local_apic.c revision 210298
1/*-
2 * Copyright (c) 2003 John Baldwin <jhb@FreeBSD.org>
3 * Copyright (c) 1996, by Steve Passe
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 *    notice, this list of conditions and the following disclaimer.
11 * 2. The name of the developer may NOT be used to endorse or promote products
12 *    derived from this software without specific prior written permission.
13 * 3. Neither the name of the author nor the names of any co-contributors
14 *    may be used to endorse or promote products derived from this software
15 *    without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 */
29
30/*
31 * Local APIC support on Pentium and later processors.
32 */
33
34#include <sys/cdefs.h>
35__FBSDID("$FreeBSD: head/sys/x86/x86/local_apic.c 210298 2010-07-20 15:48:29Z mav $");
36
37#include "opt_hwpmc_hooks.h"
38#include "opt_kdtrace.h"
39
40#include "opt_ddb.h"
41
42#include <sys/param.h>
43#include <sys/systm.h>
44#include <sys/bus.h>
45#include <sys/kernel.h>
46#include <sys/lock.h>
47#include <sys/mutex.h>
48#include <sys/pcpu.h>
49#include <sys/proc.h>
50#include <sys/sched.h>
51#include <sys/smp.h>
52#include <sys/timeet.h>
53
54#include <vm/vm.h>
55#include <vm/pmap.h>
56
57#include <machine/apicreg.h>
58#include <machine/cpu.h>
59#include <machine/cputypes.h>
60#include <machine/frame.h>
61#include <machine/intr_machdep.h>
62#include <machine/apicvar.h>
63#include <machine/mca.h>
64#include <machine/md_var.h>
65#include <machine/smp.h>
66#include <machine/specialreg.h>
67
68#ifdef DDB
69#include <sys/interrupt.h>
70#include <ddb/ddb.h>
71#endif
72
73#ifdef __amd64__
74#define	SDT_APIC	SDT_SYSIGT
75#define	SDT_APICT	SDT_SYSIGT
76#define	GSEL_APIC	0
77#else
78#define	SDT_APIC	SDT_SYS386IGT
79#define	SDT_APICT	SDT_SYS386TGT
80#define	GSEL_APIC	GSEL(GCODE_SEL, SEL_KPL)
81#endif
82
83/* Sanity checks on IDT vectors. */
84CTASSERT(APIC_IO_INTS + APIC_NUM_IOINTS == APIC_TIMER_INT);
85CTASSERT(APIC_TIMER_INT < APIC_LOCAL_INTS);
86CTASSERT(APIC_LOCAL_INTS == 240);
87CTASSERT(IPI_STOP < APIC_SPURIOUS_INT);
88
89/* Magic IRQ values for the timer and syscalls. */
90#define	IRQ_TIMER	(NUM_IO_INTS + 1)
91#define	IRQ_SYSCALL	(NUM_IO_INTS + 2)
92
93/*
94 * Support for local APICs.  Local APICs manage interrupts on each
95 * individual processor as opposed to I/O APICs which receive interrupts
96 * from I/O devices and then forward them on to the local APICs.
97 *
98 * Local APICs can also send interrupts to each other thus providing the
99 * mechanism for IPIs.
100 */
101
102struct lvt {
103	u_int lvt_edgetrigger:1;
104	u_int lvt_activehi:1;
105	u_int lvt_masked:1;
106	u_int lvt_active:1;
107	u_int lvt_mode:16;
108	u_int lvt_vector:8;
109};
110
111struct lapic {
112	struct lvt la_lvts[LVT_MAX + 1];
113	u_int la_id:8;
114	u_int la_cluster:4;
115	u_int la_cluster_id:2;
116	u_int la_present:1;
117	u_long *la_timer_count;
118	u_long la_timer_period;
119	u_int la_timer_mode;
120	/* Include IDT_SYSCALL to make indexing easier. */
121	int la_ioint_irqs[APIC_NUM_IOINTS + 1];
122} static lapics[MAX_APIC_ID + 1];
123
124/* Global defaults for local APIC LVT entries. */
125static struct lvt lvts[LVT_MAX + 1] = {
126	{ 1, 1, 1, 1, APIC_LVT_DM_EXTINT, 0 },	/* LINT0: masked ExtINT */
127	{ 1, 1, 0, 1, APIC_LVT_DM_NMI, 0 },	/* LINT1: NMI */
128	{ 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_TIMER_INT },	/* Timer */
129	{ 1, 1, 0, 1, APIC_LVT_DM_FIXED, APIC_ERROR_INT },	/* Error */
130	{ 1, 1, 1, 1, APIC_LVT_DM_NMI, 0 },	/* PMC */
131	{ 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_THERMAL_INT },	/* Thermal */
132	{ 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_CMC_INT },	/* CMCI */
133};
134
135static inthand_t *ioint_handlers[] = {
136	NULL,			/* 0 - 31 */
137	IDTVEC(apic_isr1),	/* 32 - 63 */
138	IDTVEC(apic_isr2),	/* 64 - 95 */
139	IDTVEC(apic_isr3),	/* 96 - 127 */
140	IDTVEC(apic_isr4),	/* 128 - 159 */
141	IDTVEC(apic_isr5),	/* 160 - 191 */
142	IDTVEC(apic_isr6),	/* 192 - 223 */
143	IDTVEC(apic_isr7),	/* 224 - 255 */
144};
145
146
147static u_int32_t lapic_timer_divisors[] = {
148	APIC_TDCR_1, APIC_TDCR_2, APIC_TDCR_4, APIC_TDCR_8, APIC_TDCR_16,
149	APIC_TDCR_32, APIC_TDCR_64, APIC_TDCR_128
150};
151
152extern inthand_t IDTVEC(rsvd);
153
154volatile lapic_t *lapic;
155vm_paddr_t lapic_paddr;
156static u_long lapic_timer_divisor;
157static struct eventtimer lapic_et;
158
159static void	lapic_enable(void);
160static void	lapic_resume(struct pic *pic);
161static void	lapic_timer_enable_intr(void);
162static void	lapic_timer_oneshot(u_int count);
163static void	lapic_timer_periodic(u_int count);
164static void	lapic_timer_stop(void);
165static void	lapic_timer_set_divisor(u_int divisor);
166static uint32_t	lvt_mode(struct lapic *la, u_int pin, uint32_t value);
167static int	lapic_et_start(struct eventtimer *et,
168    struct bintime *first, struct bintime *period);
169static int	lapic_et_stop(struct eventtimer *et);
170
171struct pic lapic_pic = { .pic_resume = lapic_resume };
172
173static uint32_t
174lvt_mode(struct lapic *la, u_int pin, uint32_t value)
175{
176	struct lvt *lvt;
177
178	KASSERT(pin <= LVT_MAX, ("%s: pin %u out of range", __func__, pin));
179	if (la->la_lvts[pin].lvt_active)
180		lvt = &la->la_lvts[pin];
181	else
182		lvt = &lvts[pin];
183
184	value &= ~(APIC_LVT_M | APIC_LVT_TM | APIC_LVT_IIPP | APIC_LVT_DM |
185	    APIC_LVT_VECTOR);
186	if (lvt->lvt_edgetrigger == 0)
187		value |= APIC_LVT_TM;
188	if (lvt->lvt_activehi == 0)
189		value |= APIC_LVT_IIPP_INTALO;
190	if (lvt->lvt_masked)
191		value |= APIC_LVT_M;
192	value |= lvt->lvt_mode;
193	switch (lvt->lvt_mode) {
194	case APIC_LVT_DM_NMI:
195	case APIC_LVT_DM_SMI:
196	case APIC_LVT_DM_INIT:
197	case APIC_LVT_DM_EXTINT:
198		if (!lvt->lvt_edgetrigger) {
199			printf("lapic%u: Forcing LINT%u to edge trigger\n",
200			    la->la_id, pin);
201			value |= APIC_LVT_TM;
202		}
203		/* Use a vector of 0. */
204		break;
205	case APIC_LVT_DM_FIXED:
206		value |= lvt->lvt_vector;
207		break;
208	default:
209		panic("bad APIC LVT delivery mode: %#x\n", value);
210	}
211	return (value);
212}
213
214/*
215 * Map the local APIC and setup necessary interrupt vectors.
216 */
217void
218lapic_init(vm_paddr_t addr)
219{
220	u_int regs[4];
221	int i, arat;
222
223	/* Map the local APIC and setup the spurious interrupt handler. */
224	KASSERT(trunc_page(addr) == addr,
225	    ("local APIC not aligned on a page boundary"));
226	lapic = pmap_mapdev(addr, sizeof(lapic_t));
227	lapic_paddr = addr;
228	setidt(APIC_SPURIOUS_INT, IDTVEC(spuriousint), SDT_APIC, SEL_KPL,
229	    GSEL_APIC);
230
231	/* Perform basic initialization of the BSP's local APIC. */
232	lapic_enable();
233
234	/* Set BSP's per-CPU local APIC ID. */
235	PCPU_SET(apic_id, lapic_id());
236
237	/* Local APIC timer interrupt. */
238	setidt(APIC_TIMER_INT, IDTVEC(timerint), SDT_APIC, SEL_KPL, GSEL_APIC);
239
240	/* Local APIC error interrupt. */
241	setidt(APIC_ERROR_INT, IDTVEC(errorint), SDT_APIC, SEL_KPL, GSEL_APIC);
242
243	/* XXX: Thermal interrupt */
244
245	/* Local APIC CMCI. */
246	setidt(APIC_CMC_INT, IDTVEC(cmcint), SDT_APICT, SEL_KPL, GSEL_APIC);
247
248	if ((resource_int_value("apic", 0, "clock", &i) != 0 || i != 0)) {
249		arat = 0;
250		/* Intel CPUID 0x06 EAX[2] set if APIC timer runs in C3. */
251		if (cpu_vendor_id == CPU_VENDOR_INTEL && cpu_high >= 6) {
252			do_cpuid(0x06, regs);
253			if (regs[0] & 0x4)
254				arat = 1;
255		}
256		bzero(&lapic_et, sizeof(lapic_et));
257		lapic_et.et_name = "LAPIC";
258		lapic_et.et_flags = ET_FLAGS_PERIODIC | ET_FLAGS_ONESHOT |
259		    ET_FLAGS_PERCPU;
260		lapic_et.et_quality = 600;
261		if (!arat) {
262			lapic_et.et_flags |= ET_FLAGS_C3STOP;
263			lapic_et.et_quality -= 100;
264		}
265		lapic_et.et_frequency = 0;
266		/* We don't know frequency yet, so trying to guess. */
267		lapic_et.et_min_period.sec = 0;
268		lapic_et.et_min_period.frac = 0x00001000LL << 32;
269		lapic_et.et_max_period.sec = 1;
270		lapic_et.et_max_period.frac = 0;
271		lapic_et.et_start = lapic_et_start;
272		lapic_et.et_stop = lapic_et_stop;
273		lapic_et.et_priv = NULL;
274		et_register(&lapic_et);
275	}
276}
277
278/*
279 * Create a local APIC instance.
280 */
281void
282lapic_create(u_int apic_id, int boot_cpu)
283{
284	int i;
285
286	if (apic_id > MAX_APIC_ID) {
287		printf("APIC: Ignoring local APIC with ID %d\n", apic_id);
288		if (boot_cpu)
289			panic("Can't ignore BSP");
290		return;
291	}
292	KASSERT(!lapics[apic_id].la_present, ("duplicate local APIC %u",
293	    apic_id));
294
295	/*
296	 * Assume no local LVT overrides and a cluster of 0 and
297	 * intra-cluster ID of 0.
298	 */
299	lapics[apic_id].la_present = 1;
300	lapics[apic_id].la_id = apic_id;
301	for (i = 0; i <= LVT_MAX; i++) {
302		lapics[apic_id].la_lvts[i] = lvts[i];
303		lapics[apic_id].la_lvts[i].lvt_active = 0;
304	}
305	for (i = 0; i <= APIC_NUM_IOINTS; i++)
306	    lapics[apic_id].la_ioint_irqs[i] = -1;
307	lapics[apic_id].la_ioint_irqs[IDT_SYSCALL - APIC_IO_INTS] = IRQ_SYSCALL;
308	lapics[apic_id].la_ioint_irqs[APIC_TIMER_INT - APIC_IO_INTS] =
309	    IRQ_TIMER;
310
311#ifdef SMP
312	cpu_add(apic_id, boot_cpu);
313#endif
314}
315
316/*
317 * Dump contents of local APIC registers
318 */
319void
320lapic_dump(const char* str)
321{
322
323	printf("cpu%d %s:\n", PCPU_GET(cpuid), str);
324	printf("     ID: 0x%08x   VER: 0x%08x LDR: 0x%08x DFR: 0x%08x\n",
325	    lapic->id, lapic->version, lapic->ldr, lapic->dfr);
326	printf("  lint0: 0x%08x lint1: 0x%08x TPR: 0x%08x SVR: 0x%08x\n",
327	    lapic->lvt_lint0, lapic->lvt_lint1, lapic->tpr, lapic->svr);
328	printf("  timer: 0x%08x therm: 0x%08x err: 0x%08x pmc: 0x%08x\n",
329	    lapic->lvt_timer, lapic->lvt_thermal, lapic->lvt_error,
330	    lapic->lvt_pcint);
331	printf("   cmci: 0x%08x\n", lapic->lvt_cmci);
332}
333
334void
335lapic_setup(int boot)
336{
337	struct lapic *la;
338	u_int32_t maxlvt;
339	register_t eflags;
340	char buf[MAXCOMLEN + 1];
341
342	la = &lapics[lapic_id()];
343	KASSERT(la->la_present, ("missing APIC structure"));
344	eflags = intr_disable();
345	maxlvt = (lapic->version & APIC_VER_MAXLVT) >> MAXLVTSHIFT;
346
347	/* Initialize the TPR to allow all interrupts. */
348	lapic_set_tpr(0);
349
350	/* Setup spurious vector and enable the local APIC. */
351	lapic_enable();
352
353	/* Program LINT[01] LVT entries. */
354	lapic->lvt_lint0 = lvt_mode(la, LVT_LINT0, lapic->lvt_lint0);
355	lapic->lvt_lint1 = lvt_mode(la, LVT_LINT1, lapic->lvt_lint1);
356
357	/* Program the PMC LVT entry if present. */
358	if (maxlvt >= LVT_PMC)
359		lapic->lvt_pcint = lvt_mode(la, LVT_PMC, lapic->lvt_pcint);
360
361	/* Program timer LVT and setup handler. */
362	lapic->lvt_timer = lvt_mode(la, LVT_TIMER, lapic->lvt_timer);
363	if (boot) {
364		snprintf(buf, sizeof(buf), "cpu%d:timer", PCPU_GET(cpuid));
365		intrcnt_add(buf, &la->la_timer_count);
366	}
367
368	/* Setup the timer if configured. */
369	if (la->la_timer_mode != 0) {
370		KASSERT(la->la_timer_period != 0, ("lapic%u: zero divisor",
371		    lapic_id()));
372		lapic_timer_set_divisor(lapic_timer_divisor);
373		if (la->la_timer_mode == 1)
374			lapic_timer_periodic(la->la_timer_period);
375		else
376			lapic_timer_oneshot(la->la_timer_period);
377		lapic_timer_enable_intr();
378	}
379
380	/* Program error LVT and clear any existing errors. */
381	lapic->lvt_error = lvt_mode(la, LVT_ERROR, lapic->lvt_error);
382	lapic->esr = 0;
383
384	/* XXX: Thermal LVT */
385
386	/* Program the CMCI LVT entry if present. */
387	if (maxlvt >= LVT_CMCI)
388		lapic->lvt_cmci = lvt_mode(la, LVT_CMCI, lapic->lvt_cmci);
389
390	intr_restore(eflags);
391}
392
393void
394lapic_reenable_pmc(void)
395{
396#ifdef HWPMC_HOOKS
397	uint32_t value;
398
399	value =  lapic->lvt_pcint;
400	value &= ~APIC_LVT_M;
401	lapic->lvt_pcint = value;
402#endif
403}
404
405#ifdef HWPMC_HOOKS
406static void
407lapic_update_pmc(void *dummy)
408{
409	struct lapic *la;
410
411	la = &lapics[lapic_id()];
412	lapic->lvt_pcint = lvt_mode(la, LVT_PMC, lapic->lvt_pcint);
413}
414#endif
415
416int
417lapic_enable_pmc(void)
418{
419#ifdef HWPMC_HOOKS
420	u_int32_t maxlvt;
421
422	/* Fail if the local APIC is not present. */
423	if (lapic == NULL)
424		return (0);
425
426	/* Fail if the PMC LVT is not present. */
427	maxlvt = (lapic->version & APIC_VER_MAXLVT) >> MAXLVTSHIFT;
428	if (maxlvt < LVT_PMC)
429		return (0);
430
431	lvts[LVT_PMC].lvt_masked = 0;
432
433#ifdef SMP
434	/*
435	 * If hwpmc was loaded at boot time then the APs may not be
436	 * started yet.  In that case, don't forward the request to
437	 * them as they will program the lvt when they start.
438	 */
439	if (smp_started)
440		smp_rendezvous(NULL, lapic_update_pmc, NULL, NULL);
441	else
442#endif
443		lapic_update_pmc(NULL);
444	return (1);
445#else
446	return (0);
447#endif
448}
449
450void
451lapic_disable_pmc(void)
452{
453#ifdef HWPMC_HOOKS
454	u_int32_t maxlvt;
455
456	/* Fail if the local APIC is not present. */
457	if (lapic == NULL)
458		return;
459
460	/* Fail if the PMC LVT is not present. */
461	maxlvt = (lapic->version & APIC_VER_MAXLVT) >> MAXLVTSHIFT;
462	if (maxlvt < LVT_PMC)
463		return;
464
465	lvts[LVT_PMC].lvt_masked = 1;
466
467#ifdef SMP
468	/* The APs should always be started when hwpmc is unloaded. */
469	KASSERT(mp_ncpus == 1 || smp_started, ("hwpmc unloaded too early"));
470#endif
471	smp_rendezvous(NULL, lapic_update_pmc, NULL, NULL);
472#endif
473}
474
475static int
476lapic_et_start(struct eventtimer *et,
477    struct bintime *first, struct bintime *period)
478{
479	struct lapic *la;
480	u_long value;
481
482	if (et->et_frequency == 0) {
483		/* Start off with a divisor of 2 (power on reset default). */
484		lapic_timer_divisor = 2;
485		/* Try to calibrate the local APIC timer. */
486		do {
487			lapic_timer_set_divisor(lapic_timer_divisor);
488			lapic_timer_oneshot(APIC_TIMER_MAX_COUNT);
489			DELAY(1000000);
490			value = APIC_TIMER_MAX_COUNT - lapic->ccr_timer;
491			if (value != APIC_TIMER_MAX_COUNT)
492				break;
493			lapic_timer_divisor <<= 1;
494		} while (lapic_timer_divisor <= 128);
495		if (lapic_timer_divisor > 128)
496			panic("lapic: Divisor too big");
497		if (bootverbose)
498			printf("lapic: Divisor %lu, Frequency %lu Hz\n",
499			    lapic_timer_divisor, value);
500		et->et_frequency = value;
501		et->et_min_period.sec = 0;
502		et->et_min_period.frac =
503		    ((0x00000002LLU << 32) / et->et_frequency) << 32;
504		et->et_max_period.sec = 0xfffffffeLLU / et->et_frequency;
505		et->et_max_period.frac =
506		    ((0xfffffffeLLU << 32) / et->et_frequency) << 32;
507	}
508	la = &lapics[lapic_id()];
509	/*
510	 * Start up the timer on the BSP.  The APs will kick off their
511	 * timer during lapic_setup().
512	 */
513	lapic_timer_set_divisor(lapic_timer_divisor);
514	if (period != NULL) {
515		la->la_timer_mode = 1;
516		la->la_timer_period =
517		    (et->et_frequency * (period->frac >> 32)) >> 32;
518		if (period->sec != 0)
519			la->la_timer_period += et->et_frequency * period->sec;
520		lapic_timer_periodic(la->la_timer_period);
521	} else {
522		la->la_timer_mode = 2;
523		la->la_timer_period =
524		    (et->et_frequency * (first->frac >> 32)) >> 32;
525		if (first->sec != 0)
526			la->la_timer_period += et->et_frequency * first->sec;
527		lapic_timer_oneshot(la->la_timer_period);
528	}
529	lapic_timer_enable_intr();
530	return (0);
531}
532
533static int
534lapic_et_stop(struct eventtimer *et)
535{
536	struct lapic *la = &lapics[lapic_id()];
537
538	la->la_timer_mode = 0;
539	lapic_timer_stop();
540	return (0);
541}
542
543void
544lapic_disable(void)
545{
546	uint32_t value;
547
548	/* Software disable the local APIC. */
549	value = lapic->svr;
550	value &= ~APIC_SVR_SWEN;
551	lapic->svr = value;
552}
553
554static void
555lapic_enable(void)
556{
557	u_int32_t value;
558
559	/* Program the spurious vector to enable the local APIC. */
560	value = lapic->svr;
561	value &= ~(APIC_SVR_VECTOR | APIC_SVR_FOCUS);
562	value |= (APIC_SVR_FEN | APIC_SVR_SWEN | APIC_SPURIOUS_INT);
563	lapic->svr = value;
564}
565
566/* Reset the local APIC on the BSP during resume. */
567static void
568lapic_resume(struct pic *pic)
569{
570
571	lapic_setup(0);
572}
573
574int
575lapic_id(void)
576{
577
578	KASSERT(lapic != NULL, ("local APIC is not mapped"));
579	return (lapic->id >> APIC_ID_SHIFT);
580}
581
582int
583lapic_intr_pending(u_int vector)
584{
585	volatile u_int32_t *irr;
586
587	/*
588	 * The IRR registers are an array of 128-bit registers each of
589	 * which only describes 32 interrupts in the low 32 bits..  Thus,
590	 * we divide the vector by 32 to get the 128-bit index.  We then
591	 * multiply that index by 4 to get the equivalent index from
592	 * treating the IRR as an array of 32-bit registers.  Finally, we
593	 * modulus the vector by 32 to determine the individual bit to
594	 * test.
595	 */
596	irr = &lapic->irr0;
597	return (irr[(vector / 32) * 4] & 1 << (vector % 32));
598}
599
600void
601lapic_set_logical_id(u_int apic_id, u_int cluster, u_int cluster_id)
602{
603	struct lapic *la;
604
605	KASSERT(lapics[apic_id].la_present, ("%s: APIC %u doesn't exist",
606	    __func__, apic_id));
607	KASSERT(cluster <= APIC_MAX_CLUSTER, ("%s: cluster %u too big",
608	    __func__, cluster));
609	KASSERT(cluster_id <= APIC_MAX_INTRACLUSTER_ID,
610	    ("%s: intra cluster id %u too big", __func__, cluster_id));
611	la = &lapics[apic_id];
612	la->la_cluster = cluster;
613	la->la_cluster_id = cluster_id;
614}
615
616int
617lapic_set_lvt_mask(u_int apic_id, u_int pin, u_char masked)
618{
619
620	if (pin > LVT_MAX)
621		return (EINVAL);
622	if (apic_id == APIC_ID_ALL) {
623		lvts[pin].lvt_masked = masked;
624		if (bootverbose)
625			printf("lapic:");
626	} else {
627		KASSERT(lapics[apic_id].la_present,
628		    ("%s: missing APIC %u", __func__, apic_id));
629		lapics[apic_id].la_lvts[pin].lvt_masked = masked;
630		lapics[apic_id].la_lvts[pin].lvt_active = 1;
631		if (bootverbose)
632			printf("lapic%u:", apic_id);
633	}
634	if (bootverbose)
635		printf(" LINT%u %s\n", pin, masked ? "masked" : "unmasked");
636	return (0);
637}
638
639int
640lapic_set_lvt_mode(u_int apic_id, u_int pin, u_int32_t mode)
641{
642	struct lvt *lvt;
643
644	if (pin > LVT_MAX)
645		return (EINVAL);
646	if (apic_id == APIC_ID_ALL) {
647		lvt = &lvts[pin];
648		if (bootverbose)
649			printf("lapic:");
650	} else {
651		KASSERT(lapics[apic_id].la_present,
652		    ("%s: missing APIC %u", __func__, apic_id));
653		lvt = &lapics[apic_id].la_lvts[pin];
654		lvt->lvt_active = 1;
655		if (bootverbose)
656			printf("lapic%u:", apic_id);
657	}
658	lvt->lvt_mode = mode;
659	switch (mode) {
660	case APIC_LVT_DM_NMI:
661	case APIC_LVT_DM_SMI:
662	case APIC_LVT_DM_INIT:
663	case APIC_LVT_DM_EXTINT:
664		lvt->lvt_edgetrigger = 1;
665		lvt->lvt_activehi = 1;
666		if (mode == APIC_LVT_DM_EXTINT)
667			lvt->lvt_masked = 1;
668		else
669			lvt->lvt_masked = 0;
670		break;
671	default:
672		panic("Unsupported delivery mode: 0x%x\n", mode);
673	}
674	if (bootverbose) {
675		printf(" Routing ");
676		switch (mode) {
677		case APIC_LVT_DM_NMI:
678			printf("NMI");
679			break;
680		case APIC_LVT_DM_SMI:
681			printf("SMI");
682			break;
683		case APIC_LVT_DM_INIT:
684			printf("INIT");
685			break;
686		case APIC_LVT_DM_EXTINT:
687			printf("ExtINT");
688			break;
689		}
690		printf(" -> LINT%u\n", pin);
691	}
692	return (0);
693}
694
695int
696lapic_set_lvt_polarity(u_int apic_id, u_int pin, enum intr_polarity pol)
697{
698
699	if (pin > LVT_MAX || pol == INTR_POLARITY_CONFORM)
700		return (EINVAL);
701	if (apic_id == APIC_ID_ALL) {
702		lvts[pin].lvt_activehi = (pol == INTR_POLARITY_HIGH);
703		if (bootverbose)
704			printf("lapic:");
705	} else {
706		KASSERT(lapics[apic_id].la_present,
707		    ("%s: missing APIC %u", __func__, apic_id));
708		lapics[apic_id].la_lvts[pin].lvt_active = 1;
709		lapics[apic_id].la_lvts[pin].lvt_activehi =
710		    (pol == INTR_POLARITY_HIGH);
711		if (bootverbose)
712			printf("lapic%u:", apic_id);
713	}
714	if (bootverbose)
715		printf(" LINT%u polarity: %s\n", pin,
716		    pol == INTR_POLARITY_HIGH ? "high" : "low");
717	return (0);
718}
719
720int
721lapic_set_lvt_triggermode(u_int apic_id, u_int pin, enum intr_trigger trigger)
722{
723
724	if (pin > LVT_MAX || trigger == INTR_TRIGGER_CONFORM)
725		return (EINVAL);
726	if (apic_id == APIC_ID_ALL) {
727		lvts[pin].lvt_edgetrigger = (trigger == INTR_TRIGGER_EDGE);
728		if (bootverbose)
729			printf("lapic:");
730	} else {
731		KASSERT(lapics[apic_id].la_present,
732		    ("%s: missing APIC %u", __func__, apic_id));
733		lapics[apic_id].la_lvts[pin].lvt_edgetrigger =
734		    (trigger == INTR_TRIGGER_EDGE);
735		lapics[apic_id].la_lvts[pin].lvt_active = 1;
736		if (bootverbose)
737			printf("lapic%u:", apic_id);
738	}
739	if (bootverbose)
740		printf(" LINT%u trigger: %s\n", pin,
741		    trigger == INTR_TRIGGER_EDGE ? "edge" : "level");
742	return (0);
743}
744
745/*
746 * Adjust the TPR of the current CPU so that it blocks all interrupts below
747 * the passed in vector.
748 */
749void
750lapic_set_tpr(u_int vector)
751{
752#ifdef CHEAP_TPR
753	lapic->tpr = vector;
754#else
755	u_int32_t tpr;
756
757	tpr = lapic->tpr & ~APIC_TPR_PRIO;
758	tpr |= vector;
759	lapic->tpr = tpr;
760#endif
761}
762
763void
764lapic_eoi(void)
765{
766
767	lapic->eoi = 0;
768}
769
770void
771lapic_handle_intr(int vector, struct trapframe *frame)
772{
773	struct intsrc *isrc;
774
775	if (vector == -1)
776		panic("Couldn't get vector from ISR!");
777	isrc = intr_lookup_source(apic_idt_to_irq(PCPU_GET(apic_id),
778	    vector));
779	intr_execute_handlers(isrc, frame);
780}
781
782void
783lapic_handle_timer(struct trapframe *frame)
784{
785	struct lapic *la;
786	struct trapframe *oldframe;
787	struct thread *td;
788
789	/* Send EOI first thing. */
790	lapic_eoi();
791
792#if defined(SMP) && !defined(SCHED_ULE)
793	/*
794	 * Don't do any accounting for the disabled HTT cores, since it
795	 * will provide misleading numbers for the userland.
796	 *
797	 * No locking is necessary here, since even if we loose the race
798	 * when hlt_cpus_mask changes it is not a big deal, really.
799	 *
800	 * Don't do that for ULE, since ULE doesn't consider hlt_cpus_mask
801	 * and unlike other schedulers it actually schedules threads to
802	 * those CPUs.
803	 */
804	if ((hlt_cpus_mask & (1 << PCPU_GET(cpuid))) != 0)
805		return;
806#endif
807
808	/* Look up our local APIC structure for the tick counters. */
809	la = &lapics[PCPU_GET(apic_id)];
810	(*la->la_timer_count)++;
811	critical_enter();
812	if (lapic_et.et_active) {
813		td = curthread;
814		oldframe = td->td_intr_frame;
815		td->td_intr_frame = frame;
816		lapic_et.et_event_cb(&lapic_et, lapic_et.et_arg);
817		td->td_intr_frame = oldframe;
818	}
819	critical_exit();
820}
821
822static void
823lapic_timer_set_divisor(u_int divisor)
824{
825
826	KASSERT(powerof2(divisor), ("lapic: invalid divisor %u", divisor));
827	KASSERT(ffs(divisor) <= sizeof(lapic_timer_divisors) /
828	    sizeof(u_int32_t), ("lapic: invalid divisor %u", divisor));
829	lapic->dcr_timer = lapic_timer_divisors[ffs(divisor) - 1];
830}
831
832static void
833lapic_timer_oneshot(u_int count)
834{
835	u_int32_t value;
836
837	value = lapic->lvt_timer;
838	value &= ~APIC_LVTT_TM;
839	value |= APIC_LVTT_TM_ONE_SHOT;
840	lapic->lvt_timer = value;
841	lapic->icr_timer = count;
842}
843
844static void
845lapic_timer_periodic(u_int count)
846{
847	u_int32_t value;
848
849	value = lapic->lvt_timer;
850	value &= ~APIC_LVTT_TM;
851	value |= APIC_LVTT_TM_PERIODIC;
852	lapic->lvt_timer = value;
853	lapic->icr_timer = count;
854}
855
856static void
857lapic_timer_stop(void)
858{
859	u_int32_t value;
860
861	value = lapic->lvt_timer;
862	value &= ~APIC_LVTT_TM;
863	value &= ~APIC_LVT_M;
864	lapic->lvt_timer = value;
865}
866
867static void
868lapic_timer_enable_intr(void)
869{
870	u_int32_t value;
871
872	value = lapic->lvt_timer;
873	value &= ~APIC_LVT_M;
874	lapic->lvt_timer = value;
875}
876
877void
878lapic_handle_cmc(void)
879{
880
881	lapic_eoi();
882	cmc_intr();
883}
884
885/*
886 * Called from the mca_init() to activate the CMC interrupt if this CPU is
887 * responsible for monitoring any MC banks for CMC events.  Since mca_init()
888 * is called prior to lapic_setup() during boot, this just needs to unmask
889 * this CPU's LVT_CMCI entry.
890 */
891void
892lapic_enable_cmc(void)
893{
894	u_int apic_id;
895
896	apic_id = PCPU_GET(apic_id);
897	KASSERT(lapics[apic_id].la_present,
898	    ("%s: missing APIC %u", __func__, apic_id));
899	lapics[apic_id].la_lvts[LVT_CMCI].lvt_masked = 0;
900	lapics[apic_id].la_lvts[LVT_CMCI].lvt_active = 1;
901	if (bootverbose)
902		printf("lapic%u: CMCI unmasked\n", apic_id);
903}
904
905void
906lapic_handle_error(void)
907{
908	u_int32_t esr;
909
910	/*
911	 * Read the contents of the error status register.  Write to
912	 * the register first before reading from it to force the APIC
913	 * to update its value to indicate any errors that have
914	 * occurred since the previous write to the register.
915	 */
916	lapic->esr = 0;
917	esr = lapic->esr;
918
919	printf("CPU%d: local APIC error 0x%x\n", PCPU_GET(cpuid), esr);
920	lapic_eoi();
921}
922
923u_int
924apic_cpuid(u_int apic_id)
925{
926#ifdef SMP
927	return apic_cpuids[apic_id];
928#else
929	return 0;
930#endif
931}
932
933/* Request a free IDT vector to be used by the specified IRQ. */
934u_int
935apic_alloc_vector(u_int apic_id, u_int irq)
936{
937	u_int vector;
938
939	KASSERT(irq < NUM_IO_INTS, ("Invalid IRQ %u", irq));
940
941	/*
942	 * Search for a free vector.  Currently we just use a very simple
943	 * algorithm to find the first free vector.
944	 */
945	mtx_lock_spin(&icu_lock);
946	for (vector = 0; vector < APIC_NUM_IOINTS; vector++) {
947		if (lapics[apic_id].la_ioint_irqs[vector] != -1)
948			continue;
949		lapics[apic_id].la_ioint_irqs[vector] = irq;
950		mtx_unlock_spin(&icu_lock);
951		return (vector + APIC_IO_INTS);
952	}
953	mtx_unlock_spin(&icu_lock);
954	return (0);
955}
956
957/*
958 * Request 'count' free contiguous IDT vectors to be used by 'count'
959 * IRQs.  'count' must be a power of two and the vectors will be
960 * aligned on a boundary of 'align'.  If the request cannot be
961 * satisfied, 0 is returned.
962 */
963u_int
964apic_alloc_vectors(u_int apic_id, u_int *irqs, u_int count, u_int align)
965{
966	u_int first, run, vector;
967
968	KASSERT(powerof2(count), ("bad count"));
969	KASSERT(powerof2(align), ("bad align"));
970	KASSERT(align >= count, ("align < count"));
971#ifdef INVARIANTS
972	for (run = 0; run < count; run++)
973		KASSERT(irqs[run] < NUM_IO_INTS, ("Invalid IRQ %u at index %u",
974		    irqs[run], run));
975#endif
976
977	/*
978	 * Search for 'count' free vectors.  As with apic_alloc_vector(),
979	 * this just uses a simple first fit algorithm.
980	 */
981	run = 0;
982	first = 0;
983	mtx_lock_spin(&icu_lock);
984	for (vector = 0; vector < APIC_NUM_IOINTS; vector++) {
985
986		/* Vector is in use, end run. */
987		if (lapics[apic_id].la_ioint_irqs[vector] != -1) {
988			run = 0;
989			first = 0;
990			continue;
991		}
992
993		/* Start a new run if run == 0 and vector is aligned. */
994		if (run == 0) {
995			if ((vector & (align - 1)) != 0)
996				continue;
997			first = vector;
998		}
999		run++;
1000
1001		/* Keep looping if the run isn't long enough yet. */
1002		if (run < count)
1003			continue;
1004
1005		/* Found a run, assign IRQs and return the first vector. */
1006		for (vector = 0; vector < count; vector++)
1007			lapics[apic_id].la_ioint_irqs[first + vector] =
1008			    irqs[vector];
1009		mtx_unlock_spin(&icu_lock);
1010		return (first + APIC_IO_INTS);
1011	}
1012	mtx_unlock_spin(&icu_lock);
1013	printf("APIC: Couldn't find APIC vectors for %u IRQs\n", count);
1014	return (0);
1015}
1016
1017/*
1018 * Enable a vector for a particular apic_id.  Since all lapics share idt
1019 * entries and ioint_handlers this enables the vector on all lapics.  lapics
1020 * which do not have the vector configured would report spurious interrupts
1021 * should it fire.
1022 */
1023void
1024apic_enable_vector(u_int apic_id, u_int vector)
1025{
1026
1027	KASSERT(vector != IDT_SYSCALL, ("Attempt to overwrite syscall entry"));
1028	KASSERT(ioint_handlers[vector / 32] != NULL,
1029	    ("No ISR handler for vector %u", vector));
1030	setidt(vector, ioint_handlers[vector / 32], SDT_APIC, SEL_KPL,
1031	    GSEL_APIC);
1032}
1033
1034void
1035apic_disable_vector(u_int apic_id, u_int vector)
1036{
1037
1038	KASSERT(vector != IDT_SYSCALL, ("Attempt to overwrite syscall entry"));
1039	KASSERT(ioint_handlers[vector / 32] != NULL,
1040	    ("No ISR handler for vector %u", vector));
1041#ifdef notyet
1042	/*
1043	 * We can not currently clear the idt entry because other cpus
1044	 * may have a valid vector at this offset.
1045	 */
1046	setidt(vector, &IDTVEC(rsvd), SDT_APICT, SEL_KPL, GSEL_APIC);
1047#endif
1048}
1049
1050/* Release an APIC vector when it's no longer in use. */
1051void
1052apic_free_vector(u_int apic_id, u_int vector, u_int irq)
1053{
1054	struct thread *td;
1055
1056	KASSERT(vector >= APIC_IO_INTS && vector != IDT_SYSCALL &&
1057	    vector <= APIC_IO_INTS + APIC_NUM_IOINTS,
1058	    ("Vector %u does not map to an IRQ line", vector));
1059	KASSERT(irq < NUM_IO_INTS, ("Invalid IRQ %u", irq));
1060	KASSERT(lapics[apic_id].la_ioint_irqs[vector - APIC_IO_INTS] ==
1061	    irq, ("IRQ mismatch"));
1062
1063	/*
1064	 * Bind us to the cpu that owned the vector before freeing it so
1065	 * we don't lose an interrupt delivery race.
1066	 */
1067	td = curthread;
1068	if (!rebooting) {
1069		thread_lock(td);
1070		if (sched_is_bound(td))
1071			panic("apic_free_vector: Thread already bound.\n");
1072		sched_bind(td, apic_cpuid(apic_id));
1073		thread_unlock(td);
1074	}
1075	mtx_lock_spin(&icu_lock);
1076	lapics[apic_id].la_ioint_irqs[vector - APIC_IO_INTS] = -1;
1077	mtx_unlock_spin(&icu_lock);
1078	if (!rebooting) {
1079		thread_lock(td);
1080		sched_unbind(td);
1081		thread_unlock(td);
1082	}
1083}
1084
1085/* Map an IDT vector (APIC) to an IRQ (interrupt source). */
1086u_int
1087apic_idt_to_irq(u_int apic_id, u_int vector)
1088{
1089	int irq;
1090
1091	KASSERT(vector >= APIC_IO_INTS && vector != IDT_SYSCALL &&
1092	    vector <= APIC_IO_INTS + APIC_NUM_IOINTS,
1093	    ("Vector %u does not map to an IRQ line", vector));
1094	irq = lapics[apic_id].la_ioint_irqs[vector - APIC_IO_INTS];
1095	if (irq < 0)
1096		irq = 0;
1097	return (irq);
1098}
1099
1100#ifdef DDB
1101/*
1102 * Dump data about APIC IDT vector mappings.
1103 */
1104DB_SHOW_COMMAND(apic, db_show_apic)
1105{
1106	struct intsrc *isrc;
1107	int i, verbose;
1108	u_int apic_id;
1109	u_int irq;
1110
1111	if (strcmp(modif, "vv") == 0)
1112		verbose = 2;
1113	else if (strcmp(modif, "v") == 0)
1114		verbose = 1;
1115	else
1116		verbose = 0;
1117	for (apic_id = 0; apic_id <= MAX_APIC_ID; apic_id++) {
1118		if (lapics[apic_id].la_present == 0)
1119			continue;
1120		db_printf("Interrupts bound to lapic %u\n", apic_id);
1121		for (i = 0; i < APIC_NUM_IOINTS + 1 && !db_pager_quit; i++) {
1122			irq = lapics[apic_id].la_ioint_irqs[i];
1123			if (irq == -1 || irq == IRQ_SYSCALL)
1124				continue;
1125			db_printf("vec 0x%2x -> ", i + APIC_IO_INTS);
1126			if (irq == IRQ_TIMER)
1127				db_printf("lapic timer\n");
1128			else if (irq < NUM_IO_INTS) {
1129				isrc = intr_lookup_source(irq);
1130				if (isrc == NULL || verbose == 0)
1131					db_printf("IRQ %u\n", irq);
1132				else
1133					db_dump_intr_event(isrc->is_event,
1134					    verbose == 2);
1135			} else
1136				db_printf("IRQ %u ???\n", irq);
1137		}
1138	}
1139}
1140
1141static void
1142dump_mask(const char *prefix, uint32_t v, int base)
1143{
1144	int i, first;
1145
1146	first = 1;
1147	for (i = 0; i < 32; i++)
1148		if (v & (1 << i)) {
1149			if (first) {
1150				db_printf("%s:", prefix);
1151				first = 0;
1152			}
1153			db_printf(" %02x", base + i);
1154		}
1155	if (!first)
1156		db_printf("\n");
1157}
1158
1159/* Show info from the lapic regs for this CPU. */
1160DB_SHOW_COMMAND(lapic, db_show_lapic)
1161{
1162	uint32_t v;
1163
1164	db_printf("lapic ID = %d\n", lapic_id());
1165	v = lapic->version;
1166	db_printf("version  = %d.%d\n", (v & APIC_VER_VERSION) >> 4,
1167	    v & 0xf);
1168	db_printf("max LVT  = %d\n", (v & APIC_VER_MAXLVT) >> MAXLVTSHIFT);
1169	v = lapic->svr;
1170	db_printf("SVR      = %02x (%s)\n", v & APIC_SVR_VECTOR,
1171	    v & APIC_SVR_ENABLE ? "enabled" : "disabled");
1172	db_printf("TPR      = %02x\n", lapic->tpr);
1173
1174#define dump_field(prefix, index)					\
1175	dump_mask(__XSTRING(prefix ## index), lapic->prefix ## index,	\
1176	    index * 32)
1177
1178	db_printf("In-service Interrupts:\n");
1179	dump_field(isr, 0);
1180	dump_field(isr, 1);
1181	dump_field(isr, 2);
1182	dump_field(isr, 3);
1183	dump_field(isr, 4);
1184	dump_field(isr, 5);
1185	dump_field(isr, 6);
1186	dump_field(isr, 7);
1187
1188	db_printf("TMR Interrupts:\n");
1189	dump_field(tmr, 0);
1190	dump_field(tmr, 1);
1191	dump_field(tmr, 2);
1192	dump_field(tmr, 3);
1193	dump_field(tmr, 4);
1194	dump_field(tmr, 5);
1195	dump_field(tmr, 6);
1196	dump_field(tmr, 7);
1197
1198	db_printf("IRR Interrupts:\n");
1199	dump_field(irr, 0);
1200	dump_field(irr, 1);
1201	dump_field(irr, 2);
1202	dump_field(irr, 3);
1203	dump_field(irr, 4);
1204	dump_field(irr, 5);
1205	dump_field(irr, 6);
1206	dump_field(irr, 7);
1207
1208#undef dump_field
1209}
1210#endif
1211
1212/*
1213 * APIC probing support code.  This includes code to manage enumerators.
1214 */
1215
1216static SLIST_HEAD(, apic_enumerator) enumerators =
1217	SLIST_HEAD_INITIALIZER(enumerators);
1218static struct apic_enumerator *best_enum;
1219
1220void
1221apic_register_enumerator(struct apic_enumerator *enumerator)
1222{
1223#ifdef INVARIANTS
1224	struct apic_enumerator *apic_enum;
1225
1226	SLIST_FOREACH(apic_enum, &enumerators, apic_next) {
1227		if (apic_enum == enumerator)
1228			panic("%s: Duplicate register of %s", __func__,
1229			    enumerator->apic_name);
1230	}
1231#endif
1232	SLIST_INSERT_HEAD(&enumerators, enumerator, apic_next);
1233}
1234
1235/*
1236 * We have to look for CPU's very, very early because certain subsystems
1237 * want to know how many CPU's we have extremely early on in the boot
1238 * process.
1239 */
1240static void
1241apic_init(void *dummy __unused)
1242{
1243	struct apic_enumerator *enumerator;
1244#ifndef __amd64__
1245	uint64_t apic_base;
1246#endif
1247	int retval, best;
1248
1249	/* We only support built in local APICs. */
1250	if (!(cpu_feature & CPUID_APIC))
1251		return;
1252
1253	/* Don't probe if APIC mode is disabled. */
1254	if (resource_disabled("apic", 0))
1255		return;
1256
1257	/* First, probe all the enumerators to find the best match. */
1258	best_enum = NULL;
1259	best = 0;
1260	SLIST_FOREACH(enumerator, &enumerators, apic_next) {
1261		retval = enumerator->apic_probe();
1262		if (retval > 0)
1263			continue;
1264		if (best_enum == NULL || best < retval) {
1265			best_enum = enumerator;
1266			best = retval;
1267		}
1268	}
1269	if (best_enum == NULL) {
1270		if (bootverbose)
1271			printf("APIC: Could not find any APICs.\n");
1272		return;
1273	}
1274
1275	if (bootverbose)
1276		printf("APIC: Using the %s enumerator.\n",
1277		    best_enum->apic_name);
1278
1279#ifndef __amd64__
1280	/*
1281	 * To work around an errata, we disable the local APIC on some
1282	 * CPUs during early startup.  We need to turn the local APIC back
1283	 * on on such CPUs now.
1284	 */
1285	if (cpu == CPU_686 && cpu_vendor_id == CPU_VENDOR_INTEL &&
1286	    (cpu_id & 0xff0) == 0x610) {
1287		apic_base = rdmsr(MSR_APICBASE);
1288		apic_base |= APICBASE_ENABLED;
1289		wrmsr(MSR_APICBASE, apic_base);
1290	}
1291#endif
1292
1293	/* Second, probe the CPU's in the system. */
1294	retval = best_enum->apic_probe_cpus();
1295	if (retval != 0)
1296		printf("%s: Failed to probe CPUs: returned %d\n",
1297		    best_enum->apic_name, retval);
1298
1299#ifdef __amd64__
1300}
1301SYSINIT(apic_init, SI_SUB_TUNABLES - 1, SI_ORDER_SECOND, apic_init, NULL);
1302
1303/*
1304 * Setup the local APIC.  We have to do this prior to starting up the APs
1305 * in the SMP case.
1306 */
1307static void
1308apic_setup_local(void *dummy __unused)
1309{
1310	int retval;
1311
1312	if (best_enum == NULL)
1313		return;
1314#endif
1315	/* Third, initialize the local APIC. */
1316	retval = best_enum->apic_setup_local();
1317	if (retval != 0)
1318		printf("%s: Failed to setup the local APIC: returned %d\n",
1319		    best_enum->apic_name, retval);
1320}
1321#ifdef __amd64__
1322SYSINIT(apic_setup_local, SI_SUB_CPU, SI_ORDER_SECOND, apic_setup_local,
1323    NULL);
1324#else
1325SYSINIT(apic_init, SI_SUB_CPU, SI_ORDER_SECOND, apic_init, NULL);
1326#endif
1327
1328/*
1329 * Setup the I/O APICs.
1330 */
1331static void
1332apic_setup_io(void *dummy __unused)
1333{
1334	int retval;
1335
1336	if (best_enum == NULL)
1337		return;
1338	retval = best_enum->apic_setup_io();
1339	if (retval != 0)
1340		printf("%s: Failed to setup I/O APICs: returned %d\n",
1341		    best_enum->apic_name, retval);
1342
1343#ifdef XEN
1344	return;
1345#endif
1346	/*
1347	 * Finish setting up the local APIC on the BSP once we know how to
1348	 * properly program the LINT pins.
1349	 */
1350	lapic_setup(1);
1351	intr_register_pic(&lapic_pic);
1352	if (bootverbose)
1353		lapic_dump("BSP");
1354
1355	/* Enable the MSI "pic". */
1356	msi_init();
1357}
1358SYSINIT(apic_setup_io, SI_SUB_INTR, SI_ORDER_SECOND, apic_setup_io, NULL);
1359
1360#ifdef SMP
1361/*
1362 * Inter Processor Interrupt functions.  The lapic_ipi_*() functions are
1363 * private to the MD code.  The public interface for the rest of the
1364 * kernel is defined in mp_machdep.c.
1365 */
1366int
1367lapic_ipi_wait(int delay)
1368{
1369	int x, incr;
1370
1371	/*
1372	 * Wait delay loops for IPI to be sent.  This is highly bogus
1373	 * since this is sensitive to CPU clock speed.  If delay is
1374	 * -1, we wait forever.
1375	 */
1376	if (delay == -1) {
1377		incr = 0;
1378		delay = 1;
1379	} else
1380		incr = 1;
1381	for (x = 0; x < delay; x += incr) {
1382		if ((lapic->icr_lo & APIC_DELSTAT_MASK) == APIC_DELSTAT_IDLE)
1383			return (1);
1384		ia32_pause();
1385	}
1386	return (0);
1387}
1388
1389void
1390lapic_ipi_raw(register_t icrlo, u_int dest)
1391{
1392	register_t value, eflags;
1393
1394	/* XXX: Need more sanity checking of icrlo? */
1395	KASSERT(lapic != NULL, ("%s called too early", __func__));
1396	KASSERT((dest & ~(APIC_ID_MASK >> APIC_ID_SHIFT)) == 0,
1397	    ("%s: invalid dest field", __func__));
1398	KASSERT((icrlo & APIC_ICRLO_RESV_MASK) == 0,
1399	    ("%s: reserved bits set in ICR LO register", __func__));
1400
1401	/* Set destination in ICR HI register if it is being used. */
1402	eflags = intr_disable();
1403	if ((icrlo & APIC_DEST_MASK) == APIC_DEST_DESTFLD) {
1404		value = lapic->icr_hi;
1405		value &= ~APIC_ID_MASK;
1406		value |= dest << APIC_ID_SHIFT;
1407		lapic->icr_hi = value;
1408	}
1409
1410	/* Program the contents of the IPI and dispatch it. */
1411	value = lapic->icr_lo;
1412	value &= APIC_ICRLO_RESV_MASK;
1413	value |= icrlo;
1414	lapic->icr_lo = value;
1415	intr_restore(eflags);
1416}
1417
1418#define	BEFORE_SPIN	1000000
1419#ifdef DETECT_DEADLOCK
1420#define	AFTER_SPIN	1000
1421#endif
1422
1423void
1424lapic_ipi_vectored(u_int vector, int dest)
1425{
1426	register_t icrlo, destfield;
1427
1428	KASSERT((vector & ~APIC_VECTOR_MASK) == 0,
1429	    ("%s: invalid vector %d", __func__, vector));
1430
1431	icrlo = APIC_DESTMODE_PHY | APIC_TRIGMOD_EDGE;
1432
1433	/*
1434	 * IPI_STOP_HARD is just a "fake" vector used to send a NMI.
1435	 * Use special rules regard NMI if passed, otherwise specify
1436	 * the vector.
1437	 */
1438	if (vector == IPI_STOP_HARD)
1439		icrlo |= APIC_DELMODE_NMI | APIC_LEVEL_ASSERT;
1440	else
1441		icrlo |= vector | APIC_DELMODE_FIXED | APIC_LEVEL_DEASSERT;
1442	destfield = 0;
1443	switch (dest) {
1444	case APIC_IPI_DEST_SELF:
1445		icrlo |= APIC_DEST_SELF;
1446		break;
1447	case APIC_IPI_DEST_ALL:
1448		icrlo |= APIC_DEST_ALLISELF;
1449		break;
1450	case APIC_IPI_DEST_OTHERS:
1451		icrlo |= APIC_DEST_ALLESELF;
1452		break;
1453	default:
1454		KASSERT((dest & ~(APIC_ID_MASK >> APIC_ID_SHIFT)) == 0,
1455		    ("%s: invalid destination 0x%x", __func__, dest));
1456		destfield = dest;
1457	}
1458
1459	/* Wait for an earlier IPI to finish. */
1460	if (!lapic_ipi_wait(BEFORE_SPIN)) {
1461		if (panicstr != NULL)
1462			return;
1463		else
1464			panic("APIC: Previous IPI is stuck");
1465	}
1466
1467	lapic_ipi_raw(icrlo, destfield);
1468
1469#ifdef DETECT_DEADLOCK
1470	/* Wait for IPI to be delivered. */
1471	if (!lapic_ipi_wait(AFTER_SPIN)) {
1472#ifdef needsattention
1473		/*
1474		 * XXX FIXME:
1475		 *
1476		 * The above function waits for the message to actually be
1477		 * delivered.  It breaks out after an arbitrary timeout
1478		 * since the message should eventually be delivered (at
1479		 * least in theory) and that if it wasn't we would catch
1480		 * the failure with the check above when the next IPI is
1481		 * sent.
1482		 *
1483		 * We could skip this wait entirely, EXCEPT it probably
1484		 * protects us from other routines that assume that the
1485		 * message was delivered and acted upon when this function
1486		 * returns.
1487		 */
1488		printf("APIC: IPI might be stuck\n");
1489#else /* !needsattention */
1490		/* Wait until mesage is sent without a timeout. */
1491		while (lapic->icr_lo & APIC_DELSTAT_PEND)
1492			ia32_pause();
1493#endif /* needsattention */
1494	}
1495#endif /* DETECT_DEADLOCK */
1496}
1497#endif /* SMP */
1498