}
1726 	if (ret)
1727 		apic_debug("Local APIC Write to read-only register %x\n", reg);
1728 	return ret;
1729 }
1730 EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
1731 
1732 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1733 			    gpa_t address, int len, const void *data)
1734 {
1735 	struct kvm_lapic *apic = to_lapic(this);
1736 	unsigned int offset = address - apic->base_address;
1737 	u32 val;
1738 
1739 	if (!apic_mmio_in_range(apic, address))
1740 		return -EOPNOTSUPP;
1741 
1742 	/*
1743 	 * APIC register must be aligned on 128-bits boundary.
1744 	 * 32/64/128 bits registers must be accessed thru 32 bits.
1745 	 * Refer SDM 8.4.1
1746 	 */
1747 	if (len != 4 || (offset & 0xf)) {
1748 		/* Don't shout loud, $infamous_os would cause only noise. */
1749 		apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1750 		return 0;
1751 	}
1752 
1753 	val = *(u32*)data;
1754 
1755 	/* too common printing */
1756 	if (offset != APIC_EOI)
1757 		apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1758 			   "0x%x\n", __func__, offset, len, val);
1759 
1760 	kvm_lapic_reg_write(apic, offset & 0xff0, val);
1761 
1762 	return 0;
1763 }
1764 
1765 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1766 {
1767 	kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1768 }
1769 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1770 
1771 /* emulate APIC access in a trap manner */
1772 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1773 {
1774 	u32 val = 0;
1775 
1776 	/* hw has done the conditional check and inst decode */
1777 	offset &= 0xff0;
1778 
1779 	kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
1780 
1781 	/* TODO: optimize to just emulate side effect w/o one more write */
1782 	kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
1783 }
1784 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1785 
1786 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1787 {
1788 	struct kvm_lapic *apic = vcpu->arch.apic;
1789 
1790 	if (!vcpu->arch.apic)
1791 		return;
1792 
1793 	hrtimer_cancel(&apic->lapic_timer.timer);
1794 
1795 	if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1796 		static_key_slow_dec_deferred(&apic_hw_disabled);
1797 
1798 	if (!apic->sw_enabled)
1799 		static_key_slow_dec_deferred(&apic_sw_disabled);
1800 
1801 	if (apic->regs)
1802 		free_page((unsigned long)apic->regs);
1803 
1804 	kfree(apic);
1805 }
1806 
1807 /*
1808  *----------------------------------------------------------------------
1809  * LAPIC interface
1810  *----------------------------------------------------------------------
1811  */
1812 u64 kvm_get_lapic_target_expiration_tsc(struct kvm_vcpu *vcpu)
1813 {
1814 	struct kvm_lapic *apic = vcpu->arch.apic;
1815 
1816 	if (!lapic_in_kernel(vcpu))
1817 		return 0;
1818 
1819 	return apic->lapic_timer.tscdeadline;
1820 }
1821 
1822 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1823 {
1824 	struct kvm_lapic *apic = vcpu->arch.apic;
1825 
1826 	if (!lapic_in_kernel(vcpu) ||
1827 		!apic_lvtt_tscdeadline(apic))
1828 		return 0;
1829 
1830 	return apic->lapic_timer.tscdeadline;
1831 }
1832 
1833 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1834 {
1835 	struct kvm_lapic *apic = vcpu->arch.apic;
1836 
1837 	if (!lapic_in_kernel(vcpu) || apic_lvtt_oneshot(apic) ||
1838 			apic_lvtt_period(apic))
1839 		return;
1840 
1841 	hrtimer_cancel(&apic->lapic_timer.timer);
1842 	apic->lapic_timer.tscdeadline = data;
1843 	start_apic_timer(apic);
1844 }
1845 
1846 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1847 {
1848 	struct kvm_lapic *apic = vcpu->arch.apic;
1849 
1850 	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1851 		     | (kvm_lapic_get_reg(apic, APIC_TASKPRI) & 4));
1852 }
1853 
1854 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1855 {
1856 	u64 tpr;
1857 
1858 	tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1859 
1860 	return (tpr & 0xf0) >> 4;
1861 }
1862 
1863 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1864 {
1865 	u64 old_value = vcpu->arch.apic_base;
1866 	struct kvm_lapic *apic = vcpu->arch.apic;
1867 
1868 	if (!apic)
1869 		value |= MSR_IA32_APICBASE_BSP;
1870 
1871 	vcpu->arch.apic_base = value;
1872 
1873 	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
1874 		kvm_update_cpuid(vcpu);
1875 
1876 	if (!apic)
1877 		return;
1878 
1879 	/* update jump label if enable bit changes */
1880 	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1881 		if (value & MSR_IA32_APICBASE_ENABLE) {
1882 			kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
1883 			static_key_slow_dec_deferred(&apic_hw_disabled);
1884 		} else {
1885 			static_key_slow_inc(&apic_hw_disabled.key);
1886 			recalculate_apic_map(vcpu->kvm);
1887 		}
1888 	}
1889 
1890 	if ((old_value ^ value) & X2APIC_ENABLE) {
1891 		if (value & X2APIC_ENABLE) {
1892 			kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
1893 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1894 		} else
1895 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1896 	}
1897 
1898 	apic->base_address = apic->vcpu->arch.apic_base &
1899 			     MSR_IA32_APICBASE_BASE;
1900 
1901 	if ((value & MSR_IA32_APICBASE_ENABLE) &&
1902 	     apic->base_address != APIC_DEFAULT_PHYS_BASE)
1903 		pr_warn_once("APIC base relocation is unsupported by KVM");
1904 
1905 	/* with FSB delivery interrupt, we can restart APIC functionality */
1906 	apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1907 		   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1908 
1909 }
1910 
1911 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
1912 {
1913 	struct kvm_lapic *apic;
1914 	int i;
1915 
1916 	apic_debug("%s\n", __func__);
1917 
1918 	ASSERT(vcpu);
1919 	apic = vcpu->arch.apic;
1920 	ASSERT(apic != NULL);
1921 
1922 	/* Stop the timer in case it's a reset to an active apic */
1923 	hrtimer_cancel(&apic->lapic_timer.timer);
1924 
1925 	if (!init_event) {
1926 		kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
1927 		                         MSR_IA32_APICBASE_ENABLE);
1928 		kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
1929 	}
1930 	kvm_apic_set_version(apic->vcpu);
1931 
1932 	for (i = 0; i < KVM_APIC_LVT_NUM; i++)
1933 		kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1934 	apic_update_lvtt(apic);
1935 	if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
1936 		kvm_lapic_set_reg(apic, APIC_LVT0,
1937 			     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1938 	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
1939 
1940 	kvm_lapic_set_reg(apic, APIC_DFR, 0xffffffffU);
1941 	apic_set_spiv(apic, 0xff);
1942 	kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
1943 	if (!apic_x2apic_mode(apic))
1944 		kvm_apic_set_ldr(apic, 0);
1945 	kvm_lapic_set_reg(apic, APIC_ESR, 0);
1946 	kvm_lapic_set_reg(apic, APIC_ICR, 0);
1947 	kvm_lapic_set_reg(apic, APIC_ICR2, 0);
1948 	kvm_lapic_set_reg(apic, APIC_TDCR, 0);
1949 	kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1950 	for (i = 0; i < 8; i++) {
1951 		kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1952 		kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1953 		kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1954 	}
1955 	apic->irr_pending = vcpu->arch.apicv_active;
1956 	apic->isr_count = vcpu->arch.apicv_active ? 1 : 0;
1957 	apic->highest_isr_cache = -1;
1958 	update_divide_count(apic);
1959 	atomic_set(&apic->lapic_timer.pending, 0);
1960 	if (kvm_vcpu_is_bsp(vcpu))
1961 		kvm_lapic_set_base(vcpu,
1962 				vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1963 	vcpu->arch.pv_eoi.msr_val = 0;
1964 	apic_update_ppr(apic);
1965 
1966 	vcpu->arch.apic_arb_prio = 0;
1967 	vcpu->arch.apic_attention = 0;
1968 
1969 	apic_debug("%s: vcpu=%p, id=0x%x, base_msr="
1970 		   "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1971 		   vcpu, kvm_lapic_get_reg(apic, APIC_ID),
1972 		   vcpu->arch.apic_base, apic->base_address);
1973 }
1974 
1975 /*
1976  *----------------------------------------------------------------------
1977  * timer interface
1978  *----------------------------------------------------------------------
1979  */
1980 
1981 static bool lapic_is_periodic(struct kvm_lapic *apic)
1982 {
1983 	return apic_lvtt_period(apic);
1984 }
1985 
1986 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1987 {
1988 	struct kvm_lapic *apic = vcpu->arch.apic;
1989 
1990 	if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
1991 		return atomic_read(&apic->lapic_timer.pending);
1992 
1993 	return 0;
1994 }
1995 
1996 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1997 {
1998 	u32 reg = kvm_lapic_get_reg(apic, lvt_type);
1999 	int vector, mode, trig_mode;
2000 
2001 	if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2002 		vector = reg & APIC_VECTOR_MASK;
2003 		mode = reg & APIC_MODE_MASK;
2004 		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2005 		return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2006 					NULL);
2007 	}
2008 	return 0;
2009 }
2010 
2011 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2012 {
2013 	struct kvm_lapic *apic = vcpu->arch.apic;
2014 
2015 	if (apic)
2016 		kvm_apic_local_deliver(apic, APIC_LVT0);
2017 }
2018 
2019 static const struct kvm_io_device_ops apic_mmio_ops = {
2020 	.read     = apic_mmio_read,
2021 	.write    = apic_mmio_write,
2022 };
2023 
2024 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2025 {
2026 	struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2027 	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2028 
2029 	apic_timer_expired(apic);
2030 
2031 	if (lapic_is_periodic(apic)) {
2032 		advance_periodic_target_expiration(apic);
2033 		hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2034 		return HRTIMER_RESTART;
2035 	} else
2036 		return HRTIMER_NORESTART;
2037 }
2038 
2039 int kvm_create_lapic(struct kvm_vcpu *vcpu)
2040 {
2041 	struct kvm_lapic *apic;
2042 
2043 	ASSERT(vcpu != NULL);
2044 	apic_debug("apic_init %d\n", vcpu->vcpu_id);
2045 
2046 	apic = kzalloc(sizeof(*apic), GFP_KERNEL);
2047 	if (!apic)
2048 		goto nomem;
2049 
2050 	vcpu->arch.apic = apic;
2051 
2052 	apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
2053 	if (!apic->regs) {
2054 		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2055 		       vcpu->vcpu_id);
2056 		goto nomem_free_apic;
2057 	}
2058 	apic->vcpu = vcpu;
2059 
2060 	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2061 		     HRTIMER_MODE_ABS_PINNED);
2062 	apic->lapic_timer.timer.function = apic_timer_fn;
2063 
2064 	/*
2065 	 * APIC is created enabled. This will prevent kvm_lapic_set_base from
2066 	 * thinking that APIC satet has changed.
2067 	 */
2068 	vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2069 	static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2070 	kvm_lapic_reset(vcpu, false);
2071 	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2072 
2073 	return 0;
2074 nomem_free_apic:
2075 	kfree(apic);
2076 nomem:
2077 	return -ENOMEM;
2078 }
2079 
2080 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2081 {
2082 	struct kvm_lapic *apic = vcpu->arch.apic;
2083 	u32 ppr;
2084 
2085 	if (!apic_enabled(apic))
2086 		return -1;
2087 
2088 	__apic_update_ppr(apic, &ppr);
2089 	return apic_has_interrupt_for_ppr(apic, ppr);
2090 }
2091 
2092 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2093 {
2094 	u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2095 	int r = 0;
2096 
2097 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2098 		r = 1;
2099 	if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2100 	    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2101 		r = 1;
2102 	return r;
2103 }
2104 
2105 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2106 {
2107 	struct kvm_lapic *apic = vcpu->arch.apic;
2108 
2109 	if (atomic_read(&apic->lapic_timer.pending) > 0) {
2110 		kvm_apic_local_deliver(apic, APIC_LVTT);
2111 		if (apic_lvtt_tscdeadline(apic))
2112 			apic->lapic_timer.tscdeadline = 0;
2113 		if (apic_lvtt_oneshot(apic)) {
2114 			apic->lapic_timer.tscdeadline = 0;
2115 			apic->lapic_timer.target_expiration = 0;
2116 		}
2117 		atomic_set(&apic->lapic_timer.pending, 0);
2118 	}
2119 }
2120 
2121 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2122 {
2123 	int vector = kvm_apic_has_interrupt(vcpu);
2124 	struct kvm_lapic *apic = vcpu->arch.apic;
2125 	u32 ppr;
2126 
2127 	if (vector == -1)
2128 		return -1;
2129 
2130 	/*
2131 	 * We get here even with APIC virtualization enabled, if doing
2132 	 * nested virtualization and L1 runs with the "acknowledge interrupt
2133 	 * on exit" mode.  Then we cannot inject the interrupt via RVI,
2134 	 * because the process would deliver it through the IDT.
2135 	 */
2136 
2137 	apic_clear_irr(vector, apic);
2138 	if (test_bit(vector, vcpu_to_synic(vcpu)->auto_eoi_bitmap)) {
2139 		/*
2140 		 * For auto-EOI interrupts, there might be another pending
2141 		 * interrupt above PPR, so check whether to raise another
2142 		 * KVM_REQ_EVENT.
2143 		 */
2144 		apic_update_ppr(apic);
2145 	} else {
2146 		/*
2147 		 * For normal interrupts, PPR has been raised and there cannot
2148 		 * be a higher-priority pending interrupt---except if there was
2149 		 * a concurrent interrupt injection, but that would have
2150 		 * triggered KVM_REQ_EVENT already.
2151 		 */
2152 		apic_set_isr(vector, apic);
2153 		__apic_update_ppr(apic, &ppr);
2154 	}
2155 
2156 	return vector;
2157 }
2158 
2159 static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2160 		struct kvm_lapic_state *s, bool set)
2161 {
2162 	if (apic_x2apic_mode(vcpu->arch.apic)) {
2163 		u32 *id = (u32 *)(s->regs + APIC_ID);
2164 
2165 		if (vcpu->kvm->arch.x2apic_format) {
2166 			if (*id != vcpu->vcpu_id)
2167 				return -EINVAL;
2168 		} else {
2169 			if (set)
2170 				*id >>= 24;
2171 			else
2172 				*id <<= 24;
2173 		}
2174 	}
2175 
2176 	return 0;
2177 }
2178 
2179 int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2180 {
2181 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2182 	return kvm_apic_state_fixup(vcpu, s, false);
2183 }
2184 
2185 int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2186 {
2187 	struct kvm_lapic *apic = vcpu->arch.apic;
2188 	int r;
2189 
2190 
2191 	kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2192 	/* set SPIV separately to get count of SW disabled APICs right */
2193 	apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2194 
2195 	r = kvm_apic_state_fixup(vcpu, s, true);
2196 	if (r)
2197 		return r;
2198 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2199 
2200 	recalculate_apic_map(vcpu->kvm);
2201 	kvm_apic_set_version(vcpu);
2202 
2203 	apic_update_ppr(apic);
2204 	hrtimer_cancel(&apic->lapic_timer.timer);
2205 	apic_update_lvtt(apic);
2206 	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2207 	update_divide_count(apic);
2208 	start_apic_timer(apic);
2209 	apic->irr_pending = true;
2210 	apic->isr_count = vcpu->arch.apicv_active ?
2211 				1 : count_vectors(apic->regs + APIC_ISR);
2212 	apic->highest_isr_cache = -1;
2213 	if (vcpu->arch.apicv_active) {
2214 		kvm_x86_ops->apicv_post_state_restore(vcpu);
2215 		kvm_x86_ops->hwapic_irr_update(vcpu,
2216 				apic_find_highest_irr(apic));
2217 		kvm_x86_ops->hwapic_isr_update(vcpu,
2218 				apic_find_highest_isr(apic));
2219 	}
2220 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2221 	if (ioapic_in_kernel(vcpu->kvm))
2222 		kvm_rtc_eoi_tracking_restore_one(vcpu);
2223 
2224 	vcpu->arch.apic_arb_prio = 0;
2225 
2226 	return 0;
2227 }
2228 
2229 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2230 {
2231 	struct hrtimer *timer;
2232 
2233 	if (!lapic_in_kernel(vcpu))
2234 		return;
2235 
2236 	timer = &vcpu->arch.apic->lapic_timer.timer;
2237 	if (hrtimer_cancel(timer))
2238 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
2239 }
2240 
2241 /*
2242  * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2243  *
2244  * Detect whether guest triggered PV EOI since the
2245  * last entry. If yes, set EOI on guests's behalf.
2246  * Clear PV EOI in guest memory in any case.
2247  */
2248 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2249 					struct kvm_lapic *apic)
2250 {
2251 	bool pending;
2252 	int vector;
2253 	/*
2254 	 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2255 	 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2256 	 *
2257 	 * KVM_APIC_PV_EOI_PENDING is unset:
2258 	 * 	-> host disabled PV EOI.
2259 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2260 	 * 	-> host enabled PV EOI, guest did not execute EOI yet.
2261 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2262 	 * 	-> host enabled PV EOI, guest executed EOI.
2263 	 */
2264 	BUG_ON(!pv_eoi_enabled(vcpu));
2265 	pending = pv_eoi_get_pending(vcpu);
2266 	/*
2267 	 * Clear pending bit in any case: it will be set again on vmentry.
2268 	 * While this might not be ideal from performance point of view,
2269 	 * this makes sure pv eoi is only enabled when we know it's safe.
2270 	 */
2271 	pv_eoi_clr_pending(vcpu);
2272 	if (pending)
2273 		return;
2274 	vector = apic_set_eoi(apic);
2275 	trace_kvm_pv_eoi(apic, vector);
2276 }
2277 
2278 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2279 {
2280 	u32 data;
2281 
2282 	if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2283 		apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2284 
2285 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2286 		return;
2287 
2288 	if (kvm_vcpu_read_guest_cached(vcpu, &vcpu->arch.apic->vapic_cache, &data,
2289 				       sizeof(u32)))
2290 		return;
2291 
2292 	apic_set_tpr(vcpu->arch.apic, data & 0xff);
2293 }
2294 
2295 /*
2296  * apic_sync_pv_eoi_to_guest - called before vmentry
2297  *
2298  * Detect whether it's safe to enable PV EOI and
2299  * if yes do so.
2300  */
2301 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2302 					struct kvm_lapic *apic)
2303 {
2304 	if (!pv_eoi_enabled(vcpu) ||
2305 	    /* IRR set or many bits in ISR: could be nested. */
2306 	    apic->irr_pending ||
2307 	    /* Cache not set: could be safe but we don't bother. */
2308 	    apic->highest_isr_cache == -1 ||
2309 	    /* Need EOI to update ioapic. */
2310 	    kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2311 		/*
2312 		 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2313 		 * so we need not do anything here.
2314 		 */
2315 		return;
2316 	}
2317 
2318 	pv_eoi_set_pending(apic->vcpu);
2319 }
2320 
2321 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2322 {
2323 	u32 data, tpr;
2324 	int max_irr, max_isr;
2325 	struct kvm_lapic *apic = vcpu->arch.apic;
2326 
2327 	apic_sync_pv_eoi_to_guest(vcpu, apic);
2328 
2329 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2330 		return;
2331 
2332 	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2333 	max_irr = apic_find_highest_irr(apic);
2334 	if (max_irr < 0)
2335 		max_irr = 0;
2336 	max_isr = apic_find_highest_isr(apic);
2337 	if (max_isr < 0)
2338 		max_isr = 0;
2339 	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2340 
2341 	kvm_vcpu_write_guest_cached(vcpu, &vcpu->arch.apic->vapic_cache, &data,
2342 				    sizeof(u32));
2343 }
2344 
2345 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2346 {
2347 	if (vapic_addr) {
2348 		if (kvm_vcpu_gfn_to_hva_cache_init(vcpu,
2349 					&vcpu->arch.apic->vapic_cache,
2350 					vapic_addr, sizeof(u32)))
2351 			return -EINVAL;
2352 		__set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2353 	} else {
2354 		__clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2355 	}
2356 
2357 	vcpu->arch.apic->vapic_addr = vapic_addr;
2358 	return 0;
2359 }
2360 
2361 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2362 {
2363 	struct kvm_lapic *apic = vcpu->arch.apic;
2364 	u32 reg = (msr - APIC_BASE_MSR) << 4;
2365 
2366 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2367 		return 1;
2368 
2369 	if (reg == APIC_ICR2)
2370 		return 1;
2371 
2372 	/* if this is ICR write vector before command */
2373 	if (reg == APIC_ICR)
2374 		kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2375 	return kvm_lapic_reg_write(apic, reg, (u32)data);
2376 }
2377 
2378 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2379 {
2380 	struct kvm_lapic *apic = vcpu->arch.apic;
2381 	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2382 
2383 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2384 		return 1;
2385 
2386 	if (reg == APIC_DFR || reg == APIC_ICR2) {
2387 		apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
2388 			   reg);
2389 		return 1;
2390 	}
2391 
2392 	if (kvm_lapic_reg_read(apic, reg, 4, &low))
2393 		return 1;
2394 	if (reg == APIC_ICR)
2395 		kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2396 
2397 	*data = (((u64)high) << 32) | low;
2398 
2399 	return 0;
2400 }
2401 
2402 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2403 {
2404 	struct kvm_lapic *apic = vcpu->arch.apic;
2405 
2406 	if (!lapic_in_kernel(vcpu))
2407 		return 1;
2408 
2409 	/* if this is ICR write vector before command */
2410 	if (reg == APIC_ICR)
2411 		kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2412 	return kvm_lapic_reg_write(apic, reg, (u32)data);
2413 }
2414 
2415 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2416 {
2417 	struct kvm_lapic *apic = vcpu->arch.apic;
2418 	u32 low, high = 0;
2419 
2420 	if (!lapic_in_kernel(vcpu))
2421 		return 1;
2422 
2423 	if (kvm_lapic_reg_read(apic, reg, 4, &low))
2424 		return 1;
2425 	if (reg == APIC_ICR)
2426 		kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2427 
2428 	*data = (((u64)high) << 32) | low;
2429 
2430 	return 0;
2431 }
2432 
2433 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2434 {
2435 	u64 addr = data & ~KVM_MSR_ENABLED;
2436 	if (!IS_ALIGNED(addr, 4))
2437 		return 1;
2438 
2439 	vcpu->arch.pv_eoi.msr_val = data;
2440 	if (!pv_eoi_enabled(vcpu))
2441 		return 0;
2442 	return kvm_vcpu_gfn_to_hva_cache_init(vcpu, &vcpu->arch.pv_eoi.data,
2443 					 addr, sizeof(u8));
2444 }
2445 
2446 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2447 {
2448 	struct kvm_lapic *apic = vcpu->arch.apic;
2449 	u8 sipi_vector;
2450 	unsigned long pe;
2451 
2452 	if (!lapic_in_kernel(vcpu) || !apic->pending_events)
2453 		return;
2454 
2455 	/*
2456 	 * INITs are latched while in SMM.  Because an SMM CPU cannot
2457 	 * be in KVM_MP_STATE_INIT_RECEIVED state, just eat SIPIs
2458 	 * and delay processing of INIT until the next RSM.
2459 	 */
2460 	if (is_smm(vcpu)) {
2461 		WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2462 		if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2463 			clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2464 		return;
2465 	}
2466 
2467 	pe = xchg(&apic->pending_events, 0);
2468 	if (test_bit(KVM_APIC_INIT, &pe)) {
2469 		kvm_lapic_reset(vcpu, true);
2470 		kvm_vcpu_reset(vcpu, true);
2471 		if (kvm_vcpu_is_bsp(apic->vcpu))
2472 			vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2473 		else
2474 			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2475 	}
2476 	if (test_bit(KVM_APIC_SIPI, &pe) &&
2477 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2478 		/* evaluate pending_events before reading the vector */
2479 		smp_rmb();
2480 		sipi_vector = apic->sipi_vector;
2481 		apic_debug("vcpu %d received sipi with vector # %x\n",
2482 			 vcpu->vcpu_id, sipi_vector);
2483 		kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2484 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2485 	}
2486 }
2487 
2488 void kvm_lapic_init(void)
2489 {
2490 	/* do not patch jump label more than once per second */
2491 	jump_label_rate_limit(&apic_hw_disabled, HZ);
2492 	jump_label_rate_limit(&apic_sw_disabled, HZ);
2493 }
2494 
2495 void kvm_lapic_exit(void)
2496 {
2497 	static_key_deferred_flush(&apic_hw_disabled);
2498 	static_key_deferred_flush(&apic_sw_disabled);
2499 }
2500