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#include "opt_acpi.h"
30
31#if defined(__amd64__)
32#define	DEV_APIC
33#else
34#include "opt_apic.h"
35#endif
36#include <sys/param.h>
37#include <sys/conf.h>
38#include <sys/bus.h>
39#include <sys/kernel.h>
40#include <sys/module.h>
41#include <sys/proc.h>
42#include <sys/rman.h>
43#include <sys/mman.h>
44#include <sys/time.h>
45#include <sys/smp.h>
46#include <sys/sysctl.h>
47#include <sys/timeet.h>
48#include <sys/timetc.h>
49#include <sys/vdso.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_HYGON	0x1d94
64#define HPET_VENDID_INTEL	0x8086
65#define HPET_VENDID_NVIDIA	0x10de
66#define HPET_VENDID_SW		0x1166
67
68ACPI_SERIAL_DECL(hpet, "ACPI HPET support");
69
70/* ACPI CA debugging */
71#define _COMPONENT	ACPI_TIMER
72ACPI_MODULE_NAME("HPET")
73
74struct hpet_softc {
75	device_t		dev;
76	int			mem_rid;
77	int			intr_rid;
78	int			irq;
79	int			useirq;
80	int			legacy_route;
81	int			per_cpu;
82	uint32_t		allowed_irqs;
83	struct resource		*mem_res;
84	struct resource		*intr_res;
85	void			*intr_handle;
86	ACPI_HANDLE		handle;
87	uint32_t		acpi_uid;
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#define	TIMER_STOPPED	0
97#define	TIMER_PERIODIC	1
98#define	TIMER_ONESHOT	2
99		int			intr_rid;
100		int			irq;
101		int			pcpu_cpu;
102		int			pcpu_misrouted;
103		int			pcpu_master;
104		int			pcpu_slaves[MAXCPU];
105		struct resource		*intr_res;
106		void			*intr_handle;
107		uint32_t		caps;
108		uint32_t		vectors;
109		uint32_t		div;
110		uint32_t		next;
111		char			name[8];
112	} 			t[32];
113	int			num_timers;
114	struct cdev		*pdev;
115	int			mmap_allow;
116	int			mmap_allow_write;
117};
118
119static d_open_t hpet_open;
120static d_mmap_t hpet_mmap;
121
122static struct cdevsw hpet_cdevsw = {
123	.d_version =	D_VERSION,
124	.d_name =	"hpet",
125	.d_open =	hpet_open,
126	.d_mmap =	hpet_mmap,
127};
128
129static u_int hpet_get_timecount(struct timecounter *tc);
130static void hpet_test(struct hpet_softc *sc);
131
132static char *hpet_ids[] = { "PNP0103", NULL };
133
134/* Knob to disable acpi_hpet device */
135bool acpi_hpet_disabled = false;
136
137static u_int
138hpet_get_timecount(struct timecounter *tc)
139{
140	struct hpet_softc *sc;
141
142	sc = tc->tc_priv;
143	return (bus_read_4(sc->mem_res, HPET_MAIN_COUNTER));
144}
145
146uint32_t
147hpet_vdso_timehands(struct vdso_timehands *vdso_th, struct timecounter *tc)
148{
149	struct hpet_softc *sc;
150
151	sc = tc->tc_priv;
152	vdso_th->th_algo = VDSO_TH_ALGO_X86_HPET;
153	vdso_th->th_x86_shift = 0;
154	vdso_th->th_x86_hpet_idx = device_get_unit(sc->dev);
155	vdso_th->th_x86_pvc_last_systime = 0;
156	vdso_th->th_x86_pvc_stable_mask = 0;
157	bzero(vdso_th->th_res, sizeof(vdso_th->th_res));
158	return (sc->mmap_allow != 0);
159}
160
161#ifdef COMPAT_FREEBSD32
162uint32_t
163hpet_vdso_timehands32(struct vdso_timehands32 *vdso_th32,
164    struct timecounter *tc)
165{
166	struct hpet_softc *sc;
167
168	sc = tc->tc_priv;
169	vdso_th32->th_algo = VDSO_TH_ALGO_X86_HPET;
170	vdso_th32->th_x86_shift = 0;
171	vdso_th32->th_x86_hpet_idx = device_get_unit(sc->dev);
172	vdso_th32->th_x86_pvc_last_systime[0] = 0;
173	vdso_th32->th_x86_pvc_last_systime[1] = 0;
174	vdso_th32->th_x86_pvc_stable_mask = 0;
175	bzero(vdso_th32->th_res, sizeof(vdso_th32->th_res));
176	return (sc->mmap_allow != 0);
177}
178#endif
179
180static void
181hpet_enable(struct hpet_softc *sc)
182{
183	uint32_t val;
184
185	val = bus_read_4(sc->mem_res, HPET_CONFIG);
186	if (sc->legacy_route)
187		val |= HPET_CNF_LEG_RT;
188	else
189		val &= ~HPET_CNF_LEG_RT;
190	val |= HPET_CNF_ENABLE;
191	bus_write_4(sc->mem_res, HPET_CONFIG, val);
192}
193
194static void
195hpet_disable(struct hpet_softc *sc)
196{
197	uint32_t val;
198
199	val = bus_read_4(sc->mem_res, HPET_CONFIG);
200	val &= ~HPET_CNF_ENABLE;
201	bus_write_4(sc->mem_res, HPET_CONFIG, val);
202}
203
204static int
205hpet_start(struct eventtimer *et, sbintime_t first, sbintime_t period)
206{
207	struct hpet_timer *mt = (struct hpet_timer *)et->et_priv;
208	struct hpet_timer *t;
209	struct hpet_softc *sc = mt->sc;
210	uint32_t fdiv, now;
211
212	t = (mt->pcpu_master < 0) ? mt : &sc->t[mt->pcpu_slaves[curcpu]];
213	if (period != 0) {
214		t->mode = TIMER_PERIODIC;
215		t->div = (sc->freq * period) >> 32;
216	} else {
217		t->mode = TIMER_ONESHOT;
218		t->div = 0;
219	}
220	if (first != 0)
221		fdiv = (sc->freq * first) >> 32;
222	else
223		fdiv = t->div;
224	if (t->irq < 0)
225		bus_write_4(sc->mem_res, HPET_ISR, 1 << t->num);
226	t->caps |= HPET_TCNF_INT_ENB;
227	now = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
228restart:
229	t->next = now + fdiv;
230	if (t->mode == TIMER_PERIODIC && (t->caps & HPET_TCAP_PER_INT)) {
231		t->caps |= HPET_TCNF_TYPE;
232		bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num),
233		    t->caps | HPET_TCNF_VAL_SET);
234		bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
235		    t->next);
236		bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
237		    t->div);
238	} else {
239		t->caps &= ~HPET_TCNF_TYPE;
240		bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num),
241		    t->caps);
242		bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
243		    t->next);
244	}
245	now = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
246	if ((int32_t)(now - t->next + HPET_MIN_CYCLES) >= 0) {
247		fdiv *= 2;
248		goto restart;
249	}
250	return (0);
251}
252
253static int
254hpet_stop(struct eventtimer *et)
255{
256	struct hpet_timer *mt = (struct hpet_timer *)et->et_priv;
257	struct hpet_timer *t;
258	struct hpet_softc *sc = mt->sc;
259
260	t = (mt->pcpu_master < 0) ? mt : &sc->t[mt->pcpu_slaves[curcpu]];
261	t->mode = TIMER_STOPPED;
262	t->caps &= ~(HPET_TCNF_INT_ENB | HPET_TCNF_TYPE);
263	bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num), t->caps);
264	return (0);
265}
266
267static int
268hpet_intr_single(void *arg)
269{
270	struct hpet_timer *t = (struct hpet_timer *)arg;
271	struct hpet_timer *mt;
272	struct hpet_softc *sc = t->sc;
273	uint32_t now;
274
275	if (t->mode == TIMER_STOPPED)
276		return (FILTER_STRAY);
277	/* Check that per-CPU timer interrupt reached right CPU. */
278	if (t->pcpu_cpu >= 0 && t->pcpu_cpu != curcpu) {
279		if ((++t->pcpu_misrouted) % 32 == 0) {
280			printf("HPET interrupt routed to the wrong CPU"
281			    " (timer %d CPU %d -> %d)!\n",
282			    t->num, t->pcpu_cpu, curcpu);
283		}
284
285		/*
286		 * Reload timer, hoping that next time may be more lucky
287		 * (system will manage proper interrupt binding).
288		 */
289		if ((t->mode == TIMER_PERIODIC &&
290		    (t->caps & HPET_TCAP_PER_INT) == 0) ||
291		    t->mode == TIMER_ONESHOT) {
292			t->next = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER) +
293			    sc->freq / 8;
294			bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
295			    t->next);
296		}
297		return (FILTER_HANDLED);
298	}
299	if (t->mode == TIMER_PERIODIC &&
300	    (t->caps & HPET_TCAP_PER_INT) == 0) {
301		t->next += t->div;
302		now = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
303		if ((int32_t)((now + t->div / 2) - t->next) > 0)
304			t->next = now + t->div / 2;
305		bus_write_4(sc->mem_res,
306		    HPET_TIMER_COMPARATOR(t->num), t->next);
307	} else if (t->mode == TIMER_ONESHOT)
308		t->mode = TIMER_STOPPED;
309	mt = (t->pcpu_master < 0) ? t : &sc->t[t->pcpu_master];
310	if (mt->et.et_active)
311		mt->et.et_event_cb(&mt->et, mt->et.et_arg);
312	return (FILTER_HANDLED);
313}
314
315static int
316hpet_intr(void *arg)
317{
318	struct hpet_softc *sc = (struct hpet_softc *)arg;
319	int i;
320	uint32_t val;
321
322	val = bus_read_4(sc->mem_res, HPET_ISR);
323	if (val) {
324		bus_write_4(sc->mem_res, HPET_ISR, val);
325		val &= sc->useirq;
326		for (i = 0; i < sc->num_timers; i++) {
327			if ((val & (1 << i)) == 0)
328				continue;
329			hpet_intr_single(&sc->t[i]);
330		}
331		return (FILTER_HANDLED);
332	}
333	return (FILTER_STRAY);
334}
335
336uint32_t
337hpet_get_uid(device_t dev)
338{
339	struct hpet_softc *sc;
340
341	sc = device_get_softc(dev);
342	return (sc->acpi_uid);
343}
344
345static ACPI_STATUS
346hpet_find(ACPI_HANDLE handle, UINT32 level, void *context,
347    void **status)
348{
349	char 		**ids;
350	uint32_t	id = (uint32_t)(uintptr_t)context;
351	uint32_t	uid = 0;
352
353	for (ids = hpet_ids; *ids != NULL; ids++) {
354		if (acpi_MatchHid(handle, *ids))
355		        break;
356	}
357	if (*ids == NULL)
358		return (AE_OK);
359	if (ACPI_FAILURE(acpi_GetInteger(handle, "_UID", &uid)) ||
360	    id == uid)
361		*status = acpi_get_device(handle);
362	return (AE_OK);
363}
364
365/*
366 * Find an existing IRQ resource that matches the requested IRQ range
367 * and return its RID.  If one is not found, use a new RID.
368 */
369static int
370hpet_find_irq_rid(device_t dev, u_long start, u_long end)
371{
372	rman_res_t irq;
373	int error, rid;
374
375	for (rid = 0;; rid++) {
376		error = bus_get_resource(dev, SYS_RES_IRQ, rid, &irq, NULL);
377		if (error != 0 || (start <= irq && irq <= end))
378			return (rid);
379	}
380}
381
382static int
383hpet_open(struct cdev *cdev, int oflags, int devtype, struct thread *td)
384{
385	struct hpet_softc *sc;
386
387	sc = cdev->si_drv1;
388	if (!sc->mmap_allow)
389		return (EPERM);
390	else
391		return (0);
392}
393
394static int
395hpet_mmap(struct cdev *cdev, vm_ooffset_t offset, vm_paddr_t *paddr,
396    int nprot, vm_memattr_t *memattr)
397{
398	struct hpet_softc *sc;
399
400	sc = cdev->si_drv1;
401	if (offset >= rman_get_size(sc->mem_res))
402		return (EINVAL);
403	if (!sc->mmap_allow_write && (nprot & PROT_WRITE))
404		return (EPERM);
405	*paddr = rman_get_start(sc->mem_res) + offset;
406	*memattr = VM_MEMATTR_UNCACHEABLE;
407
408	return (0);
409}
410
411/* Discover the HPET via the ACPI table of the same name. */
412static void
413hpet_identify(driver_t *driver, device_t parent)
414{
415	ACPI_TABLE_HPET *hpet;
416	ACPI_STATUS	status;
417	device_t	child;
418	int		i;
419
420	/* Only one HPET device can be added. */
421	if (devclass_get_device(devclass_find("hpet"), 0))
422		return;
423	for (i = 1; ; i++) {
424		/* Search for HPET table. */
425		status = AcpiGetTable(ACPI_SIG_HPET, i, (ACPI_TABLE_HEADER **)&hpet);
426		if (ACPI_FAILURE(status))
427			return;
428		/* Search for HPET device with same ID. */
429		child = NULL;
430		AcpiWalkNamespace(ACPI_TYPE_DEVICE, ACPI_ROOT_OBJECT,
431		    100, hpet_find, NULL, (void *)(uintptr_t)hpet->Sequence,
432		    (void *)&child);
433		/* If found - let it be probed in normal way. */
434		if (child) {
435			if (bus_get_resource(child, SYS_RES_MEMORY, 0,
436			    NULL, NULL) != 0)
437				bus_set_resource(child, SYS_RES_MEMORY, 0,
438				    hpet->Address.Address, HPET_MEM_WIDTH);
439			continue;
440		}
441		/* If not - create it from table info. */
442		child = BUS_ADD_CHILD(parent, 2, "hpet", 0);
443		if (child == NULL) {
444			printf("%s: can't add child\n", __func__);
445			continue;
446		}
447		bus_set_resource(child, SYS_RES_MEMORY, 0, hpet->Address.Address,
448		    HPET_MEM_WIDTH);
449	}
450}
451
452static int
453hpet_probe(device_t dev)
454{
455	int rv;
456
457	ACPI_FUNCTION_TRACE((char *)(uintptr_t) __func__);
458	if (acpi_disabled("hpet") || acpi_hpet_disabled)
459		return (ENXIO);
460	if (acpi_get_handle(dev) != NULL)
461		rv = ACPI_ID_PROBE(device_get_parent(dev), dev, hpet_ids, NULL);
462	else
463		rv = 0;
464	if (rv <= 0)
465		device_set_desc(dev, "High Precision Event Timer");
466	return (rv);
467}
468
469static int
470hpet_attach(device_t dev)
471{
472	struct hpet_softc *sc;
473	struct hpet_timer *t;
474	struct make_dev_args mda;
475	int i, j, num_msi, num_timers, num_percpu_et, num_percpu_t, cur_cpu;
476	int pcpu_master, error;
477	rman_res_t hpet_region_size;
478	static int maxhpetet = 0;
479	uint32_t val, val2, cvectors, dvectors;
480	uint16_t vendor, rev;
481
482	ACPI_FUNCTION_TRACE((char *)(uintptr_t) __func__);
483
484	sc = device_get_softc(dev);
485	sc->dev = dev;
486	sc->handle = acpi_get_handle(dev);
487
488	sc->mem_rid = 0;
489	sc->mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &sc->mem_rid,
490	    RF_ACTIVE);
491	if (sc->mem_res == NULL)
492		return (ENOMEM);
493
494	hpet_region_size = rman_get_size(sc->mem_res);
495	/* Validate that the region is big enough for the control registers. */
496	if (hpet_region_size < HPET_MEM_MIN_WIDTH) {
497		device_printf(dev, "memory region width %jd too small\n",
498		    hpet_region_size);
499		bus_free_resource(dev, SYS_RES_MEMORY, sc->mem_res);
500		return (ENXIO);
501	}
502
503	/* Be sure timer is enabled. */
504	hpet_enable(sc);
505
506	/* Read basic statistics about the timer. */
507	val = bus_read_4(sc->mem_res, HPET_PERIOD);
508	if (val == 0) {
509		device_printf(dev, "invalid period\n");
510		hpet_disable(sc);
511		bus_free_resource(dev, SYS_RES_MEMORY, sc->mem_res);
512		return (ENXIO);
513	}
514
515	sc->freq = (1000000000000000LL + val / 2) / val;
516	sc->caps = bus_read_4(sc->mem_res, HPET_CAPABILITIES);
517	vendor = (sc->caps & HPET_CAP_VENDOR_ID) >> 16;
518	rev = sc->caps & HPET_CAP_REV_ID;
519	num_timers = 1 + ((sc->caps & HPET_CAP_NUM_TIM) >> 8);
520	/*
521	 * ATI/AMD violates IA-PC HPET (High Precision Event Timers)
522	 * Specification and provides an off by one number
523	 * of timers/comparators.
524	 * Additionally, they use unregistered value in VENDOR_ID field.
525	 */
526	if (vendor == HPET_VENDID_AMD && rev < 0x10 && num_timers > 0)
527		num_timers--;
528	/*
529	 * Now validate that the region is big enough to address all counters.
530	 */
531	if (hpet_region_size < HPET_TIMER_CAP_CNF(num_timers)) {
532		device_printf(dev,
533		    "memory region width %jd too small for %d timers\n",
534		    hpet_region_size, num_timers);
535		hpet_disable(sc);
536		bus_free_resource(dev, SYS_RES_MEMORY, sc->mem_res);
537		return (ENXIO);
538	}
539
540	sc->num_timers = num_timers;
541	if (bootverbose) {
542		device_printf(dev,
543		    "vendor 0x%x, rev 0x%x, %jdHz%s, %d timers,%s\n",
544		    vendor, rev, sc->freq,
545		    (sc->caps & HPET_CAP_COUNT_SIZE) ? " 64bit" : "",
546		    num_timers,
547		    (sc->caps & HPET_CAP_LEG_RT) ? " legacy route" : "");
548	}
549	for (i = 0; i < num_timers; i++) {
550		t = &sc->t[i];
551		t->sc = sc;
552		t->num = i;
553		t->mode = TIMER_STOPPED;
554		t->intr_rid = -1;
555		t->irq = -1;
556		t->pcpu_cpu = -1;
557		t->pcpu_misrouted = 0;
558		t->pcpu_master = -1;
559		t->caps = bus_read_4(sc->mem_res, HPET_TIMER_CAP_CNF(i));
560		t->vectors = bus_read_4(sc->mem_res, HPET_TIMER_CAP_CNF(i) + 4);
561		if (bootverbose) {
562			device_printf(dev,
563			    " t%d: irqs 0x%08x (%d)%s%s%s\n", i,
564			    t->vectors, (t->caps & HPET_TCNF_INT_ROUTE) >> 9,
565			    (t->caps & HPET_TCAP_FSB_INT_DEL) ? ", MSI" : "",
566			    (t->caps & HPET_TCAP_SIZE) ? ", 64bit" : "",
567			    (t->caps & HPET_TCAP_PER_INT) ? ", periodic" : "");
568		}
569	}
570	if (testenv("debug.acpi.hpet_test"))
571		hpet_test(sc);
572	/*
573	 * Don't attach if the timer never increments.  Since the spec
574	 * requires it to be at least 10 MHz, it has to change in 1 us.
575	 */
576	val = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
577	DELAY(1);
578	val2 = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
579	if (val == val2) {
580		device_printf(dev, "HPET never increments, disabling\n");
581		hpet_disable(sc);
582		bus_free_resource(dev, SYS_RES_MEMORY, sc->mem_res);
583		return (ENXIO);
584	}
585	/* Announce first HPET as timecounter. */
586	if (device_get_unit(dev) == 0) {
587		sc->tc.tc_get_timecount = hpet_get_timecount,
588		sc->tc.tc_counter_mask = ~0u,
589		sc->tc.tc_name = "HPET",
590		sc->tc.tc_quality = 950,
591		sc->tc.tc_frequency = sc->freq;
592		sc->tc.tc_priv = sc;
593		sc->tc.tc_fill_vdso_timehands = hpet_vdso_timehands;
594#ifdef COMPAT_FREEBSD32
595		sc->tc.tc_fill_vdso_timehands32 = hpet_vdso_timehands32;
596#endif
597		tc_init(&sc->tc);
598	}
599	/* If not disabled - setup and announce event timers. */
600	if (resource_int_value(device_get_name(dev), device_get_unit(dev),
601	     "clock", &i) == 0 && i == 0)
602	        return (0);
603
604	/* Check whether we can and want legacy routing. */
605	sc->legacy_route = 0;
606	resource_int_value(device_get_name(dev), device_get_unit(dev),
607	     "legacy_route", &sc->legacy_route);
608	if ((sc->caps & HPET_CAP_LEG_RT) == 0)
609		sc->legacy_route = 0;
610	if (sc->legacy_route) {
611		sc->t[0].vectors = 0;
612		sc->t[1].vectors = 0;
613	}
614
615	/* Check what IRQs we want use. */
616	/* By default allow any PCI IRQs. */
617	sc->allowed_irqs = 0xffff0000;
618	/*
619	 * HPETs in AMD chipsets before SB800 have problems with IRQs >= 16
620	 * Lower are also not always working for different reasons.
621	 * SB800 fixed it, but seems do not implements level triggering
622	 * properly, that makes it very unreliable - it freezes after any
623	 * interrupt loss. Avoid legacy IRQs for AMD.
624	 */
625	if (vendor == HPET_VENDID_AMD || vendor == HPET_VENDID_AMD2 ||
626	    vendor == HPET_VENDID_HYGON)
627		sc->allowed_irqs = 0x00000000;
628	/*
629	 * NVidia MCP5x chipsets have number of unexplained interrupt
630	 * problems. For some reason, using HPET interrupts breaks HDA sound.
631	 */
632	if (vendor == HPET_VENDID_NVIDIA && rev <= 0x01)
633		sc->allowed_irqs = 0x00000000;
634	/*
635	 * ServerWorks HT1000 reported to have problems with IRQs >= 16.
636	 * Lower IRQs are working, but allowed mask is not set correctly.
637	 * Legacy_route mode works fine.
638	 */
639	if (vendor == HPET_VENDID_SW && rev <= 0x01)
640		sc->allowed_irqs = 0x00000000;
641	/*
642	 * Neither QEMU nor VirtualBox report supported IRQs correctly.
643	 * The only way to use HPET there is to specify IRQs manually
644	 * and/or use legacy_route. Legacy_route mode works on both.
645	 */
646	if (vm_guest != VM_GUEST_NO)
647		sc->allowed_irqs = 0x00000000;
648	/* Let user override. */
649	resource_int_value(device_get_name(dev), device_get_unit(dev),
650	     "allowed_irqs", &sc->allowed_irqs);
651
652	/* Get how much per-CPU timers we should try to provide. */
653	sc->per_cpu = 1;
654	resource_int_value(device_get_name(dev), device_get_unit(dev),
655	     "per_cpu", &sc->per_cpu);
656
657	num_msi = 0;
658	sc->useirq = 0;
659	/* Find IRQ vectors for all timers. */
660	cvectors = sc->allowed_irqs & 0xffff0000;
661	dvectors = sc->allowed_irqs & 0x0000ffff;
662	if (sc->legacy_route)
663		dvectors &= 0x0000fefe;
664	for (i = 0; i < num_timers; i++) {
665		t = &sc->t[i];
666		if (sc->legacy_route && i < 2)
667			t->irq = (i == 0) ? 0 : 8;
668#ifdef DEV_APIC
669		else if (t->caps & HPET_TCAP_FSB_INT_DEL) {
670			if ((j = PCIB_ALLOC_MSIX(
671			    device_get_parent(device_get_parent(dev)), dev,
672			    &t->irq))) {
673				device_printf(dev,
674				    "Can't allocate interrupt for t%d: %d\n",
675				    i, j);
676			}
677		}
678#endif
679		else if (dvectors & t->vectors) {
680			t->irq = ffs(dvectors & t->vectors) - 1;
681			dvectors &= ~(1 << t->irq);
682		}
683		if (t->irq >= 0) {
684			t->intr_rid = hpet_find_irq_rid(dev, t->irq, t->irq);
685			t->intr_res = bus_alloc_resource(dev, SYS_RES_IRQ,
686			    &t->intr_rid, t->irq, t->irq, 1, RF_ACTIVE);
687			if (t->intr_res == NULL) {
688				t->irq = -1;
689				device_printf(dev,
690				    "Can't map interrupt for t%d.\n", i);
691			} else if (bus_setup_intr(dev, t->intr_res,
692			    INTR_TYPE_CLK, hpet_intr_single, NULL, t,
693			    &t->intr_handle) != 0) {
694				t->irq = -1;
695				device_printf(dev,
696				    "Can't setup interrupt for t%d.\n", i);
697			} else {
698				bus_describe_intr(dev, t->intr_res,
699				    t->intr_handle, "t%d", i);
700				num_msi++;
701			}
702		}
703		if (t->irq < 0 && (cvectors & t->vectors) != 0) {
704			cvectors &= t->vectors;
705			sc->useirq |= (1 << i);
706		}
707	}
708	if (sc->legacy_route && sc->t[0].irq < 0 && sc->t[1].irq < 0)
709		sc->legacy_route = 0;
710	if (sc->legacy_route)
711		hpet_enable(sc);
712	/* Group timers for per-CPU operation. */
713	num_percpu_et = min(num_msi / mp_ncpus, sc->per_cpu);
714	num_percpu_t = num_percpu_et * mp_ncpus;
715	pcpu_master = 0;
716	cur_cpu = CPU_FIRST();
717	for (i = 0; i < num_timers; i++) {
718		t = &sc->t[i];
719		if (t->irq >= 0 && num_percpu_t > 0) {
720			if (cur_cpu == CPU_FIRST())
721				pcpu_master = i;
722			t->pcpu_cpu = cur_cpu;
723			t->pcpu_master = pcpu_master;
724			sc->t[pcpu_master].
725			    pcpu_slaves[cur_cpu] = i;
726			bus_bind_intr(dev, t->intr_res, cur_cpu);
727			cur_cpu = CPU_NEXT(cur_cpu);
728			num_percpu_t--;
729		} else if (t->irq >= 0)
730			bus_bind_intr(dev, t->intr_res, CPU_FIRST());
731	}
732	bus_write_4(sc->mem_res, HPET_ISR, 0xffffffff);
733	sc->irq = -1;
734	/* If at least one timer needs legacy IRQ - set it up. */
735	if (sc->useirq) {
736		j = i = fls(cvectors) - 1;
737		while (j > 0 && (cvectors & (1 << (j - 1))) != 0)
738			j--;
739		sc->intr_rid = hpet_find_irq_rid(dev, j, i);
740		sc->intr_res = bus_alloc_resource(dev, SYS_RES_IRQ,
741		    &sc->intr_rid, j, i, 1, RF_SHAREABLE | RF_ACTIVE);
742		if (sc->intr_res == NULL)
743			device_printf(dev, "Can't map interrupt.\n");
744		else if (bus_setup_intr(dev, sc->intr_res, INTR_TYPE_CLK,
745		    hpet_intr, NULL, sc, &sc->intr_handle) != 0) {
746			device_printf(dev, "Can't setup interrupt.\n");
747		} else {
748			sc->irq = rman_get_start(sc->intr_res);
749			/* Bind IRQ to BSP to avoid live migration. */
750			bus_bind_intr(dev, sc->intr_res, CPU_FIRST());
751		}
752	}
753	/* Program and announce event timers. */
754	for (i = 0; i < num_timers; i++) {
755		t = &sc->t[i];
756		t->caps &= ~(HPET_TCNF_FSB_EN | HPET_TCNF_INT_ROUTE);
757		t->caps &= ~(HPET_TCNF_VAL_SET | HPET_TCNF_INT_ENB);
758		t->caps &= ~(HPET_TCNF_INT_TYPE);
759		t->caps |= HPET_TCNF_32MODE;
760		if (t->irq >= 0 && sc->legacy_route && i < 2) {
761			/* Legacy route doesn't need more configuration. */
762		} else
763#ifdef DEV_APIC
764		if ((t->caps & HPET_TCAP_FSB_INT_DEL) && t->irq >= 0) {
765			uint64_t addr;
766			uint32_t data;
767
768			if (PCIB_MAP_MSI(
769			    device_get_parent(device_get_parent(dev)), dev,
770			    t->irq, &addr, &data) == 0) {
771				bus_write_4(sc->mem_res,
772				    HPET_TIMER_FSB_ADDR(i), addr);
773				bus_write_4(sc->mem_res,
774				    HPET_TIMER_FSB_VAL(i), data);
775				t->caps |= HPET_TCNF_FSB_EN;
776			} else
777				t->irq = -2;
778		} else
779#endif
780		if (t->irq >= 0)
781			t->caps |= (t->irq << 9);
782		else if (sc->irq >= 0 && (t->vectors & (1 << sc->irq)))
783			t->caps |= (sc->irq << 9) | HPET_TCNF_INT_TYPE;
784		bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(i), t->caps);
785		/* Skip event timers without set up IRQ. */
786		if (t->irq < 0 &&
787		    (sc->irq < 0 || (t->vectors & (1 << sc->irq)) == 0))
788			continue;
789		/* Announce the reset. */
790		if (maxhpetet == 0)
791			t->et.et_name = "HPET";
792		else {
793			sprintf(t->name, "HPET%d", maxhpetet);
794			t->et.et_name = t->name;
795		}
796		t->et.et_flags = ET_FLAGS_PERIODIC | ET_FLAGS_ONESHOT;
797		t->et.et_quality = 450;
798		if (t->pcpu_master >= 0) {
799			t->et.et_flags |= ET_FLAGS_PERCPU;
800			t->et.et_quality += 100;
801		} else if (mp_ncpus >= 8)
802			t->et.et_quality -= 100;
803		if ((t->caps & HPET_TCAP_PER_INT) == 0)
804			t->et.et_quality -= 10;
805		t->et.et_frequency = sc->freq;
806		t->et.et_min_period =
807		    ((uint64_t)(HPET_MIN_CYCLES * 2) << 32) / sc->freq;
808		t->et.et_max_period = (0xfffffffeLLU << 32) / sc->freq;
809		t->et.et_start = hpet_start;
810		t->et.et_stop = hpet_stop;
811		t->et.et_priv = &sc->t[i];
812		if (t->pcpu_master < 0 || t->pcpu_master == i) {
813			et_register(&t->et);
814			maxhpetet++;
815		}
816	}
817	acpi_GetInteger(sc->handle, "_UID", &sc->acpi_uid);
818
819	make_dev_args_init(&mda);
820	mda.mda_devsw = &hpet_cdevsw;
821	mda.mda_uid = UID_ROOT;
822	mda.mda_gid = GID_WHEEL;
823	mda.mda_mode = 0644;
824	mda.mda_si_drv1 = sc;
825	error = make_dev_s(&mda, &sc->pdev, "hpet%d", device_get_unit(dev));
826	if (error == 0) {
827		sc->mmap_allow = 1;
828		TUNABLE_INT_FETCH("hw.acpi.hpet.mmap_allow",
829		    &sc->mmap_allow);
830		sc->mmap_allow_write = 0;
831		TUNABLE_INT_FETCH("hw.acpi.hpet.mmap_allow_write",
832		    &sc->mmap_allow_write);
833		SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
834		    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
835		    OID_AUTO, "mmap_allow",
836		    CTLFLAG_RW, &sc->mmap_allow, 0,
837		    "Allow userland to memory map HPET");
838		SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
839		    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
840		    OID_AUTO, "mmap_allow_write",
841		    CTLFLAG_RW, &sc->mmap_allow_write, 0,
842		    "Allow userland write to the HPET register space");
843	} else {
844		device_printf(dev, "could not create /dev/hpet%d, error %d\n",
845		    device_get_unit(dev), error);
846	}
847
848	return (0);
849}
850
851static int
852hpet_detach(device_t dev)
853{
854	ACPI_FUNCTION_TRACE((char *)(uintptr_t) __func__);
855
856	/* XXX Without a tc_remove() function, we can't detach. */
857	return (EBUSY);
858}
859
860static int
861hpet_suspend(device_t dev)
862{
863//	struct hpet_softc *sc;
864
865	/*
866	 * Disable the timer during suspend.  The timer will not lose
867	 * its state in S1 or S2, but we are required to disable
868	 * it.
869	 */
870//	sc = device_get_softc(dev);
871//	hpet_disable(sc);
872
873	return (0);
874}
875
876static int
877hpet_resume(device_t dev)
878{
879	struct hpet_softc *sc;
880	struct hpet_timer *t;
881	int i;
882
883	/* Re-enable the timer after a resume to keep the clock advancing. */
884	sc = device_get_softc(dev);
885	hpet_enable(sc);
886	/* Restart event timers that were running on suspend. */
887	for (i = 0; i < sc->num_timers; i++) {
888		t = &sc->t[i];
889#ifdef DEV_APIC
890		if (t->irq >= 0 && (sc->legacy_route == 0 || i >= 2)) {
891			uint64_t addr;
892			uint32_t data;
893
894			if (PCIB_MAP_MSI(
895			    device_get_parent(device_get_parent(dev)), dev,
896			    t->irq, &addr, &data) == 0) {
897				bus_write_4(sc->mem_res,
898				    HPET_TIMER_FSB_ADDR(i), addr);
899				bus_write_4(sc->mem_res,
900				    HPET_TIMER_FSB_VAL(i), data);
901			}
902		}
903#endif
904		if (t->mode == TIMER_STOPPED)
905			continue;
906		t->next = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
907		if (t->mode == TIMER_PERIODIC &&
908		    (t->caps & HPET_TCAP_PER_INT) != 0) {
909			t->caps |= HPET_TCNF_TYPE;
910			t->next += t->div;
911			bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num),
912			    t->caps | HPET_TCNF_VAL_SET);
913			bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
914			    t->next);
915			bus_read_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num));
916			bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
917			    t->div);
918		} else {
919			t->next += sc->freq / 1024;
920			bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
921			    t->next);
922		}
923		bus_write_4(sc->mem_res, HPET_ISR, 1 << t->num);
924		bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num), t->caps);
925	}
926	return (0);
927}
928
929/* Print some basic latency/rate information to assist in debugging. */
930static void
931hpet_test(struct hpet_softc *sc)
932{
933	int i;
934	uint32_t u1, u2;
935	struct bintime b0, b1, b2;
936	struct timespec ts;
937
938	binuptime(&b0);
939	binuptime(&b0);
940	binuptime(&b1);
941	u1 = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
942	for (i = 1; i < 1000; i++)
943		u2 = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
944	binuptime(&b2);
945	u2 = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
946
947	bintime_sub(&b2, &b1);
948	bintime_sub(&b1, &b0);
949	bintime_sub(&b2, &b1);
950	bintime2timespec(&b2, &ts);
951
952	device_printf(sc->dev, "%ld.%09ld: %u ... %u = %u\n",
953	    (long)ts.tv_sec, ts.tv_nsec, u1, u2, u2 - u1);
954
955	device_printf(sc->dev, "time per call: %ld ns\n", ts.tv_nsec / 1000);
956}
957
958#ifdef DEV_APIC
959static int
960hpet_remap_intr(device_t dev, device_t child, u_int irq)
961{
962	struct hpet_softc *sc = device_get_softc(dev);
963	struct hpet_timer *t;
964	uint64_t addr;
965	uint32_t data;
966	int error, i;
967
968	for (i = 0; i < sc->num_timers; i++) {
969		t = &sc->t[i];
970		if (t->irq != irq)
971			continue;
972		error = PCIB_MAP_MSI(
973		    device_get_parent(device_get_parent(dev)), dev,
974		    irq, &addr, &data);
975		if (error)
976			return (error);
977		hpet_disable(sc); /* Stop timer to avoid interrupt loss. */
978		bus_write_4(sc->mem_res, HPET_TIMER_FSB_ADDR(i), addr);
979		bus_write_4(sc->mem_res, HPET_TIMER_FSB_VAL(i), data);
980		hpet_enable(sc);
981		return (0);
982	}
983	return (ENOENT);
984}
985#endif
986
987static device_method_t hpet_methods[] = {
988	/* Device interface */
989	DEVMETHOD(device_identify, hpet_identify),
990	DEVMETHOD(device_probe, hpet_probe),
991	DEVMETHOD(device_attach, hpet_attach),
992	DEVMETHOD(device_detach, hpet_detach),
993	DEVMETHOD(device_suspend, hpet_suspend),
994	DEVMETHOD(device_resume, hpet_resume),
995
996#ifdef DEV_APIC
997	DEVMETHOD(bus_remap_intr, hpet_remap_intr),
998#endif
999
1000	DEVMETHOD_END
1001};
1002
1003static driver_t	hpet_driver = {
1004	"hpet",
1005	hpet_methods,
1006	sizeof(struct hpet_softc),
1007};
1008
1009DRIVER_MODULE(hpet, acpi, hpet_driver, 0, 0);
1010MODULE_DEPEND(hpet, acpi, 1, 1, 1);
1011