local_apic.c revision 151979
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 151979 2005-11-02 20:11:47Z jhb $");
36
37#include "opt_hwpmc_hooks.h"
38
39#include "opt_ddb.h"
40
41#include <sys/param.h>
42#include <sys/systm.h>
43#include <sys/bus.h>
44#include <sys/kernel.h>
45#include <sys/lock.h>
46#include <sys/mutex.h>
47#include <sys/pcpu.h>
48#include <sys/smp.h>
49
50#include <vm/vm.h>
51#include <vm/pmap.h>
52
53#include <machine/apicreg.h>
54#include <machine/cputypes.h>
55#include <machine/frame.h>
56#include <machine/intr_machdep.h>
57#include <machine/apicvar.h>
58#include <machine/md_var.h>
59#include <machine/smp.h>
60#include <machine/specialreg.h>
61
62#ifdef DDB
63#include <sys/interrupt.h>
64#include <ddb/ddb.h>
65#endif
66
67/*
68 * We can handle up to 60 APICs via our logical cluster IDs, but currently
69 * the physical IDs on Intel processors up to the Pentium 4 are limited to
70 * 16.
71 */
72#define	MAX_APICID	16
73
74/* Sanity checks on IDT vectors. */
75CTASSERT(APIC_IO_INTS + APIC_NUM_IOINTS == APIC_TIMER_INT);
76CTASSERT(APIC_TIMER_INT < APIC_LOCAL_INTS);
77CTASSERT(APIC_LOCAL_INTS == 240);
78CTASSERT(IPI_STOP < APIC_SPURIOUS_INT);
79
80#define	LAPIC_TIMER_HZ_DIVIDER		2
81#define	LAPIC_TIMER_STATHZ_DIVIDER	15
82#define	LAPIC_TIMER_PROFHZ_DIVIDER	3
83
84/* Magic IRQ values for the timer and syscalls. */
85#define	IRQ_TIMER	(NUM_IO_INTS + 1)
86#define	IRQ_SYSCALL	(NUM_IO_INTS + 2)
87
88/*
89 * Support for local APICs.  Local APICs manage interrupts on each
90 * individual processor as opposed to I/O APICs which receive interrupts
91 * from I/O devices and then forward them on to the local APICs.
92 *
93 * Local APICs can also send interrupts to each other thus providing the
94 * mechanism for IPIs.
95 */
96
97struct lvt {
98	u_int lvt_edgetrigger:1;
99	u_int lvt_activehi:1;
100	u_int lvt_masked:1;
101	u_int lvt_active:1;
102	u_int lvt_mode:16;
103	u_int lvt_vector:8;
104};
105
106struct lapic {
107	struct lvt la_lvts[LVT_MAX + 1];
108	u_int la_id:8;
109	u_int la_cluster:4;
110	u_int la_cluster_id:2;
111	u_int la_present:1;
112	u_long *la_timer_count;
113	u_long la_hard_ticks;
114	u_long la_stat_ticks;
115	u_long la_prof_ticks;
116} static lapics[MAX_APICID];
117
118/* XXX: should thermal be an NMI? */
119
120/* Global defaults for local APIC LVT entries. */
121static struct lvt lvts[LVT_MAX + 1] = {
122	{ 1, 1, 1, 1, APIC_LVT_DM_EXTINT, 0 },	/* LINT0: masked ExtINT */
123	{ 1, 1, 0, 1, APIC_LVT_DM_NMI, 0 },	/* LINT1: NMI */
124	{ 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_TIMER_INT },	/* Timer */
125	{ 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_ERROR_INT },	/* Error */
126	{ 1, 1, 0, 1, APIC_LVT_DM_NMI, 0 },	/* PMC */
127	{ 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_THERMAL_INT },	/* Thermal */
128};
129
130static inthand_t *ioint_handlers[] = {
131	NULL,			/* 0 - 31 */
132	IDTVEC(apic_isr1),	/* 32 - 63 */
133	IDTVEC(apic_isr2),	/* 64 - 95 */
134	IDTVEC(apic_isr3),	/* 96 - 127 */
135	IDTVEC(apic_isr4),	/* 128 - 159 */
136	IDTVEC(apic_isr5),	/* 160 - 191 */
137	IDTVEC(apic_isr6),	/* 192 - 223 */
138	IDTVEC(apic_isr7),	/* 224 - 255 */
139};
140
141/* Include IDT_SYSCALL to make indexing easier. */
142static u_int ioint_irqs[APIC_NUM_IOINTS + 1];
143
144static u_int32_t lapic_timer_divisors[] = {
145	APIC_TDCR_1, APIC_TDCR_2, APIC_TDCR_4, APIC_TDCR_8, APIC_TDCR_16,
146	APIC_TDCR_32, APIC_TDCR_64, APIC_TDCR_128
147};
148
149volatile lapic_t *lapic;
150static u_long lapic_timer_divisor, lapic_timer_period, lapic_timer_hz;
151
152static void	lapic_enable(void);
153static void	lapic_timer_enable_intr(void);
154static void	lapic_timer_oneshot(u_int count);
155static void	lapic_timer_periodic(u_int count);
156static void	lapic_timer_set_divisor(u_int divisor);
157static uint32_t	lvt_mode(struct lapic *la, u_int pin, uint32_t value);
158
159static uint32_t
160lvt_mode(struct lapic *la, u_int pin, uint32_t value)
161{
162	struct lvt *lvt;
163
164	KASSERT(pin <= LVT_MAX, ("%s: pin %u out of range", __func__, pin));
165	if (la->la_lvts[pin].lvt_active)
166		lvt = &la->la_lvts[pin];
167	else
168		lvt = &lvts[pin];
169
170	value &= ~(APIC_LVT_M | APIC_LVT_TM | APIC_LVT_IIPP | APIC_LVT_DM |
171	    APIC_LVT_VECTOR);
172	if (lvt->lvt_edgetrigger == 0)
173		value |= APIC_LVT_TM;
174	if (lvt->lvt_activehi == 0)
175		value |= APIC_LVT_IIPP_INTALO;
176	if (lvt->lvt_masked)
177		value |= APIC_LVT_M;
178	value |= lvt->lvt_mode;
179	switch (lvt->lvt_mode) {
180	case APIC_LVT_DM_NMI:
181	case APIC_LVT_DM_SMI:
182	case APIC_LVT_DM_INIT:
183	case APIC_LVT_DM_EXTINT:
184		if (!lvt->lvt_edgetrigger) {
185			printf("lapic%u: Forcing LINT%u to edge trigger\n",
186			    la->la_id, pin);
187			value |= APIC_LVT_TM;
188		}
189		/* Use a vector of 0. */
190		break;
191	case APIC_LVT_DM_FIXED:
192		value |= lvt->lvt_vector;
193		break;
194	default:
195		panic("bad APIC LVT delivery mode: %#x\n", value);
196	}
197	return (value);
198}
199
200/*
201 * Map the local APIC and setup necessary interrupt vectors.
202 */
203void
204lapic_init(uintptr_t addr)
205{
206
207	/* Map the local APIC and setup the spurious interrupt handler. */
208	KASSERT(trunc_page(addr) == addr,
209	    ("local APIC not aligned on a page boundary"));
210	lapic = (lapic_t *)pmap_mapdev(addr, sizeof(lapic_t));
211	setidt(APIC_SPURIOUS_INT, IDTVEC(spuriousint), SDT_SYS386IGT, SEL_KPL,
212	    GSEL(GCODE_SEL, SEL_KPL));
213
214	/* Perform basic initialization of the BSP's local APIC. */
215	lapic_enable();
216	ioint_irqs[IDT_SYSCALL - APIC_IO_INTS] = IRQ_SYSCALL;
217
218	/* Set BSP's per-CPU local APIC ID. */
219	PCPU_SET(apic_id, lapic_id());
220
221	/* Local APIC timer interrupt. */
222	setidt(APIC_TIMER_INT, IDTVEC(timerint), SDT_SYS386IGT, SEL_KPL,
223	    GSEL(GCODE_SEL, SEL_KPL));
224	ioint_irqs[APIC_TIMER_INT - APIC_IO_INTS] = IRQ_TIMER;
225
226	/* XXX: error/thermal interrupts */
227}
228
229/*
230 * Create a local APIC instance.
231 */
232void
233lapic_create(u_int apic_id, int boot_cpu)
234{
235	int i;
236
237	if (apic_id >= MAX_APICID) {
238		printf("APIC: Ignoring local APIC with ID %d\n", apic_id);
239		if (boot_cpu)
240			panic("Can't ignore BSP");
241		return;
242	}
243	KASSERT(!lapics[apic_id].la_present, ("duplicate local APIC %u",
244	    apic_id));
245
246	/*
247	 * Assume no local LVT overrides and a cluster of 0 and
248	 * intra-cluster ID of 0.
249	 */
250	lapics[apic_id].la_present = 1;
251	lapics[apic_id].la_id = apic_id;
252	for (i = 0; i < LVT_MAX; i++) {
253		lapics[apic_id].la_lvts[i] = lvts[i];
254		lapics[apic_id].la_lvts[i].lvt_active = 0;
255	}
256
257#ifdef SMP
258	cpu_add(apic_id, boot_cpu);
259#endif
260}
261
262/*
263 * Dump contents of local APIC registers
264 */
265void
266lapic_dump(const char* str)
267{
268
269	printf("cpu%d %s:\n", PCPU_GET(cpuid), str);
270	printf("     ID: 0x%08x   VER: 0x%08x LDR: 0x%08x DFR: 0x%08x\n",
271	    lapic->id, lapic->version, lapic->ldr, lapic->dfr);
272	printf("  lint0: 0x%08x lint1: 0x%08x TPR: 0x%08x SVR: 0x%08x\n",
273	    lapic->lvt_lint0, lapic->lvt_lint1, lapic->tpr, lapic->svr);
274	printf("  timer: 0x%08x therm: 0x%08x err: 0x%08x pcm: 0x%08x\n",
275	    lapic->lvt_timer, lapic->lvt_thermal, lapic->lvt_error,
276	    lapic->lvt_pcint);
277}
278
279void
280lapic_setup(void)
281{
282	struct lapic *la;
283	u_int32_t value, maxlvt;
284	register_t eflags;
285	char buf[MAXCOMLEN + 1];
286
287	la = &lapics[lapic_id()];
288	KASSERT(la->la_present, ("missing APIC structure"));
289	eflags = intr_disable();
290	maxlvt = (lapic->version & APIC_VER_MAXLVT) >> MAXLVTSHIFT;
291
292	/* Initialize the TPR to allow all interrupts. */
293	lapic_set_tpr(0);
294
295	/* Use the cluster model for logical IDs. */
296	value = lapic->dfr;
297	value &= ~APIC_DFR_MODEL_MASK;
298	value |= APIC_DFR_MODEL_CLUSTER;
299	lapic->dfr = value;
300
301	/* Set this APIC's logical ID. */
302	value = lapic->ldr;
303	value &= ~APIC_ID_MASK;
304	value |= (la->la_cluster << APIC_ID_CLUSTER_SHIFT |
305	    1 << la->la_cluster_id) << APIC_ID_SHIFT;
306	lapic->ldr = value;
307
308	/* Setup spurious vector and enable the local APIC. */
309	lapic_enable();
310
311	/* Program LINT[01] LVT entries. */
312	lapic->lvt_lint0 = lvt_mode(la, LVT_LINT0, lapic->lvt_lint0);
313	lapic->lvt_lint1 = lvt_mode(la, LVT_LINT1, lapic->lvt_lint1);
314#ifdef	HWPMC_HOOKS
315	/* Program the PMC LVT entry if present. */
316	if (maxlvt >= LVT_PMC)
317		lapic->lvt_pcint = lvt_mode(la, LVT_PMC, lapic->lvt_pcint);
318#endif
319
320	/* Program timer LVT and setup handler. */
321	lapic->lvt_timer = lvt_mode(la, LVT_TIMER, lapic->lvt_timer);
322	snprintf(buf, sizeof(buf), "cpu%d: timer", PCPU_GET(cpuid));
323	intrcnt_add(buf, &la->la_timer_count);
324	if (PCPU_GET(cpuid) != 0) {
325		KASSERT(lapic_timer_period != 0, ("lapic%u: zero divisor",
326		    lapic_id()));
327		lapic_timer_set_divisor(lapic_timer_divisor);
328		lapic_timer_periodic(lapic_timer_period);
329		lapic_timer_enable_intr();
330	}
331
332	/* XXX: Error and thermal LVTs */
333
334	intr_restore(eflags);
335}
336
337/*
338 * Called by cpu_initclocks() on the BSP to setup the local APIC timer so
339 * that it can drive hardclock, statclock, and profclock.  This function
340 * returns true if it is able to use the local APIC timer to drive the
341 * clocks and false if it is not able.
342 */
343int
344lapic_setup_clock(void)
345{
346	u_long value;
347
348	/* Can't drive the timer without a local APIC. */
349	if (lapic == NULL)
350		return (0);
351
352	/* Start off with a divisor of 2 (power on reset default). */
353	lapic_timer_divisor = 2;
354
355	/* Try to calibrate the local APIC timer. */
356	do {
357		lapic_timer_set_divisor(lapic_timer_divisor);
358		lapic_timer_oneshot(APIC_TIMER_MAX_COUNT);
359		DELAY(2000000);
360		value = APIC_TIMER_MAX_COUNT - lapic->ccr_timer;
361		if (value != APIC_TIMER_MAX_COUNT)
362			break;
363		lapic_timer_divisor <<= 1;
364	} while (lapic_timer_divisor <= 128);
365	if (lapic_timer_divisor > 128)
366		panic("lapic: Divisor too big");
367	value /= 2;
368	if (bootverbose)
369		printf("lapic: Divisor %lu, Frequency %lu hz\n",
370		    lapic_timer_divisor, value);
371
372	/*
373	 * We will drive the timer at a small multiple of hz and drive
374	 * both of the other timers with similarly small but relatively
375	 * prime divisors.
376	 */
377	lapic_timer_hz = hz * LAPIC_TIMER_HZ_DIVIDER;
378	stathz = lapic_timer_hz / LAPIC_TIMER_STATHZ_DIVIDER;
379	profhz = lapic_timer_hz / LAPIC_TIMER_PROFHZ_DIVIDER;
380	lapic_timer_period = value / lapic_timer_hz;
381
382	/*
383	 * Start up the timer on the BSP.  The APs will kick off their
384	 * timer during lapic_setup().
385	 */
386	lapic_timer_periodic(lapic_timer_period);
387	lapic_timer_enable_intr();
388	return (1);
389}
390
391void
392lapic_disable(void)
393{
394	uint32_t value;
395
396	/* Software disable the local APIC. */
397	value = lapic->svr;
398	value &= ~APIC_SVR_SWEN;
399	lapic->svr = value;
400}
401
402static void
403lapic_enable(void)
404{
405	u_int32_t value;
406
407	/* Program the spurious vector to enable the local APIC. */
408	value = lapic->svr;
409	value &= ~(APIC_SVR_VECTOR | APIC_SVR_FOCUS);
410	value |= (APIC_SVR_FEN | APIC_SVR_SWEN | APIC_SPURIOUS_INT);
411	lapic->svr = value;
412}
413
414int
415lapic_id(void)
416{
417
418	KASSERT(lapic != NULL, ("local APIC is not mapped"));
419	return (lapic->id >> APIC_ID_SHIFT);
420}
421
422int
423lapic_intr_pending(u_int vector)
424{
425	volatile u_int32_t *irr;
426
427	/*
428	 * The IRR registers are an array of 128-bit registers each of
429	 * which only describes 32 interrupts in the low 32 bits..  Thus,
430	 * we divide the vector by 32 to get the 128-bit index.  We then
431	 * multiply that index by 4 to get the equivalent index from
432	 * treating the IRR as an array of 32-bit registers.  Finally, we
433	 * modulus the vector by 32 to determine the individual bit to
434	 * test.
435	 */
436	irr = &lapic->irr0;
437	return (irr[(vector / 32) * 4] & 1 << (vector % 32));
438}
439
440void
441lapic_set_logical_id(u_int apic_id, u_int cluster, u_int cluster_id)
442{
443	struct lapic *la;
444
445	KASSERT(lapics[apic_id].la_present, ("%s: APIC %u doesn't exist",
446	    __func__, apic_id));
447	KASSERT(cluster <= APIC_MAX_CLUSTER, ("%s: cluster %u too big",
448	    __func__, cluster));
449	KASSERT(cluster_id <= APIC_MAX_INTRACLUSTER_ID,
450	    ("%s: intra cluster id %u too big", __func__, cluster_id));
451	la = &lapics[apic_id];
452	la->la_cluster = cluster;
453	la->la_cluster_id = cluster_id;
454}
455
456int
457lapic_set_lvt_mask(u_int apic_id, u_int pin, u_char masked)
458{
459
460	if (pin > LVT_MAX)
461		return (EINVAL);
462	if (apic_id == APIC_ID_ALL) {
463		lvts[pin].lvt_masked = masked;
464		if (bootverbose)
465			printf("lapic:");
466	} else {
467		KASSERT(lapics[apic_id].la_present,
468		    ("%s: missing APIC %u", __func__, apic_id));
469		lapics[apic_id].la_lvts[pin].lvt_masked = masked;
470		lapics[apic_id].la_lvts[pin].lvt_active = 1;
471		if (bootverbose)
472			printf("lapic%u:", apic_id);
473	}
474	if (bootverbose)
475		printf(" LINT%u %s\n", pin, masked ? "masked" : "unmasked");
476	return (0);
477}
478
479int
480lapic_set_lvt_mode(u_int apic_id, u_int pin, u_int32_t mode)
481{
482	struct lvt *lvt;
483
484	if (pin > LVT_MAX)
485		return (EINVAL);
486	if (apic_id == APIC_ID_ALL) {
487		lvt = &lvts[pin];
488		if (bootverbose)
489			printf("lapic:");
490	} else {
491		KASSERT(lapics[apic_id].la_present,
492		    ("%s: missing APIC %u", __func__, apic_id));
493		lvt = &lapics[apic_id].la_lvts[pin];
494		lvt->lvt_active = 1;
495		if (bootverbose)
496			printf("lapic%u:", apic_id);
497	}
498	lvt->lvt_mode = mode;
499	switch (mode) {
500	case APIC_LVT_DM_NMI:
501	case APIC_LVT_DM_SMI:
502	case APIC_LVT_DM_INIT:
503	case APIC_LVT_DM_EXTINT:
504		lvt->lvt_edgetrigger = 1;
505		lvt->lvt_activehi = 1;
506		if (mode == APIC_LVT_DM_EXTINT)
507			lvt->lvt_masked = 1;
508		else
509			lvt->lvt_masked = 0;
510		break;
511	default:
512		panic("Unsupported delivery mode: 0x%x\n", mode);
513	}
514	if (bootverbose) {
515		printf(" Routing ");
516		switch (mode) {
517		case APIC_LVT_DM_NMI:
518			printf("NMI");
519			break;
520		case APIC_LVT_DM_SMI:
521			printf("SMI");
522			break;
523		case APIC_LVT_DM_INIT:
524			printf("INIT");
525			break;
526		case APIC_LVT_DM_EXTINT:
527			printf("ExtINT");
528			break;
529		}
530		printf(" -> LINT%u\n", pin);
531	}
532	return (0);
533}
534
535int
536lapic_set_lvt_polarity(u_int apic_id, u_int pin, enum intr_polarity pol)
537{
538
539	if (pin > LVT_MAX || pol == INTR_POLARITY_CONFORM)
540		return (EINVAL);
541	if (apic_id == APIC_ID_ALL) {
542		lvts[pin].lvt_activehi = (pol == INTR_POLARITY_HIGH);
543		if (bootverbose)
544			printf("lapic:");
545	} else {
546		KASSERT(lapics[apic_id].la_present,
547		    ("%s: missing APIC %u", __func__, apic_id));
548		lapics[apic_id].la_lvts[pin].lvt_active = 1;
549		lapics[apic_id].la_lvts[pin].lvt_activehi =
550		    (pol == INTR_POLARITY_HIGH);
551		if (bootverbose)
552			printf("lapic%u:", apic_id);
553	}
554	if (bootverbose)
555		printf(" LINT%u polarity: %s\n", pin,
556		    pol == INTR_POLARITY_HIGH ? "high" : "low");
557	return (0);
558}
559
560int
561lapic_set_lvt_triggermode(u_int apic_id, u_int pin, enum intr_trigger trigger)
562{
563
564	if (pin > LVT_MAX || trigger == INTR_TRIGGER_CONFORM)
565		return (EINVAL);
566	if (apic_id == APIC_ID_ALL) {
567		lvts[pin].lvt_edgetrigger = (trigger == INTR_TRIGGER_EDGE);
568		if (bootverbose)
569			printf("lapic:");
570	} else {
571		KASSERT(lapics[apic_id].la_present,
572		    ("%s: missing APIC %u", __func__, apic_id));
573		lapics[apic_id].la_lvts[pin].lvt_edgetrigger =
574		    (trigger == INTR_TRIGGER_EDGE);
575		lapics[apic_id].la_lvts[pin].lvt_active = 1;
576		if (bootverbose)
577			printf("lapic%u:", apic_id);
578	}
579	if (bootverbose)
580		printf(" LINT%u trigger: %s\n", pin,
581		    trigger == INTR_TRIGGER_EDGE ? "edge" : "level");
582	return (0);
583}
584
585/*
586 * Adjust the TPR of the current CPU so that it blocks all interrupts below
587 * the passed in vector.
588 */
589void
590lapic_set_tpr(u_int vector)
591{
592#ifdef CHEAP_TPR
593	lapic->tpr = vector;
594#else
595	u_int32_t tpr;
596
597	tpr = lapic->tpr & ~APIC_TPR_PRIO;
598	tpr |= vector;
599	lapic->tpr = tpr;
600#endif
601}
602
603void
604lapic_eoi(void)
605{
606
607	lapic->eoi = 0;
608}
609
610void
611lapic_handle_intr(struct intrframe frame)
612{
613	struct intsrc *isrc;
614
615	if (frame.if_vec == -1)
616		panic("Couldn't get vector from ISR!");
617	isrc = intr_lookup_source(apic_idt_to_irq(frame.if_vec));
618	intr_execute_handlers(isrc, &frame);
619}
620
621void
622lapic_handle_timer(struct clockframe frame)
623{
624	struct lapic *la;
625
626	la = &lapics[PCPU_GET(apic_id)];
627	(*la->la_timer_count)++;
628	critical_enter();
629
630	/* Fire hardclock at hz. */
631	la->la_hard_ticks += hz;
632	if (la->la_hard_ticks >= lapic_timer_hz) {
633		la->la_hard_ticks -= lapic_timer_hz;
634		if (PCPU_GET(cpuid) == 0)
635			hardclock(&frame);
636		else
637			hardclock_process(&frame);
638	}
639
640	/* Fire statclock at stathz. */
641	la->la_stat_ticks += stathz;
642	if (la->la_stat_ticks >= lapic_timer_hz) {
643		la->la_stat_ticks -= lapic_timer_hz;
644		statclock(&frame);
645	}
646
647	/* Fire profclock at profhz, but only when needed. */
648	la->la_prof_ticks += profhz;
649	if (la->la_prof_ticks >= lapic_timer_hz) {
650		la->la_prof_ticks -= lapic_timer_hz;
651		if (profprocs != 0)
652			profclock(&frame);
653	}
654	critical_exit();
655}
656
657static void
658lapic_timer_set_divisor(u_int divisor)
659{
660
661	KASSERT(powerof2(divisor), ("lapic: invalid divisor %u", divisor));
662	KASSERT(ffs(divisor) <= sizeof(lapic_timer_divisors) /
663	    sizeof(u_int32_t), ("lapic: invalid divisor %u", divisor));
664	lapic->dcr_timer = lapic_timer_divisors[ffs(divisor) - 1];
665}
666
667static void
668lapic_timer_oneshot(u_int count)
669{
670	u_int32_t value;
671
672	value = lapic->lvt_timer;
673	value &= ~APIC_LVTT_TM;
674	value |= APIC_LVTT_TM_ONE_SHOT;
675	lapic->lvt_timer = value;
676	lapic->icr_timer = count;
677}
678
679static void
680lapic_timer_periodic(u_int count)
681{
682	u_int32_t value;
683
684	value = lapic->lvt_timer;
685	value &= ~APIC_LVTT_TM;
686	value |= APIC_LVTT_TM_PERIODIC;
687	lapic->lvt_timer = value;
688	lapic->icr_timer = count;
689}
690
691static void
692lapic_timer_enable_intr(void)
693{
694	u_int32_t value;
695
696	value = lapic->lvt_timer;
697	value &= ~APIC_LVT_M;
698	lapic->lvt_timer = value;
699}
700
701/* Request a free IDT vector to be used by the specified IRQ. */
702u_int
703apic_alloc_vector(u_int irq)
704{
705	u_int vector;
706
707	KASSERT(irq < NUM_IO_INTS, ("Invalid IRQ %u", irq));
708
709	/*
710	 * Search for a free vector.  Currently we just use a very simple
711	 * algorithm to find the first free vector.
712	 */
713	mtx_lock_spin(&icu_lock);
714	for (vector = 0; vector < APIC_NUM_IOINTS; vector++) {
715		if (ioint_irqs[vector] != 0)
716			continue;
717		ioint_irqs[vector] = irq;
718		mtx_unlock_spin(&icu_lock);
719		return (vector + APIC_IO_INTS);
720	}
721	mtx_unlock_spin(&icu_lock);
722	panic("Couldn't find an APIC vector for IRQ %u", irq);
723}
724
725void
726apic_enable_vector(u_int vector)
727{
728
729	KASSERT(vector != IDT_SYSCALL, ("Attempt to overwrite syscall entry"));
730	KASSERT(ioint_handlers[vector / 32] != NULL,
731	    ("No ISR handler for vector %u", vector));
732	setidt(vector, ioint_handlers[vector / 32], SDT_SYS386IGT, SEL_KPL,
733	    GSEL(GCODE_SEL, SEL_KPL));
734}
735
736/* Release an APIC vector when it's no longer in use. */
737void
738apic_free_vector(u_int vector, u_int irq)
739{
740	KASSERT(vector >= APIC_IO_INTS && vector != IDT_SYSCALL &&
741	    vector <= APIC_IO_INTS + APIC_NUM_IOINTS,
742	    ("Vector %u does not map to an IRQ line", vector));
743	KASSERT(irq < NUM_IO_INTS, ("Invalid IRQ %u", irq));
744	KASSERT(ioint_irqs[vector - APIC_IO_INTS] == irq, ("IRQ mismatch"));
745	mtx_lock_spin(&icu_lock);
746	ioint_irqs[vector - APIC_IO_INTS] = 0;
747	mtx_unlock_spin(&icu_lock);
748}
749
750/* Map an IDT vector (APIC) to an IRQ (interrupt source). */
751u_int
752apic_idt_to_irq(u_int vector)
753{
754
755	KASSERT(vector >= APIC_IO_INTS && vector != IDT_SYSCALL &&
756	    vector <= APIC_IO_INTS + APIC_NUM_IOINTS,
757	    ("Vector %u does not map to an IRQ line", vector));
758	return (ioint_irqs[vector - APIC_IO_INTS]);
759}
760
761#ifdef DDB
762/*
763 * Dump data about APIC IDT vector mappings.
764 */
765DB_SHOW_COMMAND(apic, db_show_apic)
766{
767	struct intsrc *isrc;
768	int quit, i, verbose;
769	u_int irq;
770
771	quit = 0;
772	if (strcmp(modif, "vv") == 0)
773		verbose = 2;
774	else if (strcmp(modif, "v") == 0)
775		verbose = 1;
776	else
777		verbose = 0;
778	db_setup_paging(db_simple_pager, &quit, db_lines_per_page);
779	for (i = 0; i < APIC_NUM_IOINTS + 1 && !quit; i++) {
780		irq = ioint_irqs[i];
781		if (irq != 0 && irq != IRQ_SYSCALL) {
782			db_printf("vec 0x%2x -> ", i + APIC_IO_INTS);
783			if (irq == IRQ_TIMER)
784				db_printf("lapic timer\n");
785			else if (irq < NUM_IO_INTS) {
786				isrc = intr_lookup_source(irq);
787				if (isrc == NULL || verbose == 0)
788					db_printf("IRQ %u\n", irq);
789				else
790					db_dump_intr_event(isrc->is_event,
791					    verbose == 2);
792			} else
793				db_printf("IRQ %u ???\n", irq);
794		}
795	}
796}
797#endif
798
799/*
800 * APIC probing support code.  This includes code to manage enumerators.
801 */
802
803static SLIST_HEAD(, apic_enumerator) enumerators =
804	SLIST_HEAD_INITIALIZER(enumerators);
805static struct apic_enumerator *best_enum;
806
807void
808apic_register_enumerator(struct apic_enumerator *enumerator)
809{
810#ifdef INVARIANTS
811	struct apic_enumerator *apic_enum;
812
813	SLIST_FOREACH(apic_enum, &enumerators, apic_next) {
814		if (apic_enum == enumerator)
815			panic("%s: Duplicate register of %s", __func__,
816			    enumerator->apic_name);
817	}
818#endif
819	SLIST_INSERT_HEAD(&enumerators, enumerator, apic_next);
820}
821
822/*
823 * Probe the APIC enumerators, enumerate CPUs, and initialize the
824 * local APIC.
825 */
826static void
827apic_init(void *dummy __unused)
828{
829	struct apic_enumerator *enumerator;
830	uint64_t apic_base;
831	int retval, best;
832
833	/* We only support built in local APICs. */
834	if (!(cpu_feature & CPUID_APIC))
835		return;
836
837	/* Don't probe if APIC mode is disabled. */
838	if (resource_disabled("apic", 0))
839		return;
840
841	/* First, probe all the enumerators to find the best match. */
842	best_enum = NULL;
843	best = 0;
844	SLIST_FOREACH(enumerator, &enumerators, apic_next) {
845		retval = enumerator->apic_probe();
846		if (retval > 0)
847			continue;
848		if (best_enum == NULL || best < retval) {
849			best_enum = enumerator;
850			best = retval;
851		}
852	}
853	if (best_enum == NULL) {
854		if (bootverbose)
855			printf("APIC: Could not find any APICs.\n");
856		return;
857	}
858
859	if (bootverbose)
860		printf("APIC: Using the %s enumerator.\n",
861		    best_enum->apic_name);
862
863	/*
864	 * To work around an errata, we disable the local APIC on some
865	 * CPUs during early startup.  We need to turn the local APIC back
866	 * on on such CPUs now.
867	 */
868	if (cpu == CPU_686 && strcmp(cpu_vendor, "GenuineIntel") == 0 &&
869	    (cpu_id & 0xff0) == 0x610) {
870		apic_base = rdmsr(MSR_APICBASE);
871		apic_base |= APICBASE_ENABLED;
872		wrmsr(MSR_APICBASE, apic_base);
873	}
874
875	/* Second, probe the CPU's in the system. */
876	retval = best_enum->apic_probe_cpus();
877	if (retval != 0)
878		printf("%s: Failed to probe CPUs: returned %d\n",
879		    best_enum->apic_name, retval);
880
881	/* Third, initialize the local APIC. */
882	retval = best_enum->apic_setup_local();
883	if (retval != 0)
884		printf("%s: Failed to setup the local APIC: returned %d\n",
885		    best_enum->apic_name, retval);
886#ifdef SMP
887	/* Last, setup the cpu topology now that we have probed CPUs */
888	mp_topology();
889#endif
890}
891SYSINIT(apic_init, SI_SUB_CPU, SI_ORDER_FIRST, apic_init, NULL)
892
893/*
894 * Setup the I/O APICs.
895 */
896static void
897apic_setup_io(void *dummy __unused)
898{
899	int retval;
900
901	if (best_enum == NULL)
902		return;
903	retval = best_enum->apic_setup_io();
904	if (retval != 0)
905		printf("%s: Failed to setup I/O APICs: returned %d\n",
906		    best_enum->apic_name, retval);
907
908	/*
909	 * Finish setting up the local APIC on the BSP once we know how to
910	 * properly program the LINT pins.
911	 */
912	lapic_setup();
913	if (bootverbose)
914		lapic_dump("BSP");
915}
916SYSINIT(apic_setup_io, SI_SUB_INTR, SI_ORDER_SECOND, apic_setup_io, NULL)
917
918#ifdef SMP
919/*
920 * Inter Processor Interrupt functions.  The lapic_ipi_*() functions are
921 * private to the sys/i386 code.  The public interface for the rest of the
922 * kernel is defined in mp_machdep.c.
923 */
924int
925lapic_ipi_wait(int delay)
926{
927	int x, incr;
928
929	/*
930	 * Wait delay loops for IPI to be sent.  This is highly bogus
931	 * since this is sensitive to CPU clock speed.  If delay is
932	 * -1, we wait forever.
933	 */
934	if (delay == -1) {
935		incr = 0;
936		delay = 1;
937	} else
938		incr = 1;
939	for (x = 0; x < delay; x += incr) {
940		if ((lapic->icr_lo & APIC_DELSTAT_MASK) == APIC_DELSTAT_IDLE)
941			return (1);
942		ia32_pause();
943	}
944	return (0);
945}
946
947void
948lapic_ipi_raw(register_t icrlo, u_int dest)
949{
950	register_t value, eflags;
951
952	/* XXX: Need more sanity checking of icrlo? */
953	KASSERT(lapic != NULL, ("%s called too early", __func__));
954	KASSERT((dest & ~(APIC_ID_MASK >> APIC_ID_SHIFT)) == 0,
955	    ("%s: invalid dest field", __func__));
956	KASSERT((icrlo & APIC_ICRLO_RESV_MASK) == 0,
957	    ("%s: reserved bits set in ICR LO register", __func__));
958
959	/* Set destination in ICR HI register if it is being used. */
960	eflags = intr_disable();
961	if ((icrlo & APIC_DEST_MASK) == APIC_DEST_DESTFLD) {
962		value = lapic->icr_hi;
963		value &= ~APIC_ID_MASK;
964		value |= dest << APIC_ID_SHIFT;
965		lapic->icr_hi = value;
966	}
967
968	/* Program the contents of the IPI and dispatch it. */
969	value = lapic->icr_lo;
970	value &= APIC_ICRLO_RESV_MASK;
971	value |= icrlo;
972	lapic->icr_lo = value;
973	intr_restore(eflags);
974}
975
976#define	BEFORE_SPIN	1000000
977#ifdef DETECT_DEADLOCK
978#define	AFTER_SPIN	1000
979#endif
980
981void
982lapic_ipi_vectored(u_int vector, int dest)
983{
984	register_t icrlo, destfield;
985
986	KASSERT((vector & ~APIC_VECTOR_MASK) == 0,
987	    ("%s: invalid vector %d", __func__, vector));
988
989	icrlo = vector | APIC_DELMODE_FIXED | APIC_DESTMODE_PHY |
990	    APIC_LEVEL_DEASSERT | APIC_TRIGMOD_EDGE;
991	destfield = 0;
992	switch (dest) {
993	case APIC_IPI_DEST_SELF:
994		icrlo |= APIC_DEST_SELF;
995		break;
996	case APIC_IPI_DEST_ALL:
997		icrlo |= APIC_DEST_ALLISELF;
998		break;
999	case APIC_IPI_DEST_OTHERS:
1000		icrlo |= APIC_DEST_ALLESELF;
1001		break;
1002	default:
1003		KASSERT((dest & ~(APIC_ID_MASK >> APIC_ID_SHIFT)) == 0,
1004		    ("%s: invalid destination 0x%x", __func__, dest));
1005		destfield = dest;
1006	}
1007
1008	/* Wait for an earlier IPI to finish. */
1009	if (!lapic_ipi_wait(BEFORE_SPIN)) {
1010		if (panicstr != NULL)
1011			return;
1012		else
1013			panic("APIC: Previous IPI is stuck");
1014	}
1015
1016	lapic_ipi_raw(icrlo, destfield);
1017
1018#ifdef DETECT_DEADLOCK
1019	/* Wait for IPI to be delivered. */
1020	if (!lapic_ipi_wait(AFTER_SPIN)) {
1021#ifdef needsattention
1022		/*
1023		 * XXX FIXME:
1024		 *
1025		 * The above function waits for the message to actually be
1026		 * delivered.  It breaks out after an arbitrary timeout
1027		 * since the message should eventually be delivered (at
1028		 * least in theory) and that if it wasn't we would catch
1029		 * the failure with the check above when the next IPI is
1030		 * sent.
1031		 *
1032		 * We could skip this wait entirely, EXCEPT it probably
1033		 * protects us from other routines that assume that the
1034		 * message was delivered and acted upon when this function
1035		 * returns.
1036		 */
1037		printf("APIC: IPI might be stuck\n");
1038#else /* !needsattention */
1039		/* Wait until mesage is sent without a timeout. */
1040		while (lapic->icr_lo & APIC_DELSTAT_PEND)
1041			ia32_pause();
1042#endif /* needsattention */
1043	}
1044#endif /* DETECT_DEADLOCK */
1045}
1046#endif /* SMP */
1047