local_apic.c revision 150176
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/i386/i386/local_apic.c 150176 2005-09-15 19:02:01Z jhb $");
36
37#include "opt_hwpmc_hooks.h"
38
39#include <sys/param.h>
40#include <sys/systm.h>
41#include <sys/bus.h>
42#include <sys/kernel.h>
43#include <sys/pcpu.h>
44#include <sys/smp.h>
45
46#include <vm/vm.h>
47#include <vm/pmap.h>
48
49#include <machine/apicreg.h>
50#include <machine/cputypes.h>
51#include <machine/frame.h>
52#include <machine/intr_machdep.h>
53#include <machine/apicvar.h>
54#include <machine/md_var.h>
55#include <machine/smp.h>
56#include <machine/specialreg.h>
57
58/*
59 * We can handle up to 60 APICs via our logical cluster IDs, but currently
60 * the physical IDs on Intel processors up to the Pentium 4 are limited to
61 * 16.
62 */
63#define	MAX_APICID	16
64
65/* Sanity checks on IDT vectors. */
66CTASSERT(APIC_IO_INTS + APIC_NUM_IOINTS == APIC_TIMER_INT);
67CTASSERT(APIC_TIMER_INT < APIC_LOCAL_INTS);
68CTASSERT(APIC_LOCAL_INTS == 240);
69CTASSERT(IPI_STOP < APIC_SPURIOUS_INT);
70
71#define	LAPIC_TIMER_HZ_DIVIDER		2
72#define	LAPIC_TIMER_STATHZ_DIVIDER	15
73#define	LAPIC_TIMER_PROFHZ_DIVIDER	3
74
75/*
76 * Support for local APICs.  Local APICs manage interrupts on each
77 * individual processor as opposed to I/O APICs which receive interrupts
78 * from I/O devices and then forward them on to the local APICs.
79 *
80 * Local APICs can also send interrupts to each other thus providing the
81 * mechanism for IPIs.
82 */
83
84struct lvt {
85	u_int lvt_edgetrigger:1;
86	u_int lvt_activehi:1;
87	u_int lvt_masked:1;
88	u_int lvt_active:1;
89	u_int lvt_mode:16;
90	u_int lvt_vector:8;
91};
92
93struct lapic {
94	struct lvt la_lvts[LVT_MAX + 1];
95	u_int la_id:8;
96	u_int la_cluster:4;
97	u_int la_cluster_id:2;
98	u_int la_present:1;
99	u_long *la_timer_count;
100	u_long la_hard_ticks;
101	u_long la_stat_ticks;
102	u_long la_prof_ticks;
103} static lapics[MAX_APICID];
104
105/* XXX: should thermal be an NMI? */
106
107/* Global defaults for local APIC LVT entries. */
108static struct lvt lvts[LVT_MAX + 1] = {
109	{ 1, 1, 1, 1, APIC_LVT_DM_EXTINT, 0 },	/* LINT0: masked ExtINT */
110	{ 1, 1, 0, 1, APIC_LVT_DM_NMI, 0 },	/* LINT1: NMI */
111	{ 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_TIMER_INT },	/* Timer */
112	{ 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_ERROR_INT },	/* Error */
113	{ 1, 1, 0, 1, APIC_LVT_DM_NMI, 0 },	/* PMC */
114	{ 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_THERMAL_INT },	/* Thermal */
115};
116
117static inthand_t *ioint_handlers[] = {
118	NULL,			/* 0 - 31 */
119	IDTVEC(apic_isr1),	/* 32 - 63 */
120	IDTVEC(apic_isr2),	/* 64 - 95 */
121	IDTVEC(apic_isr3),	/* 96 - 127 */
122	IDTVEC(apic_isr4),	/* 128 - 159 */
123	IDTVEC(apic_isr5),	/* 160 - 191 */
124	IDTVEC(apic_isr6),	/* 192 - 223 */
125	IDTVEC(apic_isr7),	/* 224 - 255 */
126};
127
128static u_int32_t lapic_timer_divisors[] = {
129	APIC_TDCR_1, APIC_TDCR_2, APIC_TDCR_4, APIC_TDCR_8, APIC_TDCR_16,
130	APIC_TDCR_32, APIC_TDCR_64, APIC_TDCR_128
131};
132
133volatile lapic_t *lapic;
134static u_long lapic_timer_divisor, lapic_timer_period, lapic_timer_hz;
135
136static void	lapic_enable(void);
137static void	lapic_timer_enable_intr(void);
138static void	lapic_timer_oneshot(u_int count);
139static void	lapic_timer_periodic(u_int count);
140static void	lapic_timer_set_divisor(u_int divisor);
141static uint32_t	lvt_mode(struct lapic *la, u_int pin, uint32_t value);
142
143static uint32_t
144lvt_mode(struct lapic *la, u_int pin, uint32_t value)
145{
146	struct lvt *lvt;
147
148	KASSERT(pin <= LVT_MAX, ("%s: pin %u out of range", __func__, pin));
149	if (la->la_lvts[pin].lvt_active)
150		lvt = &la->la_lvts[pin];
151	else
152		lvt = &lvts[pin];
153
154	value &= ~(APIC_LVT_M | APIC_LVT_TM | APIC_LVT_IIPP | APIC_LVT_DM |
155	    APIC_LVT_VECTOR);
156	if (lvt->lvt_edgetrigger == 0)
157		value |= APIC_LVT_TM;
158	if (lvt->lvt_activehi == 0)
159		value |= APIC_LVT_IIPP_INTALO;
160	if (lvt->lvt_masked)
161		value |= APIC_LVT_M;
162	value |= lvt->lvt_mode;
163	switch (lvt->lvt_mode) {
164	case APIC_LVT_DM_NMI:
165	case APIC_LVT_DM_SMI:
166	case APIC_LVT_DM_INIT:
167	case APIC_LVT_DM_EXTINT:
168		if (!lvt->lvt_edgetrigger) {
169			printf("lapic%u: Forcing LINT%u to edge trigger\n",
170			    la->la_id, pin);
171			value |= APIC_LVT_TM;
172		}
173		/* Use a vector of 0. */
174		break;
175	case APIC_LVT_DM_FIXED:
176		value |= lvt->lvt_vector;
177		break;
178	default:
179		panic("bad APIC LVT delivery mode: %#x\n", value);
180	}
181	return (value);
182}
183
184/*
185 * Map the local APIC and setup necessary interrupt vectors.
186 */
187void
188lapic_init(uintptr_t addr)
189{
190
191	/* Map the local APIC and setup the spurious interrupt handler. */
192	KASSERT(trunc_page(addr) == addr,
193	    ("local APIC not aligned on a page boundary"));
194	lapic = (lapic_t *)pmap_mapdev(addr, sizeof(lapic_t));
195	setidt(APIC_SPURIOUS_INT, IDTVEC(spuriousint), SDT_SYS386IGT, SEL_KPL,
196	    GSEL(GCODE_SEL, SEL_KPL));
197
198	/* Perform basic initialization of the BSP's local APIC. */
199	lapic_enable();
200
201	/* Set BSP's per-CPU local APIC ID. */
202	PCPU_SET(apic_id, lapic_id());
203
204	/* Local APIC timer interrupt. */
205	setidt(APIC_TIMER_INT, IDTVEC(timerint), SDT_SYS386IGT, SEL_KPL,
206	    GSEL(GCODE_SEL, SEL_KPL));
207
208	/* XXX: error/thermal interrupts */
209}
210
211/*
212 * Create a local APIC instance.
213 */
214void
215lapic_create(u_int apic_id, int boot_cpu)
216{
217	int i;
218
219	if (apic_id >= MAX_APICID) {
220		printf("APIC: Ignoring local APIC with ID %d\n", apic_id);
221		if (boot_cpu)
222			panic("Can't ignore BSP");
223		return;
224	}
225	KASSERT(!lapics[apic_id].la_present, ("duplicate local APIC %u",
226	    apic_id));
227
228	/*
229	 * Assume no local LVT overrides and a cluster of 0 and
230	 * intra-cluster ID of 0.
231	 */
232	lapics[apic_id].la_present = 1;
233	lapics[apic_id].la_id = apic_id;
234	for (i = 0; i < LVT_MAX; i++) {
235		lapics[apic_id].la_lvts[i] = lvts[i];
236		lapics[apic_id].la_lvts[i].lvt_active = 0;
237	}
238
239#ifdef SMP
240	cpu_add(apic_id, boot_cpu);
241#endif
242}
243
244/*
245 * Dump contents of local APIC registers
246 */
247void
248lapic_dump(const char* str)
249{
250
251	printf("cpu%d %s:\n", PCPU_GET(cpuid), str);
252	printf("     ID: 0x%08x   VER: 0x%08x LDR: 0x%08x DFR: 0x%08x\n",
253	    lapic->id, lapic->version, lapic->ldr, lapic->dfr);
254	printf("  lint0: 0x%08x lint1: 0x%08x TPR: 0x%08x SVR: 0x%08x\n",
255	    lapic->lvt_lint0, lapic->lvt_lint1, lapic->tpr, lapic->svr);
256	printf("  timer: 0x%08x therm: 0x%08x err: 0x%08x pcm: 0x%08x\n",
257	    lapic->lvt_timer, lapic->lvt_thermal, lapic->lvt_error,
258	    lapic->lvt_pcint);
259}
260
261void
262lapic_enable_intr(u_int irq)
263{
264	u_int vector;
265
266	vector = apic_irq_to_idt(irq);
267	KASSERT(vector != IDT_SYSCALL, ("Attempt to overwrite syscall entry"));
268	KASSERT(ioint_handlers[vector / 32] != NULL,
269	    ("No ISR handler for IRQ %u", irq));
270	setidt(vector, ioint_handlers[vector / 32], SDT_SYS386IGT, SEL_KPL,
271	    GSEL(GCODE_SEL, SEL_KPL));
272}
273
274void
275lapic_setup(void)
276{
277	struct lapic *la;
278	u_int32_t value, maxlvt;
279	register_t eflags;
280	char buf[MAXCOMLEN + 1];
281
282	la = &lapics[lapic_id()];
283	KASSERT(la->la_present, ("missing APIC structure"));
284	eflags = intr_disable();
285	maxlvt = (lapic->version & APIC_VER_MAXLVT) >> MAXLVTSHIFT;
286
287	/* Initialize the TPR to allow all interrupts. */
288	lapic_set_tpr(0);
289
290	/* Use the cluster model for logical IDs. */
291	value = lapic->dfr;
292	value &= ~APIC_DFR_MODEL_MASK;
293	value |= APIC_DFR_MODEL_CLUSTER;
294	lapic->dfr = value;
295
296	/* Set this APIC's logical ID. */
297	value = lapic->ldr;
298	value &= ~APIC_ID_MASK;
299	value |= (la->la_cluster << APIC_ID_CLUSTER_SHIFT |
300	    1 << la->la_cluster_id) << APIC_ID_SHIFT;
301	lapic->ldr = value;
302
303	/* Setup spurious vector and enable the local APIC. */
304	lapic_enable();
305
306	/* Program LINT[01] LVT entries. */
307	lapic->lvt_lint0 = lvt_mode(la, LVT_LINT0, lapic->lvt_lint0);
308	lapic->lvt_lint1 = lvt_mode(la, LVT_LINT1, lapic->lvt_lint1);
309#ifdef	HWPMC_HOOKS
310	/* Program the PMC LVT entry if present. */
311	if (maxlvt >= LVT_PMC)
312		lapic->lvt_pcint = lvt_mode(la, LVT_PMC, lapic->lvt_pcint);
313#endif
314
315	/* Program timer LVT and setup handler. */
316	lapic->lvt_timer = lvt_mode(la, LVT_TIMER, lapic->lvt_timer);
317	snprintf(buf, sizeof(buf), "lapic%d: timer", lapic_id());
318	intrcnt_add(buf, &la->la_timer_count);
319	if (PCPU_GET(cpuid) != 0) {
320		KASSERT(lapic_timer_period != 0, ("lapic%u: zero divisor",
321		    lapic_id()));
322		lapic_timer_set_divisor(lapic_timer_divisor);
323		lapic_timer_periodic(lapic_timer_period);
324		lapic_timer_enable_intr();
325	}
326
327	/* XXX: Error and thermal LVTs */
328
329	intr_restore(eflags);
330}
331
332/*
333 * Called by cpu_initclocks() on the BSP to setup the local APIC timer so
334 * that it can drive hardclock, statclock, and profclock.  This function
335 * returns true if it is able to use the local APIC timer to drive the
336 * clocks and false if it is not able.
337 */
338int
339lapic_setup_clock(void)
340{
341	u_long value;
342
343	/* Can't drive the timer without a local APIC. */
344	if (lapic == NULL)
345		return (0);
346
347	/* Start off with a divisor of 2 (power on reset default). */
348	lapic_timer_divisor = 2;
349
350	/* Try to calibrate the local APIC timer. */
351	do {
352		lapic_timer_set_divisor(lapic_timer_divisor);
353		lapic_timer_oneshot(APIC_TIMER_MAX_COUNT);
354		DELAY(2000000);
355		value = APIC_TIMER_MAX_COUNT - lapic->ccr_timer;
356		if (value != APIC_TIMER_MAX_COUNT)
357			break;
358		lapic_timer_divisor <<= 1;
359	} while (lapic_timer_divisor <= 128);
360	if (lapic_timer_divisor > 128)
361		panic("lapic: Divisor too big");
362	value /= 2;
363	if (bootverbose)
364		printf("lapic: Divisor %lu, Frequency %lu hz\n",
365		    lapic_timer_divisor, value);
366
367	/*
368	 * We will drive the timer at a small multiple of hz and drive
369	 * both of the other timers with similarly small but relatively
370	 * prime divisors.
371	 */
372	lapic_timer_hz = hz * LAPIC_TIMER_HZ_DIVIDER;
373	stathz = lapic_timer_hz / LAPIC_TIMER_STATHZ_DIVIDER;
374	profhz = lapic_timer_hz / LAPIC_TIMER_PROFHZ_DIVIDER;
375	lapic_timer_period = value / lapic_timer_hz;
376
377	/*
378	 * Start up the timer on the BSP.  The APs will kick off their
379	 * timer during lapic_setup().
380	 */
381	lapic_timer_periodic(lapic_timer_period);
382	lapic_timer_enable_intr();
383	return (1);
384}
385
386void
387lapic_disable(void)
388{
389	uint32_t value;
390
391	/* Software disable the local APIC. */
392	value = lapic->svr;
393	value &= ~APIC_SVR_SWEN;
394	lapic->svr = value;
395}
396
397static void
398lapic_enable(void)
399{
400	u_int32_t value;
401
402	/* Program the spurious vector to enable the local APIC. */
403	value = lapic->svr;
404	value &= ~(APIC_SVR_VECTOR | APIC_SVR_FOCUS);
405	value |= (APIC_SVR_FEN | APIC_SVR_SWEN | APIC_SPURIOUS_INT);
406	lapic->svr = value;
407}
408
409int
410lapic_id(void)
411{
412
413	KASSERT(lapic != NULL, ("local APIC is not mapped"));
414	return (lapic->id >> APIC_ID_SHIFT);
415}
416
417int
418lapic_intr_pending(u_int vector)
419{
420	volatile u_int32_t *irr;
421
422	/*
423	 * The IRR registers are an array of 128-bit registers each of
424	 * which only describes 32 interrupts in the low 32 bits..  Thus,
425	 * we divide the vector by 32 to get the 128-bit index.  We then
426	 * multiply that index by 4 to get the equivalent index from
427	 * treating the IRR as an array of 32-bit registers.  Finally, we
428	 * modulus the vector by 32 to determine the individual bit to
429	 * test.
430	 */
431	irr = &lapic->irr0;
432	return (irr[(vector / 32) * 4] & 1 << (vector % 32));
433}
434
435void
436lapic_set_logical_id(u_int apic_id, u_int cluster, u_int cluster_id)
437{
438	struct lapic *la;
439
440	KASSERT(lapics[apic_id].la_present, ("%s: APIC %u doesn't exist",
441	    __func__, apic_id));
442	KASSERT(cluster <= APIC_MAX_CLUSTER, ("%s: cluster %u too big",
443	    __func__, cluster));
444	KASSERT(cluster_id <= APIC_MAX_INTRACLUSTER_ID,
445	    ("%s: intra cluster id %u too big", __func__, cluster_id));
446	la = &lapics[apic_id];
447	la->la_cluster = cluster;
448	la->la_cluster_id = cluster_id;
449}
450
451int
452lapic_set_lvt_mask(u_int apic_id, u_int pin, u_char masked)
453{
454
455	if (pin > LVT_MAX)
456		return (EINVAL);
457	if (apic_id == APIC_ID_ALL) {
458		lvts[pin].lvt_masked = masked;
459		if (bootverbose)
460			printf("lapic:");
461	} else {
462		KASSERT(lapics[apic_id].la_present,
463		    ("%s: missing APIC %u", __func__, apic_id));
464		lapics[apic_id].la_lvts[pin].lvt_masked = masked;
465		lapics[apic_id].la_lvts[pin].lvt_active = 1;
466		if (bootverbose)
467			printf("lapic%u:", apic_id);
468	}
469	if (bootverbose)
470		printf(" LINT%u %s\n", pin, masked ? "masked" : "unmasked");
471	return (0);
472}
473
474int
475lapic_set_lvt_mode(u_int apic_id, u_int pin, u_int32_t mode)
476{
477	struct lvt *lvt;
478
479	if (pin > LVT_MAX)
480		return (EINVAL);
481	if (apic_id == APIC_ID_ALL) {
482		lvt = &lvts[pin];
483		if (bootverbose)
484			printf("lapic:");
485	} else {
486		KASSERT(lapics[apic_id].la_present,
487		    ("%s: missing APIC %u", __func__, apic_id));
488		lvt = &lapics[apic_id].la_lvts[pin];
489		lvt->lvt_active = 1;
490		if (bootverbose)
491			printf("lapic%u:", apic_id);
492	}
493	lvt->lvt_mode = mode;
494	switch (mode) {
495	case APIC_LVT_DM_NMI:
496	case APIC_LVT_DM_SMI:
497	case APIC_LVT_DM_INIT:
498	case APIC_LVT_DM_EXTINT:
499		lvt->lvt_edgetrigger = 1;
500		lvt->lvt_activehi = 1;
501		if (mode == APIC_LVT_DM_EXTINT)
502			lvt->lvt_masked = 1;
503		else
504			lvt->lvt_masked = 0;
505		break;
506	default:
507		panic("Unsupported delivery mode: 0x%x\n", mode);
508	}
509	if (bootverbose) {
510		printf(" Routing ");
511		switch (mode) {
512		case APIC_LVT_DM_NMI:
513			printf("NMI");
514			break;
515		case APIC_LVT_DM_SMI:
516			printf("SMI");
517			break;
518		case APIC_LVT_DM_INIT:
519			printf("INIT");
520			break;
521		case APIC_LVT_DM_EXTINT:
522			printf("ExtINT");
523			break;
524		}
525		printf(" -> LINT%u\n", pin);
526	}
527	return (0);
528}
529
530int
531lapic_set_lvt_polarity(u_int apic_id, u_int pin, enum intr_polarity pol)
532{
533
534	if (pin > LVT_MAX || pol == INTR_POLARITY_CONFORM)
535		return (EINVAL);
536	if (apic_id == APIC_ID_ALL) {
537		lvts[pin].lvt_activehi = (pol == INTR_POLARITY_HIGH);
538		if (bootverbose)
539			printf("lapic:");
540	} else {
541		KASSERT(lapics[apic_id].la_present,
542		    ("%s: missing APIC %u", __func__, apic_id));
543		lapics[apic_id].la_lvts[pin].lvt_active = 1;
544		lapics[apic_id].la_lvts[pin].lvt_activehi =
545		    (pol == INTR_POLARITY_HIGH);
546		if (bootverbose)
547			printf("lapic%u:", apic_id);
548	}
549	if (bootverbose)
550		printf(" LINT%u polarity: %s\n", pin,
551		    pol == INTR_POLARITY_HIGH ? "high" : "low");
552	return (0);
553}
554
555int
556lapic_set_lvt_triggermode(u_int apic_id, u_int pin, enum intr_trigger trigger)
557{
558
559	if (pin > LVT_MAX || trigger == INTR_TRIGGER_CONFORM)
560		return (EINVAL);
561	if (apic_id == APIC_ID_ALL) {
562		lvts[pin].lvt_edgetrigger = (trigger == INTR_TRIGGER_EDGE);
563		if (bootverbose)
564			printf("lapic:");
565	} else {
566		KASSERT(lapics[apic_id].la_present,
567		    ("%s: missing APIC %u", __func__, apic_id));
568		lapics[apic_id].la_lvts[pin].lvt_edgetrigger =
569		    (trigger == INTR_TRIGGER_EDGE);
570		lapics[apic_id].la_lvts[pin].lvt_active = 1;
571		if (bootverbose)
572			printf("lapic%u:", apic_id);
573	}
574	if (bootverbose)
575		printf(" LINT%u trigger: %s\n", pin,
576		    trigger == INTR_TRIGGER_EDGE ? "edge" : "level");
577	return (0);
578}
579
580/*
581 * Adjust the TPR of the current CPU so that it blocks all interrupts below
582 * the passed in vector.
583 */
584void
585lapic_set_tpr(u_int vector)
586{
587#ifdef CHEAP_TPR
588	lapic->tpr = vector;
589#else
590	u_int32_t tpr;
591
592	tpr = lapic->tpr & ~APIC_TPR_PRIO;
593	tpr |= vector;
594	lapic->tpr = tpr;
595#endif
596}
597
598void
599lapic_eoi(void)
600{
601
602	lapic->eoi = 0;
603}
604
605void
606lapic_handle_intr(struct intrframe frame)
607{
608	struct intsrc *isrc;
609
610	if (frame.if_vec == -1)
611		panic("Couldn't get vector from ISR!");
612	isrc = intr_lookup_source(apic_idt_to_irq(frame.if_vec));
613	intr_execute_handlers(isrc, &frame);
614}
615
616void
617lapic_handle_timer(struct clockframe frame)
618{
619	struct lapic *la;
620
621	la = &lapics[PCPU_GET(apic_id)];
622	(*la->la_timer_count)++;
623	critical_enter();
624
625	/* Fire hardclock at hz. */
626	la->la_hard_ticks += hz;
627	if (la->la_hard_ticks >= lapic_timer_hz) {
628		la->la_hard_ticks -= lapic_timer_hz;
629		if (PCPU_GET(cpuid) == 0)
630			hardclock(&frame);
631		else
632			hardclock_process(&frame);
633	}
634
635	/* Fire statclock at stathz. */
636	la->la_stat_ticks += stathz;
637	if (la->la_stat_ticks >= lapic_timer_hz) {
638		la->la_stat_ticks -= lapic_timer_hz;
639		statclock(&frame);
640	}
641
642	/* Fire profclock at profhz, but only when needed. */
643	la->la_prof_ticks += profhz;
644	if (la->la_prof_ticks >= lapic_timer_hz) {
645		la->la_prof_ticks -= lapic_timer_hz;
646		if (profprocs != 0)
647			profclock(&frame);
648	}
649	critical_exit();
650}
651
652static void
653lapic_timer_set_divisor(u_int divisor)
654{
655
656	KASSERT(powerof2(divisor), ("lapic: invalid divisor %u", divisor));
657	KASSERT(ffs(divisor) <= sizeof(lapic_timer_divisors) /
658	    sizeof(u_int32_t), ("lapic: invalid divisor %u", divisor));
659	lapic->dcr_timer = lapic_timer_divisors[ffs(divisor) - 1];
660}
661
662static void
663lapic_timer_oneshot(u_int count)
664{
665	u_int32_t value;
666
667	value = lapic->lvt_timer;
668	value &= ~APIC_LVTT_TM;
669	value |= APIC_LVTT_TM_ONE_SHOT;
670	lapic->lvt_timer = value;
671	lapic->icr_timer = count;
672}
673
674static void
675lapic_timer_periodic(u_int count)
676{
677	u_int32_t value;
678
679	value = lapic->lvt_timer;
680	value &= ~APIC_LVTT_TM;
681	value |= APIC_LVTT_TM_PERIODIC;
682	lapic->lvt_timer = value;
683	lapic->icr_timer = count;
684}
685
686static void
687lapic_timer_enable_intr(void)
688{
689	u_int32_t value;
690
691	value = lapic->lvt_timer;
692	value &= ~APIC_LVT_M;
693	lapic->lvt_timer = value;
694}
695
696/* Translate between IDT vectors and IRQ vectors. */
697u_int
698apic_irq_to_idt(u_int irq)
699{
700	u_int vector;
701
702	KASSERT(irq < NUM_IO_INTS, ("Invalid IRQ %u", irq));
703	vector = irq + APIC_IO_INTS;
704	if (vector >= IDT_SYSCALL)
705		vector++;
706	return (vector);
707}
708
709u_int
710apic_idt_to_irq(u_int vector)
711{
712
713	KASSERT(vector >= APIC_IO_INTS && vector != IDT_SYSCALL &&
714	    vector <= APIC_IO_INTS + NUM_IO_INTS,
715	    ("Vector %u does not map to an IRQ line", vector));
716	if (vector > IDT_SYSCALL)
717		vector--;
718	return (vector - APIC_IO_INTS);
719}
720
721/*
722 * APIC probing support code.  This includes code to manage enumerators.
723 */
724
725static SLIST_HEAD(, apic_enumerator) enumerators =
726	SLIST_HEAD_INITIALIZER(enumerators);
727static struct apic_enumerator *best_enum;
728
729void
730apic_register_enumerator(struct apic_enumerator *enumerator)
731{
732#ifdef INVARIANTS
733	struct apic_enumerator *apic_enum;
734
735	SLIST_FOREACH(apic_enum, &enumerators, apic_next) {
736		if (apic_enum == enumerator)
737			panic("%s: Duplicate register of %s", __func__,
738			    enumerator->apic_name);
739	}
740#endif
741	SLIST_INSERT_HEAD(&enumerators, enumerator, apic_next);
742}
743
744/*
745 * Probe the APIC enumerators, enumerate CPUs, and initialize the
746 * local APIC.
747 */
748static void
749apic_init(void *dummy __unused)
750{
751	struct apic_enumerator *enumerator;
752	uint64_t apic_base;
753	int retval, best;
754
755	/* We only support built in local APICs. */
756	if (!(cpu_feature & CPUID_APIC))
757		return;
758
759	/* Don't probe if APIC mode is disabled. */
760	if (resource_disabled("apic", 0))
761		return;
762
763	/* First, probe all the enumerators to find the best match. */
764	best_enum = NULL;
765	best = 0;
766	SLIST_FOREACH(enumerator, &enumerators, apic_next) {
767		retval = enumerator->apic_probe();
768		if (retval > 0)
769			continue;
770		if (best_enum == NULL || best < retval) {
771			best_enum = enumerator;
772			best = retval;
773		}
774	}
775	if (best_enum == NULL) {
776		if (bootverbose)
777			printf("APIC: Could not find any APICs.\n");
778		return;
779	}
780
781	if (bootverbose)
782		printf("APIC: Using the %s enumerator.\n",
783		    best_enum->apic_name);
784
785	/*
786	 * To work around an errata, we disable the local APIC on some
787	 * CPUs during early startup.  We need to turn the local APIC back
788	 * on on such CPUs now.
789	 */
790	if (cpu == CPU_686 && strcmp(cpu_vendor, "GenuineIntel") == 0 &&
791	    (cpu_id & 0xff0) == 0x610) {
792		apic_base = rdmsr(MSR_APICBASE);
793		apic_base |= APICBASE_ENABLED;
794		wrmsr(MSR_APICBASE, apic_base);
795	}
796
797	/* Second, probe the CPU's in the system. */
798	retval = best_enum->apic_probe_cpus();
799	if (retval != 0)
800		printf("%s: Failed to probe CPUs: returned %d\n",
801		    best_enum->apic_name, retval);
802
803	/* Third, initialize the local APIC. */
804	retval = best_enum->apic_setup_local();
805	if (retval != 0)
806		printf("%s: Failed to setup the local APIC: returned %d\n",
807		    best_enum->apic_name, retval);
808#ifdef SMP
809	/* Last, setup the cpu topology now that we have probed CPUs */
810	mp_topology();
811#endif
812}
813SYSINIT(apic_init, SI_SUB_CPU, SI_ORDER_FIRST, apic_init, NULL)
814
815/*
816 * Setup the I/O APICs.
817 */
818static void
819apic_setup_io(void *dummy __unused)
820{
821	int retval;
822
823	if (best_enum == NULL)
824		return;
825	retval = best_enum->apic_setup_io();
826	if (retval != 0)
827		printf("%s: Failed to setup I/O APICs: returned %d\n",
828		    best_enum->apic_name, retval);
829
830	/*
831	 * Finish setting up the local APIC on the BSP once we know how to
832	 * properly program the LINT pins.
833	 */
834	lapic_setup();
835	if (bootverbose)
836		lapic_dump("BSP");
837}
838SYSINIT(apic_setup_io, SI_SUB_INTR, SI_ORDER_SECOND, apic_setup_io, NULL)
839
840#ifdef SMP
841/*
842 * Inter Processor Interrupt functions.  The lapic_ipi_*() functions are
843 * private to the sys/i386 code.  The public interface for the rest of the
844 * kernel is defined in mp_machdep.c.
845 */
846int
847lapic_ipi_wait(int delay)
848{
849	int x, incr;
850
851	/*
852	 * Wait delay loops for IPI to be sent.  This is highly bogus
853	 * since this is sensitive to CPU clock speed.  If delay is
854	 * -1, we wait forever.
855	 */
856	if (delay == -1) {
857		incr = 0;
858		delay = 1;
859	} else
860		incr = 1;
861	for (x = 0; x < delay; x += incr) {
862		if ((lapic->icr_lo & APIC_DELSTAT_MASK) == APIC_DELSTAT_IDLE)
863			return (1);
864		ia32_pause();
865	}
866	return (0);
867}
868
869void
870lapic_ipi_raw(register_t icrlo, u_int dest)
871{
872	register_t value, eflags;
873
874	/* XXX: Need more sanity checking of icrlo? */
875	KASSERT(lapic != NULL, ("%s called too early", __func__));
876	KASSERT((dest & ~(APIC_ID_MASK >> APIC_ID_SHIFT)) == 0,
877	    ("%s: invalid dest field", __func__));
878	KASSERT((icrlo & APIC_ICRLO_RESV_MASK) == 0,
879	    ("%s: reserved bits set in ICR LO register", __func__));
880
881	/* Set destination in ICR HI register if it is being used. */
882	eflags = intr_disable();
883	if ((icrlo & APIC_DEST_MASK) == APIC_DEST_DESTFLD) {
884		value = lapic->icr_hi;
885		value &= ~APIC_ID_MASK;
886		value |= dest << APIC_ID_SHIFT;
887		lapic->icr_hi = value;
888	}
889
890	/* Program the contents of the IPI and dispatch it. */
891	value = lapic->icr_lo;
892	value &= APIC_ICRLO_RESV_MASK;
893	value |= icrlo;
894	lapic->icr_lo = value;
895	intr_restore(eflags);
896}
897
898#define	BEFORE_SPIN	1000000
899#ifdef DETECT_DEADLOCK
900#define	AFTER_SPIN	1000
901#endif
902
903void
904lapic_ipi_vectored(u_int vector, int dest)
905{
906	register_t icrlo, destfield;
907
908	KASSERT((vector & ~APIC_VECTOR_MASK) == 0,
909	    ("%s: invalid vector %d", __func__, vector));
910
911	icrlo = vector | APIC_DELMODE_FIXED | APIC_DESTMODE_PHY |
912	    APIC_LEVEL_DEASSERT | APIC_TRIGMOD_EDGE;
913	destfield = 0;
914	switch (dest) {
915	case APIC_IPI_DEST_SELF:
916		icrlo |= APIC_DEST_SELF;
917		break;
918	case APIC_IPI_DEST_ALL:
919		icrlo |= APIC_DEST_ALLISELF;
920		break;
921	case APIC_IPI_DEST_OTHERS:
922		icrlo |= APIC_DEST_ALLESELF;
923		break;
924	default:
925		KASSERT((dest & ~(APIC_ID_MASK >> APIC_ID_SHIFT)) == 0,
926		    ("%s: invalid destination 0x%x", __func__, dest));
927		destfield = dest;
928	}
929
930	/* Wait for an earlier IPI to finish. */
931	if (!lapic_ipi_wait(BEFORE_SPIN)) {
932		if (panicstr != NULL)
933			return;
934		else
935			panic("APIC: Previous IPI is stuck");
936	}
937
938	lapic_ipi_raw(icrlo, destfield);
939
940#ifdef DETECT_DEADLOCK
941	/* Wait for IPI to be delivered. */
942	if (!lapic_ipi_wait(AFTER_SPIN)) {
943#ifdef needsattention
944		/*
945		 * XXX FIXME:
946		 *
947		 * The above function waits for the message to actually be
948		 * delivered.  It breaks out after an arbitrary timeout
949		 * since the message should eventually be delivered (at
950		 * least in theory) and that if it wasn't we would catch
951		 * the failure with the check above when the next IPI is
952		 * sent.
953		 *
954		 * We could skip this wait entirely, EXCEPT it probably
955		 * protects us from other routines that assume that the
956		 * message was delivered and acted upon when this function
957		 * returns.
958		 */
959		printf("APIC: IPI might be stuck\n");
960#else /* !needsattention */
961		/* Wait until mesage is sent without a timeout. */
962		while (lapic->icr_lo & APIC_DELSTAT_PEND)
963			ia32_pause();
964#endif /* needsattention */
965	}
966#endif /* DETECT_DEADLOCK */
967}
968#endif /* SMP */
969