xref: /openbmc/linux/arch/x86/kvm/lapic.c (revision e9d90d472da97e1b1560bffb89578ba082c88a69)
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 "kvm_cache_regs.h"
38 #include "irq.h"
39 #include "trace.h"
40 #include "x86.h"
41 #include "cpuid.h"
42 
43 #ifndef CONFIG_X86_64
44 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
45 #else
46 #define mod_64(x, y) ((x) % (y))
47 #endif
48 
49 #define PRId64 "d"
50 #define PRIx64 "llx"
51 #define PRIu64 "u"
52 #define PRIo64 "o"
53 
54 #define APIC_BUS_CYCLE_NS 1
55 
56 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
57 #define apic_debug(fmt, arg...)
58 
59 #define APIC_LVT_NUM			6
60 /* 14 is the version for Xeon and Pentium 8.4.8*/
61 #define APIC_VERSION			(0x14UL | ((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 
69 #define VEC_POS(v) ((v) & (32 - 1))
70 #define REG_POS(v) (((v) >> 5) << 4)
71 
72 static unsigned int min_timer_period_us = 500;
73 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
74 
75 static inline u32 apic_get_reg(struct kvm_lapic *apic, int reg_off)
76 {
77 	return *((u32 *) (apic->regs + reg_off));
78 }
79 
80 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
81 {
82 	*((u32 *) (apic->regs + reg_off)) = val;
83 }
84 
85 static inline int apic_test_and_set_vector(int vec, void *bitmap)
86 {
87 	return test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
88 }
89 
90 static inline int apic_test_and_clear_vector(int vec, void *bitmap)
91 {
92 	return test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
93 }
94 
95 static inline int apic_test_vector(int vec, void *bitmap)
96 {
97 	return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
98 }
99 
100 static inline void apic_set_vector(int vec, void *bitmap)
101 {
102 	set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
103 }
104 
105 static inline void apic_clear_vector(int vec, void *bitmap)
106 {
107 	clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
108 }
109 
110 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
111 {
112 	return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
113 }
114 
115 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
116 {
117 	return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
118 }
119 
120 static inline int apic_hw_enabled(struct kvm_lapic *apic)
121 {
122 	return (apic)->vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE;
123 }
124 
125 static inline int  apic_sw_enabled(struct kvm_lapic *apic)
126 {
127 	return apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED;
128 }
129 
130 static inline int apic_enabled(struct kvm_lapic *apic)
131 {
132 	return apic_sw_enabled(apic) &&	apic_hw_enabled(apic);
133 }
134 
135 #define LVT_MASK	\
136 	(APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
137 
138 #define LINT_MASK	\
139 	(LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
140 	 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
141 
142 static inline int kvm_apic_id(struct kvm_lapic *apic)
143 {
144 	return (apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
145 }
146 
147 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
148 {
149 	return !(apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
150 }
151 
152 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
153 {
154 	return apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
155 }
156 
157 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
158 {
159 	return ((apic_get_reg(apic, APIC_LVTT) &
160 		apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_ONESHOT);
161 }
162 
163 static inline int apic_lvtt_period(struct kvm_lapic *apic)
164 {
165 	return ((apic_get_reg(apic, APIC_LVTT) &
166 		apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_PERIODIC);
167 }
168 
169 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
170 {
171 	return ((apic_get_reg(apic, APIC_LVTT) &
172 		apic->lapic_timer.timer_mode_mask) ==
173 			APIC_LVT_TIMER_TSCDEADLINE);
174 }
175 
176 static inline int apic_lvt_nmi_mode(u32 lvt_val)
177 {
178 	return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
179 }
180 
181 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
182 {
183 	struct kvm_lapic *apic = vcpu->arch.apic;
184 	struct kvm_cpuid_entry2 *feat;
185 	u32 v = APIC_VERSION;
186 
187 	if (!irqchip_in_kernel(vcpu->kvm))
188 		return;
189 
190 	feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
191 	if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
192 		v |= APIC_LVR_DIRECTED_EOI;
193 	apic_set_reg(apic, APIC_LVR, v);
194 }
195 
196 static inline int apic_x2apic_mode(struct kvm_lapic *apic)
197 {
198 	return apic->vcpu->arch.apic_base & X2APIC_ENABLE;
199 }
200 
201 static unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
202 	LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
203 	LVT_MASK | APIC_MODE_MASK,	/* LVTTHMR */
204 	LVT_MASK | APIC_MODE_MASK,	/* LVTPC */
205 	LINT_MASK, LINT_MASK,	/* LVT0-1 */
206 	LVT_MASK		/* LVTERR */
207 };
208 
209 static int find_highest_vector(void *bitmap)
210 {
211 	u32 *word = bitmap;
212 	int word_offset = MAX_APIC_VECTOR >> 5;
213 
214 	while ((word_offset != 0) && (word[(--word_offset) << 2] == 0))
215 		continue;
216 
217 	if (likely(!word_offset && !word[0]))
218 		return -1;
219 	else
220 		return fls(word[word_offset << 2]) - 1 + (word_offset << 5);
221 }
222 
223 static u8 count_vectors(void *bitmap)
224 {
225 	u32 *word = bitmap;
226 	int word_offset;
227 	u8 count = 0;
228 	for (word_offset = 0; word_offset < MAX_APIC_VECTOR >> 5; ++word_offset)
229 		count += hweight32(word[word_offset << 2]);
230 	return count;
231 }
232 
233 static inline int apic_test_and_set_irr(int vec, struct kvm_lapic *apic)
234 {
235 	apic->irr_pending = true;
236 	return apic_test_and_set_vector(vec, apic->regs + APIC_IRR);
237 }
238 
239 static inline int apic_search_irr(struct kvm_lapic *apic)
240 {
241 	return find_highest_vector(apic->regs + APIC_IRR);
242 }
243 
244 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
245 {
246 	int result;
247 
248 	if (!apic->irr_pending)
249 		return -1;
250 
251 	result = apic_search_irr(apic);
252 	ASSERT(result == -1 || result >= 16);
253 
254 	return result;
255 }
256 
257 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
258 {
259 	apic->irr_pending = false;
260 	apic_clear_vector(vec, apic->regs + APIC_IRR);
261 	if (apic_search_irr(apic) != -1)
262 		apic->irr_pending = true;
263 }
264 
265 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
266 {
267 	if (!__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
268 		++apic->isr_count;
269 	BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
270 	/*
271 	 * ISR (in service register) bit is set when injecting an interrupt.
272 	 * The highest vector is injected. Thus the latest bit set matches
273 	 * the highest bit in ISR.
274 	 */
275 	apic->highest_isr_cache = vec;
276 }
277 
278 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
279 {
280 	if (__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
281 		--apic->isr_count;
282 	BUG_ON(apic->isr_count < 0);
283 	apic->highest_isr_cache = -1;
284 }
285 
286 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
287 {
288 	struct kvm_lapic *apic = vcpu->arch.apic;
289 	int highest_irr;
290 
291 	/* This may race with setting of irr in __apic_accept_irq() and
292 	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
293 	 * will cause vmexit immediately and the value will be recalculated
294 	 * on the next vmentry.
295 	 */
296 	if (!apic)
297 		return 0;
298 	highest_irr = apic_find_highest_irr(apic);
299 
300 	return highest_irr;
301 }
302 
303 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
304 			     int vector, int level, int trig_mode);
305 
306 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq)
307 {
308 	struct kvm_lapic *apic = vcpu->arch.apic;
309 
310 	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
311 			irq->level, irq->trig_mode);
312 }
313 
314 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
315 {
316 
317 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
318 				      sizeof(val));
319 }
320 
321 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
322 {
323 
324 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
325 				      sizeof(*val));
326 }
327 
328 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
329 {
330 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
331 }
332 
333 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
334 {
335 	u8 val;
336 	if (pv_eoi_get_user(vcpu, &val) < 0)
337 		apic_debug("Can't read EOI MSR value: 0x%llx\n",
338 			   (unsigned long long)vcpi->arch.pv_eoi.msr_val);
339 	return val & 0x1;
340 }
341 
342 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
343 {
344 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
345 		apic_debug("Can't set EOI MSR value: 0x%llx\n",
346 			   (unsigned long long)vcpi->arch.pv_eoi.msr_val);
347 		return;
348 	}
349 	__set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
350 }
351 
352 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
353 {
354 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
355 		apic_debug("Can't clear EOI MSR value: 0x%llx\n",
356 			   (unsigned long long)vcpi->arch.pv_eoi.msr_val);
357 		return;
358 	}
359 	__clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
360 }
361 
362 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
363 {
364 	int result;
365 	if (!apic->isr_count)
366 		return -1;
367 	if (likely(apic->highest_isr_cache != -1))
368 		return apic->highest_isr_cache;
369 
370 	result = find_highest_vector(apic->regs + APIC_ISR);
371 	ASSERT(result == -1 || result >= 16);
372 
373 	return result;
374 }
375 
376 static void apic_update_ppr(struct kvm_lapic *apic)
377 {
378 	u32 tpr, isrv, ppr, old_ppr;
379 	int isr;
380 
381 	old_ppr = apic_get_reg(apic, APIC_PROCPRI);
382 	tpr = apic_get_reg(apic, APIC_TASKPRI);
383 	isr = apic_find_highest_isr(apic);
384 	isrv = (isr != -1) ? isr : 0;
385 
386 	if ((tpr & 0xf0) >= (isrv & 0xf0))
387 		ppr = tpr & 0xff;
388 	else
389 		ppr = isrv & 0xf0;
390 
391 	apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
392 		   apic, ppr, isr, isrv);
393 
394 	if (old_ppr != ppr) {
395 		apic_set_reg(apic, APIC_PROCPRI, ppr);
396 		if (ppr < old_ppr)
397 			kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
398 	}
399 }
400 
401 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
402 {
403 	apic_set_reg(apic, APIC_TASKPRI, tpr);
404 	apic_update_ppr(apic);
405 }
406 
407 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
408 {
409 	return dest == 0xff || kvm_apic_id(apic) == dest;
410 }
411 
412 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
413 {
414 	int result = 0;
415 	u32 logical_id;
416 
417 	if (apic_x2apic_mode(apic)) {
418 		logical_id = apic_get_reg(apic, APIC_LDR);
419 		return logical_id & mda;
420 	}
421 
422 	logical_id = GET_APIC_LOGICAL_ID(apic_get_reg(apic, APIC_LDR));
423 
424 	switch (apic_get_reg(apic, APIC_DFR)) {
425 	case APIC_DFR_FLAT:
426 		if (logical_id & mda)
427 			result = 1;
428 		break;
429 	case APIC_DFR_CLUSTER:
430 		if (((logical_id >> 4) == (mda >> 0x4))
431 		    && (logical_id & mda & 0xf))
432 			result = 1;
433 		break;
434 	default:
435 		apic_debug("Bad DFR vcpu %d: %08x\n",
436 			   apic->vcpu->vcpu_id, apic_get_reg(apic, APIC_DFR));
437 		break;
438 	}
439 
440 	return result;
441 }
442 
443 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
444 			   int short_hand, int dest, int dest_mode)
445 {
446 	int result = 0;
447 	struct kvm_lapic *target = vcpu->arch.apic;
448 
449 	apic_debug("target %p, source %p, dest 0x%x, "
450 		   "dest_mode 0x%x, short_hand 0x%x\n",
451 		   target, source, dest, dest_mode, short_hand);
452 
453 	ASSERT(target);
454 	switch (short_hand) {
455 	case APIC_DEST_NOSHORT:
456 		if (dest_mode == 0)
457 			/* Physical mode. */
458 			result = kvm_apic_match_physical_addr(target, dest);
459 		else
460 			/* Logical mode. */
461 			result = kvm_apic_match_logical_addr(target, dest);
462 		break;
463 	case APIC_DEST_SELF:
464 		result = (target == source);
465 		break;
466 	case APIC_DEST_ALLINC:
467 		result = 1;
468 		break;
469 	case APIC_DEST_ALLBUT:
470 		result = (target != source);
471 		break;
472 	default:
473 		apic_debug("kvm: apic: Bad dest shorthand value %x\n",
474 			   short_hand);
475 		break;
476 	}
477 
478 	return result;
479 }
480 
481 /*
482  * Add a pending IRQ into lapic.
483  * Return 1 if successfully added and 0 if discarded.
484  */
485 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
486 			     int vector, int level, int trig_mode)
487 {
488 	int result = 0;
489 	struct kvm_vcpu *vcpu = apic->vcpu;
490 
491 	switch (delivery_mode) {
492 	case APIC_DM_LOWEST:
493 		vcpu->arch.apic_arb_prio++;
494 	case APIC_DM_FIXED:
495 		/* FIXME add logic for vcpu on reset */
496 		if (unlikely(!apic_enabled(apic)))
497 			break;
498 
499 		if (trig_mode) {
500 			apic_debug("level trig mode for vector %d", vector);
501 			apic_set_vector(vector, apic->regs + APIC_TMR);
502 		} else
503 			apic_clear_vector(vector, apic->regs + APIC_TMR);
504 
505 		result = !apic_test_and_set_irr(vector, apic);
506 		trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
507 					  trig_mode, vector, !result);
508 		if (!result) {
509 			if (trig_mode)
510 				apic_debug("level trig mode repeatedly for "
511 						"vector %d", vector);
512 			break;
513 		}
514 
515 		kvm_make_request(KVM_REQ_EVENT, vcpu);
516 		kvm_vcpu_kick(vcpu);
517 		break;
518 
519 	case APIC_DM_REMRD:
520 		apic_debug("Ignoring delivery mode 3\n");
521 		break;
522 
523 	case APIC_DM_SMI:
524 		apic_debug("Ignoring guest SMI\n");
525 		break;
526 
527 	case APIC_DM_NMI:
528 		result = 1;
529 		kvm_inject_nmi(vcpu);
530 		kvm_vcpu_kick(vcpu);
531 		break;
532 
533 	case APIC_DM_INIT:
534 		if (!trig_mode || level) {
535 			result = 1;
536 			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
537 			kvm_make_request(KVM_REQ_EVENT, vcpu);
538 			kvm_vcpu_kick(vcpu);
539 		} else {
540 			apic_debug("Ignoring de-assert INIT to vcpu %d\n",
541 				   vcpu->vcpu_id);
542 		}
543 		break;
544 
545 	case APIC_DM_STARTUP:
546 		apic_debug("SIPI to vcpu %d vector 0x%02x\n",
547 			   vcpu->vcpu_id, vector);
548 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
549 			result = 1;
550 			vcpu->arch.sipi_vector = vector;
551 			vcpu->arch.mp_state = KVM_MP_STATE_SIPI_RECEIVED;
552 			kvm_make_request(KVM_REQ_EVENT, vcpu);
553 			kvm_vcpu_kick(vcpu);
554 		}
555 		break;
556 
557 	case APIC_DM_EXTINT:
558 		/*
559 		 * Should only be called by kvm_apic_local_deliver() with LVT0,
560 		 * before NMI watchdog was enabled. Already handled by
561 		 * kvm_apic_accept_pic_intr().
562 		 */
563 		break;
564 
565 	default:
566 		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
567 		       delivery_mode);
568 		break;
569 	}
570 	return result;
571 }
572 
573 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
574 {
575 	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
576 }
577 
578 static int apic_set_eoi(struct kvm_lapic *apic)
579 {
580 	int vector = apic_find_highest_isr(apic);
581 
582 	trace_kvm_eoi(apic, vector);
583 
584 	/*
585 	 * Not every write EOI will has corresponding ISR,
586 	 * one example is when Kernel check timer on setup_IO_APIC
587 	 */
588 	if (vector == -1)
589 		return vector;
590 
591 	apic_clear_isr(vector, apic);
592 	apic_update_ppr(apic);
593 
594 	if (!(apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI) &&
595 	    kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
596 		int trigger_mode;
597 		if (apic_test_vector(vector, apic->regs + APIC_TMR))
598 			trigger_mode = IOAPIC_LEVEL_TRIG;
599 		else
600 			trigger_mode = IOAPIC_EDGE_TRIG;
601 		kvm_ioapic_update_eoi(apic->vcpu->kvm, vector, trigger_mode);
602 	}
603 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
604 	return vector;
605 }
606 
607 static void apic_send_ipi(struct kvm_lapic *apic)
608 {
609 	u32 icr_low = apic_get_reg(apic, APIC_ICR);
610 	u32 icr_high = apic_get_reg(apic, APIC_ICR2);
611 	struct kvm_lapic_irq irq;
612 
613 	irq.vector = icr_low & APIC_VECTOR_MASK;
614 	irq.delivery_mode = icr_low & APIC_MODE_MASK;
615 	irq.dest_mode = icr_low & APIC_DEST_MASK;
616 	irq.level = icr_low & APIC_INT_ASSERT;
617 	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
618 	irq.shorthand = icr_low & APIC_SHORT_MASK;
619 	if (apic_x2apic_mode(apic))
620 		irq.dest_id = icr_high;
621 	else
622 		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
623 
624 	trace_kvm_apic_ipi(icr_low, irq.dest_id);
625 
626 	apic_debug("icr_high 0x%x, icr_low 0x%x, "
627 		   "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
628 		   "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
629 		   icr_high, icr_low, irq.shorthand, irq.dest_id,
630 		   irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
631 		   irq.vector);
632 
633 	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq);
634 }
635 
636 static u32 apic_get_tmcct(struct kvm_lapic *apic)
637 {
638 	ktime_t remaining;
639 	s64 ns;
640 	u32 tmcct;
641 
642 	ASSERT(apic != NULL);
643 
644 	/* if initial count is 0, current count should also be 0 */
645 	if (apic_get_reg(apic, APIC_TMICT) == 0)
646 		return 0;
647 
648 	remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
649 	if (ktime_to_ns(remaining) < 0)
650 		remaining = ktime_set(0, 0);
651 
652 	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
653 	tmcct = div64_u64(ns,
654 			 (APIC_BUS_CYCLE_NS * apic->divide_count));
655 
656 	return tmcct;
657 }
658 
659 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
660 {
661 	struct kvm_vcpu *vcpu = apic->vcpu;
662 	struct kvm_run *run = vcpu->run;
663 
664 	kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
665 	run->tpr_access.rip = kvm_rip_read(vcpu);
666 	run->tpr_access.is_write = write;
667 }
668 
669 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
670 {
671 	if (apic->vcpu->arch.tpr_access_reporting)
672 		__report_tpr_access(apic, write);
673 }
674 
675 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
676 {
677 	u32 val = 0;
678 
679 	if (offset >= LAPIC_MMIO_LENGTH)
680 		return 0;
681 
682 	switch (offset) {
683 	case APIC_ID:
684 		if (apic_x2apic_mode(apic))
685 			val = kvm_apic_id(apic);
686 		else
687 			val = kvm_apic_id(apic) << 24;
688 		break;
689 	case APIC_ARBPRI:
690 		apic_debug("Access APIC ARBPRI register which is for P6\n");
691 		break;
692 
693 	case APIC_TMCCT:	/* Timer CCR */
694 		if (apic_lvtt_tscdeadline(apic))
695 			return 0;
696 
697 		val = apic_get_tmcct(apic);
698 		break;
699 	case APIC_PROCPRI:
700 		apic_update_ppr(apic);
701 		val = apic_get_reg(apic, offset);
702 		break;
703 	case APIC_TASKPRI:
704 		report_tpr_access(apic, false);
705 		/* fall thru */
706 	default:
707 		val = apic_get_reg(apic, offset);
708 		break;
709 	}
710 
711 	return val;
712 }
713 
714 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
715 {
716 	return container_of(dev, struct kvm_lapic, dev);
717 }
718 
719 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
720 		void *data)
721 {
722 	unsigned char alignment = offset & 0xf;
723 	u32 result;
724 	/* this bitmask has a bit cleared for each reserved register */
725 	static const u64 rmask = 0x43ff01ffffffe70cULL;
726 
727 	if ((alignment + len) > 4) {
728 		apic_debug("KVM_APIC_READ: alignment error %x %d\n",
729 			   offset, len);
730 		return 1;
731 	}
732 
733 	if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
734 		apic_debug("KVM_APIC_READ: read reserved register %x\n",
735 			   offset);
736 		return 1;
737 	}
738 
739 	result = __apic_read(apic, offset & ~0xf);
740 
741 	trace_kvm_apic_read(offset, result);
742 
743 	switch (len) {
744 	case 1:
745 	case 2:
746 	case 4:
747 		memcpy(data, (char *)&result + alignment, len);
748 		break;
749 	default:
750 		printk(KERN_ERR "Local APIC read with len = %x, "
751 		       "should be 1,2, or 4 instead\n", len);
752 		break;
753 	}
754 	return 0;
755 }
756 
757 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
758 {
759 	return apic_hw_enabled(apic) &&
760 	    addr >= apic->base_address &&
761 	    addr < apic->base_address + LAPIC_MMIO_LENGTH;
762 }
763 
764 static int apic_mmio_read(struct kvm_io_device *this,
765 			   gpa_t address, int len, void *data)
766 {
767 	struct kvm_lapic *apic = to_lapic(this);
768 	u32 offset = address - apic->base_address;
769 
770 	if (!apic_mmio_in_range(apic, address))
771 		return -EOPNOTSUPP;
772 
773 	apic_reg_read(apic, offset, len, data);
774 
775 	return 0;
776 }
777 
778 static void update_divide_count(struct kvm_lapic *apic)
779 {
780 	u32 tmp1, tmp2, tdcr;
781 
782 	tdcr = apic_get_reg(apic, APIC_TDCR);
783 	tmp1 = tdcr & 0xf;
784 	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
785 	apic->divide_count = 0x1 << (tmp2 & 0x7);
786 
787 	apic_debug("timer divide count is 0x%x\n",
788 				   apic->divide_count);
789 }
790 
791 static void start_apic_timer(struct kvm_lapic *apic)
792 {
793 	ktime_t now;
794 	atomic_set(&apic->lapic_timer.pending, 0);
795 
796 	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
797 		/* lapic timer in oneshot or periodic mode */
798 		now = apic->lapic_timer.timer.base->get_time();
799 		apic->lapic_timer.period = (u64)apic_get_reg(apic, APIC_TMICT)
800 			    * APIC_BUS_CYCLE_NS * apic->divide_count;
801 
802 		if (!apic->lapic_timer.period)
803 			return;
804 		/*
805 		 * Do not allow the guest to program periodic timers with small
806 		 * interval, since the hrtimers are not throttled by the host
807 		 * scheduler.
808 		 */
809 		if (apic_lvtt_period(apic)) {
810 			s64 min_period = min_timer_period_us * 1000LL;
811 
812 			if (apic->lapic_timer.period < min_period) {
813 				pr_info_ratelimited(
814 				    "kvm: vcpu %i: requested %lld ns "
815 				    "lapic timer period limited to %lld ns\n",
816 				    apic->vcpu->vcpu_id,
817 				    apic->lapic_timer.period, min_period);
818 				apic->lapic_timer.period = min_period;
819 			}
820 		}
821 
822 		hrtimer_start(&apic->lapic_timer.timer,
823 			      ktime_add_ns(now, apic->lapic_timer.period),
824 			      HRTIMER_MODE_ABS);
825 
826 		apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
827 			   PRIx64 ", "
828 			   "timer initial count 0x%x, period %lldns, "
829 			   "expire @ 0x%016" PRIx64 ".\n", __func__,
830 			   APIC_BUS_CYCLE_NS, ktime_to_ns(now),
831 			   apic_get_reg(apic, APIC_TMICT),
832 			   apic->lapic_timer.period,
833 			   ktime_to_ns(ktime_add_ns(now,
834 					apic->lapic_timer.period)));
835 	} else if (apic_lvtt_tscdeadline(apic)) {
836 		/* lapic timer in tsc deadline mode */
837 		u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
838 		u64 ns = 0;
839 		struct kvm_vcpu *vcpu = apic->vcpu;
840 		unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
841 		unsigned long flags;
842 
843 		if (unlikely(!tscdeadline || !this_tsc_khz))
844 			return;
845 
846 		local_irq_save(flags);
847 
848 		now = apic->lapic_timer.timer.base->get_time();
849 		guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu);
850 		if (likely(tscdeadline > guest_tsc)) {
851 			ns = (tscdeadline - guest_tsc) * 1000000ULL;
852 			do_div(ns, this_tsc_khz);
853 		}
854 		hrtimer_start(&apic->lapic_timer.timer,
855 			ktime_add_ns(now, ns), HRTIMER_MODE_ABS);
856 
857 		local_irq_restore(flags);
858 	}
859 }
860 
861 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
862 {
863 	int nmi_wd_enabled = apic_lvt_nmi_mode(apic_get_reg(apic, APIC_LVT0));
864 
865 	if (apic_lvt_nmi_mode(lvt0_val)) {
866 		if (!nmi_wd_enabled) {
867 			apic_debug("Receive NMI setting on APIC_LVT0 "
868 				   "for cpu %d\n", apic->vcpu->vcpu_id);
869 			apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
870 		}
871 	} else if (nmi_wd_enabled)
872 		apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
873 }
874 
875 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
876 {
877 	int ret = 0;
878 
879 	trace_kvm_apic_write(reg, val);
880 
881 	switch (reg) {
882 	case APIC_ID:		/* Local APIC ID */
883 		if (!apic_x2apic_mode(apic))
884 			apic_set_reg(apic, APIC_ID, val);
885 		else
886 			ret = 1;
887 		break;
888 
889 	case APIC_TASKPRI:
890 		report_tpr_access(apic, true);
891 		apic_set_tpr(apic, val & 0xff);
892 		break;
893 
894 	case APIC_EOI:
895 		apic_set_eoi(apic);
896 		break;
897 
898 	case APIC_LDR:
899 		if (!apic_x2apic_mode(apic))
900 			apic_set_reg(apic, APIC_LDR, val & APIC_LDR_MASK);
901 		else
902 			ret = 1;
903 		break;
904 
905 	case APIC_DFR:
906 		if (!apic_x2apic_mode(apic))
907 			apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
908 		else
909 			ret = 1;
910 		break;
911 
912 	case APIC_SPIV: {
913 		u32 mask = 0x3ff;
914 		if (apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
915 			mask |= APIC_SPIV_DIRECTED_EOI;
916 		apic_set_reg(apic, APIC_SPIV, val & mask);
917 		if (!(val & APIC_SPIV_APIC_ENABLED)) {
918 			int i;
919 			u32 lvt_val;
920 
921 			for (i = 0; i < APIC_LVT_NUM; i++) {
922 				lvt_val = apic_get_reg(apic,
923 						       APIC_LVTT + 0x10 * i);
924 				apic_set_reg(apic, APIC_LVTT + 0x10 * i,
925 					     lvt_val | APIC_LVT_MASKED);
926 			}
927 			atomic_set(&apic->lapic_timer.pending, 0);
928 
929 		}
930 		break;
931 	}
932 	case APIC_ICR:
933 		/* No delay here, so we always clear the pending bit */
934 		apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
935 		apic_send_ipi(apic);
936 		break;
937 
938 	case APIC_ICR2:
939 		if (!apic_x2apic_mode(apic))
940 			val &= 0xff000000;
941 		apic_set_reg(apic, APIC_ICR2, val);
942 		break;
943 
944 	case APIC_LVT0:
945 		apic_manage_nmi_watchdog(apic, val);
946 	case APIC_LVTTHMR:
947 	case APIC_LVTPC:
948 	case APIC_LVT1:
949 	case APIC_LVTERR:
950 		/* TODO: Check vector */
951 		if (!apic_sw_enabled(apic))
952 			val |= APIC_LVT_MASKED;
953 
954 		val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
955 		apic_set_reg(apic, reg, val);
956 
957 		break;
958 
959 	case APIC_LVTT:
960 		if ((apic_get_reg(apic, APIC_LVTT) &
961 		    apic->lapic_timer.timer_mode_mask) !=
962 		   (val & apic->lapic_timer.timer_mode_mask))
963 			hrtimer_cancel(&apic->lapic_timer.timer);
964 
965 		if (!apic_sw_enabled(apic))
966 			val |= APIC_LVT_MASKED;
967 		val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
968 		apic_set_reg(apic, APIC_LVTT, val);
969 		break;
970 
971 	case APIC_TMICT:
972 		if (apic_lvtt_tscdeadline(apic))
973 			break;
974 
975 		hrtimer_cancel(&apic->lapic_timer.timer);
976 		apic_set_reg(apic, APIC_TMICT, val);
977 		start_apic_timer(apic);
978 		break;
979 
980 	case APIC_TDCR:
981 		if (val & 4)
982 			apic_debug("KVM_WRITE:TDCR %x\n", val);
983 		apic_set_reg(apic, APIC_TDCR, val);
984 		update_divide_count(apic);
985 		break;
986 
987 	case APIC_ESR:
988 		if (apic_x2apic_mode(apic) && val != 0) {
989 			apic_debug("KVM_WRITE:ESR not zero %x\n", val);
990 			ret = 1;
991 		}
992 		break;
993 
994 	case APIC_SELF_IPI:
995 		if (apic_x2apic_mode(apic)) {
996 			apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
997 		} else
998 			ret = 1;
999 		break;
1000 	default:
1001 		ret = 1;
1002 		break;
1003 	}
1004 	if (ret)
1005 		apic_debug("Local APIC Write to read-only register %x\n", reg);
1006 	return ret;
1007 }
1008 
1009 static int apic_mmio_write(struct kvm_io_device *this,
1010 			    gpa_t address, int len, const void *data)
1011 {
1012 	struct kvm_lapic *apic = to_lapic(this);
1013 	unsigned int offset = address - apic->base_address;
1014 	u32 val;
1015 
1016 	if (!apic_mmio_in_range(apic, address))
1017 		return -EOPNOTSUPP;
1018 
1019 	/*
1020 	 * APIC register must be aligned on 128-bits boundary.
1021 	 * 32/64/128 bits registers must be accessed thru 32 bits.
1022 	 * Refer SDM 8.4.1
1023 	 */
1024 	if (len != 4 || (offset & 0xf)) {
1025 		/* Don't shout loud, $infamous_os would cause only noise. */
1026 		apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1027 		return 0;
1028 	}
1029 
1030 	val = *(u32*)data;
1031 
1032 	/* too common printing */
1033 	if (offset != APIC_EOI)
1034 		apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1035 			   "0x%x\n", __func__, offset, len, val);
1036 
1037 	apic_reg_write(apic, offset & 0xff0, val);
1038 
1039 	return 0;
1040 }
1041 
1042 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1043 {
1044 	struct kvm_lapic *apic = vcpu->arch.apic;
1045 
1046 	if (apic)
1047 		apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1048 }
1049 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1050 
1051 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1052 {
1053 	if (!vcpu->arch.apic)
1054 		return;
1055 
1056 	hrtimer_cancel(&vcpu->arch.apic->lapic_timer.timer);
1057 
1058 	if (vcpu->arch.apic->regs)
1059 		free_page((unsigned long)vcpu->arch.apic->regs);
1060 
1061 	kfree(vcpu->arch.apic);
1062 }
1063 
1064 /*
1065  *----------------------------------------------------------------------
1066  * LAPIC interface
1067  *----------------------------------------------------------------------
1068  */
1069 
1070 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1071 {
1072 	struct kvm_lapic *apic = vcpu->arch.apic;
1073 	if (!apic)
1074 		return 0;
1075 
1076 	if (apic_lvtt_oneshot(apic) || apic_lvtt_period(apic))
1077 		return 0;
1078 
1079 	return apic->lapic_timer.tscdeadline;
1080 }
1081 
1082 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1083 {
1084 	struct kvm_lapic *apic = vcpu->arch.apic;
1085 	if (!apic)
1086 		return;
1087 
1088 	if (apic_lvtt_oneshot(apic) || apic_lvtt_period(apic))
1089 		return;
1090 
1091 	hrtimer_cancel(&apic->lapic_timer.timer);
1092 	apic->lapic_timer.tscdeadline = data;
1093 	start_apic_timer(apic);
1094 }
1095 
1096 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1097 {
1098 	struct kvm_lapic *apic = vcpu->arch.apic;
1099 
1100 	if (!apic)
1101 		return;
1102 	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1103 		     | (apic_get_reg(apic, APIC_TASKPRI) & 4));
1104 }
1105 
1106 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1107 {
1108 	struct kvm_lapic *apic = vcpu->arch.apic;
1109 	u64 tpr;
1110 
1111 	if (!apic)
1112 		return 0;
1113 	tpr = (u64) apic_get_reg(apic, APIC_TASKPRI);
1114 
1115 	return (tpr & 0xf0) >> 4;
1116 }
1117 
1118 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1119 {
1120 	struct kvm_lapic *apic = vcpu->arch.apic;
1121 
1122 	if (!apic) {
1123 		value |= MSR_IA32_APICBASE_BSP;
1124 		vcpu->arch.apic_base = value;
1125 		return;
1126 	}
1127 
1128 	if (!kvm_vcpu_is_bsp(apic->vcpu))
1129 		value &= ~MSR_IA32_APICBASE_BSP;
1130 
1131 	vcpu->arch.apic_base = value;
1132 	if (apic_x2apic_mode(apic)) {
1133 		u32 id = kvm_apic_id(apic);
1134 		u32 ldr = ((id & ~0xf) << 16) | (1 << (id & 0xf));
1135 		apic_set_reg(apic, APIC_LDR, ldr);
1136 	}
1137 	apic->base_address = apic->vcpu->arch.apic_base &
1138 			     MSR_IA32_APICBASE_BASE;
1139 
1140 	/* with FSB delivery interrupt, we can restart APIC functionality */
1141 	apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1142 		   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1143 
1144 }
1145 
1146 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1147 {
1148 	struct kvm_lapic *apic;
1149 	int i;
1150 
1151 	apic_debug("%s\n", __func__);
1152 
1153 	ASSERT(vcpu);
1154 	apic = vcpu->arch.apic;
1155 	ASSERT(apic != NULL);
1156 
1157 	/* Stop the timer in case it's a reset to an active apic */
1158 	hrtimer_cancel(&apic->lapic_timer.timer);
1159 
1160 	apic_set_reg(apic, APIC_ID, vcpu->vcpu_id << 24);
1161 	kvm_apic_set_version(apic->vcpu);
1162 
1163 	for (i = 0; i < APIC_LVT_NUM; i++)
1164 		apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1165 	apic_set_reg(apic, APIC_LVT0,
1166 		     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1167 
1168 	apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1169 	apic_set_reg(apic, APIC_SPIV, 0xff);
1170 	apic_set_reg(apic, APIC_TASKPRI, 0);
1171 	apic_set_reg(apic, APIC_LDR, 0);
1172 	apic_set_reg(apic, APIC_ESR, 0);
1173 	apic_set_reg(apic, APIC_ICR, 0);
1174 	apic_set_reg(apic, APIC_ICR2, 0);
1175 	apic_set_reg(apic, APIC_TDCR, 0);
1176 	apic_set_reg(apic, APIC_TMICT, 0);
1177 	for (i = 0; i < 8; i++) {
1178 		apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1179 		apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1180 		apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1181 	}
1182 	apic->irr_pending = false;
1183 	apic->isr_count = 0;
1184 	apic->highest_isr_cache = -1;
1185 	update_divide_count(apic);
1186 	atomic_set(&apic->lapic_timer.pending, 0);
1187 	if (kvm_vcpu_is_bsp(vcpu))
1188 		vcpu->arch.apic_base |= MSR_IA32_APICBASE_BSP;
1189 	vcpu->arch.pv_eoi.msr_val = 0;
1190 	apic_update_ppr(apic);
1191 
1192 	vcpu->arch.apic_arb_prio = 0;
1193 	vcpu->arch.apic_attention = 0;
1194 
1195 	apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
1196 		   "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1197 		   vcpu, kvm_apic_id(apic),
1198 		   vcpu->arch.apic_base, apic->base_address);
1199 }
1200 
1201 bool kvm_apic_present(struct kvm_vcpu *vcpu)
1202 {
1203 	return vcpu->arch.apic && apic_hw_enabled(vcpu->arch.apic);
1204 }
1205 
1206 int kvm_lapic_enabled(struct kvm_vcpu *vcpu)
1207 {
1208 	return kvm_apic_present(vcpu) && apic_sw_enabled(vcpu->arch.apic);
1209 }
1210 
1211 /*
1212  *----------------------------------------------------------------------
1213  * timer interface
1214  *----------------------------------------------------------------------
1215  */
1216 
1217 static bool lapic_is_periodic(struct kvm_timer *ktimer)
1218 {
1219 	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic,
1220 					      lapic_timer);
1221 	return apic_lvtt_period(apic);
1222 }
1223 
1224 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1225 {
1226 	struct kvm_lapic *lapic = vcpu->arch.apic;
1227 
1228 	if (lapic && apic_enabled(lapic) && apic_lvt_enabled(lapic, APIC_LVTT))
1229 		return atomic_read(&lapic->lapic_timer.pending);
1230 
1231 	return 0;
1232 }
1233 
1234 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1235 {
1236 	u32 reg = apic_get_reg(apic, lvt_type);
1237 	int vector, mode, trig_mode;
1238 
1239 	if (apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1240 		vector = reg & APIC_VECTOR_MASK;
1241 		mode = reg & APIC_MODE_MASK;
1242 		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1243 		return __apic_accept_irq(apic, mode, vector, 1, trig_mode);
1244 	}
1245 	return 0;
1246 }
1247 
1248 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1249 {
1250 	struct kvm_lapic *apic = vcpu->arch.apic;
1251 
1252 	if (apic)
1253 		kvm_apic_local_deliver(apic, APIC_LVT0);
1254 }
1255 
1256 static struct kvm_timer_ops lapic_timer_ops = {
1257 	.is_periodic = lapic_is_periodic,
1258 };
1259 
1260 static const struct kvm_io_device_ops apic_mmio_ops = {
1261 	.read     = apic_mmio_read,
1262 	.write    = apic_mmio_write,
1263 };
1264 
1265 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1266 {
1267 	struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1268 	struct kvm_vcpu *vcpu = ktimer->vcpu;
1269 	wait_queue_head_t *q = &vcpu->wq;
1270 
1271 	/*
1272 	 * There is a race window between reading and incrementing, but we do
1273 	 * not care about potentially losing timer events in the !reinject
1274 	 * case anyway. Note: KVM_REQ_PENDING_TIMER is implicitly checked
1275 	 * in vcpu_enter_guest.
1276 	 */
1277 	if (ktimer->reinject || !atomic_read(&ktimer->pending)) {
1278 		atomic_inc(&ktimer->pending);
1279 		/* FIXME: this code should not know anything about vcpus */
1280 		kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1281 	}
1282 
1283 	if (waitqueue_active(q))
1284 		wake_up_interruptible(q);
1285 
1286 	if (ktimer->t_ops->is_periodic(ktimer)) {
1287 		hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1288 		return HRTIMER_RESTART;
1289 	} else
1290 		return HRTIMER_NORESTART;
1291 }
1292 
1293 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1294 {
1295 	struct kvm_lapic *apic;
1296 
1297 	ASSERT(vcpu != NULL);
1298 	apic_debug("apic_init %d\n", vcpu->vcpu_id);
1299 
1300 	apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1301 	if (!apic)
1302 		goto nomem;
1303 
1304 	vcpu->arch.apic = apic;
1305 
1306 	apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1307 	if (!apic->regs) {
1308 		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1309 		       vcpu->vcpu_id);
1310 		goto nomem_free_apic;
1311 	}
1312 	apic->vcpu = vcpu;
1313 
1314 	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1315 		     HRTIMER_MODE_ABS);
1316 	apic->lapic_timer.timer.function = apic_timer_fn;
1317 	apic->lapic_timer.t_ops = &lapic_timer_ops;
1318 	apic->lapic_timer.kvm = vcpu->kvm;
1319 	apic->lapic_timer.vcpu = vcpu;
1320 
1321 	apic->base_address = APIC_DEFAULT_PHYS_BASE;
1322 	vcpu->arch.apic_base = APIC_DEFAULT_PHYS_BASE;
1323 
1324 	kvm_lapic_reset(vcpu);
1325 	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1326 
1327 	return 0;
1328 nomem_free_apic:
1329 	kfree(apic);
1330 nomem:
1331 	return -ENOMEM;
1332 }
1333 
1334 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1335 {
1336 	struct kvm_lapic *apic = vcpu->arch.apic;
1337 	int highest_irr;
1338 
1339 	if (!apic || !apic_enabled(apic))
1340 		return -1;
1341 
1342 	apic_update_ppr(apic);
1343 	highest_irr = apic_find_highest_irr(apic);
1344 	if ((highest_irr == -1) ||
1345 	    ((highest_irr & 0xF0) <= apic_get_reg(apic, APIC_PROCPRI)))
1346 		return -1;
1347 	return highest_irr;
1348 }
1349 
1350 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1351 {
1352 	u32 lvt0 = apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1353 	int r = 0;
1354 
1355 	if (!apic_hw_enabled(vcpu->arch.apic))
1356 		r = 1;
1357 	if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1358 	    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1359 		r = 1;
1360 	return r;
1361 }
1362 
1363 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1364 {
1365 	struct kvm_lapic *apic = vcpu->arch.apic;
1366 
1367 	if (apic && atomic_read(&apic->lapic_timer.pending) > 0) {
1368 		if (kvm_apic_local_deliver(apic, APIC_LVTT))
1369 			atomic_dec(&apic->lapic_timer.pending);
1370 	}
1371 }
1372 
1373 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1374 {
1375 	int vector = kvm_apic_has_interrupt(vcpu);
1376 	struct kvm_lapic *apic = vcpu->arch.apic;
1377 
1378 	if (vector == -1)
1379 		return -1;
1380 
1381 	apic_set_isr(vector, apic);
1382 	apic_update_ppr(apic);
1383 	apic_clear_irr(vector, apic);
1384 	return vector;
1385 }
1386 
1387 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu)
1388 {
1389 	struct kvm_lapic *apic = vcpu->arch.apic;
1390 
1391 	apic->base_address = vcpu->arch.apic_base &
1392 			     MSR_IA32_APICBASE_BASE;
1393 	kvm_apic_set_version(vcpu);
1394 
1395 	apic_update_ppr(apic);
1396 	hrtimer_cancel(&apic->lapic_timer.timer);
1397 	update_divide_count(apic);
1398 	start_apic_timer(apic);
1399 	apic->irr_pending = true;
1400 	apic->isr_count = count_vectors(apic->regs + APIC_ISR);
1401 	apic->highest_isr_cache = -1;
1402 	kvm_make_request(KVM_REQ_EVENT, vcpu);
1403 }
1404 
1405 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1406 {
1407 	struct kvm_lapic *apic = vcpu->arch.apic;
1408 	struct hrtimer *timer;
1409 
1410 	if (!apic)
1411 		return;
1412 
1413 	timer = &apic->lapic_timer.timer;
1414 	if (hrtimer_cancel(timer))
1415 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1416 }
1417 
1418 /*
1419  * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1420  *
1421  * Detect whether guest triggered PV EOI since the
1422  * last entry. If yes, set EOI on guests's behalf.
1423  * Clear PV EOI in guest memory in any case.
1424  */
1425 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1426 					struct kvm_lapic *apic)
1427 {
1428 	bool pending;
1429 	int vector;
1430 	/*
1431 	 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1432 	 * and KVM_PV_EOI_ENABLED in guest memory as follows:
1433 	 *
1434 	 * KVM_APIC_PV_EOI_PENDING is unset:
1435 	 * 	-> host disabled PV EOI.
1436 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1437 	 * 	-> host enabled PV EOI, guest did not execute EOI yet.
1438 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1439 	 * 	-> host enabled PV EOI, guest executed EOI.
1440 	 */
1441 	BUG_ON(!pv_eoi_enabled(vcpu));
1442 	pending = pv_eoi_get_pending(vcpu);
1443 	/*
1444 	 * Clear pending bit in any case: it will be set again on vmentry.
1445 	 * While this might not be ideal from performance point of view,
1446 	 * this makes sure pv eoi is only enabled when we know it's safe.
1447 	 */
1448 	pv_eoi_clr_pending(vcpu);
1449 	if (pending)
1450 		return;
1451 	vector = apic_set_eoi(apic);
1452 	trace_kvm_pv_eoi(apic, vector);
1453 }
1454 
1455 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1456 {
1457 	u32 data;
1458 	void *vapic;
1459 
1460 	if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1461 		apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1462 
1463 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1464 		return;
1465 
1466 	vapic = kmap_atomic(vcpu->arch.apic->vapic_page);
1467 	data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr));
1468 	kunmap_atomic(vapic);
1469 
1470 	apic_set_tpr(vcpu->arch.apic, data & 0xff);
1471 }
1472 
1473 /*
1474  * apic_sync_pv_eoi_to_guest - called before vmentry
1475  *
1476  * Detect whether it's safe to enable PV EOI and
1477  * if yes do so.
1478  */
1479 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
1480 					struct kvm_lapic *apic)
1481 {
1482 	if (!pv_eoi_enabled(vcpu) ||
1483 	    /* IRR set or many bits in ISR: could be nested. */
1484 	    apic->irr_pending ||
1485 	    /* Cache not set: could be safe but we don't bother. */
1486 	    apic->highest_isr_cache == -1 ||
1487 	    /* Need EOI to update ioapic. */
1488 	    kvm_ioapic_handles_vector(vcpu->kvm, apic->highest_isr_cache)) {
1489 		/*
1490 		 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1491 		 * so we need not do anything here.
1492 		 */
1493 		return;
1494 	}
1495 
1496 	pv_eoi_set_pending(apic->vcpu);
1497 }
1498 
1499 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1500 {
1501 	u32 data, tpr;
1502 	int max_irr, max_isr;
1503 	struct kvm_lapic *apic = vcpu->arch.apic;
1504 	void *vapic;
1505 
1506 	apic_sync_pv_eoi_to_guest(vcpu, apic);
1507 
1508 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1509 		return;
1510 
1511 	tpr = apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1512 	max_irr = apic_find_highest_irr(apic);
1513 	if (max_irr < 0)
1514 		max_irr = 0;
1515 	max_isr = apic_find_highest_isr(apic);
1516 	if (max_isr < 0)
1517 		max_isr = 0;
1518 	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1519 
1520 	vapic = kmap_atomic(vcpu->arch.apic->vapic_page);
1521 	*(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data;
1522 	kunmap_atomic(vapic);
1523 }
1524 
1525 void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1526 {
1527 	vcpu->arch.apic->vapic_addr = vapic_addr;
1528 	if (vapic_addr)
1529 		__set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1530 	else
1531 		__clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1532 }
1533 
1534 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1535 {
1536 	struct kvm_lapic *apic = vcpu->arch.apic;
1537 	u32 reg = (msr - APIC_BASE_MSR) << 4;
1538 
1539 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1540 		return 1;
1541 
1542 	/* if this is ICR write vector before command */
1543 	if (msr == 0x830)
1544 		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1545 	return apic_reg_write(apic, reg, (u32)data);
1546 }
1547 
1548 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1549 {
1550 	struct kvm_lapic *apic = vcpu->arch.apic;
1551 	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1552 
1553 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1554 		return 1;
1555 
1556 	if (apic_reg_read(apic, reg, 4, &low))
1557 		return 1;
1558 	if (msr == 0x830)
1559 		apic_reg_read(apic, APIC_ICR2, 4, &high);
1560 
1561 	*data = (((u64)high) << 32) | low;
1562 
1563 	return 0;
1564 }
1565 
1566 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1567 {
1568 	struct kvm_lapic *apic = vcpu->arch.apic;
1569 
1570 	if (!irqchip_in_kernel(vcpu->kvm))
1571 		return 1;
1572 
1573 	/* if this is ICR write vector before command */
1574 	if (reg == APIC_ICR)
1575 		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1576 	return apic_reg_write(apic, reg, (u32)data);
1577 }
1578 
1579 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
1580 {
1581 	struct kvm_lapic *apic = vcpu->arch.apic;
1582 	u32 low, high = 0;
1583 
1584 	if (!irqchip_in_kernel(vcpu->kvm))
1585 		return 1;
1586 
1587 	if (apic_reg_read(apic, reg, 4, &low))
1588 		return 1;
1589 	if (reg == APIC_ICR)
1590 		apic_reg_read(apic, APIC_ICR2, 4, &high);
1591 
1592 	*data = (((u64)high) << 32) | low;
1593 
1594 	return 0;
1595 }
1596 
1597 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
1598 {
1599 	u64 addr = data & ~KVM_MSR_ENABLED;
1600 	if (!IS_ALIGNED(addr, 4))
1601 		return 1;
1602 
1603 	vcpu->arch.pv_eoi.msr_val = data;
1604 	if (!pv_eoi_enabled(vcpu))
1605 		return 0;
1606 	return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
1607 					 addr);
1608 }
1609