xref: /linux/arch/x86/kvm/lapic.c (revision fd639726bf15fca8ee1a00dce8e0096d0ad9bd18)
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 int __kvm_apic_update_irr(u32 *pir, void *regs)
368 {
369 	u32 i, vec;
370 	u32 pir_val, irr_val;
371 	int max_irr = -1;
372 
373 	for (i = vec = 0; i <= 7; i++, vec += 32) {
374 		pir_val = READ_ONCE(pir[i]);
375 		irr_val = *((u32 *)(regs + APIC_IRR + i * 0x10));
376 		if (pir_val) {
377 			irr_val |= xchg(&pir[i], 0);
378 			*((u32 *)(regs + APIC_IRR + i * 0x10)) = irr_val;
379 		}
380 		if (irr_val)
381 			max_irr = __fls(irr_val) + vec;
382 	}
383 
384 	return max_irr;
385 }
386 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
387 
388 int kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
389 {
390 	struct kvm_lapic *apic = vcpu->arch.apic;
391 
392 	return __kvm_apic_update_irr(pir, apic->regs);
393 }
394 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
395 
396 static inline int apic_search_irr(struct kvm_lapic *apic)
397 {
398 	return find_highest_vector(apic->regs + APIC_IRR);
399 }
400 
401 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
402 {
403 	int result;
404 
405 	/*
406 	 * Note that irr_pending is just a hint. It will be always
407 	 * true with virtual interrupt delivery enabled.
408 	 */
409 	if (!apic->irr_pending)
410 		return -1;
411 
412 	result = apic_search_irr(apic);
413 	ASSERT(result == -1 || result >= 16);
414 
415 	return result;
416 }
417 
418 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
419 {
420 	struct kvm_vcpu *vcpu;
421 
422 	vcpu = apic->vcpu;
423 
424 	if (unlikely(vcpu->arch.apicv_active)) {
425 		/* need to update RVI */
426 		apic_clear_vector(vec, apic->regs + APIC_IRR);
427 		kvm_x86_ops->hwapic_irr_update(vcpu,
428 				apic_find_highest_irr(apic));
429 	} else {
430 		apic->irr_pending = false;
431 		apic_clear_vector(vec, apic->regs + APIC_IRR);
432 		if (apic_search_irr(apic) != -1)
433 			apic->irr_pending = true;
434 	}
435 }
436 
437 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
438 {
439 	struct kvm_vcpu *vcpu;
440 
441 	if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
442 		return;
443 
444 	vcpu = apic->vcpu;
445 
446 	/*
447 	 * With APIC virtualization enabled, all caching is disabled
448 	 * because the processor can modify ISR under the hood.  Instead
449 	 * just set SVI.
450 	 */
451 	if (unlikely(vcpu->arch.apicv_active))
452 		kvm_x86_ops->hwapic_isr_update(vcpu, vec);
453 	else {
454 		++apic->isr_count;
455 		BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
456 		/*
457 		 * ISR (in service register) bit is set when injecting an interrupt.
458 		 * The highest vector is injected. Thus the latest bit set matches
459 		 * the highest bit in ISR.
460 		 */
461 		apic->highest_isr_cache = vec;
462 	}
463 }
464 
465 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
466 {
467 	int result;
468 
469 	/*
470 	 * Note that isr_count is always 1, and highest_isr_cache
471 	 * is always -1, with APIC virtualization enabled.
472 	 */
473 	if (!apic->isr_count)
474 		return -1;
475 	if (likely(apic->highest_isr_cache != -1))
476 		return apic->highest_isr_cache;
477 
478 	result = find_highest_vector(apic->regs + APIC_ISR);
479 	ASSERT(result == -1 || result >= 16);
480 
481 	return result;
482 }
483 
484 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
485 {
486 	struct kvm_vcpu *vcpu;
487 	if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
488 		return;
489 
490 	vcpu = apic->vcpu;
491 
492 	/*
493 	 * We do get here for APIC virtualization enabled if the guest
494 	 * uses the Hyper-V APIC enlightenment.  In this case we may need
495 	 * to trigger a new interrupt delivery by writing the SVI field;
496 	 * on the other hand isr_count and highest_isr_cache are unused
497 	 * and must be left alone.
498 	 */
499 	if (unlikely(vcpu->arch.apicv_active))
500 		kvm_x86_ops->hwapic_isr_update(vcpu,
501 					       apic_find_highest_isr(apic));
502 	else {
503 		--apic->isr_count;
504 		BUG_ON(apic->isr_count < 0);
505 		apic->highest_isr_cache = -1;
506 	}
507 }
508 
509 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
510 {
511 	/* This may race with setting of irr in __apic_accept_irq() and
512 	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
513 	 * will cause vmexit immediately and the value will be recalculated
514 	 * on the next vmentry.
515 	 */
516 	return apic_find_highest_irr(vcpu->arch.apic);
517 }
518 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
519 
520 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
521 			     int vector, int level, int trig_mode,
522 			     struct dest_map *dest_map);
523 
524 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
525 		     struct dest_map *dest_map)
526 {
527 	struct kvm_lapic *apic = vcpu->arch.apic;
528 
529 	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
530 			irq->level, irq->trig_mode, dest_map);
531 }
532 
533 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
534 {
535 
536 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
537 				      sizeof(val));
538 }
539 
540 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
541 {
542 
543 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
544 				      sizeof(*val));
545 }
546 
547 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
548 {
549 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
550 }
551 
552 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
553 {
554 	u8 val;
555 	if (pv_eoi_get_user(vcpu, &val) < 0)
556 		apic_debug("Can't read EOI MSR value: 0x%llx\n",
557 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
558 	return val & 0x1;
559 }
560 
561 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
562 {
563 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
564 		apic_debug("Can't set EOI MSR value: 0x%llx\n",
565 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
566 		return;
567 	}
568 	__set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
569 }
570 
571 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
572 {
573 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
574 		apic_debug("Can't clear EOI MSR value: 0x%llx\n",
575 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
576 		return;
577 	}
578 	__clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
579 }
580 
581 static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
582 {
583 	int highest_irr;
584 	if (kvm_x86_ops->sync_pir_to_irr && apic->vcpu->arch.apicv_active)
585 		highest_irr = kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
586 	else
587 		highest_irr = apic_find_highest_irr(apic);
588 	if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
589 		return -1;
590 	return highest_irr;
591 }
592 
593 static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
594 {
595 	u32 tpr, isrv, ppr, old_ppr;
596 	int isr;
597 
598 	old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
599 	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
600 	isr = apic_find_highest_isr(apic);
601 	isrv = (isr != -1) ? isr : 0;
602 
603 	if ((tpr & 0xf0) >= (isrv & 0xf0))
604 		ppr = tpr & 0xff;
605 	else
606 		ppr = isrv & 0xf0;
607 
608 	apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
609 		   apic, ppr, isr, isrv);
610 
611 	*new_ppr = ppr;
612 	if (old_ppr != ppr)
613 		kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
614 
615 	return ppr < old_ppr;
616 }
617 
618 static void apic_update_ppr(struct kvm_lapic *apic)
619 {
620 	u32 ppr;
621 
622 	if (__apic_update_ppr(apic, &ppr) &&
623 	    apic_has_interrupt_for_ppr(apic, ppr) != -1)
624 		kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
625 }
626 
627 void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
628 {
629 	apic_update_ppr(vcpu->arch.apic);
630 }
631 EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
632 
633 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
634 {
635 	kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
636 	apic_update_ppr(apic);
637 }
638 
639 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
640 {
641 	return mda == (apic_x2apic_mode(apic) ?
642 			X2APIC_BROADCAST : APIC_BROADCAST);
643 }
644 
645 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
646 {
647 	if (kvm_apic_broadcast(apic, mda))
648 		return true;
649 
650 	if (apic_x2apic_mode(apic))
651 		return mda == kvm_x2apic_id(apic);
652 
653 	/*
654 	 * Hotplug hack: Make LAPIC in xAPIC mode also accept interrupts as if
655 	 * it were in x2APIC mode.  Hotplugged VCPUs start in xAPIC mode and
656 	 * this allows unique addressing of VCPUs with APIC ID over 0xff.
657 	 * The 0xff condition is needed because writeable xAPIC ID.
658 	 */
659 	if (kvm_x2apic_id(apic) > 0xff && mda == kvm_x2apic_id(apic))
660 		return true;
661 
662 	return mda == kvm_xapic_id(apic);
663 }
664 
665 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
666 {
667 	u32 logical_id;
668 
669 	if (kvm_apic_broadcast(apic, mda))
670 		return true;
671 
672 	logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
673 
674 	if (apic_x2apic_mode(apic))
675 		return ((logical_id >> 16) == (mda >> 16))
676 		       && (logical_id & mda & 0xffff) != 0;
677 
678 	logical_id = GET_APIC_LOGICAL_ID(logical_id);
679 
680 	switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
681 	case APIC_DFR_FLAT:
682 		return (logical_id & mda) != 0;
683 	case APIC_DFR_CLUSTER:
684 		return ((logical_id >> 4) == (mda >> 4))
685 		       && (logical_id & mda & 0xf) != 0;
686 	default:
687 		apic_debug("Bad DFR vcpu %d: %08x\n",
688 			   apic->vcpu->vcpu_id, kvm_lapic_get_reg(apic, APIC_DFR));
689 		return false;
690 	}
691 }
692 
693 /* The KVM local APIC implementation has two quirks:
694  *
695  *  - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
696  *    in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
697  *    KVM doesn't do that aliasing.
698  *
699  *  - in-kernel IOAPIC messages have to be delivered directly to
700  *    x2APIC, because the kernel does not support interrupt remapping.
701  *    In order to support broadcast without interrupt remapping, x2APIC
702  *    rewrites the destination of non-IPI messages from APIC_BROADCAST
703  *    to X2APIC_BROADCAST.
704  *
705  * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API.  This is
706  * important when userspace wants to use x2APIC-format MSIs, because
707  * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
708  */
709 static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
710 		struct kvm_lapic *source, struct kvm_lapic *target)
711 {
712 	bool ipi = source != NULL;
713 
714 	if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
715 	    !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
716 		return X2APIC_BROADCAST;
717 
718 	return dest_id;
719 }
720 
721 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
722 			   int short_hand, unsigned int dest, int dest_mode)
723 {
724 	struct kvm_lapic *target = vcpu->arch.apic;
725 	u32 mda = kvm_apic_mda(vcpu, dest, source, target);
726 
727 	apic_debug("target %p, source %p, dest 0x%x, "
728 		   "dest_mode 0x%x, short_hand 0x%x\n",
729 		   target, source, dest, dest_mode, short_hand);
730 
731 	ASSERT(target);
732 	switch (short_hand) {
733 	case APIC_DEST_NOSHORT:
734 		if (dest_mode == APIC_DEST_PHYSICAL)
735 			return kvm_apic_match_physical_addr(target, mda);
736 		else
737 			return kvm_apic_match_logical_addr(target, mda);
738 	case APIC_DEST_SELF:
739 		return target == source;
740 	case APIC_DEST_ALLINC:
741 		return true;
742 	case APIC_DEST_ALLBUT:
743 		return target != source;
744 	default:
745 		apic_debug("kvm: apic: Bad dest shorthand value %x\n",
746 			   short_hand);
747 		return false;
748 	}
749 }
750 EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
751 
752 int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
753 		       const unsigned long *bitmap, u32 bitmap_size)
754 {
755 	u32 mod;
756 	int i, idx = -1;
757 
758 	mod = vector % dest_vcpus;
759 
760 	for (i = 0; i <= mod; i++) {
761 		idx = find_next_bit(bitmap, bitmap_size, idx + 1);
762 		BUG_ON(idx == bitmap_size);
763 	}
764 
765 	return idx;
766 }
767 
768 static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
769 {
770 	if (!kvm->arch.disabled_lapic_found) {
771 		kvm->arch.disabled_lapic_found = true;
772 		printk(KERN_INFO
773 		       "Disabled LAPIC found during irq injection\n");
774 	}
775 }
776 
777 static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
778 		struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
779 {
780 	if (kvm->arch.x2apic_broadcast_quirk_disabled) {
781 		if ((irq->dest_id == APIC_BROADCAST &&
782 				map->mode != KVM_APIC_MODE_X2APIC))
783 			return true;
784 		if (irq->dest_id == X2APIC_BROADCAST)
785 			return true;
786 	} else {
787 		bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
788 		if (irq->dest_id == (x2apic_ipi ?
789 		                     X2APIC_BROADCAST : APIC_BROADCAST))
790 			return true;
791 	}
792 
793 	return false;
794 }
795 
796 /* Return true if the interrupt can be handled by using *bitmap as index mask
797  * for valid destinations in *dst array.
798  * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
799  * Note: we may have zero kvm_lapic destinations when we return true, which
800  * means that the interrupt should be dropped.  In this case, *bitmap would be
801  * zero and *dst undefined.
802  */
803 static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
804 		struct kvm_lapic **src, struct kvm_lapic_irq *irq,
805 		struct kvm_apic_map *map, struct kvm_lapic ***dst,
806 		unsigned long *bitmap)
807 {
808 	int i, lowest;
809 
810 	if (irq->shorthand == APIC_DEST_SELF && src) {
811 		*dst = src;
812 		*bitmap = 1;
813 		return true;
814 	} else if (irq->shorthand)
815 		return false;
816 
817 	if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
818 		return false;
819 
820 	if (irq->dest_mode == APIC_DEST_PHYSICAL) {
821 		if (irq->dest_id > map->max_apic_id) {
822 			*bitmap = 0;
823 		} else {
824 			*dst = &map->phys_map[irq->dest_id];
825 			*bitmap = 1;
826 		}
827 		return true;
828 	}
829 
830 	*bitmap = 0;
831 	if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
832 				(u16 *)bitmap))
833 		return false;
834 
835 	if (!kvm_lowest_prio_delivery(irq))
836 		return true;
837 
838 	if (!kvm_vector_hashing_enabled()) {
839 		lowest = -1;
840 		for_each_set_bit(i, bitmap, 16) {
841 			if (!(*dst)[i])
842 				continue;
843 			if (lowest < 0)
844 				lowest = i;
845 			else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
846 						(*dst)[lowest]->vcpu) < 0)
847 				lowest = i;
848 		}
849 	} else {
850 		if (!*bitmap)
851 			return true;
852 
853 		lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
854 				bitmap, 16);
855 
856 		if (!(*dst)[lowest]) {
857 			kvm_apic_disabled_lapic_found(kvm);
858 			*bitmap = 0;
859 			return true;
860 		}
861 	}
862 
863 	*bitmap = (lowest >= 0) ? 1 << lowest : 0;
864 
865 	return true;
866 }
867 
868 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
869 		struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
870 {
871 	struct kvm_apic_map *map;
872 	unsigned long bitmap;
873 	struct kvm_lapic **dst = NULL;
874 	int i;
875 	bool ret;
876 
877 	*r = -1;
878 
879 	if (irq->shorthand == APIC_DEST_SELF) {
880 		*r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
881 		return true;
882 	}
883 
884 	rcu_read_lock();
885 	map = rcu_dereference(kvm->arch.apic_map);
886 
887 	ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
888 	if (ret)
889 		for_each_set_bit(i, &bitmap, 16) {
890 			if (!dst[i])
891 				continue;
892 			if (*r < 0)
893 				*r = 0;
894 			*r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
895 		}
896 
897 	rcu_read_unlock();
898 	return ret;
899 }
900 
901 /*
902  * This routine tries to handler interrupts in posted mode, here is how
903  * it deals with different cases:
904  * - For single-destination interrupts, handle it in posted mode
905  * - Else if vector hashing is enabled and it is a lowest-priority
906  *   interrupt, handle it in posted mode and use the following mechanism
907  *   to find the destinaiton vCPU.
908  *	1. For lowest-priority interrupts, store all the possible
909  *	   destination vCPUs in an array.
910  *	2. Use "guest vector % max number of destination vCPUs" to find
911  *	   the right destination vCPU in the array for the lowest-priority
912  *	   interrupt.
913  * - Otherwise, use remapped mode to inject the interrupt.
914  */
915 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
916 			struct kvm_vcpu **dest_vcpu)
917 {
918 	struct kvm_apic_map *map;
919 	unsigned long bitmap;
920 	struct kvm_lapic **dst = NULL;
921 	bool ret = false;
922 
923 	if (irq->shorthand)
924 		return false;
925 
926 	rcu_read_lock();
927 	map = rcu_dereference(kvm->arch.apic_map);
928 
929 	if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
930 			hweight16(bitmap) == 1) {
931 		unsigned long i = find_first_bit(&bitmap, 16);
932 
933 		if (dst[i]) {
934 			*dest_vcpu = dst[i]->vcpu;
935 			ret = true;
936 		}
937 	}
938 
939 	rcu_read_unlock();
940 	return ret;
941 }
942 
943 /*
944  * Add a pending IRQ into lapic.
945  * Return 1 if successfully added and 0 if discarded.
946  */
947 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
948 			     int vector, int level, int trig_mode,
949 			     struct dest_map *dest_map)
950 {
951 	int result = 0;
952 	struct kvm_vcpu *vcpu = apic->vcpu;
953 
954 	trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
955 				  trig_mode, vector);
956 	switch (delivery_mode) {
957 	case APIC_DM_LOWEST:
958 		vcpu->arch.apic_arb_prio++;
959 	case APIC_DM_FIXED:
960 		if (unlikely(trig_mode && !level))
961 			break;
962 
963 		/* FIXME add logic for vcpu on reset */
964 		if (unlikely(!apic_enabled(apic)))
965 			break;
966 
967 		result = 1;
968 
969 		if (dest_map) {
970 			__set_bit(vcpu->vcpu_id, dest_map->map);
971 			dest_map->vectors[vcpu->vcpu_id] = vector;
972 		}
973 
974 		if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
975 			if (trig_mode)
976 				kvm_lapic_set_vector(vector, apic->regs + APIC_TMR);
977 			else
978 				apic_clear_vector(vector, apic->regs + APIC_TMR);
979 		}
980 
981 		if (vcpu->arch.apicv_active)
982 			kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
983 		else {
984 			kvm_lapic_set_irr(vector, apic);
985 
986 			kvm_make_request(KVM_REQ_EVENT, vcpu);
987 			kvm_vcpu_kick(vcpu);
988 		}
989 		break;
990 
991 	case APIC_DM_REMRD:
992 		result = 1;
993 		vcpu->arch.pv.pv_unhalted = 1;
994 		kvm_make_request(KVM_REQ_EVENT, vcpu);
995 		kvm_vcpu_kick(vcpu);
996 		break;
997 
998 	case APIC_DM_SMI:
999 		result = 1;
1000 		kvm_make_request(KVM_REQ_SMI, vcpu);
1001 		kvm_vcpu_kick(vcpu);
1002 		break;
1003 
1004 	case APIC_DM_NMI:
1005 		result = 1;
1006 		kvm_inject_nmi(vcpu);
1007 		kvm_vcpu_kick(vcpu);
1008 		break;
1009 
1010 	case APIC_DM_INIT:
1011 		if (!trig_mode || level) {
1012 			result = 1;
1013 			/* assumes that there are only KVM_APIC_INIT/SIPI */
1014 			apic->pending_events = (1UL << KVM_APIC_INIT);
1015 			/* make sure pending_events is visible before sending
1016 			 * the request */
1017 			smp_wmb();
1018 			kvm_make_request(KVM_REQ_EVENT, vcpu);
1019 			kvm_vcpu_kick(vcpu);
1020 		} else {
1021 			apic_debug("Ignoring de-assert INIT to vcpu %d\n",
1022 				   vcpu->vcpu_id);
1023 		}
1024 		break;
1025 
1026 	case APIC_DM_STARTUP:
1027 		apic_debug("SIPI to vcpu %d vector 0x%02x\n",
1028 			   vcpu->vcpu_id, vector);
1029 		result = 1;
1030 		apic->sipi_vector = vector;
1031 		/* make sure sipi_vector is visible for the receiver */
1032 		smp_wmb();
1033 		set_bit(KVM_APIC_SIPI, &apic->pending_events);
1034 		kvm_make_request(KVM_REQ_EVENT, vcpu);
1035 		kvm_vcpu_kick(vcpu);
1036 		break;
1037 
1038 	case APIC_DM_EXTINT:
1039 		/*
1040 		 * Should only be called by kvm_apic_local_deliver() with LVT0,
1041 		 * before NMI watchdog was enabled. Already handled by
1042 		 * kvm_apic_accept_pic_intr().
1043 		 */
1044 		break;
1045 
1046 	default:
1047 		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1048 		       delivery_mode);
1049 		break;
1050 	}
1051 	return result;
1052 }
1053 
1054 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1055 {
1056 	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1057 }
1058 
1059 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1060 {
1061 	return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1062 }
1063 
1064 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1065 {
1066 	int trigger_mode;
1067 
1068 	/* Eoi the ioapic only if the ioapic doesn't own the vector. */
1069 	if (!kvm_ioapic_handles_vector(apic, vector))
1070 		return;
1071 
1072 	/* Request a KVM exit to inform the userspace IOAPIC. */
1073 	if (irqchip_split(apic->vcpu->kvm)) {
1074 		apic->vcpu->arch.pending_ioapic_eoi = vector;
1075 		kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1076 		return;
1077 	}
1078 
1079 	if (apic_test_vector(vector, apic->regs + APIC_TMR))
1080 		trigger_mode = IOAPIC_LEVEL_TRIG;
1081 	else
1082 		trigger_mode = IOAPIC_EDGE_TRIG;
1083 
1084 	kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1085 }
1086 
1087 static int apic_set_eoi(struct kvm_lapic *apic)
1088 {
1089 	int vector = apic_find_highest_isr(apic);
1090 
1091 	trace_kvm_eoi(apic, vector);
1092 
1093 	/*
1094 	 * Not every write EOI will has corresponding ISR,
1095 	 * one example is when Kernel check timer on setup_IO_APIC
1096 	 */
1097 	if (vector == -1)
1098 		return vector;
1099 
1100 	apic_clear_isr(vector, apic);
1101 	apic_update_ppr(apic);
1102 
1103 	if (test_bit(vector, vcpu_to_synic(apic->vcpu)->vec_bitmap))
1104 		kvm_hv_synic_send_eoi(apic->vcpu, vector);
1105 
1106 	kvm_ioapic_send_eoi(apic, vector);
1107 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1108 	return vector;
1109 }
1110 
1111 /*
1112  * this interface assumes a trap-like exit, which has already finished
1113  * desired side effect including vISR and vPPR update.
1114  */
1115 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1116 {
1117 	struct kvm_lapic *apic = vcpu->arch.apic;
1118 
1119 	trace_kvm_eoi(apic, vector);
1120 
1121 	kvm_ioapic_send_eoi(apic, vector);
1122 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1123 }
1124 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1125 
1126 static void apic_send_ipi(struct kvm_lapic *apic)
1127 {
1128 	u32 icr_low = kvm_lapic_get_reg(apic, APIC_ICR);
1129 	u32 icr_high = kvm_lapic_get_reg(apic, APIC_ICR2);
1130 	struct kvm_lapic_irq irq;
1131 
1132 	irq.vector = icr_low & APIC_VECTOR_MASK;
1133 	irq.delivery_mode = icr_low & APIC_MODE_MASK;
1134 	irq.dest_mode = icr_low & APIC_DEST_MASK;
1135 	irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1136 	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1137 	irq.shorthand = icr_low & APIC_SHORT_MASK;
1138 	irq.msi_redir_hint = false;
1139 	if (apic_x2apic_mode(apic))
1140 		irq.dest_id = icr_high;
1141 	else
1142 		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1143 
1144 	trace_kvm_apic_ipi(icr_low, irq.dest_id);
1145 
1146 	apic_debug("icr_high 0x%x, icr_low 0x%x, "
1147 		   "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
1148 		   "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x, "
1149 		   "msi_redir_hint 0x%x\n",
1150 		   icr_high, icr_low, irq.shorthand, irq.dest_id,
1151 		   irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
1152 		   irq.vector, irq.msi_redir_hint);
1153 
1154 	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1155 }
1156 
1157 static u32 apic_get_tmcct(struct kvm_lapic *apic)
1158 {
1159 	ktime_t remaining, now;
1160 	s64 ns;
1161 	u32 tmcct;
1162 
1163 	ASSERT(apic != NULL);
1164 
1165 	/* if initial count is 0, current count should also be 0 */
1166 	if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1167 		apic->lapic_timer.period == 0)
1168 		return 0;
1169 
1170 	now = ktime_get();
1171 	remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1172 	if (ktime_to_ns(remaining) < 0)
1173 		remaining = 0;
1174 
1175 	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1176 	tmcct = div64_u64(ns,
1177 			 (APIC_BUS_CYCLE_NS * apic->divide_count));
1178 
1179 	return tmcct;
1180 }
1181 
1182 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1183 {
1184 	struct kvm_vcpu *vcpu = apic->vcpu;
1185 	struct kvm_run *run = vcpu->run;
1186 
1187 	kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1188 	run->tpr_access.rip = kvm_rip_read(vcpu);
1189 	run->tpr_access.is_write = write;
1190 }
1191 
1192 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1193 {
1194 	if (apic->vcpu->arch.tpr_access_reporting)
1195 		__report_tpr_access(apic, write);
1196 }
1197 
1198 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1199 {
1200 	u32 val = 0;
1201 
1202 	if (offset >= LAPIC_MMIO_LENGTH)
1203 		return 0;
1204 
1205 	switch (offset) {
1206 	case APIC_ARBPRI:
1207 		apic_debug("Access APIC ARBPRI register which is for P6\n");
1208 		break;
1209 
1210 	case APIC_TMCCT:	/* Timer CCR */
1211 		if (apic_lvtt_tscdeadline(apic))
1212 			return 0;
1213 
1214 		val = apic_get_tmcct(apic);
1215 		break;
1216 	case APIC_PROCPRI:
1217 		apic_update_ppr(apic);
1218 		val = kvm_lapic_get_reg(apic, offset);
1219 		break;
1220 	case APIC_TASKPRI:
1221 		report_tpr_access(apic, false);
1222 		/* fall thru */
1223 	default:
1224 		val = kvm_lapic_get_reg(apic, offset);
1225 		break;
1226 	}
1227 
1228 	return val;
1229 }
1230 
1231 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1232 {
1233 	return container_of(dev, struct kvm_lapic, dev);
1234 }
1235 
1236 int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1237 		void *data)
1238 {
1239 	unsigned char alignment = offset & 0xf;
1240 	u32 result;
1241 	/* this bitmask has a bit cleared for each reserved register */
1242 	static const u64 rmask = 0x43ff01ffffffe70cULL;
1243 
1244 	if ((alignment + len) > 4) {
1245 		apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1246 			   offset, len);
1247 		return 1;
1248 	}
1249 
1250 	if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1251 		apic_debug("KVM_APIC_READ: read reserved register %x\n",
1252 			   offset);
1253 		return 1;
1254 	}
1255 
1256 	result = __apic_read(apic, offset & ~0xf);
1257 
1258 	trace_kvm_apic_read(offset, result);
1259 
1260 	switch (len) {
1261 	case 1:
1262 	case 2:
1263 	case 4:
1264 		memcpy(data, (char *)&result + alignment, len);
1265 		break;
1266 	default:
1267 		printk(KERN_ERR "Local APIC read with len = %x, "
1268 		       "should be 1,2, or 4 instead\n", len);
1269 		break;
1270 	}
1271 	return 0;
1272 }
1273 EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1274 
1275 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1276 {
1277 	return kvm_apic_hw_enabled(apic) &&
1278 	    addr >= apic->base_address &&
1279 	    addr < apic->base_address + LAPIC_MMIO_LENGTH;
1280 }
1281 
1282 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1283 			   gpa_t address, int len, void *data)
1284 {
1285 	struct kvm_lapic *apic = to_lapic(this);
1286 	u32 offset = address - apic->base_address;
1287 
1288 	if (!apic_mmio_in_range(apic, address))
1289 		return -EOPNOTSUPP;
1290 
1291 	kvm_lapic_reg_read(apic, offset, len, data);
1292 
1293 	return 0;
1294 }
1295 
1296 static void update_divide_count(struct kvm_lapic *apic)
1297 {
1298 	u32 tmp1, tmp2, tdcr;
1299 
1300 	tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1301 	tmp1 = tdcr & 0xf;
1302 	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1303 	apic->divide_count = 0x1 << (tmp2 & 0x7);
1304 
1305 	apic_debug("timer divide count is 0x%x\n",
1306 				   apic->divide_count);
1307 }
1308 
1309 static void limit_periodic_timer_frequency(struct kvm_lapic *apic)
1310 {
1311 	/*
1312 	 * Do not allow the guest to program periodic timers with small
1313 	 * interval, since the hrtimers are not throttled by the host
1314 	 * scheduler.
1315 	 */
1316 	if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1317 		s64 min_period = min_timer_period_us * 1000LL;
1318 
1319 		if (apic->lapic_timer.period < min_period) {
1320 			pr_info_ratelimited(
1321 			    "kvm: vcpu %i: requested %lld ns "
1322 			    "lapic timer period limited to %lld ns\n",
1323 			    apic->vcpu->vcpu_id,
1324 			    apic->lapic_timer.period, min_period);
1325 			apic->lapic_timer.period = min_period;
1326 		}
1327 	}
1328 }
1329 
1330 static void apic_update_lvtt(struct kvm_lapic *apic)
1331 {
1332 	u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1333 			apic->lapic_timer.timer_mode_mask;
1334 
1335 	if (apic->lapic_timer.timer_mode != timer_mode) {
1336 		if (apic_lvtt_tscdeadline(apic) != (timer_mode ==
1337 				APIC_LVT_TIMER_TSCDEADLINE)) {
1338 			hrtimer_cancel(&apic->lapic_timer.timer);
1339 			kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1340 			apic->lapic_timer.period = 0;
1341 			apic->lapic_timer.tscdeadline = 0;
1342 		}
1343 		apic->lapic_timer.timer_mode = timer_mode;
1344 		limit_periodic_timer_frequency(apic);
1345 	}
1346 }
1347 
1348 static void apic_timer_expired(struct kvm_lapic *apic)
1349 {
1350 	struct kvm_vcpu *vcpu = apic->vcpu;
1351 	struct swait_queue_head *q = &vcpu->wq;
1352 	struct kvm_timer *ktimer = &apic->lapic_timer;
1353 
1354 	if (atomic_read(&apic->lapic_timer.pending))
1355 		return;
1356 
1357 	atomic_inc(&apic->lapic_timer.pending);
1358 	kvm_set_pending_timer(vcpu);
1359 
1360 	/*
1361 	 * For x86, the atomic_inc() is serialized, thus
1362 	 * using swait_active() is safe.
1363 	 */
1364 	if (swait_active(q))
1365 		swake_up(q);
1366 
1367 	if (apic_lvtt_tscdeadline(apic))
1368 		ktimer->expired_tscdeadline = ktimer->tscdeadline;
1369 }
1370 
1371 /*
1372  * On APICv, this test will cause a busy wait
1373  * during a higher-priority task.
1374  */
1375 
1376 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1377 {
1378 	struct kvm_lapic *apic = vcpu->arch.apic;
1379 	u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1380 
1381 	if (kvm_apic_hw_enabled(apic)) {
1382 		int vec = reg & APIC_VECTOR_MASK;
1383 		void *bitmap = apic->regs + APIC_ISR;
1384 
1385 		if (vcpu->arch.apicv_active)
1386 			bitmap = apic->regs + APIC_IRR;
1387 
1388 		if (apic_test_vector(vec, bitmap))
1389 			return true;
1390 	}
1391 	return false;
1392 }
1393 
1394 void wait_lapic_expire(struct kvm_vcpu *vcpu)
1395 {
1396 	struct kvm_lapic *apic = vcpu->arch.apic;
1397 	u64 guest_tsc, tsc_deadline;
1398 
1399 	if (!lapic_in_kernel(vcpu))
1400 		return;
1401 
1402 	if (apic->lapic_timer.expired_tscdeadline == 0)
1403 		return;
1404 
1405 	if (!lapic_timer_int_injected(vcpu))
1406 		return;
1407 
1408 	tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1409 	apic->lapic_timer.expired_tscdeadline = 0;
1410 	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1411 	trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1412 
1413 	/* __delay is delay_tsc whenever the hardware has TSC, thus always.  */
1414 	if (guest_tsc < tsc_deadline)
1415 		__delay(min(tsc_deadline - guest_tsc,
1416 			nsec_to_cycles(vcpu, lapic_timer_advance_ns)));
1417 }
1418 
1419 static void start_sw_tscdeadline(struct kvm_lapic *apic)
1420 {
1421 	u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1422 	u64 ns = 0;
1423 	ktime_t expire;
1424 	struct kvm_vcpu *vcpu = apic->vcpu;
1425 	unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1426 	unsigned long flags;
1427 	ktime_t now;
1428 
1429 	if (unlikely(!tscdeadline || !this_tsc_khz))
1430 		return;
1431 
1432 	local_irq_save(flags);
1433 
1434 	now = ktime_get();
1435 	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1436 	if (likely(tscdeadline > guest_tsc)) {
1437 		ns = (tscdeadline - guest_tsc) * 1000000ULL;
1438 		do_div(ns, this_tsc_khz);
1439 		expire = ktime_add_ns(now, ns);
1440 		expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1441 		hrtimer_start(&apic->lapic_timer.timer,
1442 				expire, HRTIMER_MODE_ABS_PINNED);
1443 	} else
1444 		apic_timer_expired(apic);
1445 
1446 	local_irq_restore(flags);
1447 }
1448 
1449 static void start_sw_period(struct kvm_lapic *apic)
1450 {
1451 	if (!apic->lapic_timer.period)
1452 		return;
1453 
1454 	if (apic_lvtt_oneshot(apic) &&
1455 	    ktime_after(ktime_get(),
1456 			apic->lapic_timer.target_expiration)) {
1457 		apic_timer_expired(apic);
1458 		return;
1459 	}
1460 
1461 	hrtimer_start(&apic->lapic_timer.timer,
1462 		apic->lapic_timer.target_expiration,
1463 		HRTIMER_MODE_ABS_PINNED);
1464 }
1465 
1466 static void update_target_expiration(struct kvm_lapic *apic, uint32_t old_divisor)
1467 {
1468 	ktime_t now, remaining;
1469 	u64 ns_remaining_old, ns_remaining_new;
1470 
1471 	apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1472 		* APIC_BUS_CYCLE_NS * apic->divide_count;
1473 	limit_periodic_timer_frequency(apic);
1474 
1475 	now = ktime_get();
1476 	remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1477 	if (ktime_to_ns(remaining) < 0)
1478 		remaining = 0;
1479 
1480 	ns_remaining_old = ktime_to_ns(remaining);
1481 	ns_remaining_new = mul_u64_u32_div(ns_remaining_old,
1482 	                                   apic->divide_count, old_divisor);
1483 
1484 	apic->lapic_timer.tscdeadline +=
1485 		nsec_to_cycles(apic->vcpu, ns_remaining_new) -
1486 		nsec_to_cycles(apic->vcpu, ns_remaining_old);
1487 	apic->lapic_timer.target_expiration = ktime_add_ns(now, ns_remaining_new);
1488 }
1489 
1490 static bool set_target_expiration(struct kvm_lapic *apic)
1491 {
1492 	ktime_t now;
1493 	u64 tscl = rdtsc();
1494 
1495 	now = ktime_get();
1496 	apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1497 		* APIC_BUS_CYCLE_NS * apic->divide_count;
1498 
1499 	if (!apic->lapic_timer.period) {
1500 		apic->lapic_timer.tscdeadline = 0;
1501 		return false;
1502 	}
1503 
1504 	limit_periodic_timer_frequency(apic);
1505 
1506 	apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1507 		   PRIx64 ", "
1508 		   "timer initial count 0x%x, period %lldns, "
1509 		   "expire @ 0x%016" PRIx64 ".\n", __func__,
1510 		   APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1511 		   kvm_lapic_get_reg(apic, APIC_TMICT),
1512 		   apic->lapic_timer.period,
1513 		   ktime_to_ns(ktime_add_ns(now,
1514 				apic->lapic_timer.period)));
1515 
1516 	apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1517 		nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1518 	apic->lapic_timer.target_expiration = ktime_add_ns(now, apic->lapic_timer.period);
1519 
1520 	return true;
1521 }
1522 
1523 static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1524 {
1525 	apic->lapic_timer.tscdeadline +=
1526 		nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1527 	apic->lapic_timer.target_expiration =
1528 		ktime_add_ns(apic->lapic_timer.target_expiration,
1529 				apic->lapic_timer.period);
1530 }
1531 
1532 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1533 {
1534 	if (!lapic_in_kernel(vcpu))
1535 		return false;
1536 
1537 	return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1538 }
1539 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1540 
1541 static void cancel_hv_timer(struct kvm_lapic *apic)
1542 {
1543 	WARN_ON(preemptible());
1544 	WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1545 	kvm_x86_ops->cancel_hv_timer(apic->vcpu);
1546 	apic->lapic_timer.hv_timer_in_use = false;
1547 }
1548 
1549 static bool start_hv_timer(struct kvm_lapic *apic)
1550 {
1551 	struct kvm_timer *ktimer = &apic->lapic_timer;
1552 	int r;
1553 
1554 	WARN_ON(preemptible());
1555 	if (!kvm_x86_ops->set_hv_timer)
1556 		return false;
1557 
1558 	if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1559 		return false;
1560 
1561 	if (!ktimer->tscdeadline)
1562 		return false;
1563 
1564 	r = kvm_x86_ops->set_hv_timer(apic->vcpu, ktimer->tscdeadline);
1565 	if (r < 0)
1566 		return false;
1567 
1568 	ktimer->hv_timer_in_use = true;
1569 	hrtimer_cancel(&ktimer->timer);
1570 
1571 	/*
1572 	 * Also recheck ktimer->pending, in case the sw timer triggered in
1573 	 * the window.  For periodic timer, leave the hv timer running for
1574 	 * simplicity, and the deadline will be recomputed on the next vmexit.
1575 	 */
1576 	if (!apic_lvtt_period(apic) && (r || atomic_read(&ktimer->pending))) {
1577 		if (r)
1578 			apic_timer_expired(apic);
1579 		return false;
1580 	}
1581 
1582 	trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, true);
1583 	return true;
1584 }
1585 
1586 static void start_sw_timer(struct kvm_lapic *apic)
1587 {
1588 	struct kvm_timer *ktimer = &apic->lapic_timer;
1589 
1590 	WARN_ON(preemptible());
1591 	if (apic->lapic_timer.hv_timer_in_use)
1592 		cancel_hv_timer(apic);
1593 	if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1594 		return;
1595 
1596 	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1597 		start_sw_period(apic);
1598 	else if (apic_lvtt_tscdeadline(apic))
1599 		start_sw_tscdeadline(apic);
1600 	trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, false);
1601 }
1602 
1603 static void restart_apic_timer(struct kvm_lapic *apic)
1604 {
1605 	preempt_disable();
1606 	if (!start_hv_timer(apic))
1607 		start_sw_timer(apic);
1608 	preempt_enable();
1609 }
1610 
1611 void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1612 {
1613 	struct kvm_lapic *apic = vcpu->arch.apic;
1614 
1615 	preempt_disable();
1616 	/* If the preempt notifier has already run, it also called apic_timer_expired */
1617 	if (!apic->lapic_timer.hv_timer_in_use)
1618 		goto out;
1619 	WARN_ON(swait_active(&vcpu->wq));
1620 	cancel_hv_timer(apic);
1621 	apic_timer_expired(apic);
1622 
1623 	if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1624 		advance_periodic_target_expiration(apic);
1625 		restart_apic_timer(apic);
1626 	}
1627 out:
1628 	preempt_enable();
1629 }
1630 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1631 
1632 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1633 {
1634 	restart_apic_timer(vcpu->arch.apic);
1635 }
1636 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1637 
1638 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1639 {
1640 	struct kvm_lapic *apic = vcpu->arch.apic;
1641 
1642 	preempt_disable();
1643 	/* Possibly the TSC deadline timer is not enabled yet */
1644 	if (apic->lapic_timer.hv_timer_in_use)
1645 		start_sw_timer(apic);
1646 	preempt_enable();
1647 }
1648 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1649 
1650 void kvm_lapic_restart_hv_timer(struct kvm_vcpu *vcpu)
1651 {
1652 	struct kvm_lapic *apic = vcpu->arch.apic;
1653 
1654 	WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1655 	restart_apic_timer(apic);
1656 }
1657 
1658 static void start_apic_timer(struct kvm_lapic *apic)
1659 {
1660 	atomic_set(&apic->lapic_timer.pending, 0);
1661 
1662 	if ((apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1663 	    && !set_target_expiration(apic))
1664 		return;
1665 
1666 	restart_apic_timer(apic);
1667 }
1668 
1669 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1670 {
1671 	bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1672 
1673 	if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1674 		apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1675 		if (lvt0_in_nmi_mode) {
1676 			apic_debug("Receive NMI setting on APIC_LVT0 "
1677 				   "for cpu %d\n", apic->vcpu->vcpu_id);
1678 			atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1679 		} else
1680 			atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1681 	}
1682 }
1683 
1684 int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1685 {
1686 	int ret = 0;
1687 
1688 	trace_kvm_apic_write(reg, val);
1689 
1690 	switch (reg) {
1691 	case APIC_ID:		/* Local APIC ID */
1692 		if (!apic_x2apic_mode(apic))
1693 			kvm_apic_set_xapic_id(apic, val >> 24);
1694 		else
1695 			ret = 1;
1696 		break;
1697 
1698 	case APIC_TASKPRI:
1699 		report_tpr_access(apic, true);
1700 		apic_set_tpr(apic, val & 0xff);
1701 		break;
1702 
1703 	case APIC_EOI:
1704 		apic_set_eoi(apic);
1705 		break;
1706 
1707 	case APIC_LDR:
1708 		if (!apic_x2apic_mode(apic))
1709 			kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1710 		else
1711 			ret = 1;
1712 		break;
1713 
1714 	case APIC_DFR:
1715 		if (!apic_x2apic_mode(apic)) {
1716 			kvm_lapic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1717 			recalculate_apic_map(apic->vcpu->kvm);
1718 		} else
1719 			ret = 1;
1720 		break;
1721 
1722 	case APIC_SPIV: {
1723 		u32 mask = 0x3ff;
1724 		if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1725 			mask |= APIC_SPIV_DIRECTED_EOI;
1726 		apic_set_spiv(apic, val & mask);
1727 		if (!(val & APIC_SPIV_APIC_ENABLED)) {
1728 			int i;
1729 			u32 lvt_val;
1730 
1731 			for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
1732 				lvt_val = kvm_lapic_get_reg(apic,
1733 						       APIC_LVTT + 0x10 * i);
1734 				kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
1735 					     lvt_val | APIC_LVT_MASKED);
1736 			}
1737 			apic_update_lvtt(apic);
1738 			atomic_set(&apic->lapic_timer.pending, 0);
1739 
1740 		}
1741 		break;
1742 	}
1743 	case APIC_ICR:
1744 		/* No delay here, so we always clear the pending bit */
1745 		kvm_lapic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1746 		apic_send_ipi(apic);
1747 		break;
1748 
1749 	case APIC_ICR2:
1750 		if (!apic_x2apic_mode(apic))
1751 			val &= 0xff000000;
1752 		kvm_lapic_set_reg(apic, APIC_ICR2, val);
1753 		break;
1754 
1755 	case APIC_LVT0:
1756 		apic_manage_nmi_watchdog(apic, val);
1757 	case APIC_LVTTHMR:
1758 	case APIC_LVTPC:
1759 	case APIC_LVT1:
1760 	case APIC_LVTERR:
1761 		/* TODO: Check vector */
1762 		if (!kvm_apic_sw_enabled(apic))
1763 			val |= APIC_LVT_MASKED;
1764 
1765 		val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1766 		kvm_lapic_set_reg(apic, reg, val);
1767 
1768 		break;
1769 
1770 	case APIC_LVTT:
1771 		if (!kvm_apic_sw_enabled(apic))
1772 			val |= APIC_LVT_MASKED;
1773 		val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1774 		kvm_lapic_set_reg(apic, APIC_LVTT, val);
1775 		apic_update_lvtt(apic);
1776 		break;
1777 
1778 	case APIC_TMICT:
1779 		if (apic_lvtt_tscdeadline(apic))
1780 			break;
1781 
1782 		hrtimer_cancel(&apic->lapic_timer.timer);
1783 		kvm_lapic_set_reg(apic, APIC_TMICT, val);
1784 		start_apic_timer(apic);
1785 		break;
1786 
1787 	case APIC_TDCR: {
1788 		uint32_t old_divisor = apic->divide_count;
1789 
1790 		if (val & 4)
1791 			apic_debug("KVM_WRITE:TDCR %x\n", val);
1792 		kvm_lapic_set_reg(apic, APIC_TDCR, val);
1793 		update_divide_count(apic);
1794 		if (apic->divide_count != old_divisor &&
1795 				apic->lapic_timer.period) {
1796 			hrtimer_cancel(&apic->lapic_timer.timer);
1797 			update_target_expiration(apic, old_divisor);
1798 			restart_apic_timer(apic);
1799 		}
1800 		break;
1801 	}
1802 	case APIC_ESR:
1803 		if (apic_x2apic_mode(apic) && val != 0) {
1804 			apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1805 			ret = 1;
1806 		}
1807 		break;
1808 
1809 	case APIC_SELF_IPI:
1810 		if (apic_x2apic_mode(apic)) {
1811 			kvm_lapic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1812 		} else
1813 			ret = 1;
1814 		break;
1815 	default:
1816 		ret = 1;
1817 		break;
1818 	}
1819 	if (ret)
1820 		apic_debug("Local APIC Write to read-only register %x\n", reg);
1821 	return ret;
1822 }
1823 EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
1824 
1825 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1826 			    gpa_t address, int len, const void *data)
1827 {
1828 	struct kvm_lapic *apic = to_lapic(this);
1829 	unsigned int offset = address - apic->base_address;
1830 	u32 val;
1831 
1832 	if (!apic_mmio_in_range(apic, address))
1833 		return -EOPNOTSUPP;
1834 
1835 	/*
1836 	 * APIC register must be aligned on 128-bits boundary.
1837 	 * 32/64/128 bits registers must be accessed thru 32 bits.
1838 	 * Refer SDM 8.4.1
1839 	 */
1840 	if (len != 4 || (offset & 0xf)) {
1841 		/* Don't shout loud, $infamous_os would cause only noise. */
1842 		apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1843 		return 0;
1844 	}
1845 
1846 	val = *(u32*)data;
1847 
1848 	/* too common printing */
1849 	if (offset != APIC_EOI)
1850 		apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1851 			   "0x%x\n", __func__, offset, len, val);
1852 
1853 	kvm_lapic_reg_write(apic, offset & 0xff0, val);
1854 
1855 	return 0;
1856 }
1857 
1858 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1859 {
1860 	kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1861 }
1862 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1863 
1864 /* emulate APIC access in a trap manner */
1865 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1866 {
1867 	u32 val = 0;
1868 
1869 	/* hw has done the conditional check and inst decode */
1870 	offset &= 0xff0;
1871 
1872 	kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
1873 
1874 	/* TODO: optimize to just emulate side effect w/o one more write */
1875 	kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
1876 }
1877 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1878 
1879 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1880 {
1881 	struct kvm_lapic *apic = vcpu->arch.apic;
1882 
1883 	if (!vcpu->arch.apic)
1884 		return;
1885 
1886 	hrtimer_cancel(&apic->lapic_timer.timer);
1887 
1888 	if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1889 		static_key_slow_dec_deferred(&apic_hw_disabled);
1890 
1891 	if (!apic->sw_enabled)
1892 		static_key_slow_dec_deferred(&apic_sw_disabled);
1893 
1894 	if (apic->regs)
1895 		free_page((unsigned long)apic->regs);
1896 
1897 	kfree(apic);
1898 }
1899 
1900 /*
1901  *----------------------------------------------------------------------
1902  * LAPIC interface
1903  *----------------------------------------------------------------------
1904  */
1905 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1906 {
1907 	struct kvm_lapic *apic = vcpu->arch.apic;
1908 
1909 	if (!lapic_in_kernel(vcpu) ||
1910 		!apic_lvtt_tscdeadline(apic))
1911 		return 0;
1912 
1913 	return apic->lapic_timer.tscdeadline;
1914 }
1915 
1916 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1917 {
1918 	struct kvm_lapic *apic = vcpu->arch.apic;
1919 
1920 	if (!lapic_in_kernel(vcpu) || apic_lvtt_oneshot(apic) ||
1921 			apic_lvtt_period(apic))
1922 		return;
1923 
1924 	hrtimer_cancel(&apic->lapic_timer.timer);
1925 	apic->lapic_timer.tscdeadline = data;
1926 	start_apic_timer(apic);
1927 }
1928 
1929 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1930 {
1931 	struct kvm_lapic *apic = vcpu->arch.apic;
1932 
1933 	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1934 		     | (kvm_lapic_get_reg(apic, APIC_TASKPRI) & 4));
1935 }
1936 
1937 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1938 {
1939 	u64 tpr;
1940 
1941 	tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1942 
1943 	return (tpr & 0xf0) >> 4;
1944 }
1945 
1946 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1947 {
1948 	u64 old_value = vcpu->arch.apic_base;
1949 	struct kvm_lapic *apic = vcpu->arch.apic;
1950 
1951 	if (!apic)
1952 		value |= MSR_IA32_APICBASE_BSP;
1953 
1954 	vcpu->arch.apic_base = value;
1955 
1956 	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
1957 		kvm_update_cpuid(vcpu);
1958 
1959 	if (!apic)
1960 		return;
1961 
1962 	/* update jump label if enable bit changes */
1963 	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1964 		if (value & MSR_IA32_APICBASE_ENABLE) {
1965 			kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
1966 			static_key_slow_dec_deferred(&apic_hw_disabled);
1967 		} else {
1968 			static_key_slow_inc(&apic_hw_disabled.key);
1969 			recalculate_apic_map(vcpu->kvm);
1970 		}
1971 	}
1972 
1973 	if ((old_value ^ value) & X2APIC_ENABLE) {
1974 		if (value & X2APIC_ENABLE) {
1975 			kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
1976 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1977 		} else
1978 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1979 	}
1980 
1981 	apic->base_address = apic->vcpu->arch.apic_base &
1982 			     MSR_IA32_APICBASE_BASE;
1983 
1984 	if ((value & MSR_IA32_APICBASE_ENABLE) &&
1985 	     apic->base_address != APIC_DEFAULT_PHYS_BASE)
1986 		pr_warn_once("APIC base relocation is unsupported by KVM");
1987 
1988 	/* with FSB delivery interrupt, we can restart APIC functionality */
1989 	apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1990 		   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1991 
1992 }
1993 
1994 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
1995 {
1996 	struct kvm_lapic *apic;
1997 	int i;
1998 
1999 	apic_debug("%s\n", __func__);
2000 
2001 	ASSERT(vcpu);
2002 	apic = vcpu->arch.apic;
2003 	ASSERT(apic != NULL);
2004 
2005 	/* Stop the timer in case it's a reset to an active apic */
2006 	hrtimer_cancel(&apic->lapic_timer.timer);
2007 
2008 	if (!init_event) {
2009 		kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
2010 		                         MSR_IA32_APICBASE_ENABLE);
2011 		kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2012 	}
2013 	kvm_apic_set_version(apic->vcpu);
2014 
2015 	for (i = 0; i < KVM_APIC_LVT_NUM; i++)
2016 		kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
2017 	apic_update_lvtt(apic);
2018 	if (kvm_vcpu_is_reset_bsp(vcpu) &&
2019 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
2020 		kvm_lapic_set_reg(apic, APIC_LVT0,
2021 			     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
2022 	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2023 
2024 	kvm_lapic_set_reg(apic, APIC_DFR, 0xffffffffU);
2025 	apic_set_spiv(apic, 0xff);
2026 	kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
2027 	if (!apic_x2apic_mode(apic))
2028 		kvm_apic_set_ldr(apic, 0);
2029 	kvm_lapic_set_reg(apic, APIC_ESR, 0);
2030 	kvm_lapic_set_reg(apic, APIC_ICR, 0);
2031 	kvm_lapic_set_reg(apic, APIC_ICR2, 0);
2032 	kvm_lapic_set_reg(apic, APIC_TDCR, 0);
2033 	kvm_lapic_set_reg(apic, APIC_TMICT, 0);
2034 	for (i = 0; i < 8; i++) {
2035 		kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
2036 		kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
2037 		kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
2038 	}
2039 	apic->irr_pending = vcpu->arch.apicv_active;
2040 	apic->isr_count = vcpu->arch.apicv_active ? 1 : 0;
2041 	apic->highest_isr_cache = -1;
2042 	update_divide_count(apic);
2043 	atomic_set(&apic->lapic_timer.pending, 0);
2044 	if (kvm_vcpu_is_bsp(vcpu))
2045 		kvm_lapic_set_base(vcpu,
2046 				vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
2047 	vcpu->arch.pv_eoi.msr_val = 0;
2048 	apic_update_ppr(apic);
2049 	if (vcpu->arch.apicv_active) {
2050 		kvm_x86_ops->apicv_post_state_restore(vcpu);
2051 		kvm_x86_ops->hwapic_irr_update(vcpu, -1);
2052 		kvm_x86_ops->hwapic_isr_update(vcpu, -1);
2053 	}
2054 
2055 	vcpu->arch.apic_arb_prio = 0;
2056 	vcpu->arch.apic_attention = 0;
2057 
2058 	apic_debug("%s: vcpu=%p, id=0x%x, base_msr="
2059 		   "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
2060 		   vcpu, kvm_lapic_get_reg(apic, APIC_ID),
2061 		   vcpu->arch.apic_base, apic->base_address);
2062 }
2063 
2064 /*
2065  *----------------------------------------------------------------------
2066  * timer interface
2067  *----------------------------------------------------------------------
2068  */
2069 
2070 static bool lapic_is_periodic(struct kvm_lapic *apic)
2071 {
2072 	return apic_lvtt_period(apic);
2073 }
2074 
2075 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
2076 {
2077 	struct kvm_lapic *apic = vcpu->arch.apic;
2078 
2079 	if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
2080 		return atomic_read(&apic->lapic_timer.pending);
2081 
2082 	return 0;
2083 }
2084 
2085 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
2086 {
2087 	u32 reg = kvm_lapic_get_reg(apic, lvt_type);
2088 	int vector, mode, trig_mode;
2089 
2090 	if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2091 		vector = reg & APIC_VECTOR_MASK;
2092 		mode = reg & APIC_MODE_MASK;
2093 		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2094 		return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2095 					NULL);
2096 	}
2097 	return 0;
2098 }
2099 
2100 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2101 {
2102 	struct kvm_lapic *apic = vcpu->arch.apic;
2103 
2104 	if (apic)
2105 		kvm_apic_local_deliver(apic, APIC_LVT0);
2106 }
2107 
2108 static const struct kvm_io_device_ops apic_mmio_ops = {
2109 	.read     = apic_mmio_read,
2110 	.write    = apic_mmio_write,
2111 };
2112 
2113 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2114 {
2115 	struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2116 	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2117 
2118 	apic_timer_expired(apic);
2119 
2120 	if (lapic_is_periodic(apic)) {
2121 		advance_periodic_target_expiration(apic);
2122 		hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2123 		return HRTIMER_RESTART;
2124 	} else
2125 		return HRTIMER_NORESTART;
2126 }
2127 
2128 int kvm_create_lapic(struct kvm_vcpu *vcpu)
2129 {
2130 	struct kvm_lapic *apic;
2131 
2132 	ASSERT(vcpu != NULL);
2133 	apic_debug("apic_init %d\n", vcpu->vcpu_id);
2134 
2135 	apic = kzalloc(sizeof(*apic), GFP_KERNEL);
2136 	if (!apic)
2137 		goto nomem;
2138 
2139 	vcpu->arch.apic = apic;
2140 
2141 	apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
2142 	if (!apic->regs) {
2143 		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2144 		       vcpu->vcpu_id);
2145 		goto nomem_free_apic;
2146 	}
2147 	apic->vcpu = vcpu;
2148 
2149 	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2150 		     HRTIMER_MODE_ABS_PINNED);
2151 	apic->lapic_timer.timer.function = apic_timer_fn;
2152 
2153 	/*
2154 	 * APIC is created enabled. This will prevent kvm_lapic_set_base from
2155 	 * thinking that APIC satet has changed.
2156 	 */
2157 	vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2158 	static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2159 	kvm_lapic_reset(vcpu, false);
2160 	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2161 
2162 	return 0;
2163 nomem_free_apic:
2164 	kfree(apic);
2165 nomem:
2166 	return -ENOMEM;
2167 }
2168 
2169 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2170 {
2171 	struct kvm_lapic *apic = vcpu->arch.apic;
2172 	u32 ppr;
2173 
2174 	if (!apic_enabled(apic))
2175 		return -1;
2176 
2177 	__apic_update_ppr(apic, &ppr);
2178 	return apic_has_interrupt_for_ppr(apic, ppr);
2179 }
2180 
2181 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2182 {
2183 	u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2184 	int r = 0;
2185 
2186 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2187 		r = 1;
2188 	if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2189 	    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2190 		r = 1;
2191 	return r;
2192 }
2193 
2194 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2195 {
2196 	struct kvm_lapic *apic = vcpu->arch.apic;
2197 
2198 	if (atomic_read(&apic->lapic_timer.pending) > 0) {
2199 		kvm_apic_local_deliver(apic, APIC_LVTT);
2200 		if (apic_lvtt_tscdeadline(apic))
2201 			apic->lapic_timer.tscdeadline = 0;
2202 		if (apic_lvtt_oneshot(apic)) {
2203 			apic->lapic_timer.tscdeadline = 0;
2204 			apic->lapic_timer.target_expiration = 0;
2205 		}
2206 		atomic_set(&apic->lapic_timer.pending, 0);
2207 	}
2208 }
2209 
2210 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2211 {
2212 	int vector = kvm_apic_has_interrupt(vcpu);
2213 	struct kvm_lapic *apic = vcpu->arch.apic;
2214 	u32 ppr;
2215 
2216 	if (vector == -1)
2217 		return -1;
2218 
2219 	/*
2220 	 * We get here even with APIC virtualization enabled, if doing
2221 	 * nested virtualization and L1 runs with the "acknowledge interrupt
2222 	 * on exit" mode.  Then we cannot inject the interrupt via RVI,
2223 	 * because the process would deliver it through the IDT.
2224 	 */
2225 
2226 	apic_clear_irr(vector, apic);
2227 	if (test_bit(vector, vcpu_to_synic(vcpu)->auto_eoi_bitmap)) {
2228 		/*
2229 		 * For auto-EOI interrupts, there might be another pending
2230 		 * interrupt above PPR, so check whether to raise another
2231 		 * KVM_REQ_EVENT.
2232 		 */
2233 		apic_update_ppr(apic);
2234 	} else {
2235 		/*
2236 		 * For normal interrupts, PPR has been raised and there cannot
2237 		 * be a higher-priority pending interrupt---except if there was
2238 		 * a concurrent interrupt injection, but that would have
2239 		 * triggered KVM_REQ_EVENT already.
2240 		 */
2241 		apic_set_isr(vector, apic);
2242 		__apic_update_ppr(apic, &ppr);
2243 	}
2244 
2245 	return vector;
2246 }
2247 
2248 static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2249 		struct kvm_lapic_state *s, bool set)
2250 {
2251 	if (apic_x2apic_mode(vcpu->arch.apic)) {
2252 		u32 *id = (u32 *)(s->regs + APIC_ID);
2253 		u32 *ldr = (u32 *)(s->regs + APIC_LDR);
2254 
2255 		if (vcpu->kvm->arch.x2apic_format) {
2256 			if (*id != vcpu->vcpu_id)
2257 				return -EINVAL;
2258 		} else {
2259 			if (set)
2260 				*id >>= 24;
2261 			else
2262 				*id <<= 24;
2263 		}
2264 
2265 		/* In x2APIC mode, the LDR is fixed and based on the id */
2266 		if (set)
2267 			*ldr = kvm_apic_calc_x2apic_ldr(*id);
2268 	}
2269 
2270 	return 0;
2271 }
2272 
2273 int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2274 {
2275 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2276 	return kvm_apic_state_fixup(vcpu, s, false);
2277 }
2278 
2279 int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2280 {
2281 	struct kvm_lapic *apic = vcpu->arch.apic;
2282 	int r;
2283 
2284 
2285 	kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2286 	/* set SPIV separately to get count of SW disabled APICs right */
2287 	apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2288 
2289 	r = kvm_apic_state_fixup(vcpu, s, true);
2290 	if (r)
2291 		return r;
2292 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2293 
2294 	recalculate_apic_map(vcpu->kvm);
2295 	kvm_apic_set_version(vcpu);
2296 
2297 	apic_update_ppr(apic);
2298 	hrtimer_cancel(&apic->lapic_timer.timer);
2299 	apic_update_lvtt(apic);
2300 	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2301 	update_divide_count(apic);
2302 	start_apic_timer(apic);
2303 	apic->irr_pending = true;
2304 	apic->isr_count = vcpu->arch.apicv_active ?
2305 				1 : count_vectors(apic->regs + APIC_ISR);
2306 	apic->highest_isr_cache = -1;
2307 	if (vcpu->arch.apicv_active) {
2308 		kvm_x86_ops->apicv_post_state_restore(vcpu);
2309 		kvm_x86_ops->hwapic_irr_update(vcpu,
2310 				apic_find_highest_irr(apic));
2311 		kvm_x86_ops->hwapic_isr_update(vcpu,
2312 				apic_find_highest_isr(apic));
2313 	}
2314 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2315 	if (ioapic_in_kernel(vcpu->kvm))
2316 		kvm_rtc_eoi_tracking_restore_one(vcpu);
2317 
2318 	vcpu->arch.apic_arb_prio = 0;
2319 
2320 	return 0;
2321 }
2322 
2323 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2324 {
2325 	struct hrtimer *timer;
2326 
2327 	if (!lapic_in_kernel(vcpu))
2328 		return;
2329 
2330 	timer = &vcpu->arch.apic->lapic_timer.timer;
2331 	if (hrtimer_cancel(timer))
2332 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
2333 }
2334 
2335 /*
2336  * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2337  *
2338  * Detect whether guest triggered PV EOI since the
2339  * last entry. If yes, set EOI on guests's behalf.
2340  * Clear PV EOI in guest memory in any case.
2341  */
2342 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2343 					struct kvm_lapic *apic)
2344 {
2345 	bool pending;
2346 	int vector;
2347 	/*
2348 	 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2349 	 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2350 	 *
2351 	 * KVM_APIC_PV_EOI_PENDING is unset:
2352 	 * 	-> host disabled PV EOI.
2353 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2354 	 * 	-> host enabled PV EOI, guest did not execute EOI yet.
2355 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2356 	 * 	-> host enabled PV EOI, guest executed EOI.
2357 	 */
2358 	BUG_ON(!pv_eoi_enabled(vcpu));
2359 	pending = pv_eoi_get_pending(vcpu);
2360 	/*
2361 	 * Clear pending bit in any case: it will be set again on vmentry.
2362 	 * While this might not be ideal from performance point of view,
2363 	 * this makes sure pv eoi is only enabled when we know it's safe.
2364 	 */
2365 	pv_eoi_clr_pending(vcpu);
2366 	if (pending)
2367 		return;
2368 	vector = apic_set_eoi(apic);
2369 	trace_kvm_pv_eoi(apic, vector);
2370 }
2371 
2372 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2373 {
2374 	u32 data;
2375 
2376 	if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2377 		apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2378 
2379 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2380 		return;
2381 
2382 	if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2383 				  sizeof(u32)))
2384 		return;
2385 
2386 	apic_set_tpr(vcpu->arch.apic, data & 0xff);
2387 }
2388 
2389 /*
2390  * apic_sync_pv_eoi_to_guest - called before vmentry
2391  *
2392  * Detect whether it's safe to enable PV EOI and
2393  * if yes do so.
2394  */
2395 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2396 					struct kvm_lapic *apic)
2397 {
2398 	if (!pv_eoi_enabled(vcpu) ||
2399 	    /* IRR set or many bits in ISR: could be nested. */
2400 	    apic->irr_pending ||
2401 	    /* Cache not set: could be safe but we don't bother. */
2402 	    apic->highest_isr_cache == -1 ||
2403 	    /* Need EOI to update ioapic. */
2404 	    kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2405 		/*
2406 		 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2407 		 * so we need not do anything here.
2408 		 */
2409 		return;
2410 	}
2411 
2412 	pv_eoi_set_pending(apic->vcpu);
2413 }
2414 
2415 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2416 {
2417 	u32 data, tpr;
2418 	int max_irr, max_isr;
2419 	struct kvm_lapic *apic = vcpu->arch.apic;
2420 
2421 	apic_sync_pv_eoi_to_guest(vcpu, apic);
2422 
2423 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2424 		return;
2425 
2426 	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2427 	max_irr = apic_find_highest_irr(apic);
2428 	if (max_irr < 0)
2429 		max_irr = 0;
2430 	max_isr = apic_find_highest_isr(apic);
2431 	if (max_isr < 0)
2432 		max_isr = 0;
2433 	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2434 
2435 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2436 				sizeof(u32));
2437 }
2438 
2439 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2440 {
2441 	if (vapic_addr) {
2442 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2443 					&vcpu->arch.apic->vapic_cache,
2444 					vapic_addr, sizeof(u32)))
2445 			return -EINVAL;
2446 		__set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2447 	} else {
2448 		__clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2449 	}
2450 
2451 	vcpu->arch.apic->vapic_addr = vapic_addr;
2452 	return 0;
2453 }
2454 
2455 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2456 {
2457 	struct kvm_lapic *apic = vcpu->arch.apic;
2458 	u32 reg = (msr - APIC_BASE_MSR) << 4;
2459 
2460 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2461 		return 1;
2462 
2463 	if (reg == APIC_ICR2)
2464 		return 1;
2465 
2466 	/* if this is ICR write vector before command */
2467 	if (reg == APIC_ICR)
2468 		kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2469 	return kvm_lapic_reg_write(apic, reg, (u32)data);
2470 }
2471 
2472 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2473 {
2474 	struct kvm_lapic *apic = vcpu->arch.apic;
2475 	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2476 
2477 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2478 		return 1;
2479 
2480 	if (reg == APIC_DFR || reg == APIC_ICR2) {
2481 		apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
2482 			   reg);
2483 		return 1;
2484 	}
2485 
2486 	if (kvm_lapic_reg_read(apic, reg, 4, &low))
2487 		return 1;
2488 	if (reg == APIC_ICR)
2489 		kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2490 
2491 	*data = (((u64)high) << 32) | low;
2492 
2493 	return 0;
2494 }
2495 
2496 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2497 {
2498 	struct kvm_lapic *apic = vcpu->arch.apic;
2499 
2500 	if (!lapic_in_kernel(vcpu))
2501 		return 1;
2502 
2503 	/* if this is ICR write vector before command */
2504 	if (reg == APIC_ICR)
2505 		kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2506 	return kvm_lapic_reg_write(apic, reg, (u32)data);
2507 }
2508 
2509 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2510 {
2511 	struct kvm_lapic *apic = vcpu->arch.apic;
2512 	u32 low, high = 0;
2513 
2514 	if (!lapic_in_kernel(vcpu))
2515 		return 1;
2516 
2517 	if (kvm_lapic_reg_read(apic, reg, 4, &low))
2518 		return 1;
2519 	if (reg == APIC_ICR)
2520 		kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2521 
2522 	*data = (((u64)high) << 32) | low;
2523 
2524 	return 0;
2525 }
2526 
2527 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2528 {
2529 	u64 addr = data & ~KVM_MSR_ENABLED;
2530 	if (!IS_ALIGNED(addr, 4))
2531 		return 1;
2532 
2533 	vcpu->arch.pv_eoi.msr_val = data;
2534 	if (!pv_eoi_enabled(vcpu))
2535 		return 0;
2536 	return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
2537 					 addr, sizeof(u8));
2538 }
2539 
2540 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2541 {
2542 	struct kvm_lapic *apic = vcpu->arch.apic;
2543 	u8 sipi_vector;
2544 	unsigned long pe;
2545 
2546 	if (!lapic_in_kernel(vcpu) || !apic->pending_events)
2547 		return;
2548 
2549 	/*
2550 	 * INITs are latched while in SMM.  Because an SMM CPU cannot
2551 	 * be in KVM_MP_STATE_INIT_RECEIVED state, just eat SIPIs
2552 	 * and delay processing of INIT until the next RSM.
2553 	 */
2554 	if (is_smm(vcpu)) {
2555 		WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2556 		if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2557 			clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2558 		return;
2559 	}
2560 
2561 	pe = xchg(&apic->pending_events, 0);
2562 	if (test_bit(KVM_APIC_INIT, &pe)) {
2563 		kvm_lapic_reset(vcpu, true);
2564 		kvm_vcpu_reset(vcpu, true);
2565 		if (kvm_vcpu_is_bsp(apic->vcpu))
2566 			vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2567 		else
2568 			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2569 	}
2570 	if (test_bit(KVM_APIC_SIPI, &pe) &&
2571 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2572 		/* evaluate pending_events before reading the vector */
2573 		smp_rmb();
2574 		sipi_vector = apic->sipi_vector;
2575 		apic_debug("vcpu %d received sipi with vector # %x\n",
2576 			 vcpu->vcpu_id, sipi_vector);
2577 		kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2578 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2579 	}
2580 }
2581 
2582 void kvm_lapic_init(void)
2583 {
2584 	/* do not patch jump label more than once per second */
2585 	jump_label_rate_limit(&apic_hw_disabled, HZ);
2586 	jump_label_rate_limit(&apic_sw_disabled, HZ);
2587 }
2588 
2589 void kvm_lapic_exit(void)
2590 {
2591 	static_key_deferred_flush(&apic_hw_disabled);
2592 	static_key_deferred_flush(&apic_sw_disabled);
2593 }
2594