acpi_hpet.c revision 274001
1/*-
2 * Copyright (c) 2005 Poul-Henning Kamp
3 * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org>
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. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25 * SUCH DAMAGE.
26 */
27
28#include <sys/cdefs.h>
29__FBSDID("$FreeBSD: stable/10/sys/dev/acpica/acpi_hpet.c 274001 2014-11-03 03:17:58Z rpaulo $");
30
31#include "opt_acpi.h"
32#if defined(__amd64__) || defined(__ia64__)
33#define	DEV_APIC
34#else
35#include "opt_apic.h"
36#endif
37#include <sys/param.h>
38#include <sys/conf.h>
39#include <sys/bus.h>
40#include <sys/kernel.h>
41#include <sys/module.h>
42#include <sys/proc.h>
43#include <sys/rman.h>
44#include <sys/mman.h>
45#include <sys/time.h>
46#include <sys/smp.h>
47#include <sys/sysctl.h>
48#include <sys/timeet.h>
49#include <sys/timetc.h>
50
51#include <contrib/dev/acpica/include/acpi.h>
52#include <contrib/dev/acpica/include/accommon.h>
53
54#include <dev/acpica/acpivar.h>
55#include <dev/acpica/acpi_hpet.h>
56
57#ifdef DEV_APIC
58#include "pcib_if.h"
59#endif
60
61#define HPET_VENDID_AMD		0x4353
62#define HPET_VENDID_AMD2	0x1022
63#define HPET_VENDID_INTEL	0x8086
64#define HPET_VENDID_NVIDIA	0x10de
65#define HPET_VENDID_SW		0x1166
66
67ACPI_SERIAL_DECL(hpet, "ACPI HPET support");
68
69static devclass_t hpet_devclass;
70
71/* ACPI CA debugging */
72#define _COMPONENT	ACPI_TIMER
73ACPI_MODULE_NAME("HPET")
74
75struct hpet_softc {
76	device_t		dev;
77	int			mem_rid;
78	int			intr_rid;
79	int			irq;
80	int			useirq;
81	int			legacy_route;
82	int			per_cpu;
83	uint32_t		allowed_irqs;
84	struct resource		*mem_res;
85	struct resource		*intr_res;
86	void			*intr_handle;
87	ACPI_HANDLE		handle;
88	uint64_t		freq;
89	uint32_t		caps;
90	struct timecounter	tc;
91	struct hpet_timer {
92		struct eventtimer	et;
93		struct hpet_softc	*sc;
94		int			num;
95		int			mode;
96		int			intr_rid;
97		int			irq;
98		int			pcpu_cpu;
99		int			pcpu_misrouted;
100		int			pcpu_master;
101		int			pcpu_slaves[MAXCPU];
102		struct resource		*intr_res;
103		void			*intr_handle;
104		uint32_t		caps;
105		uint32_t		vectors;
106		uint32_t		div;
107		uint32_t		next;
108		char			name[8];
109	} 			t[32];
110	int			num_timers;
111	struct cdev		*pdev;
112	int			mmap_allow;
113	int			mmap_allow_write;
114};
115
116static d_open_t hpet_open;
117static d_mmap_t hpet_mmap;
118
119static struct cdevsw hpet_cdevsw = {
120	.d_version =	D_VERSION,
121	.d_name =	"hpet",
122	.d_open =	hpet_open,
123	.d_mmap =	hpet_mmap,
124};
125
126static u_int hpet_get_timecount(struct timecounter *tc);
127static void hpet_test(struct hpet_softc *sc);
128
129static char *hpet_ids[] = { "PNP0103", NULL };
130
131static u_int
132hpet_get_timecount(struct timecounter *tc)
133{
134	struct hpet_softc *sc;
135
136	sc = tc->tc_priv;
137	return (bus_read_4(sc->mem_res, HPET_MAIN_COUNTER));
138}
139
140static void
141hpet_enable(struct hpet_softc *sc)
142{
143	uint32_t val;
144
145	val = bus_read_4(sc->mem_res, HPET_CONFIG);
146	if (sc->legacy_route)
147		val |= HPET_CNF_LEG_RT;
148	else
149		val &= ~HPET_CNF_LEG_RT;
150	val |= HPET_CNF_ENABLE;
151	bus_write_4(sc->mem_res, HPET_CONFIG, val);
152}
153
154static void
155hpet_disable(struct hpet_softc *sc)
156{
157	uint32_t val;
158
159	val = bus_read_4(sc->mem_res, HPET_CONFIG);
160	val &= ~HPET_CNF_ENABLE;
161	bus_write_4(sc->mem_res, HPET_CONFIG, val);
162}
163
164static int
165hpet_start(struct eventtimer *et, sbintime_t first, sbintime_t period)
166{
167	struct hpet_timer *mt = (struct hpet_timer *)et->et_priv;
168	struct hpet_timer *t;
169	struct hpet_softc *sc = mt->sc;
170	uint32_t fdiv, now;
171
172	t = (mt->pcpu_master < 0) ? mt : &sc->t[mt->pcpu_slaves[curcpu]];
173	if (period != 0) {
174		t->mode = 1;
175		t->div = (sc->freq * period) >> 32;
176	} else {
177		t->mode = 2;
178		t->div = 0;
179	}
180	if (first != 0)
181		fdiv = (sc->freq * first) >> 32;
182	else
183		fdiv = t->div;
184	if (t->irq < 0)
185		bus_write_4(sc->mem_res, HPET_ISR, 1 << t->num);
186	t->caps |= HPET_TCNF_INT_ENB;
187	now = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
188restart:
189	t->next = now + fdiv;
190	if (t->mode == 1 && (t->caps & HPET_TCAP_PER_INT)) {
191		t->caps |= HPET_TCNF_TYPE;
192		bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num),
193		    t->caps | HPET_TCNF_VAL_SET);
194		bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
195		    t->next);
196		bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
197		    t->div);
198	} else {
199		t->caps &= ~HPET_TCNF_TYPE;
200		bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num),
201		    t->caps);
202		bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
203		    t->next);
204	}
205	now = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
206	if ((int32_t)(now - t->next + HPET_MIN_CYCLES) >= 0) {
207		fdiv *= 2;
208		goto restart;
209	}
210	return (0);
211}
212
213static int
214hpet_stop(struct eventtimer *et)
215{
216	struct hpet_timer *mt = (struct hpet_timer *)et->et_priv;
217	struct hpet_timer *t;
218	struct hpet_softc *sc = mt->sc;
219
220	t = (mt->pcpu_master < 0) ? mt : &sc->t[mt->pcpu_slaves[curcpu]];
221	t->mode = 0;
222	t->caps &= ~(HPET_TCNF_INT_ENB | HPET_TCNF_TYPE);
223	bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num), t->caps);
224	return (0);
225}
226
227static int
228hpet_intr_single(void *arg)
229{
230	struct hpet_timer *t = (struct hpet_timer *)arg;
231	struct hpet_timer *mt;
232	struct hpet_softc *sc = t->sc;
233	uint32_t now;
234
235	if (t->mode == 0)
236		return (FILTER_STRAY);
237	/* Check that per-CPU timer interrupt reached right CPU. */
238	if (t->pcpu_cpu >= 0 && t->pcpu_cpu != curcpu) {
239		if ((++t->pcpu_misrouted) % 32 == 0) {
240			printf("HPET interrupt routed to the wrong CPU"
241			    " (timer %d CPU %d -> %d)!\n",
242			    t->num, t->pcpu_cpu, curcpu);
243		}
244
245		/*
246		 * Reload timer, hoping that next time may be more lucky
247		 * (system will manage proper interrupt binding).
248		 */
249		if ((t->mode == 1 && (t->caps & HPET_TCAP_PER_INT) == 0) ||
250		    t->mode == 2) {
251			t->next = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER) +
252			    sc->freq / 8;
253			bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
254			    t->next);
255		}
256		return (FILTER_HANDLED);
257	}
258	if (t->mode == 1 &&
259	    (t->caps & HPET_TCAP_PER_INT) == 0) {
260		t->next += t->div;
261		now = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
262		if ((int32_t)((now + t->div / 2) - t->next) > 0)
263			t->next = now + t->div / 2;
264		bus_write_4(sc->mem_res,
265		    HPET_TIMER_COMPARATOR(t->num), t->next);
266	} else if (t->mode == 2)
267		t->mode = 0;
268	mt = (t->pcpu_master < 0) ? t : &sc->t[t->pcpu_master];
269	if (mt->et.et_active)
270		mt->et.et_event_cb(&mt->et, mt->et.et_arg);
271	return (FILTER_HANDLED);
272}
273
274static int
275hpet_intr(void *arg)
276{
277	struct hpet_softc *sc = (struct hpet_softc *)arg;
278	int i;
279	uint32_t val;
280
281	val = bus_read_4(sc->mem_res, HPET_ISR);
282	if (val) {
283		bus_write_4(sc->mem_res, HPET_ISR, val);
284		val &= sc->useirq;
285		for (i = 0; i < sc->num_timers; i++) {
286			if ((val & (1 << i)) == 0)
287				continue;
288			hpet_intr_single(&sc->t[i]);
289		}
290		return (FILTER_HANDLED);
291	}
292	return (FILTER_STRAY);
293}
294
295static ACPI_STATUS
296hpet_find(ACPI_HANDLE handle, UINT32 level, void *context,
297    void **status)
298{
299	char 		**ids;
300	uint32_t	id = (uint32_t)(uintptr_t)context;
301	uint32_t	uid = 0;
302
303	for (ids = hpet_ids; *ids != NULL; ids++) {
304		if (acpi_MatchHid(handle, *ids))
305		        break;
306	}
307	if (*ids == NULL)
308		return (AE_OK);
309	if (ACPI_FAILURE(acpi_GetInteger(handle, "_UID", &uid)) ||
310	    id == uid)
311		*status = acpi_get_device(handle);
312	return (AE_OK);
313}
314
315/*
316 * Find an existing IRQ resource that matches the requested IRQ range
317 * and return its RID.  If one is not found, use a new RID.
318 */
319static int
320hpet_find_irq_rid(device_t dev, u_long start, u_long end)
321{
322	u_long irq;
323	int error, rid;
324
325	for (rid = 0;; rid++) {
326		error = bus_get_resource(dev, SYS_RES_IRQ, rid, &irq, NULL);
327		if (error != 0 || (start <= irq && irq <= end))
328			return (rid);
329	}
330}
331
332static int
333hpet_open(struct cdev *cdev, int oflags, int devtype, struct thread *td)
334{
335	struct hpet_softc *sc;
336
337	sc = cdev->si_drv1;
338	if (!sc->mmap_allow)
339		return (EPERM);
340	else
341		return (0);
342}
343
344static int
345hpet_mmap(struct cdev *cdev, vm_ooffset_t offset, vm_paddr_t *paddr,
346    int nprot, vm_memattr_t *memattr)
347{
348	struct hpet_softc *sc;
349
350	sc = cdev->si_drv1;
351	if (offset > rman_get_size(sc->mem_res))
352		return (EINVAL);
353	if (!sc->mmap_allow_write && (nprot & PROT_WRITE))
354		return (EPERM);
355	*paddr = rman_get_start(sc->mem_res) + offset;
356	*memattr = VM_MEMATTR_UNCACHEABLE;
357
358	return (0);
359}
360
361/* Discover the HPET via the ACPI table of the same name. */
362static void
363hpet_identify(driver_t *driver, device_t parent)
364{
365	ACPI_TABLE_HPET *hpet;
366	ACPI_STATUS	status;
367	device_t	child;
368	int		i;
369
370	/* Only one HPET device can be added. */
371	if (devclass_get_device(hpet_devclass, 0))
372		return;
373	for (i = 1; ; i++) {
374		/* Search for HPET table. */
375		status = AcpiGetTable(ACPI_SIG_HPET, i, (ACPI_TABLE_HEADER **)&hpet);
376		if (ACPI_FAILURE(status))
377			return;
378		/* Search for HPET device with same ID. */
379		child = NULL;
380		AcpiWalkNamespace(ACPI_TYPE_DEVICE, ACPI_ROOT_OBJECT,
381		    100, hpet_find, NULL, (void *)(uintptr_t)hpet->Sequence,
382		    (void *)&child);
383		/* If found - let it be probed in normal way. */
384		if (child) {
385			if (bus_get_resource(child, SYS_RES_MEMORY, 0,
386			    NULL, NULL) != 0)
387				bus_set_resource(child, SYS_RES_MEMORY, 0,
388				    hpet->Address.Address, HPET_MEM_WIDTH);
389			continue;
390		}
391		/* If not - create it from table info. */
392		child = BUS_ADD_CHILD(parent, 2, "hpet", 0);
393		if (child == NULL) {
394			printf("%s: can't add child\n", __func__);
395			continue;
396		}
397		bus_set_resource(child, SYS_RES_MEMORY, 0, hpet->Address.Address,
398		    HPET_MEM_WIDTH);
399	}
400}
401
402static int
403hpet_probe(device_t dev)
404{
405	ACPI_FUNCTION_TRACE((char *)(uintptr_t) __func__);
406
407	if (acpi_disabled("hpet"))
408		return (ENXIO);
409	if (acpi_get_handle(dev) != NULL &&
410	    ACPI_ID_PROBE(device_get_parent(dev), dev, hpet_ids) == NULL)
411		return (ENXIO);
412
413	device_set_desc(dev, "High Precision Event Timer");
414	return (0);
415}
416
417static int
418hpet_attach(device_t dev)
419{
420	struct hpet_softc *sc;
421	struct hpet_timer *t;
422	int i, j, num_msi, num_timers, num_percpu_et, num_percpu_t, cur_cpu;
423	int pcpu_master;
424	static int maxhpetet = 0;
425	uint32_t val, val2, cvectors, dvectors;
426	uint16_t vendor, rev;
427
428	ACPI_FUNCTION_TRACE((char *)(uintptr_t) __func__);
429
430	sc = device_get_softc(dev);
431	sc->dev = dev;
432	sc->handle = acpi_get_handle(dev);
433
434	sc->mem_rid = 0;
435	sc->mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &sc->mem_rid,
436	    RF_ACTIVE);
437	if (sc->mem_res == NULL)
438		return (ENOMEM);
439
440	/* Validate that we can access the whole region. */
441	if (rman_get_size(sc->mem_res) < HPET_MEM_WIDTH) {
442		device_printf(dev, "memory region width %ld too small\n",
443		    rman_get_size(sc->mem_res));
444		bus_free_resource(dev, SYS_RES_MEMORY, sc->mem_res);
445		return (ENXIO);
446	}
447
448	/* Be sure timer is enabled. */
449	hpet_enable(sc);
450
451	/* Read basic statistics about the timer. */
452	val = bus_read_4(sc->mem_res, HPET_PERIOD);
453	if (val == 0) {
454		device_printf(dev, "invalid period\n");
455		hpet_disable(sc);
456		bus_free_resource(dev, SYS_RES_MEMORY, sc->mem_res);
457		return (ENXIO);
458	}
459
460	sc->freq = (1000000000000000LL + val / 2) / val;
461	sc->caps = bus_read_4(sc->mem_res, HPET_CAPABILITIES);
462	vendor = (sc->caps & HPET_CAP_VENDOR_ID) >> 16;
463	rev = sc->caps & HPET_CAP_REV_ID;
464	num_timers = 1 + ((sc->caps & HPET_CAP_NUM_TIM) >> 8);
465	/*
466	 * ATI/AMD violates IA-PC HPET (High Precision Event Timers)
467	 * Specification and provides an off by one number
468	 * of timers/comparators.
469	 * Additionally, they use unregistered value in VENDOR_ID field.
470	 */
471	if (vendor == HPET_VENDID_AMD && rev < 0x10 && num_timers > 0)
472		num_timers--;
473	sc->num_timers = num_timers;
474	if (bootverbose) {
475		device_printf(dev,
476		    "vendor 0x%x, rev 0x%x, %jdHz%s, %d timers,%s\n",
477		    vendor, rev, sc->freq,
478		    (sc->caps & HPET_CAP_COUNT_SIZE) ? " 64bit" : "",
479		    num_timers,
480		    (sc->caps & HPET_CAP_LEG_RT) ? " legacy route" : "");
481	}
482	for (i = 0; i < num_timers; i++) {
483		t = &sc->t[i];
484		t->sc = sc;
485		t->num = i;
486		t->mode = 0;
487		t->intr_rid = -1;
488		t->irq = -1;
489		t->pcpu_cpu = -1;
490		t->pcpu_misrouted = 0;
491		t->pcpu_master = -1;
492		t->caps = bus_read_4(sc->mem_res, HPET_TIMER_CAP_CNF(i));
493		t->vectors = bus_read_4(sc->mem_res, HPET_TIMER_CAP_CNF(i) + 4);
494		if (bootverbose) {
495			device_printf(dev,
496			    " t%d: irqs 0x%08x (%d)%s%s%s\n", i,
497			    t->vectors, (t->caps & HPET_TCNF_INT_ROUTE) >> 9,
498			    (t->caps & HPET_TCAP_FSB_INT_DEL) ? ", MSI" : "",
499			    (t->caps & HPET_TCAP_SIZE) ? ", 64bit" : "",
500			    (t->caps & HPET_TCAP_PER_INT) ? ", periodic" : "");
501		}
502	}
503	if (testenv("debug.acpi.hpet_test"))
504		hpet_test(sc);
505	/*
506	 * Don't attach if the timer never increments.  Since the spec
507	 * requires it to be at least 10 MHz, it has to change in 1 us.
508	 */
509	val = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
510	DELAY(1);
511	val2 = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
512	if (val == val2) {
513		device_printf(dev, "HPET never increments, disabling\n");
514		hpet_disable(sc);
515		bus_free_resource(dev, SYS_RES_MEMORY, sc->mem_res);
516		return (ENXIO);
517	}
518	/* Announce first HPET as timecounter. */
519	if (device_get_unit(dev) == 0) {
520		sc->tc.tc_get_timecount = hpet_get_timecount,
521		sc->tc.tc_counter_mask = ~0u,
522		sc->tc.tc_name = "HPET",
523		sc->tc.tc_quality = 950,
524		sc->tc.tc_frequency = sc->freq;
525		sc->tc.tc_priv = sc;
526		tc_init(&sc->tc);
527	}
528	/* If not disabled - setup and announce event timers. */
529	if (resource_int_value(device_get_name(dev), device_get_unit(dev),
530	     "clock", &i) == 0 && i == 0)
531	        return (0);
532
533	/* Check whether we can and want legacy routing. */
534	sc->legacy_route = 0;
535	resource_int_value(device_get_name(dev), device_get_unit(dev),
536	     "legacy_route", &sc->legacy_route);
537	if ((sc->caps & HPET_CAP_LEG_RT) == 0)
538		sc->legacy_route = 0;
539	if (sc->legacy_route) {
540		sc->t[0].vectors = 0;
541		sc->t[1].vectors = 0;
542	}
543
544	/* Check what IRQs we want use. */
545	/* By default allow any PCI IRQs. */
546	sc->allowed_irqs = 0xffff0000;
547	/*
548	 * HPETs in AMD chipsets before SB800 have problems with IRQs >= 16
549	 * Lower are also not always working for different reasons.
550	 * SB800 fixed it, but seems do not implements level triggering
551	 * properly, that makes it very unreliable - it freezes after any
552	 * interrupt loss. Avoid legacy IRQs for AMD.
553	 */
554	if (vendor == HPET_VENDID_AMD || vendor == HPET_VENDID_AMD2)
555		sc->allowed_irqs = 0x00000000;
556	/*
557	 * NVidia MCP5x chipsets have number of unexplained interrupt
558	 * problems. For some reason, using HPET interrupts breaks HDA sound.
559	 */
560	if (vendor == HPET_VENDID_NVIDIA && rev <= 0x01)
561		sc->allowed_irqs = 0x00000000;
562	/*
563	 * ServerWorks HT1000 reported to have problems with IRQs >= 16.
564	 * Lower IRQs are working, but allowed mask is not set correctly.
565	 * Legacy_route mode works fine.
566	 */
567	if (vendor == HPET_VENDID_SW && rev <= 0x01)
568		sc->allowed_irqs = 0x00000000;
569	/*
570	 * Neither QEMU nor VirtualBox report supported IRQs correctly.
571	 * The only way to use HPET there is to specify IRQs manually
572	 * and/or use legacy_route. Legacy_route mode works on both.
573	 */
574	if (vm_guest)
575		sc->allowed_irqs = 0x00000000;
576	/* Let user override. */
577	resource_int_value(device_get_name(dev), device_get_unit(dev),
578	     "allowed_irqs", &sc->allowed_irqs);
579
580	/* Get how much per-CPU timers we should try to provide. */
581	sc->per_cpu = 1;
582	resource_int_value(device_get_name(dev), device_get_unit(dev),
583	     "per_cpu", &sc->per_cpu);
584
585	num_msi = 0;
586	sc->useirq = 0;
587	/* Find IRQ vectors for all timers. */
588	cvectors = sc->allowed_irqs & 0xffff0000;
589	dvectors = sc->allowed_irqs & 0x0000ffff;
590	if (sc->legacy_route)
591		dvectors &= 0x0000fefe;
592	for (i = 0; i < num_timers; i++) {
593		t = &sc->t[i];
594		if (sc->legacy_route && i < 2)
595			t->irq = (i == 0) ? 0 : 8;
596#ifdef DEV_APIC
597		else if (t->caps & HPET_TCAP_FSB_INT_DEL) {
598			if ((j = PCIB_ALLOC_MSIX(
599			    device_get_parent(device_get_parent(dev)), dev,
600			    &t->irq))) {
601				device_printf(dev,
602				    "Can't allocate interrupt for t%d.\n", j);
603			}
604		}
605#endif
606		else if (dvectors & t->vectors) {
607			t->irq = ffs(dvectors & t->vectors) - 1;
608			dvectors &= ~(1 << t->irq);
609		}
610		if (t->irq >= 0) {
611			t->intr_rid = hpet_find_irq_rid(dev, t->irq, t->irq);
612			t->intr_res = bus_alloc_resource(dev, SYS_RES_IRQ,
613			    &t->intr_rid, t->irq, t->irq, 1, RF_ACTIVE);
614			if (t->intr_res == NULL) {
615				t->irq = -1;
616				device_printf(dev,
617				    "Can't map interrupt for t%d.\n", i);
618			} else if (bus_setup_intr(dev, t->intr_res,
619			    INTR_TYPE_CLK, hpet_intr_single, NULL, t,
620			    &t->intr_handle) != 0) {
621				t->irq = -1;
622				device_printf(dev,
623				    "Can't setup interrupt for t%d.\n", i);
624			} else {
625				bus_describe_intr(dev, t->intr_res,
626				    t->intr_handle, "t%d", i);
627				num_msi++;
628			}
629		}
630		if (t->irq < 0 && (cvectors & t->vectors) != 0) {
631			cvectors &= t->vectors;
632			sc->useirq |= (1 << i);
633		}
634	}
635	if (sc->legacy_route && sc->t[0].irq < 0 && sc->t[1].irq < 0)
636		sc->legacy_route = 0;
637	if (sc->legacy_route)
638		hpet_enable(sc);
639	/* Group timers for per-CPU operation. */
640	num_percpu_et = min(num_msi / mp_ncpus, sc->per_cpu);
641	num_percpu_t = num_percpu_et * mp_ncpus;
642	pcpu_master = 0;
643	cur_cpu = CPU_FIRST();
644	for (i = 0; i < num_timers; i++) {
645		t = &sc->t[i];
646		if (t->irq >= 0 && num_percpu_t > 0) {
647			if (cur_cpu == CPU_FIRST())
648				pcpu_master = i;
649			t->pcpu_cpu = cur_cpu;
650			t->pcpu_master = pcpu_master;
651			sc->t[pcpu_master].
652			    pcpu_slaves[cur_cpu] = i;
653			bus_bind_intr(dev, t->intr_res, cur_cpu);
654			cur_cpu = CPU_NEXT(cur_cpu);
655			num_percpu_t--;
656		} else if (t->irq >= 0)
657			bus_bind_intr(dev, t->intr_res, CPU_FIRST());
658	}
659	bus_write_4(sc->mem_res, HPET_ISR, 0xffffffff);
660	sc->irq = -1;
661	/* If at least one timer needs legacy IRQ - set it up. */
662	if (sc->useirq) {
663		j = i = fls(cvectors) - 1;
664		while (j > 0 && (cvectors & (1 << (j - 1))) != 0)
665			j--;
666		sc->intr_rid = hpet_find_irq_rid(dev, j, i);
667		sc->intr_res = bus_alloc_resource(dev, SYS_RES_IRQ,
668		    &sc->intr_rid, j, i, 1, RF_SHAREABLE | RF_ACTIVE);
669		if (sc->intr_res == NULL)
670			device_printf(dev, "Can't map interrupt.\n");
671		else if (bus_setup_intr(dev, sc->intr_res, INTR_TYPE_CLK,
672		    hpet_intr, NULL, sc, &sc->intr_handle) != 0) {
673			device_printf(dev, "Can't setup interrupt.\n");
674		} else {
675			sc->irq = rman_get_start(sc->intr_res);
676			/* Bind IRQ to BSP to avoid live migration. */
677			bus_bind_intr(dev, sc->intr_res, CPU_FIRST());
678		}
679	}
680	/* Program and announce event timers. */
681	for (i = 0; i < num_timers; i++) {
682		t = &sc->t[i];
683		t->caps &= ~(HPET_TCNF_FSB_EN | HPET_TCNF_INT_ROUTE);
684		t->caps &= ~(HPET_TCNF_VAL_SET | HPET_TCNF_INT_ENB);
685		t->caps &= ~(HPET_TCNF_INT_TYPE);
686		t->caps |= HPET_TCNF_32MODE;
687		if (t->irq >= 0 && sc->legacy_route && i < 2) {
688			/* Legacy route doesn't need more configuration. */
689		} else
690#ifdef DEV_APIC
691		if ((t->caps & HPET_TCAP_FSB_INT_DEL) && t->irq >= 0) {
692			uint64_t addr;
693			uint32_t data;
694
695			if (PCIB_MAP_MSI(
696			    device_get_parent(device_get_parent(dev)), dev,
697			    t->irq, &addr, &data) == 0) {
698				bus_write_4(sc->mem_res,
699				    HPET_TIMER_FSB_ADDR(i), addr);
700				bus_write_4(sc->mem_res,
701				    HPET_TIMER_FSB_VAL(i), data);
702				t->caps |= HPET_TCNF_FSB_EN;
703			} else
704				t->irq = -2;
705		} else
706#endif
707		if (t->irq >= 0)
708			t->caps |= (t->irq << 9);
709		else if (sc->irq >= 0 && (t->vectors & (1 << sc->irq)))
710			t->caps |= (sc->irq << 9) | HPET_TCNF_INT_TYPE;
711		bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(i), t->caps);
712		/* Skip event timers without set up IRQ. */
713		if (t->irq < 0 &&
714		    (sc->irq < 0 || (t->vectors & (1 << sc->irq)) == 0))
715			continue;
716		/* Announce the reset. */
717		if (maxhpetet == 0)
718			t->et.et_name = "HPET";
719		else {
720			sprintf(t->name, "HPET%d", maxhpetet);
721			t->et.et_name = t->name;
722		}
723		t->et.et_flags = ET_FLAGS_PERIODIC | ET_FLAGS_ONESHOT;
724		t->et.et_quality = 450;
725		if (t->pcpu_master >= 0) {
726			t->et.et_flags |= ET_FLAGS_PERCPU;
727			t->et.et_quality += 100;
728		} else if (mp_ncpus >= 8)
729			t->et.et_quality -= 100;
730		if ((t->caps & HPET_TCAP_PER_INT) == 0)
731			t->et.et_quality -= 10;
732		t->et.et_frequency = sc->freq;
733		t->et.et_min_period =
734		    ((uint64_t)(HPET_MIN_CYCLES * 2) << 32) / sc->freq;
735		t->et.et_max_period = (0xfffffffeLLU << 32) / sc->freq;
736		t->et.et_start = hpet_start;
737		t->et.et_stop = hpet_stop;
738		t->et.et_priv = &sc->t[i];
739		if (t->pcpu_master < 0 || t->pcpu_master == i) {
740			et_register(&t->et);
741			maxhpetet++;
742		}
743	}
744
745	sc->pdev = make_dev(&hpet_cdevsw, 0, UID_ROOT, GID_WHEEL,
746	    0600, "hpet%d", device_get_unit(dev));
747	if (sc->pdev) {
748		sc->pdev->si_drv1 = sc;
749		sc->mmap_allow = 1;
750		TUNABLE_INT_FETCH("hw.acpi.hpet.mmap_allow",
751		    &sc->mmap_allow);
752		sc->mmap_allow_write = 1;
753		TUNABLE_INT_FETCH("hw.acpi.hpet.mmap_allow_write",
754		    &sc->mmap_allow_write);
755		SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
756		    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
757		    OID_AUTO, "mmap_allow",
758		    CTLFLAG_RW, &sc->mmap_allow, 0,
759		    "Allow userland to memory map HPET");
760		SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
761		    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
762		    OID_AUTO, "mmap_allow_write",
763		    CTLFLAG_RW, &sc->mmap_allow_write, 0,
764		    "Allow userland write to the HPET register space");
765	} else
766		device_printf(dev, "could not create /dev/hpet%d\n",
767		    device_get_unit(dev));
768
769	return (0);
770}
771
772static int
773hpet_detach(device_t dev)
774{
775	ACPI_FUNCTION_TRACE((char *)(uintptr_t) __func__);
776
777	/* XXX Without a tc_remove() function, we can't detach. */
778	return (EBUSY);
779}
780
781static int
782hpet_suspend(device_t dev)
783{
784//	struct hpet_softc *sc;
785
786	/*
787	 * Disable the timer during suspend.  The timer will not lose
788	 * its state in S1 or S2, but we are required to disable
789	 * it.
790	 */
791//	sc = device_get_softc(dev);
792//	hpet_disable(sc);
793
794	return (0);
795}
796
797static int
798hpet_resume(device_t dev)
799{
800	struct hpet_softc *sc;
801	struct hpet_timer *t;
802	int i;
803
804	/* Re-enable the timer after a resume to keep the clock advancing. */
805	sc = device_get_softc(dev);
806	hpet_enable(sc);
807	/* Restart event timers that were running on suspend. */
808	for (i = 0; i < sc->num_timers; i++) {
809		t = &sc->t[i];
810#ifdef DEV_APIC
811		if (t->irq >= 0 && (sc->legacy_route == 0 || i >= 2)) {
812			uint64_t addr;
813			uint32_t data;
814
815			if (PCIB_MAP_MSI(
816			    device_get_parent(device_get_parent(dev)), dev,
817			    t->irq, &addr, &data) == 0) {
818				bus_write_4(sc->mem_res,
819				    HPET_TIMER_FSB_ADDR(i), addr);
820				bus_write_4(sc->mem_res,
821				    HPET_TIMER_FSB_VAL(i), data);
822			}
823		}
824#endif
825		if (t->mode == 0)
826			continue;
827		t->next = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
828		if (t->mode == 1 && (t->caps & HPET_TCAP_PER_INT)) {
829			t->caps |= HPET_TCNF_TYPE;
830			t->next += t->div;
831			bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num),
832			    t->caps | HPET_TCNF_VAL_SET);
833			bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
834			    t->next);
835			bus_read_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num));
836			bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
837			    t->div);
838		} else {
839			t->next += sc->freq / 1024;
840			bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
841			    t->next);
842		}
843		bus_write_4(sc->mem_res, HPET_ISR, 1 << t->num);
844		bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num), t->caps);
845	}
846	return (0);
847}
848
849/* Print some basic latency/rate information to assist in debugging. */
850static void
851hpet_test(struct hpet_softc *sc)
852{
853	int i;
854	uint32_t u1, u2;
855	struct bintime b0, b1, b2;
856	struct timespec ts;
857
858	binuptime(&b0);
859	binuptime(&b0);
860	binuptime(&b1);
861	u1 = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
862	for (i = 1; i < 1000; i++)
863		u2 = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
864	binuptime(&b2);
865	u2 = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
866
867	bintime_sub(&b2, &b1);
868	bintime_sub(&b1, &b0);
869	bintime_sub(&b2, &b1);
870	bintime2timespec(&b2, &ts);
871
872	device_printf(sc->dev, "%ld.%09ld: %u ... %u = %u\n",
873	    (long)ts.tv_sec, ts.tv_nsec, u1, u2, u2 - u1);
874
875	device_printf(sc->dev, "time per call: %ld ns\n", ts.tv_nsec / 1000);
876}
877
878#ifdef DEV_APIC
879static int
880hpet_remap_intr(device_t dev, device_t child, u_int irq)
881{
882	struct hpet_softc *sc = device_get_softc(dev);
883	struct hpet_timer *t;
884	uint64_t addr;
885	uint32_t data;
886	int error, i;
887
888	for (i = 0; i < sc->num_timers; i++) {
889		t = &sc->t[i];
890		if (t->irq != irq)
891			continue;
892		error = PCIB_MAP_MSI(
893		    device_get_parent(device_get_parent(dev)), dev,
894		    irq, &addr, &data);
895		if (error)
896			return (error);
897		hpet_disable(sc); /* Stop timer to avoid interrupt loss. */
898		bus_write_4(sc->mem_res, HPET_TIMER_FSB_ADDR(i), addr);
899		bus_write_4(sc->mem_res, HPET_TIMER_FSB_VAL(i), data);
900		hpet_enable(sc);
901		return (0);
902	}
903	return (ENOENT);
904}
905#endif
906
907static device_method_t hpet_methods[] = {
908	/* Device interface */
909	DEVMETHOD(device_identify, hpet_identify),
910	DEVMETHOD(device_probe, hpet_probe),
911	DEVMETHOD(device_attach, hpet_attach),
912	DEVMETHOD(device_detach, hpet_detach),
913	DEVMETHOD(device_suspend, hpet_suspend),
914	DEVMETHOD(device_resume, hpet_resume),
915
916#ifdef DEV_APIC
917	DEVMETHOD(bus_remap_intr, hpet_remap_intr),
918#endif
919
920	DEVMETHOD_END
921};
922
923static driver_t	hpet_driver = {
924	"hpet",
925	hpet_methods,
926	sizeof(struct hpet_softc),
927};
928
929DRIVER_MODULE(hpet, acpi, hpet_driver, hpet_devclass, 0, 0);
930MODULE_DEPEND(hpet, acpi, 1, 1, 1);
931