xref: /openbmc/linux/arch/x86/kvm/lapic.c (revision a09d2831)
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 		result = !apic_test_and_set_irr(vector, apic);
377 		trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
378 					  trig_mode, vector, !result);
379 		if (!result) {
380 			if (trig_mode)
381 				apic_debug("level trig mode repeatedly for "
382 						"vector %d", vector);
383 			break;
384 		}
385 
386 		if (trig_mode) {
387 			apic_debug("level trig mode for vector %d", vector);
388 			apic_set_vector(vector, apic->regs + APIC_TMR);
389 		} else
390 			apic_clear_vector(vector, apic->regs + APIC_TMR);
391 		kvm_vcpu_kick(vcpu);
392 		break;
393 
394 	case APIC_DM_REMRD:
395 		printk(KERN_DEBUG "Ignoring delivery mode 3\n");
396 		break;
397 
398 	case APIC_DM_SMI:
399 		printk(KERN_DEBUG "Ignoring guest SMI\n");
400 		break;
401 
402 	case APIC_DM_NMI:
403 		result = 1;
404 		kvm_inject_nmi(vcpu);
405 		kvm_vcpu_kick(vcpu);
406 		break;
407 
408 	case APIC_DM_INIT:
409 		if (level) {
410 			result = 1;
411 			if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
412 				printk(KERN_DEBUG
413 				       "INIT on a runnable vcpu %d\n",
414 				       vcpu->vcpu_id);
415 			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
416 			kvm_vcpu_kick(vcpu);
417 		} else {
418 			apic_debug("Ignoring de-assert INIT to vcpu %d\n",
419 				   vcpu->vcpu_id);
420 		}
421 		break;
422 
423 	case APIC_DM_STARTUP:
424 		apic_debug("SIPI to vcpu %d vector 0x%02x\n",
425 			   vcpu->vcpu_id, vector);
426 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
427 			result = 1;
428 			vcpu->arch.sipi_vector = vector;
429 			vcpu->arch.mp_state = KVM_MP_STATE_SIPI_RECEIVED;
430 			kvm_vcpu_kick(vcpu);
431 		}
432 		break;
433 
434 	case APIC_DM_EXTINT:
435 		/*
436 		 * Should only be called by kvm_apic_local_deliver() with LVT0,
437 		 * before NMI watchdog was enabled. Already handled by
438 		 * kvm_apic_accept_pic_intr().
439 		 */
440 		break;
441 
442 	default:
443 		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
444 		       delivery_mode);
445 		break;
446 	}
447 	return result;
448 }
449 
450 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
451 {
452 	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
453 }
454 
455 static void apic_set_eoi(struct kvm_lapic *apic)
456 {
457 	int vector = apic_find_highest_isr(apic);
458 	int trigger_mode;
459 	/*
460 	 * Not every write EOI will has corresponding ISR,
461 	 * one example is when Kernel check timer on setup_IO_APIC
462 	 */
463 	if (vector == -1)
464 		return;
465 
466 	apic_clear_vector(vector, apic->regs + APIC_ISR);
467 	apic_update_ppr(apic);
468 
469 	if (apic_test_and_clear_vector(vector, apic->regs + APIC_TMR))
470 		trigger_mode = IOAPIC_LEVEL_TRIG;
471 	else
472 		trigger_mode = IOAPIC_EDGE_TRIG;
473 	if (!(apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI))
474 		kvm_ioapic_update_eoi(apic->vcpu->kvm, vector, trigger_mode);
475 }
476 
477 static void apic_send_ipi(struct kvm_lapic *apic)
478 {
479 	u32 icr_low = apic_get_reg(apic, APIC_ICR);
480 	u32 icr_high = apic_get_reg(apic, APIC_ICR2);
481 	struct kvm_lapic_irq irq;
482 
483 	irq.vector = icr_low & APIC_VECTOR_MASK;
484 	irq.delivery_mode = icr_low & APIC_MODE_MASK;
485 	irq.dest_mode = icr_low & APIC_DEST_MASK;
486 	irq.level = icr_low & APIC_INT_ASSERT;
487 	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
488 	irq.shorthand = icr_low & APIC_SHORT_MASK;
489 	if (apic_x2apic_mode(apic))
490 		irq.dest_id = icr_high;
491 	else
492 		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
493 
494 	trace_kvm_apic_ipi(icr_low, irq.dest_id);
495 
496 	apic_debug("icr_high 0x%x, icr_low 0x%x, "
497 		   "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
498 		   "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
499 		   icr_high, icr_low, irq.shorthand, irq.dest_id,
500 		   irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
501 		   irq.vector);
502 
503 	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq);
504 }
505 
506 static u32 apic_get_tmcct(struct kvm_lapic *apic)
507 {
508 	ktime_t remaining;
509 	s64 ns;
510 	u32 tmcct;
511 
512 	ASSERT(apic != NULL);
513 
514 	/* if initial count is 0, current count should also be 0 */
515 	if (apic_get_reg(apic, APIC_TMICT) == 0)
516 		return 0;
517 
518 	remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
519 	if (ktime_to_ns(remaining) < 0)
520 		remaining = ktime_set(0, 0);
521 
522 	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
523 	tmcct = div64_u64(ns,
524 			 (APIC_BUS_CYCLE_NS * apic->divide_count));
525 
526 	return tmcct;
527 }
528 
529 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
530 {
531 	struct kvm_vcpu *vcpu = apic->vcpu;
532 	struct kvm_run *run = vcpu->run;
533 
534 	set_bit(KVM_REQ_REPORT_TPR_ACCESS, &vcpu->requests);
535 	run->tpr_access.rip = kvm_rip_read(vcpu);
536 	run->tpr_access.is_write = write;
537 }
538 
539 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
540 {
541 	if (apic->vcpu->arch.tpr_access_reporting)
542 		__report_tpr_access(apic, write);
543 }
544 
545 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
546 {
547 	u32 val = 0;
548 
549 	if (offset >= LAPIC_MMIO_LENGTH)
550 		return 0;
551 
552 	switch (offset) {
553 	case APIC_ID:
554 		if (apic_x2apic_mode(apic))
555 			val = kvm_apic_id(apic);
556 		else
557 			val = kvm_apic_id(apic) << 24;
558 		break;
559 	case APIC_ARBPRI:
560 		printk(KERN_WARNING "Access APIC ARBPRI register "
561 		       "which is for P6\n");
562 		break;
563 
564 	case APIC_TMCCT:	/* Timer CCR */
565 		val = apic_get_tmcct(apic);
566 		break;
567 
568 	case APIC_TASKPRI:
569 		report_tpr_access(apic, false);
570 		/* fall thru */
571 	default:
572 		apic_update_ppr(apic);
573 		val = apic_get_reg(apic, offset);
574 		break;
575 	}
576 
577 	return val;
578 }
579 
580 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
581 {
582 	return container_of(dev, struct kvm_lapic, dev);
583 }
584 
585 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
586 		void *data)
587 {
588 	unsigned char alignment = offset & 0xf;
589 	u32 result;
590 	/* this bitmask has a bit cleared for each reserver register */
591 	static const u64 rmask = 0x43ff01ffffffe70cULL;
592 
593 	if ((alignment + len) > 4) {
594 		apic_debug("KVM_APIC_READ: alignment error %x %d\n",
595 			   offset, len);
596 		return 1;
597 	}
598 
599 	if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
600 		apic_debug("KVM_APIC_READ: read reserved register %x\n",
601 			   offset);
602 		return 1;
603 	}
604 
605 	result = __apic_read(apic, offset & ~0xf);
606 
607 	trace_kvm_apic_read(offset, result);
608 
609 	switch (len) {
610 	case 1:
611 	case 2:
612 	case 4:
613 		memcpy(data, (char *)&result + alignment, len);
614 		break;
615 	default:
616 		printk(KERN_ERR "Local APIC read with len = %x, "
617 		       "should be 1,2, or 4 instead\n", len);
618 		break;
619 	}
620 	return 0;
621 }
622 
623 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
624 {
625 	return apic_hw_enabled(apic) &&
626 	    addr >= apic->base_address &&
627 	    addr < apic->base_address + LAPIC_MMIO_LENGTH;
628 }
629 
630 static int apic_mmio_read(struct kvm_io_device *this,
631 			   gpa_t address, int len, void *data)
632 {
633 	struct kvm_lapic *apic = to_lapic(this);
634 	u32 offset = address - apic->base_address;
635 
636 	if (!apic_mmio_in_range(apic, address))
637 		return -EOPNOTSUPP;
638 
639 	apic_reg_read(apic, offset, len, data);
640 
641 	return 0;
642 }
643 
644 static void update_divide_count(struct kvm_lapic *apic)
645 {
646 	u32 tmp1, tmp2, tdcr;
647 
648 	tdcr = apic_get_reg(apic, APIC_TDCR);
649 	tmp1 = tdcr & 0xf;
650 	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
651 	apic->divide_count = 0x1 << (tmp2 & 0x7);
652 
653 	apic_debug("timer divide count is 0x%x\n",
654 				   apic->divide_count);
655 }
656 
657 static void start_apic_timer(struct kvm_lapic *apic)
658 {
659 	ktime_t now = apic->lapic_timer.timer.base->get_time();
660 
661 	apic->lapic_timer.period = (u64)apic_get_reg(apic, APIC_TMICT) *
662 		    APIC_BUS_CYCLE_NS * apic->divide_count;
663 	atomic_set(&apic->lapic_timer.pending, 0);
664 
665 	if (!apic->lapic_timer.period)
666 		return;
667 	/*
668 	 * Do not allow the guest to program periodic timers with small
669 	 * interval, since the hrtimers are not throttled by the host
670 	 * scheduler.
671 	 */
672 	if (apic_lvtt_period(apic)) {
673 		if (apic->lapic_timer.period < NSEC_PER_MSEC/2)
674 			apic->lapic_timer.period = NSEC_PER_MSEC/2;
675 	}
676 
677 	hrtimer_start(&apic->lapic_timer.timer,
678 		      ktime_add_ns(now, apic->lapic_timer.period),
679 		      HRTIMER_MODE_ABS);
680 
681 	apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
682 			   PRIx64 ", "
683 			   "timer initial count 0x%x, period %lldns, "
684 			   "expire @ 0x%016" PRIx64 ".\n", __func__,
685 			   APIC_BUS_CYCLE_NS, ktime_to_ns(now),
686 			   apic_get_reg(apic, APIC_TMICT),
687 			   apic->lapic_timer.period,
688 			   ktime_to_ns(ktime_add_ns(now,
689 					apic->lapic_timer.period)));
690 }
691 
692 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
693 {
694 	int nmi_wd_enabled = apic_lvt_nmi_mode(apic_get_reg(apic, APIC_LVT0));
695 
696 	if (apic_lvt_nmi_mode(lvt0_val)) {
697 		if (!nmi_wd_enabled) {
698 			apic_debug("Receive NMI setting on APIC_LVT0 "
699 				   "for cpu %d\n", apic->vcpu->vcpu_id);
700 			apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
701 		}
702 	} else if (nmi_wd_enabled)
703 		apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
704 }
705 
706 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
707 {
708 	int ret = 0;
709 
710 	trace_kvm_apic_write(reg, val);
711 
712 	switch (reg) {
713 	case APIC_ID:		/* Local APIC ID */
714 		if (!apic_x2apic_mode(apic))
715 			apic_set_reg(apic, APIC_ID, val);
716 		else
717 			ret = 1;
718 		break;
719 
720 	case APIC_TASKPRI:
721 		report_tpr_access(apic, true);
722 		apic_set_tpr(apic, val & 0xff);
723 		break;
724 
725 	case APIC_EOI:
726 		apic_set_eoi(apic);
727 		break;
728 
729 	case APIC_LDR:
730 		if (!apic_x2apic_mode(apic))
731 			apic_set_reg(apic, APIC_LDR, val & APIC_LDR_MASK);
732 		else
733 			ret = 1;
734 		break;
735 
736 	case APIC_DFR:
737 		if (!apic_x2apic_mode(apic))
738 			apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
739 		else
740 			ret = 1;
741 		break;
742 
743 	case APIC_SPIV: {
744 		u32 mask = 0x3ff;
745 		if (apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
746 			mask |= APIC_SPIV_DIRECTED_EOI;
747 		apic_set_reg(apic, APIC_SPIV, val & mask);
748 		if (!(val & APIC_SPIV_APIC_ENABLED)) {
749 			int i;
750 			u32 lvt_val;
751 
752 			for (i = 0; i < APIC_LVT_NUM; i++) {
753 				lvt_val = apic_get_reg(apic,
754 						       APIC_LVTT + 0x10 * i);
755 				apic_set_reg(apic, APIC_LVTT + 0x10 * i,
756 					     lvt_val | APIC_LVT_MASKED);
757 			}
758 			atomic_set(&apic->lapic_timer.pending, 0);
759 
760 		}
761 		break;
762 	}
763 	case APIC_ICR:
764 		/* No delay here, so we always clear the pending bit */
765 		apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
766 		apic_send_ipi(apic);
767 		break;
768 
769 	case APIC_ICR2:
770 		if (!apic_x2apic_mode(apic))
771 			val &= 0xff000000;
772 		apic_set_reg(apic, APIC_ICR2, val);
773 		break;
774 
775 	case APIC_LVT0:
776 		apic_manage_nmi_watchdog(apic, val);
777 	case APIC_LVTT:
778 	case APIC_LVTTHMR:
779 	case APIC_LVTPC:
780 	case APIC_LVT1:
781 	case APIC_LVTERR:
782 		/* TODO: Check vector */
783 		if (!apic_sw_enabled(apic))
784 			val |= APIC_LVT_MASKED;
785 
786 		val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
787 		apic_set_reg(apic, reg, val);
788 
789 		break;
790 
791 	case APIC_TMICT:
792 		hrtimer_cancel(&apic->lapic_timer.timer);
793 		apic_set_reg(apic, APIC_TMICT, val);
794 		start_apic_timer(apic);
795 		break;
796 
797 	case APIC_TDCR:
798 		if (val & 4)
799 			printk(KERN_ERR "KVM_WRITE:TDCR %x\n", val);
800 		apic_set_reg(apic, APIC_TDCR, val);
801 		update_divide_count(apic);
802 		break;
803 
804 	case APIC_ESR:
805 		if (apic_x2apic_mode(apic) && val != 0) {
806 			printk(KERN_ERR "KVM_WRITE:ESR not zero %x\n", val);
807 			ret = 1;
808 		}
809 		break;
810 
811 	case APIC_SELF_IPI:
812 		if (apic_x2apic_mode(apic)) {
813 			apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
814 		} else
815 			ret = 1;
816 		break;
817 	default:
818 		ret = 1;
819 		break;
820 	}
821 	if (ret)
822 		apic_debug("Local APIC Write to read-only register %x\n", reg);
823 	return ret;
824 }
825 
826 static int apic_mmio_write(struct kvm_io_device *this,
827 			    gpa_t address, int len, const void *data)
828 {
829 	struct kvm_lapic *apic = to_lapic(this);
830 	unsigned int offset = address - apic->base_address;
831 	u32 val;
832 
833 	if (!apic_mmio_in_range(apic, address))
834 		return -EOPNOTSUPP;
835 
836 	/*
837 	 * APIC register must be aligned on 128-bits boundary.
838 	 * 32/64/128 bits registers must be accessed thru 32 bits.
839 	 * Refer SDM 8.4.1
840 	 */
841 	if (len != 4 || (offset & 0xf)) {
842 		/* Don't shout loud, $infamous_os would cause only noise. */
843 		apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
844 		return 0;
845 	}
846 
847 	val = *(u32*)data;
848 
849 	/* too common printing */
850 	if (offset != APIC_EOI)
851 		apic_debug("%s: offset 0x%x with length 0x%x, and value is "
852 			   "0x%x\n", __func__, offset, len, val);
853 
854 	apic_reg_write(apic, offset & 0xff0, val);
855 
856 	return 0;
857 }
858 
859 void kvm_free_lapic(struct kvm_vcpu *vcpu)
860 {
861 	if (!vcpu->arch.apic)
862 		return;
863 
864 	hrtimer_cancel(&vcpu->arch.apic->lapic_timer.timer);
865 
866 	if (vcpu->arch.apic->regs_page)
867 		__free_page(vcpu->arch.apic->regs_page);
868 
869 	kfree(vcpu->arch.apic);
870 }
871 
872 /*
873  *----------------------------------------------------------------------
874  * LAPIC interface
875  *----------------------------------------------------------------------
876  */
877 
878 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
879 {
880 	struct kvm_lapic *apic = vcpu->arch.apic;
881 
882 	if (!apic)
883 		return;
884 	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
885 		     | (apic_get_reg(apic, APIC_TASKPRI) & 4));
886 }
887 
888 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
889 {
890 	struct kvm_lapic *apic = vcpu->arch.apic;
891 	u64 tpr;
892 
893 	if (!apic)
894 		return 0;
895 	tpr = (u64) apic_get_reg(apic, APIC_TASKPRI);
896 
897 	return (tpr & 0xf0) >> 4;
898 }
899 
900 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
901 {
902 	struct kvm_lapic *apic = vcpu->arch.apic;
903 
904 	if (!apic) {
905 		value |= MSR_IA32_APICBASE_BSP;
906 		vcpu->arch.apic_base = value;
907 		return;
908 	}
909 
910 	if (!kvm_vcpu_is_bsp(apic->vcpu))
911 		value &= ~MSR_IA32_APICBASE_BSP;
912 
913 	vcpu->arch.apic_base = value;
914 	if (apic_x2apic_mode(apic)) {
915 		u32 id = kvm_apic_id(apic);
916 		u32 ldr = ((id & ~0xf) << 16) | (1 << (id & 0xf));
917 		apic_set_reg(apic, APIC_LDR, ldr);
918 	}
919 	apic->base_address = apic->vcpu->arch.apic_base &
920 			     MSR_IA32_APICBASE_BASE;
921 
922 	/* with FSB delivery interrupt, we can restart APIC functionality */
923 	apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
924 		   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
925 
926 }
927 
928 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
929 {
930 	struct kvm_lapic *apic;
931 	int i;
932 
933 	apic_debug("%s\n", __func__);
934 
935 	ASSERT(vcpu);
936 	apic = vcpu->arch.apic;
937 	ASSERT(apic != NULL);
938 
939 	/* Stop the timer in case it's a reset to an active apic */
940 	hrtimer_cancel(&apic->lapic_timer.timer);
941 
942 	apic_set_reg(apic, APIC_ID, vcpu->vcpu_id << 24);
943 	kvm_apic_set_version(apic->vcpu);
944 
945 	for (i = 0; i < APIC_LVT_NUM; i++)
946 		apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
947 	apic_set_reg(apic, APIC_LVT0,
948 		     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
949 
950 	apic_set_reg(apic, APIC_DFR, 0xffffffffU);
951 	apic_set_reg(apic, APIC_SPIV, 0xff);
952 	apic_set_reg(apic, APIC_TASKPRI, 0);
953 	apic_set_reg(apic, APIC_LDR, 0);
954 	apic_set_reg(apic, APIC_ESR, 0);
955 	apic_set_reg(apic, APIC_ICR, 0);
956 	apic_set_reg(apic, APIC_ICR2, 0);
957 	apic_set_reg(apic, APIC_TDCR, 0);
958 	apic_set_reg(apic, APIC_TMICT, 0);
959 	for (i = 0; i < 8; i++) {
960 		apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
961 		apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
962 		apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
963 	}
964 	apic->irr_pending = false;
965 	update_divide_count(apic);
966 	atomic_set(&apic->lapic_timer.pending, 0);
967 	if (kvm_vcpu_is_bsp(vcpu))
968 		vcpu->arch.apic_base |= MSR_IA32_APICBASE_BSP;
969 	apic_update_ppr(apic);
970 
971 	vcpu->arch.apic_arb_prio = 0;
972 
973 	apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
974 		   "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
975 		   vcpu, kvm_apic_id(apic),
976 		   vcpu->arch.apic_base, apic->base_address);
977 }
978 
979 bool kvm_apic_present(struct kvm_vcpu *vcpu)
980 {
981 	return vcpu->arch.apic && apic_hw_enabled(vcpu->arch.apic);
982 }
983 
984 int kvm_lapic_enabled(struct kvm_vcpu *vcpu)
985 {
986 	return kvm_apic_present(vcpu) && apic_sw_enabled(vcpu->arch.apic);
987 }
988 
989 /*
990  *----------------------------------------------------------------------
991  * timer interface
992  *----------------------------------------------------------------------
993  */
994 
995 static bool lapic_is_periodic(struct kvm_timer *ktimer)
996 {
997 	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic,
998 					      lapic_timer);
999 	return apic_lvtt_period(apic);
1000 }
1001 
1002 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1003 {
1004 	struct kvm_lapic *lapic = vcpu->arch.apic;
1005 
1006 	if (lapic && apic_enabled(lapic) && apic_lvt_enabled(lapic, APIC_LVTT))
1007 		return atomic_read(&lapic->lapic_timer.pending);
1008 
1009 	return 0;
1010 }
1011 
1012 static int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1013 {
1014 	u32 reg = apic_get_reg(apic, lvt_type);
1015 	int vector, mode, trig_mode;
1016 
1017 	if (apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1018 		vector = reg & APIC_VECTOR_MASK;
1019 		mode = reg & APIC_MODE_MASK;
1020 		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1021 		return __apic_accept_irq(apic, mode, vector, 1, trig_mode);
1022 	}
1023 	return 0;
1024 }
1025 
1026 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1027 {
1028 	struct kvm_lapic *apic = vcpu->arch.apic;
1029 
1030 	if (apic)
1031 		kvm_apic_local_deliver(apic, APIC_LVT0);
1032 }
1033 
1034 static struct kvm_timer_ops lapic_timer_ops = {
1035 	.is_periodic = lapic_is_periodic,
1036 };
1037 
1038 static const struct kvm_io_device_ops apic_mmio_ops = {
1039 	.read     = apic_mmio_read,
1040 	.write    = apic_mmio_write,
1041 };
1042 
1043 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1044 {
1045 	struct kvm_lapic *apic;
1046 
1047 	ASSERT(vcpu != NULL);
1048 	apic_debug("apic_init %d\n", vcpu->vcpu_id);
1049 
1050 	apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1051 	if (!apic)
1052 		goto nomem;
1053 
1054 	vcpu->arch.apic = apic;
1055 
1056 	apic->regs_page = alloc_page(GFP_KERNEL);
1057 	if (apic->regs_page == NULL) {
1058 		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1059 		       vcpu->vcpu_id);
1060 		goto nomem_free_apic;
1061 	}
1062 	apic->regs = page_address(apic->regs_page);
1063 	memset(apic->regs, 0, PAGE_SIZE);
1064 	apic->vcpu = vcpu;
1065 
1066 	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1067 		     HRTIMER_MODE_ABS);
1068 	apic->lapic_timer.timer.function = kvm_timer_fn;
1069 	apic->lapic_timer.t_ops = &lapic_timer_ops;
1070 	apic->lapic_timer.kvm = vcpu->kvm;
1071 	apic->lapic_timer.vcpu = vcpu;
1072 
1073 	apic->base_address = APIC_DEFAULT_PHYS_BASE;
1074 	vcpu->arch.apic_base = APIC_DEFAULT_PHYS_BASE;
1075 
1076 	kvm_lapic_reset(vcpu);
1077 	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1078 
1079 	return 0;
1080 nomem_free_apic:
1081 	kfree(apic);
1082 nomem:
1083 	return -ENOMEM;
1084 }
1085 
1086 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1087 {
1088 	struct kvm_lapic *apic = vcpu->arch.apic;
1089 	int highest_irr;
1090 
1091 	if (!apic || !apic_enabled(apic))
1092 		return -1;
1093 
1094 	apic_update_ppr(apic);
1095 	highest_irr = apic_find_highest_irr(apic);
1096 	if ((highest_irr == -1) ||
1097 	    ((highest_irr & 0xF0) <= apic_get_reg(apic, APIC_PROCPRI)))
1098 		return -1;
1099 	return highest_irr;
1100 }
1101 
1102 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1103 {
1104 	u32 lvt0 = apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1105 	int r = 0;
1106 
1107 	if (kvm_vcpu_is_bsp(vcpu)) {
1108 		if (!apic_hw_enabled(vcpu->arch.apic))
1109 			r = 1;
1110 		if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1111 		    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1112 			r = 1;
1113 	}
1114 	return r;
1115 }
1116 
1117 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1118 {
1119 	struct kvm_lapic *apic = vcpu->arch.apic;
1120 
1121 	if (apic && atomic_read(&apic->lapic_timer.pending) > 0) {
1122 		if (kvm_apic_local_deliver(apic, APIC_LVTT))
1123 			atomic_dec(&apic->lapic_timer.pending);
1124 	}
1125 }
1126 
1127 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1128 {
1129 	int vector = kvm_apic_has_interrupt(vcpu);
1130 	struct kvm_lapic *apic = vcpu->arch.apic;
1131 
1132 	if (vector == -1)
1133 		return -1;
1134 
1135 	apic_set_vector(vector, apic->regs + APIC_ISR);
1136 	apic_update_ppr(apic);
1137 	apic_clear_irr(vector, apic);
1138 	return vector;
1139 }
1140 
1141 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu)
1142 {
1143 	struct kvm_lapic *apic = vcpu->arch.apic;
1144 
1145 	apic->base_address = vcpu->arch.apic_base &
1146 			     MSR_IA32_APICBASE_BASE;
1147 	kvm_apic_set_version(vcpu);
1148 
1149 	apic_update_ppr(apic);
1150 	hrtimer_cancel(&apic->lapic_timer.timer);
1151 	update_divide_count(apic);
1152 	start_apic_timer(apic);
1153 	apic->irr_pending = true;
1154 }
1155 
1156 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1157 {
1158 	struct kvm_lapic *apic = vcpu->arch.apic;
1159 	struct hrtimer *timer;
1160 
1161 	if (!apic)
1162 		return;
1163 
1164 	timer = &apic->lapic_timer.timer;
1165 	if (hrtimer_cancel(timer))
1166 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1167 }
1168 
1169 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1170 {
1171 	u32 data;
1172 	void *vapic;
1173 
1174 	if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1175 		return;
1176 
1177 	vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1178 	data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr));
1179 	kunmap_atomic(vapic, KM_USER0);
1180 
1181 	apic_set_tpr(vcpu->arch.apic, data & 0xff);
1182 }
1183 
1184 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1185 {
1186 	u32 data, tpr;
1187 	int max_irr, max_isr;
1188 	struct kvm_lapic *apic;
1189 	void *vapic;
1190 
1191 	if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1192 		return;
1193 
1194 	apic = vcpu->arch.apic;
1195 	tpr = apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1196 	max_irr = apic_find_highest_irr(apic);
1197 	if (max_irr < 0)
1198 		max_irr = 0;
1199 	max_isr = apic_find_highest_isr(apic);
1200 	if (max_isr < 0)
1201 		max_isr = 0;
1202 	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1203 
1204 	vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1205 	*(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data;
1206 	kunmap_atomic(vapic, KM_USER0);
1207 }
1208 
1209 void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1210 {
1211 	if (!irqchip_in_kernel(vcpu->kvm))
1212 		return;
1213 
1214 	vcpu->arch.apic->vapic_addr = vapic_addr;
1215 }
1216 
1217 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1218 {
1219 	struct kvm_lapic *apic = vcpu->arch.apic;
1220 	u32 reg = (msr - APIC_BASE_MSR) << 4;
1221 
1222 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1223 		return 1;
1224 
1225 	/* if this is ICR write vector before command */
1226 	if (msr == 0x830)
1227 		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1228 	return apic_reg_write(apic, reg, (u32)data);
1229 }
1230 
1231 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1232 {
1233 	struct kvm_lapic *apic = vcpu->arch.apic;
1234 	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1235 
1236 	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1237 		return 1;
1238 
1239 	if (apic_reg_read(apic, reg, 4, &low))
1240 		return 1;
1241 	if (msr == 0x830)
1242 		apic_reg_read(apic, APIC_ICR2, 4, &high);
1243 
1244 	*data = (((u64)high) << 32) | low;
1245 
1246 	return 0;
1247 }
1248