xref: /openbmc/linux/arch/x86/kvm/lapic.c (revision e8e0929d)
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 <asm/apicdef.h>
36 #include "kvm_cache_regs.h"
37 #include "irq.h"
38 #include "trace.h"
39 #include "x86.h"
40 
41 #ifndef CONFIG_X86_64
42 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
43 #else
44 #define mod_64(x, y) ((x) % (y))
45 #endif
46 
47 #define PRId64 "d"
48 #define PRIx64 "llx"
49 #define PRIu64 "u"
50 #define PRIo64 "o"
51 
52 #define APIC_BUS_CYCLE_NS 1
53 
54 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
55 #define apic_debug(fmt, arg...)
56 
57 #define APIC_LVT_NUM			6
58 /* 14 is the version for Xeon and Pentium 8.4.8*/
59 #define APIC_VERSION			(0x14UL | ((APIC_LVT_NUM - 1) << 16))
60 #define LAPIC_MMIO_LENGTH		(1 << 12)
61 /* followed define is not in apicdef.h */
62 #define APIC_SHORT_MASK			0xc0000
63 #define APIC_DEST_NOSHORT		0x0
64 #define APIC_DEST_MASK			0x800
65 #define MAX_APIC_VECTOR			256
66 
67 #define VEC_POS(v) ((v) & (32 - 1))
68 #define REG_POS(v) (((v) >> 5) << 4)
69 
70 static inline u32 apic_get_reg(struct kvm_lapic *apic, int reg_off)
71 {
72 	return *((u32 *) (apic->regs + reg_off));
73 }
74 
75 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
76 {
77 	*((u32 *) (apic->regs + reg_off)) = val;
78 }
79 
80 static inline int apic_test_and_set_vector(int vec, void *bitmap)
81 {
82 	return test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
83 }
84 
85 static inline int apic_test_and_clear_vector(int vec, void *bitmap)
86 {
87 	return test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
88 }
89 
90 static inline void apic_set_vector(int vec, void *bitmap)
91 {
92 	set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
93 }
94 
95 static inline void apic_clear_vector(int vec, void *bitmap)
96 {
97 	clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
98 }
99 
100 static inline int apic_hw_enabled(struct kvm_lapic *apic)
101 {
102 	return (apic)->vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE;
103 }
104 
105 static inline int  apic_sw_enabled(struct kvm_lapic *apic)
106 {
107 	return apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED;
108 }
109 
110 static inline int apic_enabled(struct kvm_lapic *apic)
111 {
112 	return apic_sw_enabled(apic) &&	apic_hw_enabled(apic);
113 }
114 
115 #define LVT_MASK	\
116 	(APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
117 
118 #define LINT_MASK	\
119 	(LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
120 	 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
121 
122 static inline int kvm_apic_id(struct kvm_lapic *apic)
123 {
124 	return (apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
125 }
126 
127 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
128 {
129 	return !(apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
130 }
131 
132 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
133 {
134 	return apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
135 }
136 
137 static inline int apic_lvtt_period(struct kvm_lapic *apic)
138 {
139 	return apic_get_reg(apic, APIC_LVTT) & APIC_LVT_TIMER_PERIODIC;
140 }
141 
142 static inline int apic_lvt_nmi_mode(u32 lvt_val)
143 {
144 	return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
145 }
146 
147 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
148 {
149 	struct kvm_lapic *apic = vcpu->arch.apic;
150 	struct kvm_cpuid_entry2 *feat;
151 	u32 v = APIC_VERSION;
152 
153 	if (!irqchip_in_kernel(vcpu->kvm))
154 		return;
155 
156 	feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
157 	if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
158 		v |= APIC_LVR_DIRECTED_EOI;
159 	apic_set_reg(apic, APIC_LVR, v);
160 }
161 
162 static inline int apic_x2apic_mode(struct kvm_lapic *apic)
163 {
164 	return apic->vcpu->arch.apic_base & X2APIC_ENABLE;
165 }
166 
167 static unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
168 	LVT_MASK | APIC_LVT_TIMER_PERIODIC,	/* LVTT */
169 	LVT_MASK | APIC_MODE_MASK,	/* LVTTHMR */
170 	LVT_MASK | APIC_MODE_MASK,	/* LVTPC */
171 	LINT_MASK, LINT_MASK,	/* LVT0-1 */
172 	LVT_MASK		/* LVTERR */
173 };
174 
175 static int find_highest_vector(void *bitmap)
176 {
177 	u32 *word = bitmap;
178 	int word_offset = MAX_APIC_VECTOR >> 5;
179 
180 	while ((word_offset != 0) && (word[(--word_offset) << 2] == 0))
181 		continue;
182 
183 	if (likely(!word_offset && !word[0]))
184 		return -1;
185 	else
186 		return fls(word[word_offset << 2]) - 1 + (word_offset << 5);
187 }
188 
189 static inline int apic_test_and_set_irr(int vec, struct kvm_lapic *apic)
190 {
191 	apic->irr_pending = true;
192 	return apic_test_and_set_vector(vec, apic->regs + APIC_IRR);
193 }
194 
195 static inline int apic_search_irr(struct kvm_lapic *apic)
196 {
197 	return find_highest_vector(apic->regs + APIC_IRR);
198 }
199 
200 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
201 {
202 	int result;
203 
204 	if (!apic->irr_pending)
205 		return -1;
206 
207 	result = apic_search_irr(apic);
208 	ASSERT(result == -1 || result >= 16);
209 
210 	return result;
211 }
212 
213 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
214 {
215 	apic->irr_pending = false;
216 	apic_clear_vector(vec, apic->regs + APIC_IRR);
217 	if (apic_search_irr(apic) != -1)
218 		apic->irr_pending = true;
219 }
220 
221 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
222 {
223 	struct kvm_lapic *apic = vcpu->arch.apic;
224 	int highest_irr;
225 
226 	/* This may race with setting of irr in __apic_accept_irq() and
227 	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
228 	 * will cause vmexit immediately and the value will be recalculated
229 	 * on the next vmentry.
230 	 */
231 	if (!apic)
232 		return 0;
233 	highest_irr = apic_find_highest_irr(apic);
234 
235 	return highest_irr;
236 }
237 
238 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
239 			     int vector, int level, int trig_mode);
240 
241 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq)
242 {
243 	struct kvm_lapic *apic = vcpu->arch.apic;
244 
245 	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
246 			irq->level, irq->trig_mode);
247 }
248 
249 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
250 {
251 	int result;
252 
253 	result = find_highest_vector(apic->regs + APIC_ISR);
254 	ASSERT(result == -1 || result >= 16);
255 
256 	return result;
257 }
258 
259 static void apic_update_ppr(struct kvm_lapic *apic)
260 {
261 	u32 tpr, isrv, ppr;
262 	int isr;
263 
264 	tpr = apic_get_reg(apic, APIC_TASKPRI);
265 	isr = apic_find_highest_isr(apic);
266 	isrv = (isr != -1) ? isr : 0;
267 
268 	if ((tpr & 0xf0) >= (isrv & 0xf0))
269 		ppr = tpr & 0xff;
270 	else
271 		ppr = isrv & 0xf0;
272 
273 	apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
274 		   apic, ppr, isr, isrv);
275 
276 	apic_set_reg(apic, APIC_PROCPRI, ppr);
277 }
278 
279 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
280 {
281 	apic_set_reg(apic, APIC_TASKPRI, tpr);
282 	apic_update_ppr(apic);
283 }
284 
285 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
286 {
287 	return dest == 0xff || kvm_apic_id(apic) == dest;
288 }
289 
290 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
291 {
292 	int result = 0;
293 	u32 logical_id;
294 
295 	if (apic_x2apic_mode(apic)) {
296 		logical_id = apic_get_reg(apic, APIC_LDR);
297 		return logical_id & mda;
298 	}
299 
300 	logical_id = GET_APIC_LOGICAL_ID(apic_get_reg(apic, APIC_LDR));
301 
302 	switch (apic_get_reg(apic, APIC_DFR)) {
303 	case APIC_DFR_FLAT:
304 		if (logical_id & mda)
305 			result = 1;
306 		break;
307 	case APIC_DFR_CLUSTER:
308 		if (((logical_id >> 4) == (mda >> 0x4))
309 		    && (logical_id & mda & 0xf))
310 			result = 1;
311 		break;
312 	default:
313 		printk(KERN_WARNING "Bad DFR vcpu %d: %08x\n",
314 		       apic->vcpu->vcpu_id, apic_get_reg(apic, APIC_DFR));
315 		break;
316 	}
317 
318 	return result;
319 }
320 
321 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
322 			   int short_hand, int dest, int dest_mode)
323 {
324 	int result = 0;
325 	struct kvm_lapic *target = vcpu->arch.apic;
326 
327 	apic_debug("target %p, source %p, dest 0x%x, "
328 		   "dest_mode 0x%x, short_hand 0x%x\n",
329 		   target, source, dest, dest_mode, short_hand);
330 
331 	ASSERT(!target);
332 	switch (short_hand) {
333 	case APIC_DEST_NOSHORT:
334 		if (dest_mode == 0)
335 			/* Physical mode. */
336 			result = kvm_apic_match_physical_addr(target, dest);
337 		else
338 			/* Logical mode. */
339 			result = kvm_apic_match_logical_addr(target, dest);
340 		break;
341 	case APIC_DEST_SELF:
342 		result = (target == source);
343 		break;
344 	case APIC_DEST_ALLINC:
345 		result = 1;
346 		break;
347 	case APIC_DEST_ALLBUT:
348 		result = (target != source);
349 		break;
350 	default:
351 		printk(KERN_WARNING "Bad dest shorthand value %x\n",
352 		       short_hand);
353 		break;
354 	}
355 
356 	return result;
357 }
358 
359 /*
360  * Add a pending IRQ into lapic.
361  * Return 1 if successfully added and 0 if discarded.
362  */
363 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
364 			     int vector, int level, int trig_mode)
365 {
366 	int result = 0;
367 	struct kvm_vcpu *vcpu = apic->vcpu;
368 
369 	switch (delivery_mode) {
370 	case APIC_DM_LOWEST:
371 		vcpu->arch.apic_arb_prio++;
372 	case APIC_DM_FIXED:
373 		/* FIXME add logic for vcpu on reset */
374 		if (unlikely(!apic_enabled(apic)))
375 			break;
376 
377 		result = !apic_test_and_set_irr(vector, apic);
378 		trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
379 					  trig_mode, vector, !result);
380 		if (!result) {
381 			if (trig_mode)
382 				apic_debug("level trig mode repeatedly for "
383 						"vector %d", vector);
384 			break;
385 		}
386 
387 		if (trig_mode) {
388 			apic_debug("level trig mode for vector %d", vector);
389 			apic_set_vector(vector, apic->regs + APIC_TMR);
390 		} else
391 			apic_clear_vector(vector, apic->regs + APIC_TMR);
392 		kvm_vcpu_kick(vcpu);
393 		break;
394 
395 	case APIC_DM_REMRD:
396 		printk(KERN_DEBUG "Ignoring delivery mode 3\n");
397 		break;
398 
399 	case APIC_DM_SMI:
400 		printk(KERN_DEBUG "Ignoring guest SMI\n");
401 		break;
402 
403 	case APIC_DM_NMI:
404 		result = 1;
405 		kvm_inject_nmi(vcpu);
406 		kvm_vcpu_kick(vcpu);
407 		break;
408 
409 	case APIC_DM_INIT:
410 		if (level) {
411 			result = 1;
412 			if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
413 				printk(KERN_DEBUG
414 				       "INIT on a runnable vcpu %d\n",
415 				       vcpu->vcpu_id);
416 			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
417 			kvm_vcpu_kick(vcpu);
418 		} else {
419 			apic_debug("Ignoring de-assert INIT to vcpu %d\n",
420 				   vcpu->vcpu_id);
421 		}
422 		break;
423 
424 	case APIC_DM_STARTUP:
425 		apic_debug("SIPI to vcpu %d vector 0x%02x\n",
426 			   vcpu->vcpu_id, vector);
427 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
428 			result = 1;
429 			vcpu->arch.sipi_vector = vector;
430 			vcpu->arch.mp_state = KVM_MP_STATE_SIPI_RECEIVED;
431 			kvm_vcpu_kick(vcpu);
432 		}
433 		break;
434 
435 	case APIC_DM_EXTINT:
436 		/*
437 		 * Should only be called by kvm_apic_local_deliver() with LVT0,
438 		 * before NMI watchdog was enabled. Already handled by
439 		 * kvm_apic_accept_pic_intr().
440 		 */
441 		break;
442 
443 	default:
444 		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
445 		       delivery_mode);
446 		break;
447 	}
448 	return result;
449 }
450 
451 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
452 {
453 	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
454 }
455 
456 static void apic_set_eoi(struct kvm_lapic *apic)
457 {
458 	int vector = apic_find_highest_isr(apic);
459 	int trigger_mode;
460 	/*
461 	 * Not every write EOI will has corresponding ISR,
462 	 * one example is when Kernel check timer on setup_IO_APIC
463 	 */
464 	if (vector == -1)
465 		return;
466 
467 	apic_clear_vector(vector, apic->regs + APIC_ISR);
468 	apic_update_ppr(apic);
469 
470 	if (apic_test_and_clear_vector(vector, apic->regs + APIC_TMR))
471 		trigger_mode = IOAPIC_LEVEL_TRIG;
472 	else
473 		trigger_mode = IOAPIC_EDGE_TRIG;
474 	if (!(apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI)) {
475 		mutex_lock(&apic->vcpu->kvm->irq_lock);
476 		kvm_ioapic_update_eoi(apic->vcpu->kvm, vector, trigger_mode);
477 		mutex_unlock(&apic->vcpu->kvm->irq_lock);
478 	}
479 }
480 
481 static void apic_send_ipi(struct kvm_lapic *apic)
482 {
483 	u32 icr_low = apic_get_reg(apic, APIC_ICR);
484 	u32 icr_high = apic_get_reg(apic, APIC_ICR2);
485 	struct kvm_lapic_irq irq;
486 
487 	irq.vector = icr_low & APIC_VECTOR_MASK;
488 	irq.delivery_mode = icr_low & APIC_MODE_MASK;
489 	irq.dest_mode = icr_low & APIC_DEST_MASK;
490 	irq.level = icr_low & APIC_INT_ASSERT;
491 	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
492 	irq.shorthand = icr_low & APIC_SHORT_MASK;
493 	if (apic_x2apic_mode(apic))
494 		irq.dest_id = icr_high;
495 	else
496 		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
497 
498 	trace_kvm_apic_ipi(icr_low, irq.dest_id);
499 
500 	apic_debug("icr_high 0x%x, icr_low 0x%x, "
501 		   "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
502 		   "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
503 		   icr_high, icr_low, irq.shorthand, irq.dest_id,
504 		   irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
505 		   irq.vector);
506 
507 	mutex_lock(&apic->vcpu->kvm->irq_lock);
508 	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq);
509 	mutex_unlock(&apic->vcpu->kvm->irq_lock);
510 }
511 
512 static u32 apic_get_tmcct(struct kvm_lapic *apic)
513 {
514 	ktime_t remaining;
515 	s64 ns;
516 	u32 tmcct;
517 
518 	ASSERT(apic != NULL);
519 
520 	/* if initial count is 0, current count should also be 0 */
521 	if (apic_get_reg(apic, APIC_TMICT) == 0)
522 		return 0;
523 
524 	remaining = hrtimer_expires_remaining(&apic->lapic_timer.timer);
525 	if (ktime_to_ns(remaining) < 0)
526 		remaining = ktime_set(0, 0);
527 
528 	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
529 	tmcct = div64_u64(ns,
530 			 (APIC_BUS_CYCLE_NS * apic->divide_count));
531 
532 	return tmcct;
533 }
534 
535 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
536 {
537 	struct kvm_vcpu *vcpu = apic->vcpu;
538 	struct kvm_run *run = vcpu->run;
539 
540 	set_bit(KVM_REQ_REPORT_TPR_ACCESS, &vcpu->requests);
541 	run->tpr_access.rip = kvm_rip_read(vcpu);
542 	run->tpr_access.is_write = write;
543 }
544 
545 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
546 {
547 	if (apic->vcpu->arch.tpr_access_reporting)
548 		__report_tpr_access(apic, write);
549 }
550 
551 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
552 {
553 	u32 val = 0;
554 
555 	if (offset >= LAPIC_MMIO_LENGTH)
556 		return 0;
557 
558 	switch (offset) {
559 	case APIC_ID:
560 		if (apic_x2apic_mode(apic))
561 			val = kvm_apic_id(apic);
562 		else
563 			val = kvm_apic_id(apic) << 24;
564 		break;
565 	case APIC_ARBPRI:
566 		printk(KERN_WARNING "Access APIC ARBPRI register "
567 		       "which is for P6\n");
568 		break;
569 
570 	case APIC_TMCCT:	/* Timer CCR */
571 		val = apic_get_tmcct(apic);
572 		break;
573 
574 	case APIC_TASKPRI:
575 		report_tpr_access(apic, false);
576 		/* fall thru */
577 	default:
578 		apic_update_ppr(apic);
579 		val = apic_get_reg(apic, offset);
580 		break;
581 	}
582 
583 	return val;
584 }
585 
586 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
587 {
588 	return container_of(dev, struct kvm_lapic, dev);
589 }
590 
591 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
592 		void *data)
593 {
594 	unsigned char alignment = offset & 0xf;
595 	u32 result;
596 	/* this bitmask has a bit cleared for each reserver register */
597 	static const u64 rmask = 0x43ff01ffffffe70cULL;
598 
599 	if ((alignment + len) > 4) {
600 		apic_debug("KVM_APIC_READ: alignment error %x %d\n",
601 			   offset, len);
602 		return 1;
603 	}
604 
605 	if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
606 		apic_debug("KVM_APIC_READ: read reserved register %x\n",
607 			   offset);
608 		return 1;
609 	}
610 
611 	result = __apic_read(apic, offset & ~0xf);
612 
613 	trace_kvm_apic_read(offset, result);
614 
615 	switch (len) {
616 	case 1:
617 	case 2:
618 	case 4:
619 		memcpy(data, (char *)&result + alignment, len);
620 		break;
621 	default:
622 		printk(KERN_ERR "Local APIC read with len = %x, "
623 		       "should be 1,2, or 4 instead\n", len);
624 		break;
625 	}
626 	return 0;
627 }
628 
629 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
630 {
631 	return apic_hw_enabled(apic) &&
632 	    addr >= apic->base_address &&
633 	    addr < apic->base_address + LAPIC_MMIO_LENGTH;
634 }
635 
636 static int apic_mmio_read(struct kvm_io_device *this,
637 			   gpa_t address, int len, void *data)
638 {
639 	struct kvm_lapic *apic = to_lapic(this);
640 	u32 offset = address - apic->base_address;
641 
642 	if (!apic_mmio_in_range(apic, address))
643 		return -EOPNOTSUPP;
644 
645 	apic_reg_read(apic, offset, len, data);
646 
647 	return 0;
648 }
649 
650 static void update_divide_count(struct kvm_lapic *apic)
651 {
652 	u32 tmp1, tmp2, tdcr;
653 
654 	tdcr = apic_get_reg(apic, APIC_TDCR);
655 	tmp1 = tdcr & 0xf;
656 	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
657 	apic->divide_count = 0x1 << (tmp2 & 0x7);
658 
659 	apic_debug("timer divide count is 0x%x\n",
660 				   apic->divide_count);
661 }
662 
663 static void start_apic_timer(struct kvm_lapic *apic)
664 {
665 	ktime_t now = apic->lapic_timer.timer.base->get_time();
666 
667 	apic->lapic_timer.period = apic_get_reg(apic, APIC_TMICT) *
668 		    APIC_BUS_CYCLE_NS * apic->divide_count;
669 	atomic_set(&apic->lapic_timer.pending, 0);
670 
671 	if (!apic->lapic_timer.period)
672 		return;
673 	/*
674 	 * Do not allow the guest to program periodic timers with small
675 	 * interval, since the hrtimers are not throttled by the host
676 	 * scheduler.
677 	 */
678 	if (apic_lvtt_period(apic)) {
679 		if (apic->lapic_timer.period < NSEC_PER_MSEC/2)
680 			apic->lapic_timer.period = NSEC_PER_MSEC/2;
681 	}
682 
683 	hrtimer_start(&apic->lapic_timer.timer,
684 		      ktime_add_ns(now, apic->lapic_timer.period),
685 		      HRTIMER_MODE_ABS);
686 
687 	apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
688 			   PRIx64 ", "
689 			   "timer initial count 0x%x, period %lldns, "
690 			   "expire @ 0x%016" PRIx64 ".\n", __func__,
691 			   APIC_BUS_CYCLE_NS, ktime_to_ns(now),
692 			   apic_get_reg(apic, APIC_TMICT),
693 			   apic->lapic_timer.period,
694 			   ktime_to_ns(ktime_add_ns(now,
695 					apic->lapic_timer.period)));
696 }
697 
698 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
699 {
700 	int nmi_wd_enabled = apic_lvt_nmi_mode(apic_get_reg(apic, APIC_LVT0));
701 
702 	if (apic_lvt_nmi_mode(lvt0_val)) {
703 		if (!nmi_wd_enabled) {
704 			apic_debug("Receive NMI setting on APIC_LVT0 "
705 				   "for cpu %d\n", apic->vcpu->vcpu_id);
706 			apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
707 		}
708 	} else if (nmi_wd_enabled)
709 		apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
710 }
711 
712 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
713 {
714 	int ret = 0;
715 
716 	trace_kvm_apic_write(reg, val);
717 
718 	switch (reg) {
719 	case APIC_ID:		/* Local APIC ID */
720 		if (!apic_x2apic_mode(apic))
721 			apic_set_reg(apic, APIC_ID, val);
722 		else
723 			ret = 1;
724 		break;
725 
726 	case APIC_TASKPRI:
727 		report_tpr_access(apic, true);
728 		apic_set_tpr(apic, val & 0xff);
729 		break;
730 
731 	case APIC_EOI:
732 		apic_set_eoi(apic);
733 		break;
734 
735 	case APIC_LDR:
736 		if (!apic_x2apic_mode(apic))
737 			apic_set_reg(apic, APIC_LDR, val & APIC_LDR_MASK);
738 		else
739 			ret = 1;
740 		break;
741 
742 	case APIC_DFR:
743 		if (!apic_x2apic_mode(apic))
744 			apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
745 		else
746 			ret = 1;
747 		break;
748 
749 	case APIC_SPIV: {
750 		u32 mask = 0x3ff;
751 		if (apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
752 			mask |= APIC_SPIV_DIRECTED_EOI;
753 		apic_set_reg(apic, APIC_SPIV, val & mask);
754 		if (!(val & APIC_SPIV_APIC_ENABLED)) {
755 			int i;
756 			u32 lvt_val;
757 
758 			for (i = 0; i < APIC_LVT_NUM; i++) {
759 				lvt_val = apic_get_reg(apic,
760 						       APIC_LVTT + 0x10 * i);
761 				apic_set_reg(apic, APIC_LVTT + 0x10 * i,
762 					     lvt_val | APIC_LVT_MASKED);
763 			}
764 			atomic_set(&apic->lapic_timer.pending, 0);
765 
766 		}
767 		break;
768 	}
769 	case APIC_ICR:
770 		/* No delay here, so we always clear the pending bit */
771 		apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
772 		apic_send_ipi(apic);
773 		break;
774 
775 	case APIC_ICR2:
776 		if (!apic_x2apic_mode(apic))
777 			val &= 0xff000000;
778 		apic_set_reg(apic, APIC_ICR2, val);
779 		break;
780 
781 	case APIC_LVT0:
782 		apic_manage_nmi_watchdog(apic, val);
783 	case APIC_LVTT:
784 	case APIC_LVTTHMR:
785 	case APIC_LVTPC:
786 	case APIC_LVT1:
787 	case APIC_LVTERR:
788 		/* TODO: Check vector */
789 		if (!apic_sw_enabled(apic))
790 			val |= APIC_LVT_MASKED;
791 
792 		val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
793 		apic_set_reg(apic, reg, val);
794 
795 		break;
796 
797 	case APIC_TMICT:
798 		hrtimer_cancel(&apic->lapic_timer.timer);
799 		apic_set_reg(apic, APIC_TMICT, val);
800 		start_apic_timer(apic);
801 		break;
802 
803 	case APIC_TDCR:
804 		if (val & 4)
805 			printk(KERN_ERR "KVM_WRITE:TDCR %x\n", val);
806 		apic_set_reg(apic, APIC_TDCR, val);
807 		update_divide_count(apic);
808 		break;
809 
810 	case APIC_ESR:
811 		if (apic_x2apic_mode(apic) && val != 0) {
812 			printk(KERN_ERR "KVM_WRITE:ESR not zero %x\n", val);
813 			ret = 1;
814 		}
815 		break;
816 
817 	case APIC_SELF_IPI:
818 		if (apic_x2apic_mode(apic)) {
819 			apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
820 		} else
821 			ret = 1;
822 		break;
823 	default:
824 		ret = 1;
825 		break;
826 	}
827 	if (ret)
828 		apic_debug("Local APIC Write to read-only register %x\n", reg);
829 	return ret;
830 }
831 
832 static int apic_mmio_write(struct kvm_io_device *this,
833 			    gpa_t address, int len, const void *data)
834 {
835 	struct kvm_lapic *apic = to_lapic(this);
836 	unsigned int offset = address - apic->base_address;
837 	u32 val;
838 
839 	if (!apic_mmio_in_range(apic, address))
840 		return -EOPNOTSUPP;
841 
842 	/*
843 	 * APIC register must be aligned on 128-bits boundary.
844 	 * 32/64/128 bits registers must be accessed thru 32 bits.
845 	 * Refer SDM 8.4.1
846 	 */
847 	if (len != 4 || (offset & 0xf)) {
848 		/* Don't shout loud, $infamous_os would cause only noise. */
849 		apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
850 		return 0;
851 	}
852 
853 	val = *(u32*)data;
854 
855 	/* too common printing */
856 	if (offset != APIC_EOI)
857 		apic_debug("%s: offset 0x%x with length 0x%x, and value is "
858 			   "0x%x\n", __func__, offset, len, val);
859 
860 	apic_reg_write(apic, offset & 0xff0, val);
861 
862 	return 0;
863 }
864 
865 void kvm_free_lapic(struct kvm_vcpu *vcpu)
866 {
867 	if (!vcpu->arch.apic)
868 		return;
869 
870 	hrtimer_cancel(&vcpu->arch.apic->lapic_timer.timer);
871 
872 	if (vcpu->arch.apic->regs_page)
873 		__free_page(vcpu->arch.apic->regs_page);
874 
875 	kfree(vcpu->arch.apic);
876 }
877 
878 /*
879  *----------------------------------------------------------------------
880  * LAPIC interface
881  *----------------------------------------------------------------------
882  */
883 
884 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
885 {
886 	struct kvm_lapic *apic = vcpu->arch.apic;
887 
888 	if (!apic)
889 		return;
890 	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
891 		     | (apic_get_reg(apic, APIC_TASKPRI) & 4));
892 }
893 
894 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
895 {
896 	struct kvm_lapic *apic = vcpu->arch.apic;
897 	u64 tpr;
898 
899 	if (!apic)
900 		return 0;
901 	tpr = (u64) apic_get_reg(apic, APIC_TASKPRI);
902 
903 	return (tpr & 0xf0) >> 4;
904 }
905 
906 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
907 {
908 	struct kvm_lapic *apic = vcpu->arch.apic;
909 
910 	if (!apic) {
911 		value |= MSR_IA32_APICBASE_BSP;
912 		vcpu->arch.apic_base = value;
913 		return;
914 	}
915 
916 	if (!kvm_vcpu_is_bsp(apic->vcpu))
917 		value &= ~MSR_IA32_APICBASE_BSP;
918 
919 	vcpu->arch.apic_base = value;
920 	if (apic_x2apic_mode(apic)) {
921 		u32 id = kvm_apic_id(apic);
922 		u32 ldr = ((id & ~0xf) << 16) | (1 << (id & 0xf));
923 		apic_set_reg(apic, APIC_LDR, ldr);
924 	}
925 	apic->base_address = apic->vcpu->arch.apic_base &
926 			     MSR_IA32_APICBASE_BASE;
927 
928 	/* with FSB delivery interrupt, we can restart APIC functionality */
929 	apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
930 		   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
931 
932 }
933 
934 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
935 {
936 	struct kvm_lapic *apic;
937 	int i;
938 
939 	apic_debug("%s\n", __func__);
940 
941 	ASSERT(vcpu);
942 	apic = vcpu->arch.apic;
943 	ASSERT(apic != NULL);
944 
945 	/* Stop the timer in case it's a reset to an active apic */
946 	hrtimer_cancel(&apic->lapic_timer.timer);
947 
948 	apic_set_reg(apic, APIC_ID, vcpu->vcpu_id << 24);
949 	kvm_apic_set_version(apic->vcpu);
950 
951 	for (i = 0; i < APIC_LVT_NUM; i++)
952 		apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
953 	apic_set_reg(apic, APIC_LVT0,
954 		     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
955 
956 	apic_set_reg(apic, APIC_DFR, 0xffffffffU);
957 	apic_set_reg(apic, APIC_SPIV, 0xff);
958 	apic_set_reg(apic, APIC_TASKPRI, 0);
959 	apic_set_reg(apic, APIC_LDR, 0);
960 	apic_set_reg(apic, APIC_ESR, 0);
961 	apic_set_reg(apic, APIC_ICR, 0);
962 	apic_set_reg(apic, APIC_ICR2, 0);
963 	apic_set_reg(apic, APIC_TDCR, 0);
964 	apic_set_reg(apic, APIC_TMICT, 0);
965 	for (i = 0; i < 8; i++) {
966 		apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
967 		apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
968 		apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
969 	}
970 	apic->irr_pending = false;
971 	update_divide_count(apic);
972 	atomic_set(&apic->lapic_timer.pending, 0);
973 	if (kvm_vcpu_is_bsp(vcpu))
974 		vcpu->arch.apic_base |= MSR_IA32_APICBASE_BSP;
975 	apic_update_ppr(apic);
976 
977 	vcpu->arch.apic_arb_prio = 0;
978 
979 	apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
980 		   "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
981 		   vcpu, kvm_apic_id(apic),
982 		   vcpu->arch.apic_base, apic->base_address);
983 }
984 
985 bool kvm_apic_present(struct kvm_vcpu *vcpu)
986 {
987 	return vcpu->arch.apic && apic_hw_enabled(vcpu->arch.apic);
988 }
989 
990 int kvm_lapic_enabled(struct kvm_vcpu *vcpu)
991 {
992 	return kvm_apic_present(vcpu) && apic_sw_enabled(vcpu->arch.apic);
993 }
994 
995 /*
996  *----------------------------------------------------------------------
997  * timer interface
998  *----------------------------------------------------------------------
999  */
1000 
1001 static bool lapic_is_periodic(struct kvm_timer *ktimer)
1002 {
1003 	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic,
1004 					      lapic_timer);
1005 	return apic_lvtt_period(apic);
1006 }
1007 
1008 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1009 {
1010 	struct kvm_lapic *lapic = vcpu->arch.apic;
1011 
1012 	if (lapic && apic_enabled(lapic) && apic_lvt_enabled(lapic, APIC_LVTT))
1013 		return atomic_read(&lapic->lapic_timer.pending);
1014 
1015 	return 0;
1016 }
1017 
1018 static int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1019 {
1020 	u32 reg = apic_get_reg(apic, lvt_type);
1021 	int vector, mode, trig_mode;
1022 
1023 	if (apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1024 		vector = reg & APIC_VECTOR_MASK;
1025 		mode = reg & APIC_MODE_MASK;
1026 		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1027 		return __apic_accept_irq(apic, mode, vector, 1, trig_mode);
1028 	}
1029 	return 0;
1030 }
1031 
1032 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1033 {
1034 	struct kvm_lapic *apic = vcpu->arch.apic;
1035 
1036 	if (apic)
1037 		kvm_apic_local_deliver(apic, APIC_LVT0);
1038 }
1039 
1040 static struct kvm_timer_ops lapic_timer_ops = {
1041 	.is_periodic = lapic_is_periodic,
1042 };
1043 
1044 static const struct kvm_io_device_ops apic_mmio_ops = {
1045 	.read     = apic_mmio_read,
1046 	.write    = apic_mmio_write,
1047 };
1048 
1049 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1050 {
1051 	struct kvm_lapic *apic;
1052 
1053 	ASSERT(vcpu != NULL);
1054 	apic_debug("apic_init %d\n", vcpu->vcpu_id);
1055 
1056 	apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1057 	if (!apic)
1058 		goto nomem;
1059 
1060 	vcpu->arch.apic = apic;
1061 
1062 	apic->regs_page = alloc_page(GFP_KERNEL);
1063 	if (apic->regs_page == NULL) {
1064 		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1065 		       vcpu->vcpu_id);
1066 		goto nomem_free_apic;
1067 	}
1068 	apic->regs = page_address(apic->regs_page);
1069 	memset(apic->regs, 0, PAGE_SIZE);
1070 	apic->vcpu = vcpu;
1071 
1072 	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1073 		     HRTIMER_MODE_ABS);
1074 	apic->lapic_timer.timer.function = kvm_timer_fn;
1075 	apic->lapic_timer.t_ops = &lapic_timer_ops;
1076 	apic->lapic_timer.kvm = vcpu->kvm;
1077 	apic->lapic_timer.vcpu = vcpu;
1078 
1079 	apic->base_address = APIC_DEFAULT_PHYS_BASE;
1080 	vcpu->arch.apic_base = APIC_DEFAULT_PHYS_BASE;
1081 
1082 	kvm_lapic_reset(vcpu);
1083 	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1084 
1085 	return 0;
1086 nomem_free_apic:
1087 	kfree(apic);
1088 nomem:
1089 	return -ENOMEM;
1090 }
1091 
1092 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1093 {
1094 	struct kvm_lapic *apic = vcpu->arch.apic;
1095 	int highest_irr;
1096 
1097 	if (!apic || !apic_enabled(apic))
1098 		return -1;
1099 
1100 	apic_update_ppr(apic);
1101 	highest_irr = apic_find_highest_irr(apic);
1102 	if ((highest_irr == -1) ||
1103 	    ((highest_irr & 0xF0) <= apic_get_reg(apic, APIC_PROCPRI)))
1104 		return -1;
1105 	return highest_irr;
1106 }
1107 
1108 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1109 {
1110 	u32 lvt0 = apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1111 	int r = 0;
1112 
1113 	if (kvm_vcpu_is_bsp(vcpu)) {
1114 		if (!apic_hw_enabled(vcpu->arch.apic))
1115 			r = 1;
1116 		if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1117 		    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1118 			r = 1;
1119 	}
1120 	return r;
1121 }
1122 
1123 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1124 {
1125 	struct kvm_lapic *apic = vcpu->arch.apic;
1126 
1127 	if (apic && atomic_read(&apic->lapic_timer.pending) > 0) {
1128 		if (kvm_apic_local_deliver(apic, APIC_LVTT))
1129 			atomic_dec(&apic->lapic_timer.pending);
1130 	}
1131 }
1132 
1133 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1134 {
1135 	int vector = kvm_apic_has_interrupt(vcpu);
1136 	struct kvm_lapic *apic = vcpu->arch.apic;
1137 
1138 	if (vector == -1)
1139 		return -1;
1140 
1141 	apic_set_vector(vector, apic->regs + APIC_ISR);
1142 	apic_update_ppr(apic);
1143 	apic_clear_irr(vector, apic);
1144 	return vector;
1145 }
1146 
1147 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu)
1148 {
1149 	struct kvm_lapic *apic = vcpu->arch.apic;
1150 
1151 	apic->base_address = vcpu->arch.apic_base &
1152 			     MSR_IA32_APICBASE_BASE;
1153 	kvm_apic_set_version(vcpu);
1154 
1155 	apic_update_ppr(apic);
1156 	hrtimer_cancel(&apic->lapic_timer.timer);
1157 	update_divide_count(apic);
1158 	start_apic_timer(apic);
1159 }
1160 
1161 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1162 {
1163 	struct kvm_lapic *apic = vcpu->arch.apic;
1164 	struct hrtimer *timer;
1165 
1166 	if (!apic)
1167 		return;
1168 
1169 	timer = &apic->lapic_timer.timer;
1170 	if (hrtimer_cancel(timer))
1171 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1172 }
1173 
1174 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1175 {
1176 	u32 data;
1177 	void *vapic;
1178 
1179 	if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1180 		return;
1181 
1182 	vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1183 	data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr));
1184 	kunmap_atomic(vapic, KM_USER0);
1185 
1186 	apic_set_tpr(vcpu->arch.apic, data & 0xff);
1187 }
1188 
1189 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1190 {
1191 	u32 data, tpr;
1192 	int max_irr, max_isr;
1193 	struct kvm_lapic *apic;
1194 	void *vapic;
1195 
1196 	if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1197 		return;
1198 
1199 	apic = vcpu->arch.apic;
1200 	tpr = apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1201 	max_irr = apic_find_highest_irr(apic);
1202 	if (max_irr < 0)
1203 		max_irr = 0;
1204 	max_isr = apic_find_highest_isr(apic);
1205 	if (max_isr < 0)
1206 		max_isr = 0;
1207 	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1208 
1209 	vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1210 	*(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data;
1211 	kunmap_atomic(vapic, KM_USER0);
1212 }
1213 
1214 void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1215 {
1216 	if (!irqchip_in_kernel(vcpu->kvm))
1217 		return;
1218 
1219 	vcpu->arch.apic->vapic_addr = vapic_addr;
1220 }
1221 
1222 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1223 {
1224 	struct kvm_lapic *apic = vcpu->arch.apic;
1225 	u32 reg = (msr - APIC_BASE_MSR) << 4;
1226 
1227 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1228 		return 1;
1229 
1230 	/* if this is ICR write vector before command */
1231 	if (msr == 0x830)
1232 		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1233 	return apic_reg_write(apic, reg, (u32)data);
1234 }
1235 
1236 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1237 {
1238 	struct kvm_lapic *apic = vcpu->arch.apic;
1239 	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1240 
1241 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1242 		return 1;
1243 
1244 	if (apic_reg_read(apic, reg, 4, &low))
1245 		return 1;
1246 	if (msr == 0x830)
1247 		apic_reg_read(apic, APIC_ICR2, 4, &high);
1248 
1249 	*data = (((u64)high) << 32) | low;
1250 
1251 	return 0;
1252 }
1253