xref: /openbmc/linux/arch/x86/kvm/lapic.c (revision 9ac8d3fb)
1 
2 /*
3  * Local APIC virtualization
4  *
5  * Copyright (C) 2006 Qumranet, Inc.
6  * Copyright (C) 2007 Novell
7  * Copyright (C) 2007 Intel
8  *
9  * Authors:
10  *   Dor Laor <dor.laor@qumranet.com>
11  *   Gregory Haskins <ghaskins@novell.com>
12  *   Yaozu (Eddie) Dong <eddie.dong@intel.com>
13  *
14  * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
15  *
16  * This work is licensed under the terms of the GNU GPL, version 2.  See
17  * the COPYING file in the top-level directory.
18  */
19 
20 #include <linux/kvm_host.h>
21 #include <linux/kvm.h>
22 #include <linux/mm.h>
23 #include <linux/highmem.h>
24 #include <linux/smp.h>
25 #include <linux/hrtimer.h>
26 #include <linux/io.h>
27 #include <linux/module.h>
28 #include <linux/math64.h>
29 #include <asm/processor.h>
30 #include <asm/msr.h>
31 #include <asm/page.h>
32 #include <asm/current.h>
33 #include <asm/apicdef.h>
34 #include <asm/atomic.h>
35 #include "kvm_cache_regs.h"
36 #include "irq.h"
37 
38 #define PRId64 "d"
39 #define PRIx64 "llx"
40 #define PRIu64 "u"
41 #define PRIo64 "o"
42 
43 #define APIC_BUS_CYCLE_NS 1
44 
45 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
46 #define apic_debug(fmt, arg...)
47 
48 #define APIC_LVT_NUM			6
49 /* 14 is the version for Xeon and Pentium 8.4.8*/
50 #define APIC_VERSION			(0x14UL | ((APIC_LVT_NUM - 1) << 16))
51 #define LAPIC_MMIO_LENGTH		(1 << 12)
52 /* followed define is not in apicdef.h */
53 #define APIC_SHORT_MASK			0xc0000
54 #define APIC_DEST_NOSHORT		0x0
55 #define APIC_DEST_MASK			0x800
56 #define MAX_APIC_VECTOR			256
57 
58 #define VEC_POS(v) ((v) & (32 - 1))
59 #define REG_POS(v) (((v) >> 5) << 4)
60 
61 static inline u32 apic_get_reg(struct kvm_lapic *apic, int reg_off)
62 {
63 	return *((u32 *) (apic->regs + reg_off));
64 }
65 
66 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
67 {
68 	*((u32 *) (apic->regs + reg_off)) = val;
69 }
70 
71 static inline int apic_test_and_set_vector(int vec, void *bitmap)
72 {
73 	return test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
74 }
75 
76 static inline int apic_test_and_clear_vector(int vec, void *bitmap)
77 {
78 	return test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
79 }
80 
81 static inline void apic_set_vector(int vec, void *bitmap)
82 {
83 	set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
84 }
85 
86 static inline void apic_clear_vector(int vec, void *bitmap)
87 {
88 	clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
89 }
90 
91 static inline int apic_hw_enabled(struct kvm_lapic *apic)
92 {
93 	return (apic)->vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE;
94 }
95 
96 static inline int  apic_sw_enabled(struct kvm_lapic *apic)
97 {
98 	return apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED;
99 }
100 
101 static inline int apic_enabled(struct kvm_lapic *apic)
102 {
103 	return apic_sw_enabled(apic) &&	apic_hw_enabled(apic);
104 }
105 
106 #define LVT_MASK	\
107 	(APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
108 
109 #define LINT_MASK	\
110 	(LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
111 	 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
112 
113 static inline int kvm_apic_id(struct kvm_lapic *apic)
114 {
115 	return (apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
116 }
117 
118 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
119 {
120 	return !(apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
121 }
122 
123 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
124 {
125 	return apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
126 }
127 
128 static inline int apic_lvtt_period(struct kvm_lapic *apic)
129 {
130 	return apic_get_reg(apic, APIC_LVTT) & APIC_LVT_TIMER_PERIODIC;
131 }
132 
133 static unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
134 	LVT_MASK | APIC_LVT_TIMER_PERIODIC,	/* LVTT */
135 	LVT_MASK | APIC_MODE_MASK,	/* LVTTHMR */
136 	LVT_MASK | APIC_MODE_MASK,	/* LVTPC */
137 	LINT_MASK, LINT_MASK,	/* LVT0-1 */
138 	LVT_MASK		/* LVTERR */
139 };
140 
141 static int find_highest_vector(void *bitmap)
142 {
143 	u32 *word = bitmap;
144 	int word_offset = MAX_APIC_VECTOR >> 5;
145 
146 	while ((word_offset != 0) && (word[(--word_offset) << 2] == 0))
147 		continue;
148 
149 	if (likely(!word_offset && !word[0]))
150 		return -1;
151 	else
152 		return fls(word[word_offset << 2]) - 1 + (word_offset << 5);
153 }
154 
155 static inline int apic_test_and_set_irr(int vec, struct kvm_lapic *apic)
156 {
157 	return apic_test_and_set_vector(vec, apic->regs + APIC_IRR);
158 }
159 
160 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
161 {
162 	apic_clear_vector(vec, apic->regs + APIC_IRR);
163 }
164 
165 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
166 {
167 	int result;
168 
169 	result = find_highest_vector(apic->regs + APIC_IRR);
170 	ASSERT(result == -1 || result >= 16);
171 
172 	return result;
173 }
174 
175 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
176 {
177 	struct kvm_lapic *apic = vcpu->arch.apic;
178 	int highest_irr;
179 
180 	if (!apic)
181 		return 0;
182 	highest_irr = apic_find_highest_irr(apic);
183 
184 	return highest_irr;
185 }
186 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
187 
188 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, u8 vec, u8 trig)
189 {
190 	struct kvm_lapic *apic = vcpu->arch.apic;
191 
192 	if (!apic_test_and_set_irr(vec, apic)) {
193 		/* a new pending irq is set in IRR */
194 		if (trig)
195 			apic_set_vector(vec, apic->regs + APIC_TMR);
196 		else
197 			apic_clear_vector(vec, apic->regs + APIC_TMR);
198 		kvm_vcpu_kick(apic->vcpu);
199 		return 1;
200 	}
201 	return 0;
202 }
203 
204 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
205 {
206 	int result;
207 
208 	result = find_highest_vector(apic->regs + APIC_ISR);
209 	ASSERT(result == -1 || result >= 16);
210 
211 	return result;
212 }
213 
214 static void apic_update_ppr(struct kvm_lapic *apic)
215 {
216 	u32 tpr, isrv, ppr;
217 	int isr;
218 
219 	tpr = apic_get_reg(apic, APIC_TASKPRI);
220 	isr = apic_find_highest_isr(apic);
221 	isrv = (isr != -1) ? isr : 0;
222 
223 	if ((tpr & 0xf0) >= (isrv & 0xf0))
224 		ppr = tpr & 0xff;
225 	else
226 		ppr = isrv & 0xf0;
227 
228 	apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
229 		   apic, ppr, isr, isrv);
230 
231 	apic_set_reg(apic, APIC_PROCPRI, ppr);
232 }
233 
234 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
235 {
236 	apic_set_reg(apic, APIC_TASKPRI, tpr);
237 	apic_update_ppr(apic);
238 }
239 
240 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
241 {
242 	return kvm_apic_id(apic) == dest;
243 }
244 
245 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
246 {
247 	int result = 0;
248 	u8 logical_id;
249 
250 	logical_id = GET_APIC_LOGICAL_ID(apic_get_reg(apic, APIC_LDR));
251 
252 	switch (apic_get_reg(apic, APIC_DFR)) {
253 	case APIC_DFR_FLAT:
254 		if (logical_id & mda)
255 			result = 1;
256 		break;
257 	case APIC_DFR_CLUSTER:
258 		if (((logical_id >> 4) == (mda >> 0x4))
259 		    && (logical_id & mda & 0xf))
260 			result = 1;
261 		break;
262 	default:
263 		printk(KERN_WARNING "Bad DFR vcpu %d: %08x\n",
264 		       apic->vcpu->vcpu_id, apic_get_reg(apic, APIC_DFR));
265 		break;
266 	}
267 
268 	return result;
269 }
270 
271 static int apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
272 			   int short_hand, int dest, int dest_mode)
273 {
274 	int result = 0;
275 	struct kvm_lapic *target = vcpu->arch.apic;
276 
277 	apic_debug("target %p, source %p, dest 0x%x, "
278 		   "dest_mode 0x%x, short_hand 0x%x",
279 		   target, source, dest, dest_mode, short_hand);
280 
281 	ASSERT(!target);
282 	switch (short_hand) {
283 	case APIC_DEST_NOSHORT:
284 		if (dest_mode == 0) {
285 			/* Physical mode. */
286 			if ((dest == 0xFF) || (dest == kvm_apic_id(target)))
287 				result = 1;
288 		} else
289 			/* Logical mode. */
290 			result = kvm_apic_match_logical_addr(target, dest);
291 		break;
292 	case APIC_DEST_SELF:
293 		if (target == source)
294 			result = 1;
295 		break;
296 	case APIC_DEST_ALLINC:
297 		result = 1;
298 		break;
299 	case APIC_DEST_ALLBUT:
300 		if (target != source)
301 			result = 1;
302 		break;
303 	default:
304 		printk(KERN_WARNING "Bad dest shorthand value %x\n",
305 		       short_hand);
306 		break;
307 	}
308 
309 	return result;
310 }
311 
312 /*
313  * Add a pending IRQ into lapic.
314  * Return 1 if successfully added and 0 if discarded.
315  */
316 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
317 			     int vector, int level, int trig_mode)
318 {
319 	int orig_irr, result = 0;
320 	struct kvm_vcpu *vcpu = apic->vcpu;
321 
322 	switch (delivery_mode) {
323 	case APIC_DM_FIXED:
324 	case APIC_DM_LOWEST:
325 		/* FIXME add logic for vcpu on reset */
326 		if (unlikely(!apic_enabled(apic)))
327 			break;
328 
329 		orig_irr = apic_test_and_set_irr(vector, apic);
330 		if (orig_irr && trig_mode) {
331 			apic_debug("level trig mode repeatedly for vector %d",
332 				   vector);
333 			break;
334 		}
335 
336 		if (trig_mode) {
337 			apic_debug("level trig mode for vector %d", vector);
338 			apic_set_vector(vector, apic->regs + APIC_TMR);
339 		} else
340 			apic_clear_vector(vector, apic->regs + APIC_TMR);
341 
342 		kvm_vcpu_kick(vcpu);
343 
344 		result = (orig_irr == 0);
345 		break;
346 
347 	case APIC_DM_REMRD:
348 		printk(KERN_DEBUG "Ignoring delivery mode 3\n");
349 		break;
350 
351 	case APIC_DM_SMI:
352 		printk(KERN_DEBUG "Ignoring guest SMI\n");
353 		break;
354 
355 	case APIC_DM_NMI:
356 		kvm_inject_nmi(vcpu);
357 		break;
358 
359 	case APIC_DM_INIT:
360 		if (level) {
361 			if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
362 				printk(KERN_DEBUG
363 				       "INIT on a runnable vcpu %d\n",
364 				       vcpu->vcpu_id);
365 			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
366 			kvm_vcpu_kick(vcpu);
367 		} else {
368 			apic_debug("Ignoring de-assert INIT to vcpu %d\n",
369 				   vcpu->vcpu_id);
370 		}
371 		break;
372 
373 	case APIC_DM_STARTUP:
374 		apic_debug("SIPI to vcpu %d vector 0x%02x\n",
375 			   vcpu->vcpu_id, vector);
376 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
377 			vcpu->arch.sipi_vector = vector;
378 			vcpu->arch.mp_state = KVM_MP_STATE_SIPI_RECEIVED;
379 			kvm_vcpu_kick(vcpu);
380 		}
381 		break;
382 
383 	default:
384 		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
385 		       delivery_mode);
386 		break;
387 	}
388 	return result;
389 }
390 
391 static struct kvm_lapic *kvm_apic_round_robin(struct kvm *kvm, u8 vector,
392 				       unsigned long bitmap)
393 {
394 	int last;
395 	int next;
396 	struct kvm_lapic *apic = NULL;
397 
398 	last = kvm->arch.round_robin_prev_vcpu;
399 	next = last;
400 
401 	do {
402 		if (++next == KVM_MAX_VCPUS)
403 			next = 0;
404 		if (kvm->vcpus[next] == NULL || !test_bit(next, &bitmap))
405 			continue;
406 		apic = kvm->vcpus[next]->arch.apic;
407 		if (apic && apic_enabled(apic))
408 			break;
409 		apic = NULL;
410 	} while (next != last);
411 	kvm->arch.round_robin_prev_vcpu = next;
412 
413 	if (!apic)
414 		printk(KERN_DEBUG "vcpu not ready for apic_round_robin\n");
415 
416 	return apic;
417 }
418 
419 struct kvm_vcpu *kvm_get_lowest_prio_vcpu(struct kvm *kvm, u8 vector,
420 		unsigned long bitmap)
421 {
422 	struct kvm_lapic *apic;
423 
424 	apic = kvm_apic_round_robin(kvm, vector, bitmap);
425 	if (apic)
426 		return apic->vcpu;
427 	return NULL;
428 }
429 
430 static void apic_set_eoi(struct kvm_lapic *apic)
431 {
432 	int vector = apic_find_highest_isr(apic);
433 	int trigger_mode;
434 	/*
435 	 * Not every write EOI will has corresponding ISR,
436 	 * one example is when Kernel check timer on setup_IO_APIC
437 	 */
438 	if (vector == -1)
439 		return;
440 
441 	apic_clear_vector(vector, apic->regs + APIC_ISR);
442 	apic_update_ppr(apic);
443 
444 	if (apic_test_and_clear_vector(vector, apic->regs + APIC_TMR))
445 		trigger_mode = IOAPIC_LEVEL_TRIG;
446 	else
447 		trigger_mode = IOAPIC_EDGE_TRIG;
448 	kvm_ioapic_update_eoi(apic->vcpu->kvm, vector, trigger_mode);
449 }
450 
451 static void apic_send_ipi(struct kvm_lapic *apic)
452 {
453 	u32 icr_low = apic_get_reg(apic, APIC_ICR);
454 	u32 icr_high = apic_get_reg(apic, APIC_ICR2);
455 
456 	unsigned int dest = GET_APIC_DEST_FIELD(icr_high);
457 	unsigned int short_hand = icr_low & APIC_SHORT_MASK;
458 	unsigned int trig_mode = icr_low & APIC_INT_LEVELTRIG;
459 	unsigned int level = icr_low & APIC_INT_ASSERT;
460 	unsigned int dest_mode = icr_low & APIC_DEST_MASK;
461 	unsigned int delivery_mode = icr_low & APIC_MODE_MASK;
462 	unsigned int vector = icr_low & APIC_VECTOR_MASK;
463 
464 	struct kvm_vcpu *target;
465 	struct kvm_vcpu *vcpu;
466 	unsigned long lpr_map = 0;
467 	int i;
468 
469 	apic_debug("icr_high 0x%x, icr_low 0x%x, "
470 		   "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
471 		   "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
472 		   icr_high, icr_low, short_hand, dest,
473 		   trig_mode, level, dest_mode, delivery_mode, vector);
474 
475 	for (i = 0; i < KVM_MAX_VCPUS; i++) {
476 		vcpu = apic->vcpu->kvm->vcpus[i];
477 		if (!vcpu)
478 			continue;
479 
480 		if (vcpu->arch.apic &&
481 		    apic_match_dest(vcpu, apic, short_hand, dest, dest_mode)) {
482 			if (delivery_mode == APIC_DM_LOWEST)
483 				set_bit(vcpu->vcpu_id, &lpr_map);
484 			else
485 				__apic_accept_irq(vcpu->arch.apic, delivery_mode,
486 						  vector, level, trig_mode);
487 		}
488 	}
489 
490 	if (delivery_mode == APIC_DM_LOWEST) {
491 		target = kvm_get_lowest_prio_vcpu(vcpu->kvm, vector, lpr_map);
492 		if (target != NULL)
493 			__apic_accept_irq(target->arch.apic, delivery_mode,
494 					  vector, level, trig_mode);
495 	}
496 }
497 
498 static u32 apic_get_tmcct(struct kvm_lapic *apic)
499 {
500 	u64 counter_passed;
501 	ktime_t passed, now;
502 	u32 tmcct;
503 
504 	ASSERT(apic != NULL);
505 
506 	now = apic->timer.dev.base->get_time();
507 	tmcct = apic_get_reg(apic, APIC_TMICT);
508 
509 	/* if initial count is 0, current count should also be 0 */
510 	if (tmcct == 0)
511 		return 0;
512 
513 	if (unlikely(ktime_to_ns(now) <=
514 		ktime_to_ns(apic->timer.last_update))) {
515 		/* Wrap around */
516 		passed = ktime_add(( {
517 				    (ktime_t) {
518 				    .tv64 = KTIME_MAX -
519 				    (apic->timer.last_update).tv64}; }
520 				   ), now);
521 		apic_debug("time elapsed\n");
522 	} else
523 		passed = ktime_sub(now, apic->timer.last_update);
524 
525 	counter_passed = div64_u64(ktime_to_ns(passed),
526 				   (APIC_BUS_CYCLE_NS * apic->timer.divide_count));
527 
528 	if (counter_passed > tmcct) {
529 		if (unlikely(!apic_lvtt_period(apic))) {
530 			/* one-shot timers stick at 0 until reset */
531 			tmcct = 0;
532 		} else {
533 			/*
534 			 * periodic timers reset to APIC_TMICT when they
535 			 * hit 0. The while loop simulates this happening N
536 			 * times. (counter_passed %= tmcct) would also work,
537 			 * but might be slower or not work on 32-bit??
538 			 */
539 			while (counter_passed > tmcct)
540 				counter_passed -= tmcct;
541 			tmcct -= counter_passed;
542 		}
543 	} else {
544 		tmcct -= counter_passed;
545 	}
546 
547 	return tmcct;
548 }
549 
550 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
551 {
552 	struct kvm_vcpu *vcpu = apic->vcpu;
553 	struct kvm_run *run = vcpu->run;
554 
555 	set_bit(KVM_REQ_REPORT_TPR_ACCESS, &vcpu->requests);
556 	run->tpr_access.rip = kvm_rip_read(vcpu);
557 	run->tpr_access.is_write = write;
558 }
559 
560 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
561 {
562 	if (apic->vcpu->arch.tpr_access_reporting)
563 		__report_tpr_access(apic, write);
564 }
565 
566 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
567 {
568 	u32 val = 0;
569 
570 	KVMTRACE_1D(APIC_ACCESS, apic->vcpu, (u32)offset, handler);
571 
572 	if (offset >= LAPIC_MMIO_LENGTH)
573 		return 0;
574 
575 	switch (offset) {
576 	case APIC_ARBPRI:
577 		printk(KERN_WARNING "Access APIC ARBPRI register "
578 		       "which is for P6\n");
579 		break;
580 
581 	case APIC_TMCCT:	/* Timer CCR */
582 		val = apic_get_tmcct(apic);
583 		break;
584 
585 	case APIC_TASKPRI:
586 		report_tpr_access(apic, false);
587 		/* fall thru */
588 	default:
589 		apic_update_ppr(apic);
590 		val = apic_get_reg(apic, offset);
591 		break;
592 	}
593 
594 	return val;
595 }
596 
597 static void apic_mmio_read(struct kvm_io_device *this,
598 			   gpa_t address, int len, void *data)
599 {
600 	struct kvm_lapic *apic = (struct kvm_lapic *)this->private;
601 	unsigned int offset = address - apic->base_address;
602 	unsigned char alignment = offset & 0xf;
603 	u32 result;
604 
605 	if ((alignment + len) > 4) {
606 		printk(KERN_ERR "KVM_APIC_READ: alignment error %lx %d",
607 		       (unsigned long)address, len);
608 		return;
609 	}
610 	result = __apic_read(apic, offset & ~0xf);
611 
612 	switch (len) {
613 	case 1:
614 	case 2:
615 	case 4:
616 		memcpy(data, (char *)&result + alignment, len);
617 		break;
618 	default:
619 		printk(KERN_ERR "Local APIC read with len = %x, "
620 		       "should be 1,2, or 4 instead\n", len);
621 		break;
622 	}
623 }
624 
625 static void update_divide_count(struct kvm_lapic *apic)
626 {
627 	u32 tmp1, tmp2, tdcr;
628 
629 	tdcr = apic_get_reg(apic, APIC_TDCR);
630 	tmp1 = tdcr & 0xf;
631 	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
632 	apic->timer.divide_count = 0x1 << (tmp2 & 0x7);
633 
634 	apic_debug("timer divide count is 0x%x\n",
635 				   apic->timer.divide_count);
636 }
637 
638 static void start_apic_timer(struct kvm_lapic *apic)
639 {
640 	ktime_t now = apic->timer.dev.base->get_time();
641 
642 	apic->timer.last_update = now;
643 
644 	apic->timer.period = apic_get_reg(apic, APIC_TMICT) *
645 		    APIC_BUS_CYCLE_NS * apic->timer.divide_count;
646 	atomic_set(&apic->timer.pending, 0);
647 
648 	if (!apic->timer.period)
649 		return;
650 
651 	hrtimer_start(&apic->timer.dev,
652 		      ktime_add_ns(now, apic->timer.period),
653 		      HRTIMER_MODE_ABS);
654 
655 	apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
656 			   PRIx64 ", "
657 			   "timer initial count 0x%x, period %lldns, "
658 			   "expire @ 0x%016" PRIx64 ".\n", __func__,
659 			   APIC_BUS_CYCLE_NS, ktime_to_ns(now),
660 			   apic_get_reg(apic, APIC_TMICT),
661 			   apic->timer.period,
662 			   ktime_to_ns(ktime_add_ns(now,
663 					apic->timer.period)));
664 }
665 
666 static void apic_mmio_write(struct kvm_io_device *this,
667 			    gpa_t address, int len, const void *data)
668 {
669 	struct kvm_lapic *apic = (struct kvm_lapic *)this->private;
670 	unsigned int offset = address - apic->base_address;
671 	unsigned char alignment = offset & 0xf;
672 	u32 val;
673 
674 	/*
675 	 * APIC register must be aligned on 128-bits boundary.
676 	 * 32/64/128 bits registers must be accessed thru 32 bits.
677 	 * Refer SDM 8.4.1
678 	 */
679 	if (len != 4 || alignment) {
680 		/* Don't shout loud, $infamous_os would cause only noise. */
681 		apic_debug("apic write: bad size=%d %lx\n",
682 			   len, (long)address);
683 		return;
684 	}
685 
686 	val = *(u32 *) data;
687 
688 	/* too common printing */
689 	if (offset != APIC_EOI)
690 		apic_debug("%s: offset 0x%x with length 0x%x, and value is "
691 			   "0x%x\n", __func__, offset, len, val);
692 
693 	offset &= 0xff0;
694 
695 	KVMTRACE_1D(APIC_ACCESS, apic->vcpu, (u32)offset, handler);
696 
697 	switch (offset) {
698 	case APIC_ID:		/* Local APIC ID */
699 		apic_set_reg(apic, APIC_ID, val);
700 		break;
701 
702 	case APIC_TASKPRI:
703 		report_tpr_access(apic, true);
704 		apic_set_tpr(apic, val & 0xff);
705 		break;
706 
707 	case APIC_EOI:
708 		apic_set_eoi(apic);
709 		break;
710 
711 	case APIC_LDR:
712 		apic_set_reg(apic, APIC_LDR, val & APIC_LDR_MASK);
713 		break;
714 
715 	case APIC_DFR:
716 		apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
717 		break;
718 
719 	case APIC_SPIV:
720 		apic_set_reg(apic, APIC_SPIV, val & 0x3ff);
721 		if (!(val & APIC_SPIV_APIC_ENABLED)) {
722 			int i;
723 			u32 lvt_val;
724 
725 			for (i = 0; i < APIC_LVT_NUM; i++) {
726 				lvt_val = apic_get_reg(apic,
727 						       APIC_LVTT + 0x10 * i);
728 				apic_set_reg(apic, APIC_LVTT + 0x10 * i,
729 					     lvt_val | APIC_LVT_MASKED);
730 			}
731 			atomic_set(&apic->timer.pending, 0);
732 
733 		}
734 		break;
735 
736 	case APIC_ICR:
737 		/* No delay here, so we always clear the pending bit */
738 		apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
739 		apic_send_ipi(apic);
740 		break;
741 
742 	case APIC_ICR2:
743 		apic_set_reg(apic, APIC_ICR2, val & 0xff000000);
744 		break;
745 
746 	case APIC_LVTT:
747 	case APIC_LVTTHMR:
748 	case APIC_LVTPC:
749 	case APIC_LVT0:
750 	case APIC_LVT1:
751 	case APIC_LVTERR:
752 		/* TODO: Check vector */
753 		if (!apic_sw_enabled(apic))
754 			val |= APIC_LVT_MASKED;
755 
756 		val &= apic_lvt_mask[(offset - APIC_LVTT) >> 4];
757 		apic_set_reg(apic, offset, val);
758 
759 		break;
760 
761 	case APIC_TMICT:
762 		hrtimer_cancel(&apic->timer.dev);
763 		apic_set_reg(apic, APIC_TMICT, val);
764 		start_apic_timer(apic);
765 		return;
766 
767 	case APIC_TDCR:
768 		if (val & 4)
769 			printk(KERN_ERR "KVM_WRITE:TDCR %x\n", val);
770 		apic_set_reg(apic, APIC_TDCR, val);
771 		update_divide_count(apic);
772 		break;
773 
774 	default:
775 		apic_debug("Local APIC Write to read-only register %x\n",
776 			   offset);
777 		break;
778 	}
779 
780 }
781 
782 static int apic_mmio_range(struct kvm_io_device *this, gpa_t addr,
783 			   int len, int size)
784 {
785 	struct kvm_lapic *apic = (struct kvm_lapic *)this->private;
786 	int ret = 0;
787 
788 
789 	if (apic_hw_enabled(apic) &&
790 	    (addr >= apic->base_address) &&
791 	    (addr < (apic->base_address + LAPIC_MMIO_LENGTH)))
792 		ret = 1;
793 
794 	return ret;
795 }
796 
797 void kvm_free_lapic(struct kvm_vcpu *vcpu)
798 {
799 	if (!vcpu->arch.apic)
800 		return;
801 
802 	hrtimer_cancel(&vcpu->arch.apic->timer.dev);
803 
804 	if (vcpu->arch.apic->regs_page)
805 		__free_page(vcpu->arch.apic->regs_page);
806 
807 	kfree(vcpu->arch.apic);
808 }
809 
810 /*
811  *----------------------------------------------------------------------
812  * LAPIC interface
813  *----------------------------------------------------------------------
814  */
815 
816 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
817 {
818 	struct kvm_lapic *apic = vcpu->arch.apic;
819 
820 	if (!apic)
821 		return;
822 	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
823 		     | (apic_get_reg(apic, APIC_TASKPRI) & 4));
824 }
825 EXPORT_SYMBOL_GPL(kvm_lapic_set_tpr);
826 
827 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
828 {
829 	struct kvm_lapic *apic = vcpu->arch.apic;
830 	u64 tpr;
831 
832 	if (!apic)
833 		return 0;
834 	tpr = (u64) apic_get_reg(apic, APIC_TASKPRI);
835 
836 	return (tpr & 0xf0) >> 4;
837 }
838 EXPORT_SYMBOL_GPL(kvm_lapic_get_cr8);
839 
840 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
841 {
842 	struct kvm_lapic *apic = vcpu->arch.apic;
843 
844 	if (!apic) {
845 		value |= MSR_IA32_APICBASE_BSP;
846 		vcpu->arch.apic_base = value;
847 		return;
848 	}
849 	if (apic->vcpu->vcpu_id)
850 		value &= ~MSR_IA32_APICBASE_BSP;
851 
852 	vcpu->arch.apic_base = value;
853 	apic->base_address = apic->vcpu->arch.apic_base &
854 			     MSR_IA32_APICBASE_BASE;
855 
856 	/* with FSB delivery interrupt, we can restart APIC functionality */
857 	apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
858 		   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
859 
860 }
861 
862 u64 kvm_lapic_get_base(struct kvm_vcpu *vcpu)
863 {
864 	return vcpu->arch.apic_base;
865 }
866 EXPORT_SYMBOL_GPL(kvm_lapic_get_base);
867 
868 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
869 {
870 	struct kvm_lapic *apic;
871 	int i;
872 
873 	apic_debug("%s\n", __func__);
874 
875 	ASSERT(vcpu);
876 	apic = vcpu->arch.apic;
877 	ASSERT(apic != NULL);
878 
879 	/* Stop the timer in case it's a reset to an active apic */
880 	hrtimer_cancel(&apic->timer.dev);
881 
882 	apic_set_reg(apic, APIC_ID, vcpu->vcpu_id << 24);
883 	apic_set_reg(apic, APIC_LVR, APIC_VERSION);
884 
885 	for (i = 0; i < APIC_LVT_NUM; i++)
886 		apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
887 	apic_set_reg(apic, APIC_LVT0,
888 		     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
889 
890 	apic_set_reg(apic, APIC_DFR, 0xffffffffU);
891 	apic_set_reg(apic, APIC_SPIV, 0xff);
892 	apic_set_reg(apic, APIC_TASKPRI, 0);
893 	apic_set_reg(apic, APIC_LDR, 0);
894 	apic_set_reg(apic, APIC_ESR, 0);
895 	apic_set_reg(apic, APIC_ICR, 0);
896 	apic_set_reg(apic, APIC_ICR2, 0);
897 	apic_set_reg(apic, APIC_TDCR, 0);
898 	apic_set_reg(apic, APIC_TMICT, 0);
899 	for (i = 0; i < 8; i++) {
900 		apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
901 		apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
902 		apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
903 	}
904 	update_divide_count(apic);
905 	atomic_set(&apic->timer.pending, 0);
906 	if (vcpu->vcpu_id == 0)
907 		vcpu->arch.apic_base |= MSR_IA32_APICBASE_BSP;
908 	apic_update_ppr(apic);
909 
910 	apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
911 		   "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
912 		   vcpu, kvm_apic_id(apic),
913 		   vcpu->arch.apic_base, apic->base_address);
914 }
915 EXPORT_SYMBOL_GPL(kvm_lapic_reset);
916 
917 int kvm_lapic_enabled(struct kvm_vcpu *vcpu)
918 {
919 	struct kvm_lapic *apic = vcpu->arch.apic;
920 	int ret = 0;
921 
922 	if (!apic)
923 		return 0;
924 	ret = apic_enabled(apic);
925 
926 	return ret;
927 }
928 EXPORT_SYMBOL_GPL(kvm_lapic_enabled);
929 
930 /*
931  *----------------------------------------------------------------------
932  * timer interface
933  *----------------------------------------------------------------------
934  */
935 
936 /* TODO: make sure __apic_timer_fn runs in current pCPU */
937 static int __apic_timer_fn(struct kvm_lapic *apic)
938 {
939 	int result = 0;
940 	wait_queue_head_t *q = &apic->vcpu->wq;
941 
942 	if(!atomic_inc_and_test(&apic->timer.pending))
943 		set_bit(KVM_REQ_PENDING_TIMER, &apic->vcpu->requests);
944 	if (waitqueue_active(q))
945 		wake_up_interruptible(q);
946 
947 	if (apic_lvtt_period(apic)) {
948 		result = 1;
949 		hrtimer_add_expires_ns(&apic->timer.dev, apic->timer.period);
950 	}
951 	return result;
952 }
953 
954 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
955 {
956 	struct kvm_lapic *lapic = vcpu->arch.apic;
957 
958 	if (lapic && apic_enabled(lapic) && apic_lvt_enabled(lapic, APIC_LVTT))
959 		return atomic_read(&lapic->timer.pending);
960 
961 	return 0;
962 }
963 
964 static int __inject_apic_timer_irq(struct kvm_lapic *apic)
965 {
966 	int vector;
967 
968 	vector = apic_lvt_vector(apic, APIC_LVTT);
969 	return __apic_accept_irq(apic, APIC_DM_FIXED, vector, 1, 0);
970 }
971 
972 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
973 {
974 	struct kvm_lapic *apic;
975 	int restart_timer = 0;
976 
977 	apic = container_of(data, struct kvm_lapic, timer.dev);
978 
979 	restart_timer = __apic_timer_fn(apic);
980 
981 	if (restart_timer)
982 		return HRTIMER_RESTART;
983 	else
984 		return HRTIMER_NORESTART;
985 }
986 
987 int kvm_create_lapic(struct kvm_vcpu *vcpu)
988 {
989 	struct kvm_lapic *apic;
990 
991 	ASSERT(vcpu != NULL);
992 	apic_debug("apic_init %d\n", vcpu->vcpu_id);
993 
994 	apic = kzalloc(sizeof(*apic), GFP_KERNEL);
995 	if (!apic)
996 		goto nomem;
997 
998 	vcpu->arch.apic = apic;
999 
1000 	apic->regs_page = alloc_page(GFP_KERNEL);
1001 	if (apic->regs_page == NULL) {
1002 		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1003 		       vcpu->vcpu_id);
1004 		goto nomem_free_apic;
1005 	}
1006 	apic->regs = page_address(apic->regs_page);
1007 	memset(apic->regs, 0, PAGE_SIZE);
1008 	apic->vcpu = vcpu;
1009 
1010 	hrtimer_init(&apic->timer.dev, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1011 	apic->timer.dev.function = apic_timer_fn;
1012 	apic->base_address = APIC_DEFAULT_PHYS_BASE;
1013 	vcpu->arch.apic_base = APIC_DEFAULT_PHYS_BASE;
1014 
1015 	kvm_lapic_reset(vcpu);
1016 	apic->dev.read = apic_mmio_read;
1017 	apic->dev.write = apic_mmio_write;
1018 	apic->dev.in_range = apic_mmio_range;
1019 	apic->dev.private = apic;
1020 
1021 	return 0;
1022 nomem_free_apic:
1023 	kfree(apic);
1024 nomem:
1025 	return -ENOMEM;
1026 }
1027 EXPORT_SYMBOL_GPL(kvm_create_lapic);
1028 
1029 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1030 {
1031 	struct kvm_lapic *apic = vcpu->arch.apic;
1032 	int highest_irr;
1033 
1034 	if (!apic || !apic_enabled(apic))
1035 		return -1;
1036 
1037 	apic_update_ppr(apic);
1038 	highest_irr = apic_find_highest_irr(apic);
1039 	if ((highest_irr == -1) ||
1040 	    ((highest_irr & 0xF0) <= apic_get_reg(apic, APIC_PROCPRI)))
1041 		return -1;
1042 	return highest_irr;
1043 }
1044 
1045 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1046 {
1047 	u32 lvt0 = apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1048 	int r = 0;
1049 
1050 	if (vcpu->vcpu_id == 0) {
1051 		if (!apic_hw_enabled(vcpu->arch.apic))
1052 			r = 1;
1053 		if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1054 		    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1055 			r = 1;
1056 	}
1057 	return r;
1058 }
1059 
1060 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1061 {
1062 	struct kvm_lapic *apic = vcpu->arch.apic;
1063 
1064 	if (apic && apic_lvt_enabled(apic, APIC_LVTT) &&
1065 		atomic_read(&apic->timer.pending) > 0) {
1066 		if (__inject_apic_timer_irq(apic))
1067 			atomic_dec(&apic->timer.pending);
1068 	}
1069 }
1070 
1071 void kvm_apic_timer_intr_post(struct kvm_vcpu *vcpu, int vec)
1072 {
1073 	struct kvm_lapic *apic = vcpu->arch.apic;
1074 
1075 	if (apic && apic_lvt_vector(apic, APIC_LVTT) == vec)
1076 		apic->timer.last_update = ktime_add_ns(
1077 				apic->timer.last_update,
1078 				apic->timer.period);
1079 }
1080 
1081 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1082 {
1083 	int vector = kvm_apic_has_interrupt(vcpu);
1084 	struct kvm_lapic *apic = vcpu->arch.apic;
1085 
1086 	if (vector == -1)
1087 		return -1;
1088 
1089 	apic_set_vector(vector, apic->regs + APIC_ISR);
1090 	apic_update_ppr(apic);
1091 	apic_clear_irr(vector, apic);
1092 	return vector;
1093 }
1094 
1095 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu)
1096 {
1097 	struct kvm_lapic *apic = vcpu->arch.apic;
1098 
1099 	apic->base_address = vcpu->arch.apic_base &
1100 			     MSR_IA32_APICBASE_BASE;
1101 	apic_set_reg(apic, APIC_LVR, APIC_VERSION);
1102 	apic_update_ppr(apic);
1103 	hrtimer_cancel(&apic->timer.dev);
1104 	update_divide_count(apic);
1105 	start_apic_timer(apic);
1106 }
1107 
1108 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1109 {
1110 	struct kvm_lapic *apic = vcpu->arch.apic;
1111 	struct hrtimer *timer;
1112 
1113 	if (!apic)
1114 		return;
1115 
1116 	timer = &apic->timer.dev;
1117 	if (hrtimer_cancel(timer))
1118 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1119 }
1120 
1121 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1122 {
1123 	u32 data;
1124 	void *vapic;
1125 
1126 	if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1127 		return;
1128 
1129 	vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1130 	data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr));
1131 	kunmap_atomic(vapic, KM_USER0);
1132 
1133 	apic_set_tpr(vcpu->arch.apic, data & 0xff);
1134 }
1135 
1136 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1137 {
1138 	u32 data, tpr;
1139 	int max_irr, max_isr;
1140 	struct kvm_lapic *apic;
1141 	void *vapic;
1142 
1143 	if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1144 		return;
1145 
1146 	apic = vcpu->arch.apic;
1147 	tpr = apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1148 	max_irr = apic_find_highest_irr(apic);
1149 	if (max_irr < 0)
1150 		max_irr = 0;
1151 	max_isr = apic_find_highest_isr(apic);
1152 	if (max_isr < 0)
1153 		max_isr = 0;
1154 	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1155 
1156 	vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1157 	*(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data;
1158 	kunmap_atomic(vapic, KM_USER0);
1159 }
1160 
1161 void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1162 {
1163 	if (!irqchip_in_kernel(vcpu->kvm))
1164 		return;
1165 
1166 	vcpu->arch.apic->vapic_addr = vapic_addr;
1167 }
1168