xref: /openbmc/linux/arch/x86/kvm/lapic.c (revision b6dcefde)
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 #include "trace.h"
38 #include "x86.h"
39 
40 #ifndef CONFIG_X86_64
41 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
42 #else
43 #define mod_64(x, y) ((x) % (y))
44 #endif
45 
46 #define PRId64 "d"
47 #define PRIx64 "llx"
48 #define PRIu64 "u"
49 #define PRIo64 "o"
50 
51 #define APIC_BUS_CYCLE_NS 1
52 
53 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
54 #define apic_debug(fmt, arg...)
55 
56 #define APIC_LVT_NUM			6
57 /* 14 is the version for Xeon and Pentium 8.4.8*/
58 #define APIC_VERSION			(0x14UL | ((APIC_LVT_NUM - 1) << 16))
59 #define LAPIC_MMIO_LENGTH		(1 << 12)
60 /* followed define is not in apicdef.h */
61 #define APIC_SHORT_MASK			0xc0000
62 #define APIC_DEST_NOSHORT		0x0
63 #define APIC_DEST_MASK			0x800
64 #define MAX_APIC_VECTOR			256
65 
66 #define VEC_POS(v) ((v) & (32 - 1))
67 #define REG_POS(v) (((v) >> 5) << 4)
68 
69 static inline u32 apic_get_reg(struct kvm_lapic *apic, int reg_off)
70 {
71 	return *((u32 *) (apic->regs + reg_off));
72 }
73 
74 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
75 {
76 	*((u32 *) (apic->regs + reg_off)) = val;
77 }
78 
79 static inline int apic_test_and_set_vector(int vec, void *bitmap)
80 {
81 	return test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
82 }
83 
84 static inline int apic_test_and_clear_vector(int vec, void *bitmap)
85 {
86 	return test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
87 }
88 
89 static inline void apic_set_vector(int vec, void *bitmap)
90 {
91 	set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
92 }
93 
94 static inline void apic_clear_vector(int vec, void *bitmap)
95 {
96 	clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
97 }
98 
99 static inline int apic_hw_enabled(struct kvm_lapic *apic)
100 {
101 	return (apic)->vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE;
102 }
103 
104 static inline int  apic_sw_enabled(struct kvm_lapic *apic)
105 {
106 	return apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED;
107 }
108 
109 static inline int apic_enabled(struct kvm_lapic *apic)
110 {
111 	return apic_sw_enabled(apic) &&	apic_hw_enabled(apic);
112 }
113 
114 #define LVT_MASK	\
115 	(APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
116 
117 #define LINT_MASK	\
118 	(LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
119 	 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
120 
121 static inline int kvm_apic_id(struct kvm_lapic *apic)
122 {
123 	return (apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
124 }
125 
126 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
127 {
128 	return !(apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
129 }
130 
131 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
132 {
133 	return apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
134 }
135 
136 static inline int apic_lvtt_period(struct kvm_lapic *apic)
137 {
138 	return apic_get_reg(apic, APIC_LVTT) & APIC_LVT_TIMER_PERIODIC;
139 }
140 
141 static inline int apic_lvt_nmi_mode(u32 lvt_val)
142 {
143 	return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
144 }
145 
146 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
147 {
148 	struct kvm_lapic *apic = vcpu->arch.apic;
149 	struct kvm_cpuid_entry2 *feat;
150 	u32 v = APIC_VERSION;
151 
152 	if (!irqchip_in_kernel(vcpu->kvm))
153 		return;
154 
155 	feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
156 	if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
157 		v |= APIC_LVR_DIRECTED_EOI;
158 	apic_set_reg(apic, APIC_LVR, v);
159 }
160 
161 static inline int apic_x2apic_mode(struct kvm_lapic *apic)
162 {
163 	return apic->vcpu->arch.apic_base & X2APIC_ENABLE;
164 }
165 
166 static unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
167 	LVT_MASK | APIC_LVT_TIMER_PERIODIC,	/* LVTT */
168 	LVT_MASK | APIC_MODE_MASK,	/* LVTTHMR */
169 	LVT_MASK | APIC_MODE_MASK,	/* LVTPC */
170 	LINT_MASK, LINT_MASK,	/* LVT0-1 */
171 	LVT_MASK		/* LVTERR */
172 };
173 
174 static int find_highest_vector(void *bitmap)
175 {
176 	u32 *word = bitmap;
177 	int word_offset = MAX_APIC_VECTOR >> 5;
178 
179 	while ((word_offset != 0) && (word[(--word_offset) << 2] == 0))
180 		continue;
181 
182 	if (likely(!word_offset && !word[0]))
183 		return -1;
184 	else
185 		return fls(word[word_offset << 2]) - 1 + (word_offset << 5);
186 }
187 
188 static inline int apic_test_and_set_irr(int vec, struct kvm_lapic *apic)
189 {
190 	apic->irr_pending = true;
191 	return apic_test_and_set_vector(vec, apic->regs + APIC_IRR);
192 }
193 
194 static inline int apic_search_irr(struct kvm_lapic *apic)
195 {
196 	return find_highest_vector(apic->regs + APIC_IRR);
197 }
198 
199 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
200 {
201 	int result;
202 
203 	if (!apic->irr_pending)
204 		return -1;
205 
206 	result = apic_search_irr(apic);
207 	ASSERT(result == -1 || result >= 16);
208 
209 	return result;
210 }
211 
212 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
213 {
214 	apic->irr_pending = false;
215 	apic_clear_vector(vec, apic->regs + APIC_IRR);
216 	if (apic_search_irr(apic) != -1)
217 		apic->irr_pending = true;
218 }
219 
220 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
221 {
222 	struct kvm_lapic *apic = vcpu->arch.apic;
223 	int highest_irr;
224 
225 	/* This may race with setting of irr in __apic_accept_irq() and
226 	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
227 	 * will cause vmexit immediately and the value will be recalculated
228 	 * on the next vmentry.
229 	 */
230 	if (!apic)
231 		return 0;
232 	highest_irr = apic_find_highest_irr(apic);
233 
234 	return highest_irr;
235 }
236 
237 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
238 			     int vector, int level, int trig_mode);
239 
240 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq)
241 {
242 	struct kvm_lapic *apic = vcpu->arch.apic;
243 
244 	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
245 			irq->level, irq->trig_mode);
246 }
247 
248 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
249 {
250 	int result;
251 
252 	result = find_highest_vector(apic->regs + APIC_ISR);
253 	ASSERT(result == -1 || result >= 16);
254 
255 	return result;
256 }
257 
258 static void apic_update_ppr(struct kvm_lapic *apic)
259 {
260 	u32 tpr, isrv, ppr;
261 	int isr;
262 
263 	tpr = apic_get_reg(apic, APIC_TASKPRI);
264 	isr = apic_find_highest_isr(apic);
265 	isrv = (isr != -1) ? isr : 0;
266 
267 	if ((tpr & 0xf0) >= (isrv & 0xf0))
268 		ppr = tpr & 0xff;
269 	else
270 		ppr = isrv & 0xf0;
271 
272 	apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
273 		   apic, ppr, isr, isrv);
274 
275 	apic_set_reg(apic, APIC_PROCPRI, ppr);
276 }
277 
278 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
279 {
280 	apic_set_reg(apic, APIC_TASKPRI, tpr);
281 	apic_update_ppr(apic);
282 }
283 
284 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
285 {
286 	return dest == 0xff || kvm_apic_id(apic) == dest;
287 }
288 
289 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
290 {
291 	int result = 0;
292 	u32 logical_id;
293 
294 	if (apic_x2apic_mode(apic)) {
295 		logical_id = apic_get_reg(apic, APIC_LDR);
296 		return logical_id & mda;
297 	}
298 
299 	logical_id = GET_APIC_LOGICAL_ID(apic_get_reg(apic, APIC_LDR));
300 
301 	switch (apic_get_reg(apic, APIC_DFR)) {
302 	case APIC_DFR_FLAT:
303 		if (logical_id & mda)
304 			result = 1;
305 		break;
306 	case APIC_DFR_CLUSTER:
307 		if (((logical_id >> 4) == (mda >> 0x4))
308 		    && (logical_id & mda & 0xf))
309 			result = 1;
310 		break;
311 	default:
312 		printk(KERN_WARNING "Bad DFR vcpu %d: %08x\n",
313 		       apic->vcpu->vcpu_id, apic_get_reg(apic, APIC_DFR));
314 		break;
315 	}
316 
317 	return result;
318 }
319 
320 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
321 			   int short_hand, int dest, int dest_mode)
322 {
323 	int result = 0;
324 	struct kvm_lapic *target = vcpu->arch.apic;
325 
326 	apic_debug("target %p, source %p, dest 0x%x, "
327 		   "dest_mode 0x%x, short_hand 0x%x\n",
328 		   target, source, dest, dest_mode, short_hand);
329 
330 	ASSERT(!target);
331 	switch (short_hand) {
332 	case APIC_DEST_NOSHORT:
333 		if (dest_mode == 0)
334 			/* Physical mode. */
335 			result = kvm_apic_match_physical_addr(target, dest);
336 		else
337 			/* Logical mode. */
338 			result = kvm_apic_match_logical_addr(target, dest);
339 		break;
340 	case APIC_DEST_SELF:
341 		result = (target == source);
342 		break;
343 	case APIC_DEST_ALLINC:
344 		result = 1;
345 		break;
346 	case APIC_DEST_ALLBUT:
347 		result = (target != source);
348 		break;
349 	default:
350 		printk(KERN_WARNING "Bad dest shorthand value %x\n",
351 		       short_hand);
352 		break;
353 	}
354 
355 	return result;
356 }
357 
358 /*
359  * Add a pending IRQ into lapic.
360  * Return 1 if successfully added and 0 if discarded.
361  */
362 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
363 			     int vector, int level, int trig_mode)
364 {
365 	int result = 0;
366 	struct kvm_vcpu *vcpu = apic->vcpu;
367 
368 	switch (delivery_mode) {
369 	case APIC_DM_LOWEST:
370 		vcpu->arch.apic_arb_prio++;
371 	case APIC_DM_FIXED:
372 		/* FIXME add logic for vcpu on reset */
373 		if (unlikely(!apic_enabled(apic)))
374 			break;
375 
376 		if (trig_mode) {
377 			apic_debug("level trig mode for vector %d", vector);
378 			apic_set_vector(vector, apic->regs + APIC_TMR);
379 		} else
380 			apic_clear_vector(vector, apic->regs + APIC_TMR);
381 
382 		result = !apic_test_and_set_irr(vector, apic);
383 		trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
384 					  trig_mode, vector, !result);
385 		if (!result) {
386 			if (trig_mode)
387 				apic_debug("level trig mode repeatedly for "
388 						"vector %d", vector);
389 			break;
390 		}
391 
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 		kvm_ioapic_update_eoi(apic->vcpu->kvm, vector, trigger_mode);
476 }
477 
478 static void apic_send_ipi(struct kvm_lapic *apic)
479 {
480 	u32 icr_low = apic_get_reg(apic, APIC_ICR);
481 	u32 icr_high = apic_get_reg(apic, APIC_ICR2);
482 	struct kvm_lapic_irq irq;
483 
484 	irq.vector = icr_low & APIC_VECTOR_MASK;
485 	irq.delivery_mode = icr_low & APIC_MODE_MASK;
486 	irq.dest_mode = icr_low & APIC_DEST_MASK;
487 	irq.level = icr_low & APIC_INT_ASSERT;
488 	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
489 	irq.shorthand = icr_low & APIC_SHORT_MASK;
490 	if (apic_x2apic_mode(apic))
491 		irq.dest_id = icr_high;
492 	else
493 		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
494 
495 	trace_kvm_apic_ipi(icr_low, irq.dest_id);
496 
497 	apic_debug("icr_high 0x%x, icr_low 0x%x, "
498 		   "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
499 		   "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
500 		   icr_high, icr_low, irq.shorthand, irq.dest_id,
501 		   irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
502 		   irq.vector);
503 
504 	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq);
505 }
506 
507 static u32 apic_get_tmcct(struct kvm_lapic *apic)
508 {
509 	ktime_t remaining;
510 	s64 ns;
511 	u32 tmcct;
512 
513 	ASSERT(apic != NULL);
514 
515 	/* if initial count is 0, current count should also be 0 */
516 	if (apic_get_reg(apic, APIC_TMICT) == 0)
517 		return 0;
518 
519 	remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
520 	if (ktime_to_ns(remaining) < 0)
521 		remaining = ktime_set(0, 0);
522 
523 	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
524 	tmcct = div64_u64(ns,
525 			 (APIC_BUS_CYCLE_NS * apic->divide_count));
526 
527 	return tmcct;
528 }
529 
530 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
531 {
532 	struct kvm_vcpu *vcpu = apic->vcpu;
533 	struct kvm_run *run = vcpu->run;
534 
535 	set_bit(KVM_REQ_REPORT_TPR_ACCESS, &vcpu->requests);
536 	run->tpr_access.rip = kvm_rip_read(vcpu);
537 	run->tpr_access.is_write = write;
538 }
539 
540 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
541 {
542 	if (apic->vcpu->arch.tpr_access_reporting)
543 		__report_tpr_access(apic, write);
544 }
545 
546 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
547 {
548 	u32 val = 0;
549 
550 	if (offset >= LAPIC_MMIO_LENGTH)
551 		return 0;
552 
553 	switch (offset) {
554 	case APIC_ID:
555 		if (apic_x2apic_mode(apic))
556 			val = kvm_apic_id(apic);
557 		else
558 			val = kvm_apic_id(apic) << 24;
559 		break;
560 	case APIC_ARBPRI:
561 		printk(KERN_WARNING "Access APIC ARBPRI register "
562 		       "which is for P6\n");
563 		break;
564 
565 	case APIC_TMCCT:	/* Timer CCR */
566 		val = apic_get_tmcct(apic);
567 		break;
568 
569 	case APIC_TASKPRI:
570 		report_tpr_access(apic, false);
571 		/* fall thru */
572 	default:
573 		apic_update_ppr(apic);
574 		val = apic_get_reg(apic, offset);
575 		break;
576 	}
577 
578 	return val;
579 }
580 
581 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
582 {
583 	return container_of(dev, struct kvm_lapic, dev);
584 }
585 
586 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
587 		void *data)
588 {
589 	unsigned char alignment = offset & 0xf;
590 	u32 result;
591 	/* this bitmask has a bit cleared for each reserver register */
592 	static const u64 rmask = 0x43ff01ffffffe70cULL;
593 
594 	if ((alignment + len) > 4) {
595 		apic_debug("KVM_APIC_READ: alignment error %x %d\n",
596 			   offset, len);
597 		return 1;
598 	}
599 
600 	if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
601 		apic_debug("KVM_APIC_READ: read reserved register %x\n",
602 			   offset);
603 		return 1;
604 	}
605 
606 	result = __apic_read(apic, offset & ~0xf);
607 
608 	trace_kvm_apic_read(offset, result);
609 
610 	switch (len) {
611 	case 1:
612 	case 2:
613 	case 4:
614 		memcpy(data, (char *)&result + alignment, len);
615 		break;
616 	default:
617 		printk(KERN_ERR "Local APIC read with len = %x, "
618 		       "should be 1,2, or 4 instead\n", len);
619 		break;
620 	}
621 	return 0;
622 }
623 
624 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
625 {
626 	return apic_hw_enabled(apic) &&
627 	    addr >= apic->base_address &&
628 	    addr < apic->base_address + LAPIC_MMIO_LENGTH;
629 }
630 
631 static int apic_mmio_read(struct kvm_io_device *this,
632 			   gpa_t address, int len, void *data)
633 {
634 	struct kvm_lapic *apic = to_lapic(this);
635 	u32 offset = address - apic->base_address;
636 
637 	if (!apic_mmio_in_range(apic, address))
638 		return -EOPNOTSUPP;
639 
640 	apic_reg_read(apic, offset, len, data);
641 
642 	return 0;
643 }
644 
645 static void update_divide_count(struct kvm_lapic *apic)
646 {
647 	u32 tmp1, tmp2, tdcr;
648 
649 	tdcr = apic_get_reg(apic, APIC_TDCR);
650 	tmp1 = tdcr & 0xf;
651 	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
652 	apic->divide_count = 0x1 << (tmp2 & 0x7);
653 
654 	apic_debug("timer divide count is 0x%x\n",
655 				   apic->divide_count);
656 }
657 
658 static void start_apic_timer(struct kvm_lapic *apic)
659 {
660 	ktime_t now = apic->lapic_timer.timer.base->get_time();
661 
662 	apic->lapic_timer.period = (u64)apic_get_reg(apic, APIC_TMICT) *
663 		    APIC_BUS_CYCLE_NS * apic->divide_count;
664 	atomic_set(&apic->lapic_timer.pending, 0);
665 
666 	if (!apic->lapic_timer.period)
667 		return;
668 	/*
669 	 * Do not allow the guest to program periodic timers with small
670 	 * interval, since the hrtimers are not throttled by the host
671 	 * scheduler.
672 	 */
673 	if (apic_lvtt_period(apic)) {
674 		if (apic->lapic_timer.period < NSEC_PER_MSEC/2)
675 			apic->lapic_timer.period = NSEC_PER_MSEC/2;
676 	}
677 
678 	hrtimer_start(&apic->lapic_timer.timer,
679 		      ktime_add_ns(now, apic->lapic_timer.period),
680 		      HRTIMER_MODE_ABS);
681 
682 	apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
683 			   PRIx64 ", "
684 			   "timer initial count 0x%x, period %lldns, "
685 			   "expire @ 0x%016" PRIx64 ".\n", __func__,
686 			   APIC_BUS_CYCLE_NS, ktime_to_ns(now),
687 			   apic_get_reg(apic, APIC_TMICT),
688 			   apic->lapic_timer.period,
689 			   ktime_to_ns(ktime_add_ns(now,
690 					apic->lapic_timer.period)));
691 }
692 
693 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
694 {
695 	int nmi_wd_enabled = apic_lvt_nmi_mode(apic_get_reg(apic, APIC_LVT0));
696 
697 	if (apic_lvt_nmi_mode(lvt0_val)) {
698 		if (!nmi_wd_enabled) {
699 			apic_debug("Receive NMI setting on APIC_LVT0 "
700 				   "for cpu %d\n", apic->vcpu->vcpu_id);
701 			apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
702 		}
703 	} else if (nmi_wd_enabled)
704 		apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
705 }
706 
707 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
708 {
709 	int ret = 0;
710 
711 	trace_kvm_apic_write(reg, val);
712 
713 	switch (reg) {
714 	case APIC_ID:		/* Local APIC ID */
715 		if (!apic_x2apic_mode(apic))
716 			apic_set_reg(apic, APIC_ID, val);
717 		else
718 			ret = 1;
719 		break;
720 
721 	case APIC_TASKPRI:
722 		report_tpr_access(apic, true);
723 		apic_set_tpr(apic, val & 0xff);
724 		break;
725 
726 	case APIC_EOI:
727 		apic_set_eoi(apic);
728 		break;
729 
730 	case APIC_LDR:
731 		if (!apic_x2apic_mode(apic))
732 			apic_set_reg(apic, APIC_LDR, val & APIC_LDR_MASK);
733 		else
734 			ret = 1;
735 		break;
736 
737 	case APIC_DFR:
738 		if (!apic_x2apic_mode(apic))
739 			apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
740 		else
741 			ret = 1;
742 		break;
743 
744 	case APIC_SPIV: {
745 		u32 mask = 0x3ff;
746 		if (apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
747 			mask |= APIC_SPIV_DIRECTED_EOI;
748 		apic_set_reg(apic, APIC_SPIV, val & mask);
749 		if (!(val & APIC_SPIV_APIC_ENABLED)) {
750 			int i;
751 			u32 lvt_val;
752 
753 			for (i = 0; i < APIC_LVT_NUM; i++) {
754 				lvt_val = apic_get_reg(apic,
755 						       APIC_LVTT + 0x10 * i);
756 				apic_set_reg(apic, APIC_LVTT + 0x10 * i,
757 					     lvt_val | APIC_LVT_MASKED);
758 			}
759 			atomic_set(&apic->lapic_timer.pending, 0);
760 
761 		}
762 		break;
763 	}
764 	case APIC_ICR:
765 		/* No delay here, so we always clear the pending bit */
766 		apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
767 		apic_send_ipi(apic);
768 		break;
769 
770 	case APIC_ICR2:
771 		if (!apic_x2apic_mode(apic))
772 			val &= 0xff000000;
773 		apic_set_reg(apic, APIC_ICR2, val);
774 		break;
775 
776 	case APIC_LVT0:
777 		apic_manage_nmi_watchdog(apic, val);
778 	case APIC_LVTT:
779 	case APIC_LVTTHMR:
780 	case APIC_LVTPC:
781 	case APIC_LVT1:
782 	case APIC_LVTERR:
783 		/* TODO: Check vector */
784 		if (!apic_sw_enabled(apic))
785 			val |= APIC_LVT_MASKED;
786 
787 		val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
788 		apic_set_reg(apic, reg, val);
789 
790 		break;
791 
792 	case APIC_TMICT:
793 		hrtimer_cancel(&apic->lapic_timer.timer);
794 		apic_set_reg(apic, APIC_TMICT, val);
795 		start_apic_timer(apic);
796 		break;
797 
798 	case APIC_TDCR:
799 		if (val & 4)
800 			printk(KERN_ERR "KVM_WRITE:TDCR %x\n", val);
801 		apic_set_reg(apic, APIC_TDCR, val);
802 		update_divide_count(apic);
803 		break;
804 
805 	case APIC_ESR:
806 		if (apic_x2apic_mode(apic) && val != 0) {
807 			printk(KERN_ERR "KVM_WRITE:ESR not zero %x\n", val);
808 			ret = 1;
809 		}
810 		break;
811 
812 	case APIC_SELF_IPI:
813 		if (apic_x2apic_mode(apic)) {
814 			apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
815 		} else
816 			ret = 1;
817 		break;
818 	default:
819 		ret = 1;
820 		break;
821 	}
822 	if (ret)
823 		apic_debug("Local APIC Write to read-only register %x\n", reg);
824 	return ret;
825 }
826 
827 static int apic_mmio_write(struct kvm_io_device *this,
828 			    gpa_t address, int len, const void *data)
829 {
830 	struct kvm_lapic *apic = to_lapic(this);
831 	unsigned int offset = address - apic->base_address;
832 	u32 val;
833 
834 	if (!apic_mmio_in_range(apic, address))
835 		return -EOPNOTSUPP;
836 
837 	/*
838 	 * APIC register must be aligned on 128-bits boundary.
839 	 * 32/64/128 bits registers must be accessed thru 32 bits.
840 	 * Refer SDM 8.4.1
841 	 */
842 	if (len != 4 || (offset & 0xf)) {
843 		/* Don't shout loud, $infamous_os would cause only noise. */
844 		apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
845 		return 0;
846 	}
847 
848 	val = *(u32*)data;
849 
850 	/* too common printing */
851 	if (offset != APIC_EOI)
852 		apic_debug("%s: offset 0x%x with length 0x%x, and value is "
853 			   "0x%x\n", __func__, offset, len, val);
854 
855 	apic_reg_write(apic, offset & 0xff0, val);
856 
857 	return 0;
858 }
859 
860 void kvm_free_lapic(struct kvm_vcpu *vcpu)
861 {
862 	if (!vcpu->arch.apic)
863 		return;
864 
865 	hrtimer_cancel(&vcpu->arch.apic->lapic_timer.timer);
866 
867 	if (vcpu->arch.apic->regs_page)
868 		__free_page(vcpu->arch.apic->regs_page);
869 
870 	kfree(vcpu->arch.apic);
871 }
872 
873 /*
874  *----------------------------------------------------------------------
875  * LAPIC interface
876  *----------------------------------------------------------------------
877  */
878 
879 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
880 {
881 	struct kvm_lapic *apic = vcpu->arch.apic;
882 
883 	if (!apic)
884 		return;
885 	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
886 		     | (apic_get_reg(apic, APIC_TASKPRI) & 4));
887 }
888 
889 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
890 {
891 	struct kvm_lapic *apic = vcpu->arch.apic;
892 	u64 tpr;
893 
894 	if (!apic)
895 		return 0;
896 	tpr = (u64) apic_get_reg(apic, APIC_TASKPRI);
897 
898 	return (tpr & 0xf0) >> 4;
899 }
900 
901 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
902 {
903 	struct kvm_lapic *apic = vcpu->arch.apic;
904 
905 	if (!apic) {
906 		value |= MSR_IA32_APICBASE_BSP;
907 		vcpu->arch.apic_base = value;
908 		return;
909 	}
910 
911 	if (!kvm_vcpu_is_bsp(apic->vcpu))
912 		value &= ~MSR_IA32_APICBASE_BSP;
913 
914 	vcpu->arch.apic_base = value;
915 	if (apic_x2apic_mode(apic)) {
916 		u32 id = kvm_apic_id(apic);
917 		u32 ldr = ((id & ~0xf) << 16) | (1 << (id & 0xf));
918 		apic_set_reg(apic, APIC_LDR, ldr);
919 	}
920 	apic->base_address = apic->vcpu->arch.apic_base &
921 			     MSR_IA32_APICBASE_BASE;
922 
923 	/* with FSB delivery interrupt, we can restart APIC functionality */
924 	apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
925 		   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
926 
927 }
928 
929 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
930 {
931 	struct kvm_lapic *apic;
932 	int i;
933 
934 	apic_debug("%s\n", __func__);
935 
936 	ASSERT(vcpu);
937 	apic = vcpu->arch.apic;
938 	ASSERT(apic != NULL);
939 
940 	/* Stop the timer in case it's a reset to an active apic */
941 	hrtimer_cancel(&apic->lapic_timer.timer);
942 
943 	apic_set_reg(apic, APIC_ID, vcpu->vcpu_id << 24);
944 	kvm_apic_set_version(apic->vcpu);
945 
946 	for (i = 0; i < APIC_LVT_NUM; i++)
947 		apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
948 	apic_set_reg(apic, APIC_LVT0,
949 		     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
950 
951 	apic_set_reg(apic, APIC_DFR, 0xffffffffU);
952 	apic_set_reg(apic, APIC_SPIV, 0xff);
953 	apic_set_reg(apic, APIC_TASKPRI, 0);
954 	apic_set_reg(apic, APIC_LDR, 0);
955 	apic_set_reg(apic, APIC_ESR, 0);
956 	apic_set_reg(apic, APIC_ICR, 0);
957 	apic_set_reg(apic, APIC_ICR2, 0);
958 	apic_set_reg(apic, APIC_TDCR, 0);
959 	apic_set_reg(apic, APIC_TMICT, 0);
960 	for (i = 0; i < 8; i++) {
961 		apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
962 		apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
963 		apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
964 	}
965 	apic->irr_pending = false;
966 	update_divide_count(apic);
967 	atomic_set(&apic->lapic_timer.pending, 0);
968 	if (kvm_vcpu_is_bsp(vcpu))
969 		vcpu->arch.apic_base |= MSR_IA32_APICBASE_BSP;
970 	apic_update_ppr(apic);
971 
972 	vcpu->arch.apic_arb_prio = 0;
973 
974 	apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
975 		   "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
976 		   vcpu, kvm_apic_id(apic),
977 		   vcpu->arch.apic_base, apic->base_address);
978 }
979 
980 bool kvm_apic_present(struct kvm_vcpu *vcpu)
981 {
982 	return vcpu->arch.apic && apic_hw_enabled(vcpu->arch.apic);
983 }
984 
985 int kvm_lapic_enabled(struct kvm_vcpu *vcpu)
986 {
987 	return kvm_apic_present(vcpu) && apic_sw_enabled(vcpu->arch.apic);
988 }
989 
990 /*
991  *----------------------------------------------------------------------
992  * timer interface
993  *----------------------------------------------------------------------
994  */
995 
996 static bool lapic_is_periodic(struct kvm_timer *ktimer)
997 {
998 	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic,
999 					      lapic_timer);
1000 	return apic_lvtt_period(apic);
1001 }
1002 
1003 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1004 {
1005 	struct kvm_lapic *lapic = vcpu->arch.apic;
1006 
1007 	if (lapic && apic_enabled(lapic) && apic_lvt_enabled(lapic, APIC_LVTT))
1008 		return atomic_read(&lapic->lapic_timer.pending);
1009 
1010 	return 0;
1011 }
1012 
1013 static int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1014 {
1015 	u32 reg = apic_get_reg(apic, lvt_type);
1016 	int vector, mode, trig_mode;
1017 
1018 	if (apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1019 		vector = reg & APIC_VECTOR_MASK;
1020 		mode = reg & APIC_MODE_MASK;
1021 		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1022 		return __apic_accept_irq(apic, mode, vector, 1, trig_mode);
1023 	}
1024 	return 0;
1025 }
1026 
1027 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1028 {
1029 	struct kvm_lapic *apic = vcpu->arch.apic;
1030 
1031 	if (apic)
1032 		kvm_apic_local_deliver(apic, APIC_LVT0);
1033 }
1034 
1035 static struct kvm_timer_ops lapic_timer_ops = {
1036 	.is_periodic = lapic_is_periodic,
1037 };
1038 
1039 static const struct kvm_io_device_ops apic_mmio_ops = {
1040 	.read     = apic_mmio_read,
1041 	.write    = apic_mmio_write,
1042 };
1043 
1044 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1045 {
1046 	struct kvm_lapic *apic;
1047 
1048 	ASSERT(vcpu != NULL);
1049 	apic_debug("apic_init %d\n", vcpu->vcpu_id);
1050 
1051 	apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1052 	if (!apic)
1053 		goto nomem;
1054 
1055 	vcpu->arch.apic = apic;
1056 
1057 	apic->regs_page = alloc_page(GFP_KERNEL);
1058 	if (apic->regs_page == NULL) {
1059 		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1060 		       vcpu->vcpu_id);
1061 		goto nomem_free_apic;
1062 	}
1063 	apic->regs = page_address(apic->regs_page);
1064 	memset(apic->regs, 0, PAGE_SIZE);
1065 	apic->vcpu = vcpu;
1066 
1067 	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1068 		     HRTIMER_MODE_ABS);
1069 	apic->lapic_timer.timer.function = kvm_timer_fn;
1070 	apic->lapic_timer.t_ops = &lapic_timer_ops;
1071 	apic->lapic_timer.kvm = vcpu->kvm;
1072 	apic->lapic_timer.vcpu = vcpu;
1073 
1074 	apic->base_address = APIC_DEFAULT_PHYS_BASE;
1075 	vcpu->arch.apic_base = APIC_DEFAULT_PHYS_BASE;
1076 
1077 	kvm_lapic_reset(vcpu);
1078 	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1079 
1080 	return 0;
1081 nomem_free_apic:
1082 	kfree(apic);
1083 nomem:
1084 	return -ENOMEM;
1085 }
1086 
1087 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1088 {
1089 	struct kvm_lapic *apic = vcpu->arch.apic;
1090 	int highest_irr;
1091 
1092 	if (!apic || !apic_enabled(apic))
1093 		return -1;
1094 
1095 	apic_update_ppr(apic);
1096 	highest_irr = apic_find_highest_irr(apic);
1097 	if ((highest_irr == -1) ||
1098 	    ((highest_irr & 0xF0) <= apic_get_reg(apic, APIC_PROCPRI)))
1099 		return -1;
1100 	return highest_irr;
1101 }
1102 
1103 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1104 {
1105 	u32 lvt0 = apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1106 	int r = 0;
1107 
1108 	if (kvm_vcpu_is_bsp(vcpu)) {
1109 		if (!apic_hw_enabled(vcpu->arch.apic))
1110 			r = 1;
1111 		if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1112 		    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1113 			r = 1;
1114 	}
1115 	return r;
1116 }
1117 
1118 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1119 {
1120 	struct kvm_lapic *apic = vcpu->arch.apic;
1121 
1122 	if (apic && atomic_read(&apic->lapic_timer.pending) > 0) {
1123 		if (kvm_apic_local_deliver(apic, APIC_LVTT))
1124 			atomic_dec(&apic->lapic_timer.pending);
1125 	}
1126 }
1127 
1128 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1129 {
1130 	int vector = kvm_apic_has_interrupt(vcpu);
1131 	struct kvm_lapic *apic = vcpu->arch.apic;
1132 
1133 	if (vector == -1)
1134 		return -1;
1135 
1136 	apic_set_vector(vector, apic->regs + APIC_ISR);
1137 	apic_update_ppr(apic);
1138 	apic_clear_irr(vector, apic);
1139 	return vector;
1140 }
1141 
1142 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu)
1143 {
1144 	struct kvm_lapic *apic = vcpu->arch.apic;
1145 
1146 	apic->base_address = vcpu->arch.apic_base &
1147 			     MSR_IA32_APICBASE_BASE;
1148 	kvm_apic_set_version(vcpu);
1149 
1150 	apic_update_ppr(apic);
1151 	hrtimer_cancel(&apic->lapic_timer.timer);
1152 	update_divide_count(apic);
1153 	start_apic_timer(apic);
1154 	apic->irr_pending = true;
1155 }
1156 
1157 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1158 {
1159 	struct kvm_lapic *apic = vcpu->arch.apic;
1160 	struct hrtimer *timer;
1161 
1162 	if (!apic)
1163 		return;
1164 
1165 	timer = &apic->lapic_timer.timer;
1166 	if (hrtimer_cancel(timer))
1167 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1168 }
1169 
1170 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1171 {
1172 	u32 data;
1173 	void *vapic;
1174 
1175 	if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1176 		return;
1177 
1178 	vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1179 	data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr));
1180 	kunmap_atomic(vapic, KM_USER0);
1181 
1182 	apic_set_tpr(vcpu->arch.apic, data & 0xff);
1183 }
1184 
1185 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1186 {
1187 	u32 data, tpr;
1188 	int max_irr, max_isr;
1189 	struct kvm_lapic *apic;
1190 	void *vapic;
1191 
1192 	if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1193 		return;
1194 
1195 	apic = vcpu->arch.apic;
1196 	tpr = apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1197 	max_irr = apic_find_highest_irr(apic);
1198 	if (max_irr < 0)
1199 		max_irr = 0;
1200 	max_isr = apic_find_highest_isr(apic);
1201 	if (max_isr < 0)
1202 		max_isr = 0;
1203 	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1204 
1205 	vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1206 	*(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data;
1207 	kunmap_atomic(vapic, KM_USER0);
1208 }
1209 
1210 void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1211 {
1212 	if (!irqchip_in_kernel(vcpu->kvm))
1213 		return;
1214 
1215 	vcpu->arch.apic->vapic_addr = vapic_addr;
1216 }
1217 
1218 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1219 {
1220 	struct kvm_lapic *apic = vcpu->arch.apic;
1221 	u32 reg = (msr - APIC_BASE_MSR) << 4;
1222 
1223 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1224 		return 1;
1225 
1226 	/* if this is ICR write vector before command */
1227 	if (msr == 0x830)
1228 		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1229 	return apic_reg_write(apic, reg, (u32)data);
1230 }
1231 
1232 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1233 {
1234 	struct kvm_lapic *apic = vcpu->arch.apic;
1235 	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1236 
1237 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1238 		return 1;
1239 
1240 	if (apic_reg_read(apic, reg, 4, &low))
1241 		return 1;
1242 	if (msr == 0x830)
1243 		apic_reg_read(apic, APIC_ICR2, 4, &high);
1244 
1245 	*data = (((u64)high) << 32) | low;
1246 
1247 	return 0;
1248 }
1249