xref: /linux/arch/x86/kvm/lapic.c (revision 3932b9ca55b0be314a36d3e84faff3e823c081f5)
1 
2 /*
3  * Local APIC virtualization
4  *
5  * Copyright (C) 2006 Qumranet, Inc.
6  * Copyright (C) 2007 Novell
7  * Copyright (C) 2007 Intel
8  * Copyright 2009 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Dor Laor <dor.laor@qumranet.com>
12  *   Gregory Haskins <ghaskins@novell.com>
13  *   Yaozu (Eddie) Dong <eddie.dong@intel.com>
14  *
15  * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
16  *
17  * This work is licensed under the terms of the GNU GPL, version 2.  See
18  * the COPYING file in the top-level directory.
19  */
20 
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/mm.h>
24 #include <linux/highmem.h>
25 #include <linux/smp.h>
26 #include <linux/hrtimer.h>
27 #include <linux/io.h>
28 #include <linux/module.h>
29 #include <linux/math64.h>
30 #include <linux/slab.h>
31 #include <asm/processor.h>
32 #include <asm/msr.h>
33 #include <asm/page.h>
34 #include <asm/current.h>
35 #include <asm/apicdef.h>
36 #include <linux/atomic.h>
37 #include <linux/jump_label.h>
38 #include "kvm_cache_regs.h"
39 #include "irq.h"
40 #include "trace.h"
41 #include "x86.h"
42 #include "cpuid.h"
43 
44 #ifndef CONFIG_X86_64
45 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
46 #else
47 #define mod_64(x, y) ((x) % (y))
48 #endif
49 
50 #define PRId64 "d"
51 #define PRIx64 "llx"
52 #define PRIu64 "u"
53 #define PRIo64 "o"
54 
55 #define APIC_BUS_CYCLE_NS 1
56 
57 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
58 #define apic_debug(fmt, arg...)
59 
60 #define APIC_LVT_NUM			6
61 /* 14 is the version for Xeon and Pentium 8.4.8*/
62 #define APIC_VERSION			(0x14UL | ((APIC_LVT_NUM - 1) << 16))
63 #define LAPIC_MMIO_LENGTH		(1 << 12)
64 /* followed define is not in apicdef.h */
65 #define APIC_SHORT_MASK			0xc0000
66 #define APIC_DEST_NOSHORT		0x0
67 #define APIC_DEST_MASK			0x800
68 #define MAX_APIC_VECTOR			256
69 #define APIC_VECTORS_PER_REG		32
70 
71 #define VEC_POS(v) ((v) & (32 - 1))
72 #define REG_POS(v) (((v) >> 5) << 4)
73 
74 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
75 {
76 	*((u32 *) (apic->regs + reg_off)) = val;
77 }
78 
79 static inline int apic_test_vector(int vec, void *bitmap)
80 {
81 	return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
82 }
83 
84 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
85 {
86 	struct kvm_lapic *apic = vcpu->arch.apic;
87 
88 	return apic_test_vector(vector, apic->regs + APIC_ISR) ||
89 		apic_test_vector(vector, apic->regs + APIC_IRR);
90 }
91 
92 static inline void apic_set_vector(int vec, void *bitmap)
93 {
94 	set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
95 }
96 
97 static inline void apic_clear_vector(int vec, void *bitmap)
98 {
99 	clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
100 }
101 
102 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
103 {
104 	return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
105 }
106 
107 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
108 {
109 	return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
110 }
111 
112 struct static_key_deferred apic_hw_disabled __read_mostly;
113 struct static_key_deferred apic_sw_disabled __read_mostly;
114 
115 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
116 {
117 	if ((kvm_apic_get_reg(apic, APIC_SPIV) ^ val) & APIC_SPIV_APIC_ENABLED) {
118 		if (val & APIC_SPIV_APIC_ENABLED)
119 			static_key_slow_dec_deferred(&apic_sw_disabled);
120 		else
121 			static_key_slow_inc(&apic_sw_disabled.key);
122 	}
123 	apic_set_reg(apic, APIC_SPIV, val);
124 }
125 
126 static inline int apic_enabled(struct kvm_lapic *apic)
127 {
128 	return kvm_apic_sw_enabled(apic) &&	kvm_apic_hw_enabled(apic);
129 }
130 
131 #define LVT_MASK	\
132 	(APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
133 
134 #define LINT_MASK	\
135 	(LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
136 	 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
137 
138 static inline int kvm_apic_id(struct kvm_lapic *apic)
139 {
140 	return (kvm_apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
141 }
142 
143 #define KVM_X2APIC_CID_BITS 0
144 
145 static void recalculate_apic_map(struct kvm *kvm)
146 {
147 	struct kvm_apic_map *new, *old = NULL;
148 	struct kvm_vcpu *vcpu;
149 	int i;
150 
151 	new = kzalloc(sizeof(struct kvm_apic_map), GFP_KERNEL);
152 
153 	mutex_lock(&kvm->arch.apic_map_lock);
154 
155 	if (!new)
156 		goto out;
157 
158 	new->ldr_bits = 8;
159 	/* flat mode is default */
160 	new->cid_shift = 8;
161 	new->cid_mask = 0;
162 	new->lid_mask = 0xff;
163 
164 	kvm_for_each_vcpu(i, vcpu, kvm) {
165 		struct kvm_lapic *apic = vcpu->arch.apic;
166 		u16 cid, lid;
167 		u32 ldr;
168 
169 		if (!kvm_apic_present(vcpu))
170 			continue;
171 
172 		/*
173 		 * All APICs have to be configured in the same mode by an OS.
174 		 * We take advatage of this while building logical id loockup
175 		 * table. After reset APICs are in xapic/flat mode, so if we
176 		 * find apic with different setting we assume this is the mode
177 		 * OS wants all apics to be in; build lookup table accordingly.
178 		 */
179 		if (apic_x2apic_mode(apic)) {
180 			new->ldr_bits = 32;
181 			new->cid_shift = 16;
182 			new->cid_mask = (1 << KVM_X2APIC_CID_BITS) - 1;
183 			new->lid_mask = 0xffff;
184 		} else if (kvm_apic_sw_enabled(apic) &&
185 				!new->cid_mask /* flat mode */ &&
186 				kvm_apic_get_reg(apic, APIC_DFR) == APIC_DFR_CLUSTER) {
187 			new->cid_shift = 4;
188 			new->cid_mask = 0xf;
189 			new->lid_mask = 0xf;
190 		}
191 
192 		new->phys_map[kvm_apic_id(apic)] = apic;
193 
194 		ldr = kvm_apic_get_reg(apic, APIC_LDR);
195 		cid = apic_cluster_id(new, ldr);
196 		lid = apic_logical_id(new, ldr);
197 
198 		if (lid)
199 			new->logical_map[cid][ffs(lid) - 1] = apic;
200 	}
201 out:
202 	old = rcu_dereference_protected(kvm->arch.apic_map,
203 			lockdep_is_held(&kvm->arch.apic_map_lock));
204 	rcu_assign_pointer(kvm->arch.apic_map, new);
205 	mutex_unlock(&kvm->arch.apic_map_lock);
206 
207 	if (old)
208 		kfree_rcu(old, rcu);
209 
210 	kvm_vcpu_request_scan_ioapic(kvm);
211 }
212 
213 static inline void kvm_apic_set_id(struct kvm_lapic *apic, u8 id)
214 {
215 	apic_set_reg(apic, APIC_ID, id << 24);
216 	recalculate_apic_map(apic->vcpu->kvm);
217 }
218 
219 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
220 {
221 	apic_set_reg(apic, APIC_LDR, id);
222 	recalculate_apic_map(apic->vcpu->kvm);
223 }
224 
225 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
226 {
227 	return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
228 }
229 
230 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
231 {
232 	return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
233 }
234 
235 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
236 {
237 	return ((kvm_apic_get_reg(apic, APIC_LVTT) &
238 		apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_ONESHOT);
239 }
240 
241 static inline int apic_lvtt_period(struct kvm_lapic *apic)
242 {
243 	return ((kvm_apic_get_reg(apic, APIC_LVTT) &
244 		apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_PERIODIC);
245 }
246 
247 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
248 {
249 	return ((kvm_apic_get_reg(apic, APIC_LVTT) &
250 		apic->lapic_timer.timer_mode_mask) ==
251 			APIC_LVT_TIMER_TSCDEADLINE);
252 }
253 
254 static inline int apic_lvt_nmi_mode(u32 lvt_val)
255 {
256 	return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
257 }
258 
259 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
260 {
261 	struct kvm_lapic *apic = vcpu->arch.apic;
262 	struct kvm_cpuid_entry2 *feat;
263 	u32 v = APIC_VERSION;
264 
265 	if (!kvm_vcpu_has_lapic(vcpu))
266 		return;
267 
268 	feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
269 	if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
270 		v |= APIC_LVR_DIRECTED_EOI;
271 	apic_set_reg(apic, APIC_LVR, v);
272 }
273 
274 static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
275 	LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
276 	LVT_MASK | APIC_MODE_MASK,	/* LVTTHMR */
277 	LVT_MASK | APIC_MODE_MASK,	/* LVTPC */
278 	LINT_MASK, LINT_MASK,	/* LVT0-1 */
279 	LVT_MASK		/* LVTERR */
280 };
281 
282 static int find_highest_vector(void *bitmap)
283 {
284 	int vec;
285 	u32 *reg;
286 
287 	for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
288 	     vec >= 0; vec -= APIC_VECTORS_PER_REG) {
289 		reg = bitmap + REG_POS(vec);
290 		if (*reg)
291 			return fls(*reg) - 1 + vec;
292 	}
293 
294 	return -1;
295 }
296 
297 static u8 count_vectors(void *bitmap)
298 {
299 	int vec;
300 	u32 *reg;
301 	u8 count = 0;
302 
303 	for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
304 		reg = bitmap + REG_POS(vec);
305 		count += hweight32(*reg);
306 	}
307 
308 	return count;
309 }
310 
311 void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
312 {
313 	u32 i, pir_val;
314 	struct kvm_lapic *apic = vcpu->arch.apic;
315 
316 	for (i = 0; i <= 7; i++) {
317 		pir_val = xchg(&pir[i], 0);
318 		if (pir_val)
319 			*((u32 *)(apic->regs + APIC_IRR + i * 0x10)) |= pir_val;
320 	}
321 }
322 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
323 
324 static inline void apic_set_irr(int vec, struct kvm_lapic *apic)
325 {
326 	apic->irr_pending = true;
327 	apic_set_vector(vec, apic->regs + APIC_IRR);
328 }
329 
330 static inline int apic_search_irr(struct kvm_lapic *apic)
331 {
332 	return find_highest_vector(apic->regs + APIC_IRR);
333 }
334 
335 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
336 {
337 	int result;
338 
339 	/*
340 	 * Note that irr_pending is just a hint. It will be always
341 	 * true with virtual interrupt delivery enabled.
342 	 */
343 	if (!apic->irr_pending)
344 		return -1;
345 
346 	kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
347 	result = apic_search_irr(apic);
348 	ASSERT(result == -1 || result >= 16);
349 
350 	return result;
351 }
352 
353 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
354 {
355 	struct kvm_vcpu *vcpu;
356 
357 	vcpu = apic->vcpu;
358 
359 	apic_clear_vector(vec, apic->regs + APIC_IRR);
360 	if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
361 		/* try to update RVI */
362 		kvm_make_request(KVM_REQ_EVENT, vcpu);
363 	else {
364 		vec = apic_search_irr(apic);
365 		apic->irr_pending = (vec != -1);
366 	}
367 }
368 
369 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
370 {
371 	struct kvm_vcpu *vcpu;
372 
373 	if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
374 		return;
375 
376 	vcpu = apic->vcpu;
377 
378 	/*
379 	 * With APIC virtualization enabled, all caching is disabled
380 	 * because the processor can modify ISR under the hood.  Instead
381 	 * just set SVI.
382 	 */
383 	if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
384 		kvm_x86_ops->hwapic_isr_update(vcpu->kvm, vec);
385 	else {
386 		++apic->isr_count;
387 		BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
388 		/*
389 		 * ISR (in service register) bit is set when injecting an interrupt.
390 		 * The highest vector is injected. Thus the latest bit set matches
391 		 * the highest bit in ISR.
392 		 */
393 		apic->highest_isr_cache = vec;
394 	}
395 }
396 
397 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
398 {
399 	int result;
400 
401 	/*
402 	 * Note that isr_count is always 1, and highest_isr_cache
403 	 * is always -1, with APIC virtualization enabled.
404 	 */
405 	if (!apic->isr_count)
406 		return -1;
407 	if (likely(apic->highest_isr_cache != -1))
408 		return apic->highest_isr_cache;
409 
410 	result = find_highest_vector(apic->regs + APIC_ISR);
411 	ASSERT(result == -1 || result >= 16);
412 
413 	return result;
414 }
415 
416 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
417 {
418 	struct kvm_vcpu *vcpu;
419 	if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
420 		return;
421 
422 	vcpu = apic->vcpu;
423 
424 	/*
425 	 * We do get here for APIC virtualization enabled if the guest
426 	 * uses the Hyper-V APIC enlightenment.  In this case we may need
427 	 * to trigger a new interrupt delivery by writing the SVI field;
428 	 * on the other hand isr_count and highest_isr_cache are unused
429 	 * and must be left alone.
430 	 */
431 	if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
432 		kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
433 					       apic_find_highest_isr(apic));
434 	else {
435 		--apic->isr_count;
436 		BUG_ON(apic->isr_count < 0);
437 		apic->highest_isr_cache = -1;
438 	}
439 }
440 
441 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
442 {
443 	int highest_irr;
444 
445 	/* This may race with setting of irr in __apic_accept_irq() and
446 	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
447 	 * will cause vmexit immediately and the value will be recalculated
448 	 * on the next vmentry.
449 	 */
450 	if (!kvm_vcpu_has_lapic(vcpu))
451 		return 0;
452 	highest_irr = apic_find_highest_irr(vcpu->arch.apic);
453 
454 	return highest_irr;
455 }
456 
457 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
458 			     int vector, int level, int trig_mode,
459 			     unsigned long *dest_map);
460 
461 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
462 		unsigned long *dest_map)
463 {
464 	struct kvm_lapic *apic = vcpu->arch.apic;
465 
466 	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
467 			irq->level, irq->trig_mode, dest_map);
468 }
469 
470 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
471 {
472 
473 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
474 				      sizeof(val));
475 }
476 
477 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
478 {
479 
480 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
481 				      sizeof(*val));
482 }
483 
484 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
485 {
486 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
487 }
488 
489 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
490 {
491 	u8 val;
492 	if (pv_eoi_get_user(vcpu, &val) < 0)
493 		apic_debug("Can't read EOI MSR value: 0x%llx\n",
494 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
495 	return val & 0x1;
496 }
497 
498 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
499 {
500 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
501 		apic_debug("Can't set EOI MSR value: 0x%llx\n",
502 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
503 		return;
504 	}
505 	__set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
506 }
507 
508 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
509 {
510 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
511 		apic_debug("Can't clear EOI MSR value: 0x%llx\n",
512 			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
513 		return;
514 	}
515 	__clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
516 }
517 
518 void kvm_apic_update_tmr(struct kvm_vcpu *vcpu, u32 *tmr)
519 {
520 	struct kvm_lapic *apic = vcpu->arch.apic;
521 	int i;
522 
523 	for (i = 0; i < 8; i++)
524 		apic_set_reg(apic, APIC_TMR + 0x10 * i, tmr[i]);
525 }
526 
527 static void apic_update_ppr(struct kvm_lapic *apic)
528 {
529 	u32 tpr, isrv, ppr, old_ppr;
530 	int isr;
531 
532 	old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI);
533 	tpr = kvm_apic_get_reg(apic, APIC_TASKPRI);
534 	isr = apic_find_highest_isr(apic);
535 	isrv = (isr != -1) ? isr : 0;
536 
537 	if ((tpr & 0xf0) >= (isrv & 0xf0))
538 		ppr = tpr & 0xff;
539 	else
540 		ppr = isrv & 0xf0;
541 
542 	apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
543 		   apic, ppr, isr, isrv);
544 
545 	if (old_ppr != ppr) {
546 		apic_set_reg(apic, APIC_PROCPRI, ppr);
547 		if (ppr < old_ppr)
548 			kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
549 	}
550 }
551 
552 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
553 {
554 	apic_set_reg(apic, APIC_TASKPRI, tpr);
555 	apic_update_ppr(apic);
556 }
557 
558 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
559 {
560 	return dest == 0xff || kvm_apic_id(apic) == dest;
561 }
562 
563 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
564 {
565 	int result = 0;
566 	u32 logical_id;
567 
568 	if (apic_x2apic_mode(apic)) {
569 		logical_id = kvm_apic_get_reg(apic, APIC_LDR);
570 		return logical_id & mda;
571 	}
572 
573 	logical_id = GET_APIC_LOGICAL_ID(kvm_apic_get_reg(apic, APIC_LDR));
574 
575 	switch (kvm_apic_get_reg(apic, APIC_DFR)) {
576 	case APIC_DFR_FLAT:
577 		if (logical_id & mda)
578 			result = 1;
579 		break;
580 	case APIC_DFR_CLUSTER:
581 		if (((logical_id >> 4) == (mda >> 0x4))
582 		    && (logical_id & mda & 0xf))
583 			result = 1;
584 		break;
585 	default:
586 		apic_debug("Bad DFR vcpu %d: %08x\n",
587 			   apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR));
588 		break;
589 	}
590 
591 	return result;
592 }
593 
594 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
595 			   int short_hand, int dest, int dest_mode)
596 {
597 	int result = 0;
598 	struct kvm_lapic *target = vcpu->arch.apic;
599 
600 	apic_debug("target %p, source %p, dest 0x%x, "
601 		   "dest_mode 0x%x, short_hand 0x%x\n",
602 		   target, source, dest, dest_mode, short_hand);
603 
604 	ASSERT(target);
605 	switch (short_hand) {
606 	case APIC_DEST_NOSHORT:
607 		if (dest_mode == 0)
608 			/* Physical mode. */
609 			result = kvm_apic_match_physical_addr(target, dest);
610 		else
611 			/* Logical mode. */
612 			result = kvm_apic_match_logical_addr(target, dest);
613 		break;
614 	case APIC_DEST_SELF:
615 		result = (target == source);
616 		break;
617 	case APIC_DEST_ALLINC:
618 		result = 1;
619 		break;
620 	case APIC_DEST_ALLBUT:
621 		result = (target != source);
622 		break;
623 	default:
624 		apic_debug("kvm: apic: Bad dest shorthand value %x\n",
625 			   short_hand);
626 		break;
627 	}
628 
629 	return result;
630 }
631 
632 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
633 		struct kvm_lapic_irq *irq, int *r, unsigned long *dest_map)
634 {
635 	struct kvm_apic_map *map;
636 	unsigned long bitmap = 1;
637 	struct kvm_lapic **dst;
638 	int i;
639 	bool ret = false;
640 
641 	*r = -1;
642 
643 	if (irq->shorthand == APIC_DEST_SELF) {
644 		*r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
645 		return true;
646 	}
647 
648 	if (irq->shorthand)
649 		return false;
650 
651 	rcu_read_lock();
652 	map = rcu_dereference(kvm->arch.apic_map);
653 
654 	if (!map)
655 		goto out;
656 
657 	if (irq->dest_mode == 0) { /* physical mode */
658 		if (irq->delivery_mode == APIC_DM_LOWEST ||
659 				irq->dest_id == 0xff)
660 			goto out;
661 		dst = &map->phys_map[irq->dest_id & 0xff];
662 	} else {
663 		u32 mda = irq->dest_id << (32 - map->ldr_bits);
664 
665 		dst = map->logical_map[apic_cluster_id(map, mda)];
666 
667 		bitmap = apic_logical_id(map, mda);
668 
669 		if (irq->delivery_mode == APIC_DM_LOWEST) {
670 			int l = -1;
671 			for_each_set_bit(i, &bitmap, 16) {
672 				if (!dst[i])
673 					continue;
674 				if (l < 0)
675 					l = i;
676 				else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0)
677 					l = i;
678 			}
679 
680 			bitmap = (l >= 0) ? 1 << l : 0;
681 		}
682 	}
683 
684 	for_each_set_bit(i, &bitmap, 16) {
685 		if (!dst[i])
686 			continue;
687 		if (*r < 0)
688 			*r = 0;
689 		*r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
690 	}
691 
692 	ret = true;
693 out:
694 	rcu_read_unlock();
695 	return ret;
696 }
697 
698 /*
699  * Add a pending IRQ into lapic.
700  * Return 1 if successfully added and 0 if discarded.
701  */
702 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
703 			     int vector, int level, int trig_mode,
704 			     unsigned long *dest_map)
705 {
706 	int result = 0;
707 	struct kvm_vcpu *vcpu = apic->vcpu;
708 
709 	switch (delivery_mode) {
710 	case APIC_DM_LOWEST:
711 		vcpu->arch.apic_arb_prio++;
712 	case APIC_DM_FIXED:
713 		/* FIXME add logic for vcpu on reset */
714 		if (unlikely(!apic_enabled(apic)))
715 			break;
716 
717 		result = 1;
718 
719 		if (dest_map)
720 			__set_bit(vcpu->vcpu_id, dest_map);
721 
722 		if (kvm_x86_ops->deliver_posted_interrupt)
723 			kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
724 		else {
725 			apic_set_irr(vector, apic);
726 
727 			kvm_make_request(KVM_REQ_EVENT, vcpu);
728 			kvm_vcpu_kick(vcpu);
729 		}
730 		trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
731 					  trig_mode, vector, false);
732 		break;
733 
734 	case APIC_DM_REMRD:
735 		result = 1;
736 		vcpu->arch.pv.pv_unhalted = 1;
737 		kvm_make_request(KVM_REQ_EVENT, vcpu);
738 		kvm_vcpu_kick(vcpu);
739 		break;
740 
741 	case APIC_DM_SMI:
742 		apic_debug("Ignoring guest SMI\n");
743 		break;
744 
745 	case APIC_DM_NMI:
746 		result = 1;
747 		kvm_inject_nmi(vcpu);
748 		kvm_vcpu_kick(vcpu);
749 		break;
750 
751 	case APIC_DM_INIT:
752 		if (!trig_mode || level) {
753 			result = 1;
754 			/* assumes that there are only KVM_APIC_INIT/SIPI */
755 			apic->pending_events = (1UL << KVM_APIC_INIT);
756 			/* make sure pending_events is visible before sending
757 			 * the request */
758 			smp_wmb();
759 			kvm_make_request(KVM_REQ_EVENT, vcpu);
760 			kvm_vcpu_kick(vcpu);
761 		} else {
762 			apic_debug("Ignoring de-assert INIT to vcpu %d\n",
763 				   vcpu->vcpu_id);
764 		}
765 		break;
766 
767 	case APIC_DM_STARTUP:
768 		apic_debug("SIPI to vcpu %d vector 0x%02x\n",
769 			   vcpu->vcpu_id, vector);
770 		result = 1;
771 		apic->sipi_vector = vector;
772 		/* make sure sipi_vector is visible for the receiver */
773 		smp_wmb();
774 		set_bit(KVM_APIC_SIPI, &apic->pending_events);
775 		kvm_make_request(KVM_REQ_EVENT, vcpu);
776 		kvm_vcpu_kick(vcpu);
777 		break;
778 
779 	case APIC_DM_EXTINT:
780 		/*
781 		 * Should only be called by kvm_apic_local_deliver() with LVT0,
782 		 * before NMI watchdog was enabled. Already handled by
783 		 * kvm_apic_accept_pic_intr().
784 		 */
785 		break;
786 
787 	default:
788 		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
789 		       delivery_mode);
790 		break;
791 	}
792 	return result;
793 }
794 
795 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
796 {
797 	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
798 }
799 
800 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
801 {
802 	if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI) &&
803 	    kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
804 		int trigger_mode;
805 		if (apic_test_vector(vector, apic->regs + APIC_TMR))
806 			trigger_mode = IOAPIC_LEVEL_TRIG;
807 		else
808 			trigger_mode = IOAPIC_EDGE_TRIG;
809 		kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
810 	}
811 }
812 
813 static int apic_set_eoi(struct kvm_lapic *apic)
814 {
815 	int vector = apic_find_highest_isr(apic);
816 
817 	trace_kvm_eoi(apic, vector);
818 
819 	/*
820 	 * Not every write EOI will has corresponding ISR,
821 	 * one example is when Kernel check timer on setup_IO_APIC
822 	 */
823 	if (vector == -1)
824 		return vector;
825 
826 	apic_clear_isr(vector, apic);
827 	apic_update_ppr(apic);
828 
829 	kvm_ioapic_send_eoi(apic, vector);
830 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
831 	return vector;
832 }
833 
834 /*
835  * this interface assumes a trap-like exit, which has already finished
836  * desired side effect including vISR and vPPR update.
837  */
838 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
839 {
840 	struct kvm_lapic *apic = vcpu->arch.apic;
841 
842 	trace_kvm_eoi(apic, vector);
843 
844 	kvm_ioapic_send_eoi(apic, vector);
845 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
846 }
847 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
848 
849 static void apic_send_ipi(struct kvm_lapic *apic)
850 {
851 	u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR);
852 	u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2);
853 	struct kvm_lapic_irq irq;
854 
855 	irq.vector = icr_low & APIC_VECTOR_MASK;
856 	irq.delivery_mode = icr_low & APIC_MODE_MASK;
857 	irq.dest_mode = icr_low & APIC_DEST_MASK;
858 	irq.level = icr_low & APIC_INT_ASSERT;
859 	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
860 	irq.shorthand = icr_low & APIC_SHORT_MASK;
861 	if (apic_x2apic_mode(apic))
862 		irq.dest_id = icr_high;
863 	else
864 		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
865 
866 	trace_kvm_apic_ipi(icr_low, irq.dest_id);
867 
868 	apic_debug("icr_high 0x%x, icr_low 0x%x, "
869 		   "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
870 		   "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
871 		   icr_high, icr_low, irq.shorthand, irq.dest_id,
872 		   irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
873 		   irq.vector);
874 
875 	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
876 }
877 
878 static u32 apic_get_tmcct(struct kvm_lapic *apic)
879 {
880 	ktime_t remaining;
881 	s64 ns;
882 	u32 tmcct;
883 
884 	ASSERT(apic != NULL);
885 
886 	/* if initial count is 0, current count should also be 0 */
887 	if (kvm_apic_get_reg(apic, APIC_TMICT) == 0 ||
888 		apic->lapic_timer.period == 0)
889 		return 0;
890 
891 	remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
892 	if (ktime_to_ns(remaining) < 0)
893 		remaining = ktime_set(0, 0);
894 
895 	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
896 	tmcct = div64_u64(ns,
897 			 (APIC_BUS_CYCLE_NS * apic->divide_count));
898 
899 	return tmcct;
900 }
901 
902 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
903 {
904 	struct kvm_vcpu *vcpu = apic->vcpu;
905 	struct kvm_run *run = vcpu->run;
906 
907 	kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
908 	run->tpr_access.rip = kvm_rip_read(vcpu);
909 	run->tpr_access.is_write = write;
910 }
911 
912 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
913 {
914 	if (apic->vcpu->arch.tpr_access_reporting)
915 		__report_tpr_access(apic, write);
916 }
917 
918 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
919 {
920 	u32 val = 0;
921 
922 	if (offset >= LAPIC_MMIO_LENGTH)
923 		return 0;
924 
925 	switch (offset) {
926 	case APIC_ID:
927 		if (apic_x2apic_mode(apic))
928 			val = kvm_apic_id(apic);
929 		else
930 			val = kvm_apic_id(apic) << 24;
931 		break;
932 	case APIC_ARBPRI:
933 		apic_debug("Access APIC ARBPRI register which is for P6\n");
934 		break;
935 
936 	case APIC_TMCCT:	/* Timer CCR */
937 		if (apic_lvtt_tscdeadline(apic))
938 			return 0;
939 
940 		val = apic_get_tmcct(apic);
941 		break;
942 	case APIC_PROCPRI:
943 		apic_update_ppr(apic);
944 		val = kvm_apic_get_reg(apic, offset);
945 		break;
946 	case APIC_TASKPRI:
947 		report_tpr_access(apic, false);
948 		/* fall thru */
949 	default:
950 		val = kvm_apic_get_reg(apic, offset);
951 		break;
952 	}
953 
954 	return val;
955 }
956 
957 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
958 {
959 	return container_of(dev, struct kvm_lapic, dev);
960 }
961 
962 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
963 		void *data)
964 {
965 	unsigned char alignment = offset & 0xf;
966 	u32 result;
967 	/* this bitmask has a bit cleared for each reserved register */
968 	static const u64 rmask = 0x43ff01ffffffe70cULL;
969 
970 	if ((alignment + len) > 4) {
971 		apic_debug("KVM_APIC_READ: alignment error %x %d\n",
972 			   offset, len);
973 		return 1;
974 	}
975 
976 	if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
977 		apic_debug("KVM_APIC_READ: read reserved register %x\n",
978 			   offset);
979 		return 1;
980 	}
981 
982 	result = __apic_read(apic, offset & ~0xf);
983 
984 	trace_kvm_apic_read(offset, result);
985 
986 	switch (len) {
987 	case 1:
988 	case 2:
989 	case 4:
990 		memcpy(data, (char *)&result + alignment, len);
991 		break;
992 	default:
993 		printk(KERN_ERR "Local APIC read with len = %x, "
994 		       "should be 1,2, or 4 instead\n", len);
995 		break;
996 	}
997 	return 0;
998 }
999 
1000 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1001 {
1002 	return kvm_apic_hw_enabled(apic) &&
1003 	    addr >= apic->base_address &&
1004 	    addr < apic->base_address + LAPIC_MMIO_LENGTH;
1005 }
1006 
1007 static int apic_mmio_read(struct kvm_io_device *this,
1008 			   gpa_t address, int len, void *data)
1009 {
1010 	struct kvm_lapic *apic = to_lapic(this);
1011 	u32 offset = address - apic->base_address;
1012 
1013 	if (!apic_mmio_in_range(apic, address))
1014 		return -EOPNOTSUPP;
1015 
1016 	apic_reg_read(apic, offset, len, data);
1017 
1018 	return 0;
1019 }
1020 
1021 static void update_divide_count(struct kvm_lapic *apic)
1022 {
1023 	u32 tmp1, tmp2, tdcr;
1024 
1025 	tdcr = kvm_apic_get_reg(apic, APIC_TDCR);
1026 	tmp1 = tdcr & 0xf;
1027 	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1028 	apic->divide_count = 0x1 << (tmp2 & 0x7);
1029 
1030 	apic_debug("timer divide count is 0x%x\n",
1031 				   apic->divide_count);
1032 }
1033 
1034 static void start_apic_timer(struct kvm_lapic *apic)
1035 {
1036 	ktime_t now;
1037 	atomic_set(&apic->lapic_timer.pending, 0);
1038 
1039 	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1040 		/* lapic timer in oneshot or periodic mode */
1041 		now = apic->lapic_timer.timer.base->get_time();
1042 		apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT)
1043 			    * APIC_BUS_CYCLE_NS * apic->divide_count;
1044 
1045 		if (!apic->lapic_timer.period)
1046 			return;
1047 		/*
1048 		 * Do not allow the guest to program periodic timers with small
1049 		 * interval, since the hrtimers are not throttled by the host
1050 		 * scheduler.
1051 		 */
1052 		if (apic_lvtt_period(apic)) {
1053 			s64 min_period = min_timer_period_us * 1000LL;
1054 
1055 			if (apic->lapic_timer.period < min_period) {
1056 				pr_info_ratelimited(
1057 				    "kvm: vcpu %i: requested %lld ns "
1058 				    "lapic timer period limited to %lld ns\n",
1059 				    apic->vcpu->vcpu_id,
1060 				    apic->lapic_timer.period, min_period);
1061 				apic->lapic_timer.period = min_period;
1062 			}
1063 		}
1064 
1065 		hrtimer_start(&apic->lapic_timer.timer,
1066 			      ktime_add_ns(now, apic->lapic_timer.period),
1067 			      HRTIMER_MODE_ABS);
1068 
1069 		apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1070 			   PRIx64 ", "
1071 			   "timer initial count 0x%x, period %lldns, "
1072 			   "expire @ 0x%016" PRIx64 ".\n", __func__,
1073 			   APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1074 			   kvm_apic_get_reg(apic, APIC_TMICT),
1075 			   apic->lapic_timer.period,
1076 			   ktime_to_ns(ktime_add_ns(now,
1077 					apic->lapic_timer.period)));
1078 	} else if (apic_lvtt_tscdeadline(apic)) {
1079 		/* lapic timer in tsc deadline mode */
1080 		u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1081 		u64 ns = 0;
1082 		struct kvm_vcpu *vcpu = apic->vcpu;
1083 		unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1084 		unsigned long flags;
1085 
1086 		if (unlikely(!tscdeadline || !this_tsc_khz))
1087 			return;
1088 
1089 		local_irq_save(flags);
1090 
1091 		now = apic->lapic_timer.timer.base->get_time();
1092 		guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1093 		if (likely(tscdeadline > guest_tsc)) {
1094 			ns = (tscdeadline - guest_tsc) * 1000000ULL;
1095 			do_div(ns, this_tsc_khz);
1096 		}
1097 		hrtimer_start(&apic->lapic_timer.timer,
1098 			ktime_add_ns(now, ns), HRTIMER_MODE_ABS);
1099 
1100 		local_irq_restore(flags);
1101 	}
1102 }
1103 
1104 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1105 {
1106 	int nmi_wd_enabled = apic_lvt_nmi_mode(kvm_apic_get_reg(apic, APIC_LVT0));
1107 
1108 	if (apic_lvt_nmi_mode(lvt0_val)) {
1109 		if (!nmi_wd_enabled) {
1110 			apic_debug("Receive NMI setting on APIC_LVT0 "
1111 				   "for cpu %d\n", apic->vcpu->vcpu_id);
1112 			apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
1113 		}
1114 	} else if (nmi_wd_enabled)
1115 		apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
1116 }
1117 
1118 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1119 {
1120 	int ret = 0;
1121 
1122 	trace_kvm_apic_write(reg, val);
1123 
1124 	switch (reg) {
1125 	case APIC_ID:		/* Local APIC ID */
1126 		if (!apic_x2apic_mode(apic))
1127 			kvm_apic_set_id(apic, val >> 24);
1128 		else
1129 			ret = 1;
1130 		break;
1131 
1132 	case APIC_TASKPRI:
1133 		report_tpr_access(apic, true);
1134 		apic_set_tpr(apic, val & 0xff);
1135 		break;
1136 
1137 	case APIC_EOI:
1138 		apic_set_eoi(apic);
1139 		break;
1140 
1141 	case APIC_LDR:
1142 		if (!apic_x2apic_mode(apic))
1143 			kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1144 		else
1145 			ret = 1;
1146 		break;
1147 
1148 	case APIC_DFR:
1149 		if (!apic_x2apic_mode(apic)) {
1150 			apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1151 			recalculate_apic_map(apic->vcpu->kvm);
1152 		} else
1153 			ret = 1;
1154 		break;
1155 
1156 	case APIC_SPIV: {
1157 		u32 mask = 0x3ff;
1158 		if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1159 			mask |= APIC_SPIV_DIRECTED_EOI;
1160 		apic_set_spiv(apic, val & mask);
1161 		if (!(val & APIC_SPIV_APIC_ENABLED)) {
1162 			int i;
1163 			u32 lvt_val;
1164 
1165 			for (i = 0; i < APIC_LVT_NUM; i++) {
1166 				lvt_val = kvm_apic_get_reg(apic,
1167 						       APIC_LVTT + 0x10 * i);
1168 				apic_set_reg(apic, APIC_LVTT + 0x10 * i,
1169 					     lvt_val | APIC_LVT_MASKED);
1170 			}
1171 			atomic_set(&apic->lapic_timer.pending, 0);
1172 
1173 		}
1174 		break;
1175 	}
1176 	case APIC_ICR:
1177 		/* No delay here, so we always clear the pending bit */
1178 		apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1179 		apic_send_ipi(apic);
1180 		break;
1181 
1182 	case APIC_ICR2:
1183 		if (!apic_x2apic_mode(apic))
1184 			val &= 0xff000000;
1185 		apic_set_reg(apic, APIC_ICR2, val);
1186 		break;
1187 
1188 	case APIC_LVT0:
1189 		apic_manage_nmi_watchdog(apic, val);
1190 	case APIC_LVTTHMR:
1191 	case APIC_LVTPC:
1192 	case APIC_LVT1:
1193 	case APIC_LVTERR:
1194 		/* TODO: Check vector */
1195 		if (!kvm_apic_sw_enabled(apic))
1196 			val |= APIC_LVT_MASKED;
1197 
1198 		val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1199 		apic_set_reg(apic, reg, val);
1200 
1201 		break;
1202 
1203 	case APIC_LVTT:
1204 		if ((kvm_apic_get_reg(apic, APIC_LVTT) &
1205 		    apic->lapic_timer.timer_mode_mask) !=
1206 		   (val & apic->lapic_timer.timer_mode_mask))
1207 			hrtimer_cancel(&apic->lapic_timer.timer);
1208 
1209 		if (!kvm_apic_sw_enabled(apic))
1210 			val |= APIC_LVT_MASKED;
1211 		val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1212 		apic_set_reg(apic, APIC_LVTT, val);
1213 		break;
1214 
1215 	case APIC_TMICT:
1216 		if (apic_lvtt_tscdeadline(apic))
1217 			break;
1218 
1219 		hrtimer_cancel(&apic->lapic_timer.timer);
1220 		apic_set_reg(apic, APIC_TMICT, val);
1221 		start_apic_timer(apic);
1222 		break;
1223 
1224 	case APIC_TDCR:
1225 		if (val & 4)
1226 			apic_debug("KVM_WRITE:TDCR %x\n", val);
1227 		apic_set_reg(apic, APIC_TDCR, val);
1228 		update_divide_count(apic);
1229 		break;
1230 
1231 	case APIC_ESR:
1232 		if (apic_x2apic_mode(apic) && val != 0) {
1233 			apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1234 			ret = 1;
1235 		}
1236 		break;
1237 
1238 	case APIC_SELF_IPI:
1239 		if (apic_x2apic_mode(apic)) {
1240 			apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1241 		} else
1242 			ret = 1;
1243 		break;
1244 	default:
1245 		ret = 1;
1246 		break;
1247 	}
1248 	if (ret)
1249 		apic_debug("Local APIC Write to read-only register %x\n", reg);
1250 	return ret;
1251 }
1252 
1253 static int apic_mmio_write(struct kvm_io_device *this,
1254 			    gpa_t address, int len, const void *data)
1255 {
1256 	struct kvm_lapic *apic = to_lapic(this);
1257 	unsigned int offset = address - apic->base_address;
1258 	u32 val;
1259 
1260 	if (!apic_mmio_in_range(apic, address))
1261 		return -EOPNOTSUPP;
1262 
1263 	/*
1264 	 * APIC register must be aligned on 128-bits boundary.
1265 	 * 32/64/128 bits registers must be accessed thru 32 bits.
1266 	 * Refer SDM 8.4.1
1267 	 */
1268 	if (len != 4 || (offset & 0xf)) {
1269 		/* Don't shout loud, $infamous_os would cause only noise. */
1270 		apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1271 		return 0;
1272 	}
1273 
1274 	val = *(u32*)data;
1275 
1276 	/* too common printing */
1277 	if (offset != APIC_EOI)
1278 		apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1279 			   "0x%x\n", __func__, offset, len, val);
1280 
1281 	apic_reg_write(apic, offset & 0xff0, val);
1282 
1283 	return 0;
1284 }
1285 
1286 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1287 {
1288 	if (kvm_vcpu_has_lapic(vcpu))
1289 		apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1290 }
1291 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1292 
1293 /* emulate APIC access in a trap manner */
1294 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1295 {
1296 	u32 val = 0;
1297 
1298 	/* hw has done the conditional check and inst decode */
1299 	offset &= 0xff0;
1300 
1301 	apic_reg_read(vcpu->arch.apic, offset, 4, &val);
1302 
1303 	/* TODO: optimize to just emulate side effect w/o one more write */
1304 	apic_reg_write(vcpu->arch.apic, offset, val);
1305 }
1306 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1307 
1308 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1309 {
1310 	struct kvm_lapic *apic = vcpu->arch.apic;
1311 
1312 	if (!vcpu->arch.apic)
1313 		return;
1314 
1315 	hrtimer_cancel(&apic->lapic_timer.timer);
1316 
1317 	if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1318 		static_key_slow_dec_deferred(&apic_hw_disabled);
1319 
1320 	if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED))
1321 		static_key_slow_dec_deferred(&apic_sw_disabled);
1322 
1323 	if (apic->regs)
1324 		free_page((unsigned long)apic->regs);
1325 
1326 	kfree(apic);
1327 }
1328 
1329 /*
1330  *----------------------------------------------------------------------
1331  * LAPIC interface
1332  *----------------------------------------------------------------------
1333  */
1334 
1335 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1336 {
1337 	struct kvm_lapic *apic = vcpu->arch.apic;
1338 
1339 	if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1340 			apic_lvtt_period(apic))
1341 		return 0;
1342 
1343 	return apic->lapic_timer.tscdeadline;
1344 }
1345 
1346 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1347 {
1348 	struct kvm_lapic *apic = vcpu->arch.apic;
1349 
1350 	if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1351 			apic_lvtt_period(apic))
1352 		return;
1353 
1354 	hrtimer_cancel(&apic->lapic_timer.timer);
1355 	apic->lapic_timer.tscdeadline = data;
1356 	start_apic_timer(apic);
1357 }
1358 
1359 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1360 {
1361 	struct kvm_lapic *apic = vcpu->arch.apic;
1362 
1363 	if (!kvm_vcpu_has_lapic(vcpu))
1364 		return;
1365 
1366 	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1367 		     | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4));
1368 }
1369 
1370 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1371 {
1372 	u64 tpr;
1373 
1374 	if (!kvm_vcpu_has_lapic(vcpu))
1375 		return 0;
1376 
1377 	tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1378 
1379 	return (tpr & 0xf0) >> 4;
1380 }
1381 
1382 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1383 {
1384 	u64 old_value = vcpu->arch.apic_base;
1385 	struct kvm_lapic *apic = vcpu->arch.apic;
1386 
1387 	if (!apic) {
1388 		value |= MSR_IA32_APICBASE_BSP;
1389 		vcpu->arch.apic_base = value;
1390 		return;
1391 	}
1392 
1393 	if (!kvm_vcpu_is_bsp(apic->vcpu))
1394 		value &= ~MSR_IA32_APICBASE_BSP;
1395 	vcpu->arch.apic_base = value;
1396 
1397 	/* update jump label if enable bit changes */
1398 	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1399 		if (value & MSR_IA32_APICBASE_ENABLE)
1400 			static_key_slow_dec_deferred(&apic_hw_disabled);
1401 		else
1402 			static_key_slow_inc(&apic_hw_disabled.key);
1403 		recalculate_apic_map(vcpu->kvm);
1404 	}
1405 
1406 	if ((old_value ^ value) & X2APIC_ENABLE) {
1407 		if (value & X2APIC_ENABLE) {
1408 			u32 id = kvm_apic_id(apic);
1409 			u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
1410 			kvm_apic_set_ldr(apic, ldr);
1411 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1412 		} else
1413 			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1414 	}
1415 
1416 	apic->base_address = apic->vcpu->arch.apic_base &
1417 			     MSR_IA32_APICBASE_BASE;
1418 
1419 	/* with FSB delivery interrupt, we can restart APIC functionality */
1420 	apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1421 		   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1422 
1423 }
1424 
1425 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1426 {
1427 	struct kvm_lapic *apic;
1428 	int i;
1429 
1430 	apic_debug("%s\n", __func__);
1431 
1432 	ASSERT(vcpu);
1433 	apic = vcpu->arch.apic;
1434 	ASSERT(apic != NULL);
1435 
1436 	/* Stop the timer in case it's a reset to an active apic */
1437 	hrtimer_cancel(&apic->lapic_timer.timer);
1438 
1439 	kvm_apic_set_id(apic, vcpu->vcpu_id);
1440 	kvm_apic_set_version(apic->vcpu);
1441 
1442 	for (i = 0; i < APIC_LVT_NUM; i++)
1443 		apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1444 	apic_set_reg(apic, APIC_LVT0,
1445 		     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1446 
1447 	apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1448 	apic_set_spiv(apic, 0xff);
1449 	apic_set_reg(apic, APIC_TASKPRI, 0);
1450 	kvm_apic_set_ldr(apic, 0);
1451 	apic_set_reg(apic, APIC_ESR, 0);
1452 	apic_set_reg(apic, APIC_ICR, 0);
1453 	apic_set_reg(apic, APIC_ICR2, 0);
1454 	apic_set_reg(apic, APIC_TDCR, 0);
1455 	apic_set_reg(apic, APIC_TMICT, 0);
1456 	for (i = 0; i < 8; i++) {
1457 		apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1458 		apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1459 		apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1460 	}
1461 	apic->irr_pending = kvm_apic_vid_enabled(vcpu->kvm);
1462 	apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm);
1463 	apic->highest_isr_cache = -1;
1464 	update_divide_count(apic);
1465 	atomic_set(&apic->lapic_timer.pending, 0);
1466 	if (kvm_vcpu_is_bsp(vcpu))
1467 		kvm_lapic_set_base(vcpu,
1468 				vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1469 	vcpu->arch.pv_eoi.msr_val = 0;
1470 	apic_update_ppr(apic);
1471 
1472 	vcpu->arch.apic_arb_prio = 0;
1473 	vcpu->arch.apic_attention = 0;
1474 
1475 	apic_debug("%s: vcpu=%p, id=%d, base_msr="
1476 		   "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1477 		   vcpu, kvm_apic_id(apic),
1478 		   vcpu->arch.apic_base, apic->base_address);
1479 }
1480 
1481 /*
1482  *----------------------------------------------------------------------
1483  * timer interface
1484  *----------------------------------------------------------------------
1485  */
1486 
1487 static bool lapic_is_periodic(struct kvm_lapic *apic)
1488 {
1489 	return apic_lvtt_period(apic);
1490 }
1491 
1492 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1493 {
1494 	struct kvm_lapic *apic = vcpu->arch.apic;
1495 
1496 	if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) &&
1497 			apic_lvt_enabled(apic, APIC_LVTT))
1498 		return atomic_read(&apic->lapic_timer.pending);
1499 
1500 	return 0;
1501 }
1502 
1503 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1504 {
1505 	u32 reg = kvm_apic_get_reg(apic, lvt_type);
1506 	int vector, mode, trig_mode;
1507 
1508 	if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1509 		vector = reg & APIC_VECTOR_MASK;
1510 		mode = reg & APIC_MODE_MASK;
1511 		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1512 		return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1513 					NULL);
1514 	}
1515 	return 0;
1516 }
1517 
1518 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1519 {
1520 	struct kvm_lapic *apic = vcpu->arch.apic;
1521 
1522 	if (apic)
1523 		kvm_apic_local_deliver(apic, APIC_LVT0);
1524 }
1525 
1526 static const struct kvm_io_device_ops apic_mmio_ops = {
1527 	.read     = apic_mmio_read,
1528 	.write    = apic_mmio_write,
1529 };
1530 
1531 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1532 {
1533 	struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1534 	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1535 	struct kvm_vcpu *vcpu = apic->vcpu;
1536 	wait_queue_head_t *q = &vcpu->wq;
1537 
1538 	/*
1539 	 * There is a race window between reading and incrementing, but we do
1540 	 * not care about potentially losing timer events in the !reinject
1541 	 * case anyway. Note: KVM_REQ_PENDING_TIMER is implicitly checked
1542 	 * in vcpu_enter_guest.
1543 	 */
1544 	if (!atomic_read(&ktimer->pending)) {
1545 		atomic_inc(&ktimer->pending);
1546 		/* FIXME: this code should not know anything about vcpus */
1547 		kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1548 	}
1549 
1550 	if (waitqueue_active(q))
1551 		wake_up_interruptible(q);
1552 
1553 	if (lapic_is_periodic(apic)) {
1554 		hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1555 		return HRTIMER_RESTART;
1556 	} else
1557 		return HRTIMER_NORESTART;
1558 }
1559 
1560 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1561 {
1562 	struct kvm_lapic *apic;
1563 
1564 	ASSERT(vcpu != NULL);
1565 	apic_debug("apic_init %d\n", vcpu->vcpu_id);
1566 
1567 	apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1568 	if (!apic)
1569 		goto nomem;
1570 
1571 	vcpu->arch.apic = apic;
1572 
1573 	apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1574 	if (!apic->regs) {
1575 		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1576 		       vcpu->vcpu_id);
1577 		goto nomem_free_apic;
1578 	}
1579 	apic->vcpu = vcpu;
1580 
1581 	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1582 		     HRTIMER_MODE_ABS);
1583 	apic->lapic_timer.timer.function = apic_timer_fn;
1584 
1585 	/*
1586 	 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1587 	 * thinking that APIC satet has changed.
1588 	 */
1589 	vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1590 	kvm_lapic_set_base(vcpu,
1591 			APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE);
1592 
1593 	static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1594 	kvm_lapic_reset(vcpu);
1595 	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1596 
1597 	return 0;
1598 nomem_free_apic:
1599 	kfree(apic);
1600 nomem:
1601 	return -ENOMEM;
1602 }
1603 
1604 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1605 {
1606 	struct kvm_lapic *apic = vcpu->arch.apic;
1607 	int highest_irr;
1608 
1609 	if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic))
1610 		return -1;
1611 
1612 	apic_update_ppr(apic);
1613 	highest_irr = apic_find_highest_irr(apic);
1614 	if ((highest_irr == -1) ||
1615 	    ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI)))
1616 		return -1;
1617 	return highest_irr;
1618 }
1619 
1620 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1621 {
1622 	u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1623 	int r = 0;
1624 
1625 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1626 		r = 1;
1627 	if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1628 	    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1629 		r = 1;
1630 	return r;
1631 }
1632 
1633 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1634 {
1635 	struct kvm_lapic *apic = vcpu->arch.apic;
1636 
1637 	if (!kvm_vcpu_has_lapic(vcpu))
1638 		return;
1639 
1640 	if (atomic_read(&apic->lapic_timer.pending) > 0) {
1641 		kvm_apic_local_deliver(apic, APIC_LVTT);
1642 		atomic_set(&apic->lapic_timer.pending, 0);
1643 	}
1644 }
1645 
1646 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1647 {
1648 	int vector = kvm_apic_has_interrupt(vcpu);
1649 	struct kvm_lapic *apic = vcpu->arch.apic;
1650 
1651 	if (vector == -1)
1652 		return -1;
1653 
1654 	/*
1655 	 * We get here even with APIC virtualization enabled, if doing
1656 	 * nested virtualization and L1 runs with the "acknowledge interrupt
1657 	 * on exit" mode.  Then we cannot inject the interrupt via RVI,
1658 	 * because the process would deliver it through the IDT.
1659 	 */
1660 
1661 	apic_set_isr(vector, apic);
1662 	apic_update_ppr(apic);
1663 	apic_clear_irr(vector, apic);
1664 	return vector;
1665 }
1666 
1667 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
1668 		struct kvm_lapic_state *s)
1669 {
1670 	struct kvm_lapic *apic = vcpu->arch.apic;
1671 
1672 	kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
1673 	/* set SPIV separately to get count of SW disabled APICs right */
1674 	apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
1675 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1676 	/* call kvm_apic_set_id() to put apic into apic_map */
1677 	kvm_apic_set_id(apic, kvm_apic_id(apic));
1678 	kvm_apic_set_version(vcpu);
1679 
1680 	apic_update_ppr(apic);
1681 	hrtimer_cancel(&apic->lapic_timer.timer);
1682 	update_divide_count(apic);
1683 	start_apic_timer(apic);
1684 	apic->irr_pending = true;
1685 	apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm) ?
1686 				1 : count_vectors(apic->regs + APIC_ISR);
1687 	apic->highest_isr_cache = -1;
1688 	kvm_x86_ops->hwapic_isr_update(vcpu->kvm, apic_find_highest_isr(apic));
1689 	kvm_make_request(KVM_REQ_EVENT, vcpu);
1690 	kvm_rtc_eoi_tracking_restore_one(vcpu);
1691 }
1692 
1693 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1694 {
1695 	struct hrtimer *timer;
1696 
1697 	if (!kvm_vcpu_has_lapic(vcpu))
1698 		return;
1699 
1700 	timer = &vcpu->arch.apic->lapic_timer.timer;
1701 	if (hrtimer_cancel(timer))
1702 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1703 }
1704 
1705 /*
1706  * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1707  *
1708  * Detect whether guest triggered PV EOI since the
1709  * last entry. If yes, set EOI on guests's behalf.
1710  * Clear PV EOI in guest memory in any case.
1711  */
1712 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1713 					struct kvm_lapic *apic)
1714 {
1715 	bool pending;
1716 	int vector;
1717 	/*
1718 	 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1719 	 * and KVM_PV_EOI_ENABLED in guest memory as follows:
1720 	 *
1721 	 * KVM_APIC_PV_EOI_PENDING is unset:
1722 	 * 	-> host disabled PV EOI.
1723 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1724 	 * 	-> host enabled PV EOI, guest did not execute EOI yet.
1725 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1726 	 * 	-> host enabled PV EOI, guest executed EOI.
1727 	 */
1728 	BUG_ON(!pv_eoi_enabled(vcpu));
1729 	pending = pv_eoi_get_pending(vcpu);
1730 	/*
1731 	 * Clear pending bit in any case: it will be set again on vmentry.
1732 	 * While this might not be ideal from performance point of view,
1733 	 * this makes sure pv eoi is only enabled when we know it's safe.
1734 	 */
1735 	pv_eoi_clr_pending(vcpu);
1736 	if (pending)
1737 		return;
1738 	vector = apic_set_eoi(apic);
1739 	trace_kvm_pv_eoi(apic, vector);
1740 }
1741 
1742 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1743 {
1744 	u32 data;
1745 
1746 	if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1747 		apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1748 
1749 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1750 		return;
1751 
1752 	kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1753 				sizeof(u32));
1754 
1755 	apic_set_tpr(vcpu->arch.apic, data & 0xff);
1756 }
1757 
1758 /*
1759  * apic_sync_pv_eoi_to_guest - called before vmentry
1760  *
1761  * Detect whether it's safe to enable PV EOI and
1762  * if yes do so.
1763  */
1764 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
1765 					struct kvm_lapic *apic)
1766 {
1767 	if (!pv_eoi_enabled(vcpu) ||
1768 	    /* IRR set or many bits in ISR: could be nested. */
1769 	    apic->irr_pending ||
1770 	    /* Cache not set: could be safe but we don't bother. */
1771 	    apic->highest_isr_cache == -1 ||
1772 	    /* Need EOI to update ioapic. */
1773 	    kvm_ioapic_handles_vector(vcpu->kvm, apic->highest_isr_cache)) {
1774 		/*
1775 		 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1776 		 * so we need not do anything here.
1777 		 */
1778 		return;
1779 	}
1780 
1781 	pv_eoi_set_pending(apic->vcpu);
1782 }
1783 
1784 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1785 {
1786 	u32 data, tpr;
1787 	int max_irr, max_isr;
1788 	struct kvm_lapic *apic = vcpu->arch.apic;
1789 
1790 	apic_sync_pv_eoi_to_guest(vcpu, apic);
1791 
1792 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1793 		return;
1794 
1795 	tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1796 	max_irr = apic_find_highest_irr(apic);
1797 	if (max_irr < 0)
1798 		max_irr = 0;
1799 	max_isr = apic_find_highest_isr(apic);
1800 	if (max_isr < 0)
1801 		max_isr = 0;
1802 	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1803 
1804 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1805 				sizeof(u32));
1806 }
1807 
1808 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1809 {
1810 	if (vapic_addr) {
1811 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1812 					&vcpu->arch.apic->vapic_cache,
1813 					vapic_addr, sizeof(u32)))
1814 			return -EINVAL;
1815 		__set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1816 	} else {
1817 		__clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1818 	}
1819 
1820 	vcpu->arch.apic->vapic_addr = vapic_addr;
1821 	return 0;
1822 }
1823 
1824 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1825 {
1826 	struct kvm_lapic *apic = vcpu->arch.apic;
1827 	u32 reg = (msr - APIC_BASE_MSR) << 4;
1828 
1829 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1830 		return 1;
1831 
1832 	/* if this is ICR write vector before command */
1833 	if (msr == 0x830)
1834 		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1835 	return apic_reg_write(apic, reg, (u32)data);
1836 }
1837 
1838 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1839 {
1840 	struct kvm_lapic *apic = vcpu->arch.apic;
1841 	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1842 
1843 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1844 		return 1;
1845 
1846 	if (apic_reg_read(apic, reg, 4, &low))
1847 		return 1;
1848 	if (msr == 0x830)
1849 		apic_reg_read(apic, APIC_ICR2, 4, &high);
1850 
1851 	*data = (((u64)high) << 32) | low;
1852 
1853 	return 0;
1854 }
1855 
1856 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1857 {
1858 	struct kvm_lapic *apic = vcpu->arch.apic;
1859 
1860 	if (!kvm_vcpu_has_lapic(vcpu))
1861 		return 1;
1862 
1863 	/* if this is ICR write vector before command */
1864 	if (reg == APIC_ICR)
1865 		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1866 	return apic_reg_write(apic, reg, (u32)data);
1867 }
1868 
1869 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
1870 {
1871 	struct kvm_lapic *apic = vcpu->arch.apic;
1872 	u32 low, high = 0;
1873 
1874 	if (!kvm_vcpu_has_lapic(vcpu))
1875 		return 1;
1876 
1877 	if (apic_reg_read(apic, reg, 4, &low))
1878 		return 1;
1879 	if (reg == APIC_ICR)
1880 		apic_reg_read(apic, APIC_ICR2, 4, &high);
1881 
1882 	*data = (((u64)high) << 32) | low;
1883 
1884 	return 0;
1885 }
1886 
1887 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
1888 {
1889 	u64 addr = data & ~KVM_MSR_ENABLED;
1890 	if (!IS_ALIGNED(addr, 4))
1891 		return 1;
1892 
1893 	vcpu->arch.pv_eoi.msr_val = data;
1894 	if (!pv_eoi_enabled(vcpu))
1895 		return 0;
1896 	return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
1897 					 addr, sizeof(u8));
1898 }
1899 
1900 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
1901 {
1902 	struct kvm_lapic *apic = vcpu->arch.apic;
1903 	unsigned int sipi_vector;
1904 	unsigned long pe;
1905 
1906 	if (!kvm_vcpu_has_lapic(vcpu) || !apic->pending_events)
1907 		return;
1908 
1909 	pe = xchg(&apic->pending_events, 0);
1910 
1911 	if (test_bit(KVM_APIC_INIT, &pe)) {
1912 		kvm_lapic_reset(vcpu);
1913 		kvm_vcpu_reset(vcpu);
1914 		if (kvm_vcpu_is_bsp(apic->vcpu))
1915 			vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1916 		else
1917 			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
1918 	}
1919 	if (test_bit(KVM_APIC_SIPI, &pe) &&
1920 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
1921 		/* evaluate pending_events before reading the vector */
1922 		smp_rmb();
1923 		sipi_vector = apic->sipi_vector;
1924 		apic_debug("vcpu %d received sipi with vector # %x\n",
1925 			 vcpu->vcpu_id, sipi_vector);
1926 		kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
1927 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1928 	}
1929 }
1930 
1931 void kvm_lapic_init(void)
1932 {
1933 	/* do not patch jump label more than once per second */
1934 	jump_label_rate_limit(&apic_hw_disabled, HZ);
1935 	jump_label_rate_limit(&apic_sw_disabled, HZ);
1936 }
1937