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("$FreeBSD$");
30
31#include "opt_acpi.h"
32
33#if defined(__amd64__)
34#define	DEV_APIC
35#else
36#include "opt_apic.h"
37#endif
38#include <sys/param.h>
39#include <sys/conf.h>
40#include <sys/bus.h>
41#include <sys/kernel.h>
42#include <sys/module.h>
43#include <sys/proc.h>
44#include <sys/rman.h>
45#include <sys/mman.h>
46#include <sys/time.h>
47#include <sys/smp.h>
48#include <sys/sysctl.h>
49#include <sys/timeet.h>
50#include <sys/timetc.h>
51#include <sys/vdso.h>
52
53#include <contrib/dev/acpica/include/acpi.h>
54#include <contrib/dev/acpica/include/accommon.h>
55
56#include <dev/acpica/acpivar.h>
57#include <dev/acpica/acpi_hpet.h>
58
59#ifdef DEV_APIC
60#include "pcib_if.h"
61#endif
62
63#define HPET_VENDID_AMD		0x4353
64#define HPET_VENDID_AMD2	0x1022
65#define HPET_VENDID_HYGON	0x1d94
66#define HPET_VENDID_INTEL	0x8086
67#define HPET_VENDID_NVIDIA	0x10de
68#define HPET_VENDID_SW		0x1166
69
70ACPI_SERIAL_DECL(hpet, "ACPI HPET support");
71
72static devclass_t hpet_devclass;
73
74/* ACPI CA debugging */
75#define _COMPONENT	ACPI_TIMER
76ACPI_MODULE_NAME("HPET")
77
78struct hpet_softc {
79	device_t		dev;
80	int			mem_rid;
81	int			intr_rid;
82	int			irq;
83	int			useirq;
84	int			legacy_route;
85	int			per_cpu;
86	uint32_t		allowed_irqs;
87	struct resource		*mem_res;
88	struct resource		*intr_res;
89	void			*intr_handle;
90	ACPI_HANDLE		handle;
91	uint32_t		acpi_uid;
92	uint64_t		freq;
93	uint32_t		caps;
94	struct timecounter	tc;
95	struct hpet_timer {
96		struct eventtimer	et;
97		struct hpet_softc	*sc;
98		int			num;
99		int			mode;
100#define	TIMER_STOPPED	0
101#define	TIMER_PERIODIC	1
102#define	TIMER_ONESHOT	2
103		int			intr_rid;
104		int			irq;
105		int			pcpu_cpu;
106		int			pcpu_misrouted;
107		int			pcpu_master;
108		int			pcpu_slaves[MAXCPU];
109		struct resource		*intr_res;
110		void			*intr_handle;
111		uint32_t		caps;
112		uint32_t		vectors;
113		uint32_t		div;
114		uint32_t		next;
115		char			name[8];
116	} 			t[32];
117	int			num_timers;
118	struct cdev		*pdev;
119	int			mmap_allow;
120	int			mmap_allow_write;
121};
122
123static d_open_t hpet_open;
124static d_mmap_t hpet_mmap;
125
126static struct cdevsw hpet_cdevsw = {
127	.d_version =	D_VERSION,
128	.d_name =	"hpet",
129	.d_open =	hpet_open,
130	.d_mmap =	hpet_mmap,
131};
132
133static u_int hpet_get_timecount(struct timecounter *tc);
134static void hpet_test(struct hpet_softc *sc);
135
136static char *hpet_ids[] = { "PNP0103", NULL };
137
138/* Knob to disable acpi_hpet device */
139bool acpi_hpet_disabled = false;
140
141static u_int
142hpet_get_timecount(struct timecounter *tc)
143{
144	struct hpet_softc *sc;
145
146	sc = tc->tc_priv;
147	return (bus_read_4(sc->mem_res, HPET_MAIN_COUNTER));
148}
149
150uint32_t
151hpet_vdso_timehands(struct vdso_timehands *vdso_th, struct timecounter *tc)
152{
153	struct hpet_softc *sc;
154
155	sc = tc->tc_priv;
156	vdso_th->th_algo = VDSO_TH_ALGO_X86_HPET;
157	vdso_th->th_x86_shift = 0;
158	vdso_th->th_x86_hpet_idx = device_get_unit(sc->dev);
159	bzero(vdso_th->th_res, sizeof(vdso_th->th_res));
160	return (sc->mmap_allow != 0);
161}
162
163#ifdef COMPAT_FREEBSD32
164uint32_t
165hpet_vdso_timehands32(struct vdso_timehands32 *vdso_th32,
166    struct timecounter *tc)
167{
168	struct hpet_softc *sc;
169
170	sc = tc->tc_priv;
171	vdso_th32->th_algo = VDSO_TH_ALGO_X86_HPET;
172	vdso_th32->th_x86_shift = 0;
173	vdso_th32->th_x86_hpet_idx = device_get_unit(sc->dev);
174	bzero(vdso_th32->th_res, sizeof(vdso_th32->th_res));
175	return (sc->mmap_allow != 0);
176}
177#endif
178
179static void
180hpet_enable(struct hpet_softc *sc)
181{
182	uint32_t val;
183
184	val = bus_read_4(sc->mem_res, HPET_CONFIG);
185	if (sc->legacy_route)
186		val |= HPET_CNF_LEG_RT;
187	else
188		val &= ~HPET_CNF_LEG_RT;
189	val |= HPET_CNF_ENABLE;
190	bus_write_4(sc->mem_res, HPET_CONFIG, val);
191}
192
193static void
194hpet_disable(struct hpet_softc *sc)
195{
196	uint32_t val;
197
198	val = bus_read_4(sc->mem_res, HPET_CONFIG);
199	val &= ~HPET_CNF_ENABLE;
200	bus_write_4(sc->mem_res, HPET_CONFIG, val);
201}
202
203static int
204hpet_start(struct eventtimer *et, sbintime_t first, sbintime_t period)
205{
206	struct hpet_timer *mt = (struct hpet_timer *)et->et_priv;
207	struct hpet_timer *t;
208	struct hpet_softc *sc = mt->sc;
209	uint32_t fdiv, now;
210
211	t = (mt->pcpu_master < 0) ? mt : &sc->t[mt->pcpu_slaves[curcpu]];
212	if (period != 0) {
213		t->mode = TIMER_PERIODIC;
214		t->div = (sc->freq * period) >> 32;
215	} else {
216		t->mode = TIMER_ONESHOT;
217		t->div = 0;
218	}
219	if (first != 0)
220		fdiv = (sc->freq * first) >> 32;
221	else
222		fdiv = t->div;
223	if (t->irq < 0)
224		bus_write_4(sc->mem_res, HPET_ISR, 1 << t->num);
225	t->caps |= HPET_TCNF_INT_ENB;
226	now = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
227restart:
228	t->next = now + fdiv;
229	if (t->mode == TIMER_PERIODIC && (t->caps & HPET_TCAP_PER_INT)) {
230		t->caps |= HPET_TCNF_TYPE;
231		bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num),
232		    t->caps | HPET_TCNF_VAL_SET);
233		bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
234		    t->next);
235		bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
236		    t->div);
237	} else {
238		t->caps &= ~HPET_TCNF_TYPE;
239		bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num),
240		    t->caps);
241		bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
242		    t->next);
243	}
244	now = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
245	if ((int32_t)(now - t->next + HPET_MIN_CYCLES) >= 0) {
246		fdiv *= 2;
247		goto restart;
248	}
249	return (0);
250}
251
252static int
253hpet_stop(struct eventtimer *et)
254{
255	struct hpet_timer *mt = (struct hpet_timer *)et->et_priv;
256	struct hpet_timer *t;
257	struct hpet_softc *sc = mt->sc;
258
259	t = (mt->pcpu_master < 0) ? mt : &sc->t[mt->pcpu_slaves[curcpu]];
260	t->mode = TIMER_STOPPED;
261	t->caps &= ~(HPET_TCNF_INT_ENB | HPET_TCNF_TYPE);
262	bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num), t->caps);
263	return (0);
264}
265
266static int
267hpet_intr_single(void *arg)
268{
269	struct hpet_timer *t = (struct hpet_timer *)arg;
270	struct hpet_timer *mt;
271	struct hpet_softc *sc = t->sc;
272	uint32_t now;
273
274	if (t->mode == TIMER_STOPPED)
275		return (FILTER_STRAY);
276	/* Check that per-CPU timer interrupt reached right CPU. */
277	if (t->pcpu_cpu >= 0 && t->pcpu_cpu != curcpu) {
278		if ((++t->pcpu_misrouted) % 32 == 0) {
279			printf("HPET interrupt routed to the wrong CPU"
280			    " (timer %d CPU %d -> %d)!\n",
281			    t->num, t->pcpu_cpu, curcpu);
282		}
283
284		/*
285		 * Reload timer, hoping that next time may be more lucky
286		 * (system will manage proper interrupt binding).
287		 */
288		if ((t->mode == TIMER_PERIODIC &&
289		    (t->caps & HPET_TCAP_PER_INT) == 0) ||
290		    t->mode == TIMER_ONESHOT) {
291			t->next = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER) +
292			    sc->freq / 8;
293			bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
294			    t->next);
295		}
296		return (FILTER_HANDLED);
297	}
298	if (t->mode == TIMER_PERIODIC &&
299	    (t->caps & HPET_TCAP_PER_INT) == 0) {
300		t->next += t->div;
301		now = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
302		if ((int32_t)((now + t->div / 2) - t->next) > 0)
303			t->next = now + t->div / 2;
304		bus_write_4(sc->mem_res,
305		    HPET_TIMER_COMPARATOR(t->num), t->next);
306	} else if (t->mode == TIMER_ONESHOT)
307		t->mode = TIMER_STOPPED;
308	mt = (t->pcpu_master < 0) ? t : &sc->t[t->pcpu_master];
309	if (mt->et.et_active)
310		mt->et.et_event_cb(&mt->et, mt->et.et_arg);
311	return (FILTER_HANDLED);
312}
313
314static int
315hpet_intr(void *arg)
316{
317	struct hpet_softc *sc = (struct hpet_softc *)arg;
318	int i;
319	uint32_t val;
320
321	val = bus_read_4(sc->mem_res, HPET_ISR);
322	if (val) {
323		bus_write_4(sc->mem_res, HPET_ISR, val);
324		val &= sc->useirq;
325		for (i = 0; i < sc->num_timers; i++) {
326			if ((val & (1 << i)) == 0)
327				continue;
328			hpet_intr_single(&sc->t[i]);
329		}
330		return (FILTER_HANDLED);
331	}
332	return (FILTER_STRAY);
333}
334
335uint32_t
336hpet_get_uid(device_t dev)
337{
338	struct hpet_softc *sc;
339
340	sc = device_get_softc(dev);
341	return (sc->acpi_uid);
342}
343
344static ACPI_STATUS
345hpet_find(ACPI_HANDLE handle, UINT32 level, void *context,
346    void **status)
347{
348	char 		**ids;
349	uint32_t	id = (uint32_t)(uintptr_t)context;
350	uint32_t	uid = 0;
351
352	for (ids = hpet_ids; *ids != NULL; ids++) {
353		if (acpi_MatchHid(handle, *ids))
354		        break;
355	}
356	if (*ids == NULL)
357		return (AE_OK);
358	if (ACPI_FAILURE(acpi_GetInteger(handle, "_UID", &uid)) ||
359	    id == uid)
360		*status = acpi_get_device(handle);
361	return (AE_OK);
362}
363
364/*
365 * Find an existing IRQ resource that matches the requested IRQ range
366 * and return its RID.  If one is not found, use a new RID.
367 */
368static int
369hpet_find_irq_rid(device_t dev, u_long start, u_long end)
370{
371	rman_res_t irq;
372	int error, rid;
373
374	for (rid = 0;; rid++) {
375		error = bus_get_resource(dev, SYS_RES_IRQ, rid, &irq, NULL);
376		if (error != 0 || (start <= irq && irq <= end))
377			return (rid);
378	}
379}
380
381static int
382hpet_open(struct cdev *cdev, int oflags, int devtype, struct thread *td)
383{
384	struct hpet_softc *sc;
385
386	sc = cdev->si_drv1;
387	if (!sc->mmap_allow)
388		return (EPERM);
389	else
390		return (0);
391}
392
393static int
394hpet_mmap(struct cdev *cdev, vm_ooffset_t offset, vm_paddr_t *paddr,
395    int nprot, vm_memattr_t *memattr)
396{
397	struct hpet_softc *sc;
398
399	sc = cdev->si_drv1;
400	if (offset >= rman_get_size(sc->mem_res))
401		return (EINVAL);
402	if (!sc->mmap_allow_write && (nprot & PROT_WRITE))
403		return (EPERM);
404	*paddr = rman_get_start(sc->mem_res) + offset;
405	*memattr = VM_MEMATTR_UNCACHEABLE;
406
407	return (0);
408}
409
410/* Discover the HPET via the ACPI table of the same name. */
411static void
412hpet_identify(driver_t *driver, device_t parent)
413{
414	ACPI_TABLE_HPET *hpet;
415	ACPI_STATUS	status;
416	device_t	child;
417	int		i;
418
419	/* Only one HPET device can be added. */
420	if (devclass_get_device(hpet_devclass, 0))
421		return;
422	for (i = 1; ; i++) {
423		/* Search for HPET table. */
424		status = AcpiGetTable(ACPI_SIG_HPET, i, (ACPI_TABLE_HEADER **)&hpet);
425		if (ACPI_FAILURE(status))
426			return;
427		/* Search for HPET device with same ID. */
428		child = NULL;
429		AcpiWalkNamespace(ACPI_TYPE_DEVICE, ACPI_ROOT_OBJECT,
430		    100, hpet_find, NULL, (void *)(uintptr_t)hpet->Sequence,
431		    (void *)&child);
432		/* If found - let it be probed in normal way. */
433		if (child) {
434			if (bus_get_resource(child, SYS_RES_MEMORY, 0,
435			    NULL, NULL) != 0)
436				bus_set_resource(child, SYS_RES_MEMORY, 0,
437				    hpet->Address.Address, HPET_MEM_WIDTH);
438			continue;
439		}
440		/* If not - create it from table info. */
441		child = BUS_ADD_CHILD(parent, 2, "hpet", 0);
442		if (child == NULL) {
443			printf("%s: can't add child\n", __func__);
444			continue;
445		}
446		bus_set_resource(child, SYS_RES_MEMORY, 0, hpet->Address.Address,
447		    HPET_MEM_WIDTH);
448	}
449}
450
451static int
452hpet_probe(device_t dev)
453{
454	ACPI_FUNCTION_TRACE((char *)(uintptr_t) __func__);
455
456	if (acpi_disabled("hpet") || acpi_hpet_disabled)
457		return (ENXIO);
458	if (acpi_get_handle(dev) != NULL &&
459	    ACPI_ID_PROBE(device_get_parent(dev), dev, hpet_ids) == NULL)
460		return (ENXIO);
461
462	device_set_desc(dev, "High Precision Event Timer");
463	return (0);
464}
465
466static int
467hpet_attach(device_t dev)
468{
469	struct hpet_softc *sc;
470	struct hpet_timer *t;
471	struct make_dev_args mda;
472	int i, j, num_msi, num_timers, num_percpu_et, num_percpu_t, cur_cpu;
473	int pcpu_master, error;
474	static int maxhpetet = 0;
475	uint32_t val, val2, cvectors, dvectors;
476	uint16_t vendor, rev;
477
478	ACPI_FUNCTION_TRACE((char *)(uintptr_t) __func__);
479
480	sc = device_get_softc(dev);
481	sc->dev = dev;
482	sc->handle = acpi_get_handle(dev);
483
484	sc->mem_rid = 0;
485	sc->mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &sc->mem_rid,
486	    RF_ACTIVE);
487	if (sc->mem_res == NULL)
488		return (ENOMEM);
489
490	/* Validate that we can access the whole region. */
491	if (rman_get_size(sc->mem_res) < HPET_MEM_WIDTH) {
492		device_printf(dev, "memory region width %jd too small\n",
493		    rman_get_size(sc->mem_res));
494		bus_free_resource(dev, SYS_RES_MEMORY, sc->mem_res);
495		return (ENXIO);
496	}
497
498	/* Be sure timer is enabled. */
499	hpet_enable(sc);
500
501	/* Read basic statistics about the timer. */
502	val = bus_read_4(sc->mem_res, HPET_PERIOD);
503	if (val == 0) {
504		device_printf(dev, "invalid period\n");
505		hpet_disable(sc);
506		bus_free_resource(dev, SYS_RES_MEMORY, sc->mem_res);
507		return (ENXIO);
508	}
509
510	sc->freq = (1000000000000000LL + val / 2) / val;
511	sc->caps = bus_read_4(sc->mem_res, HPET_CAPABILITIES);
512	vendor = (sc->caps & HPET_CAP_VENDOR_ID) >> 16;
513	rev = sc->caps & HPET_CAP_REV_ID;
514	num_timers = 1 + ((sc->caps & HPET_CAP_NUM_TIM) >> 8);
515	/*
516	 * ATI/AMD violates IA-PC HPET (High Precision Event Timers)
517	 * Specification and provides an off by one number
518	 * of timers/comparators.
519	 * Additionally, they use unregistered value in VENDOR_ID field.
520	 */
521	if (vendor == HPET_VENDID_AMD && rev < 0x10 && num_timers > 0)
522		num_timers--;
523	sc->num_timers = num_timers;
524	if (bootverbose) {
525		device_printf(dev,
526		    "vendor 0x%x, rev 0x%x, %jdHz%s, %d timers,%s\n",
527		    vendor, rev, sc->freq,
528		    (sc->caps & HPET_CAP_COUNT_SIZE) ? " 64bit" : "",
529		    num_timers,
530		    (sc->caps & HPET_CAP_LEG_RT) ? " legacy route" : "");
531	}
532	for (i = 0; i < num_timers; i++) {
533		t = &sc->t[i];
534		t->sc = sc;
535		t->num = i;
536		t->mode = TIMER_STOPPED;
537		t->intr_rid = -1;
538		t->irq = -1;
539		t->pcpu_cpu = -1;
540		t->pcpu_misrouted = 0;
541		t->pcpu_master = -1;
542		t->caps = bus_read_4(sc->mem_res, HPET_TIMER_CAP_CNF(i));
543		t->vectors = bus_read_4(sc->mem_res, HPET_TIMER_CAP_CNF(i) + 4);
544		if (bootverbose) {
545			device_printf(dev,
546			    " t%d: irqs 0x%08x (%d)%s%s%s\n", i,
547			    t->vectors, (t->caps & HPET_TCNF_INT_ROUTE) >> 9,
548			    (t->caps & HPET_TCAP_FSB_INT_DEL) ? ", MSI" : "",
549			    (t->caps & HPET_TCAP_SIZE) ? ", 64bit" : "",
550			    (t->caps & HPET_TCAP_PER_INT) ? ", periodic" : "");
551		}
552	}
553	if (testenv("debug.acpi.hpet_test"))
554		hpet_test(sc);
555	/*
556	 * Don't attach if the timer never increments.  Since the spec
557	 * requires it to be at least 10 MHz, it has to change in 1 us.
558	 */
559	val = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
560	DELAY(1);
561	val2 = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
562	if (val == val2) {
563		device_printf(dev, "HPET never increments, disabling\n");
564		hpet_disable(sc);
565		bus_free_resource(dev, SYS_RES_MEMORY, sc->mem_res);
566		return (ENXIO);
567	}
568	/* Announce first HPET as timecounter. */
569	if (device_get_unit(dev) == 0) {
570		sc->tc.tc_get_timecount = hpet_get_timecount,
571		sc->tc.tc_counter_mask = ~0u,
572		sc->tc.tc_name = "HPET",
573		sc->tc.tc_quality = 950,
574		sc->tc.tc_frequency = sc->freq;
575		sc->tc.tc_priv = sc;
576		sc->tc.tc_fill_vdso_timehands = hpet_vdso_timehands;
577#ifdef COMPAT_FREEBSD32
578		sc->tc.tc_fill_vdso_timehands32 = hpet_vdso_timehands32;
579#endif
580		tc_init(&sc->tc);
581	}
582	/* If not disabled - setup and announce event timers. */
583	if (resource_int_value(device_get_name(dev), device_get_unit(dev),
584	     "clock", &i) == 0 && i == 0)
585	        return (0);
586
587	/* Check whether we can and want legacy routing. */
588	sc->legacy_route = 0;
589	resource_int_value(device_get_name(dev), device_get_unit(dev),
590	     "legacy_route", &sc->legacy_route);
591	if ((sc->caps & HPET_CAP_LEG_RT) == 0)
592		sc->legacy_route = 0;
593	if (sc->legacy_route) {
594		sc->t[0].vectors = 0;
595		sc->t[1].vectors = 0;
596	}
597
598	/* Check what IRQs we want use. */
599	/* By default allow any PCI IRQs. */
600	sc->allowed_irqs = 0xffff0000;
601	/*
602	 * HPETs in AMD chipsets before SB800 have problems with IRQs >= 16
603	 * Lower are also not always working for different reasons.
604	 * SB800 fixed it, but seems do not implements level triggering
605	 * properly, that makes it very unreliable - it freezes after any
606	 * interrupt loss. Avoid legacy IRQs for AMD.
607	 */
608	if (vendor == HPET_VENDID_AMD || vendor == HPET_VENDID_AMD2 ||
609	    vendor == HPET_VENDID_HYGON)
610		sc->allowed_irqs = 0x00000000;
611	/*
612	 * NVidia MCP5x chipsets have number of unexplained interrupt
613	 * problems. For some reason, using HPET interrupts breaks HDA sound.
614	 */
615	if (vendor == HPET_VENDID_NVIDIA && rev <= 0x01)
616		sc->allowed_irqs = 0x00000000;
617	/*
618	 * ServerWorks HT1000 reported to have problems with IRQs >= 16.
619	 * Lower IRQs are working, but allowed mask is not set correctly.
620	 * Legacy_route mode works fine.
621	 */
622	if (vendor == HPET_VENDID_SW && rev <= 0x01)
623		sc->allowed_irqs = 0x00000000;
624	/*
625	 * Neither QEMU nor VirtualBox report supported IRQs correctly.
626	 * The only way to use HPET there is to specify IRQs manually
627	 * and/or use legacy_route. Legacy_route mode works on both.
628	 */
629	if (vm_guest)
630		sc->allowed_irqs = 0x00000000;
631	/* Let user override. */
632	resource_int_value(device_get_name(dev), device_get_unit(dev),
633	     "allowed_irqs", &sc->allowed_irqs);
634
635	/* Get how much per-CPU timers we should try to provide. */
636	sc->per_cpu = 1;
637	resource_int_value(device_get_name(dev), device_get_unit(dev),
638	     "per_cpu", &sc->per_cpu);
639
640	num_msi = 0;
641	sc->useirq = 0;
642	/* Find IRQ vectors for all timers. */
643	cvectors = sc->allowed_irqs & 0xffff0000;
644	dvectors = sc->allowed_irqs & 0x0000ffff;
645	if (sc->legacy_route)
646		dvectors &= 0x0000fefe;
647	for (i = 0; i < num_timers; i++) {
648		t = &sc->t[i];
649		if (sc->legacy_route && i < 2)
650			t->irq = (i == 0) ? 0 : 8;
651#ifdef DEV_APIC
652		else if (t->caps & HPET_TCAP_FSB_INT_DEL) {
653			if ((j = PCIB_ALLOC_MSIX(
654			    device_get_parent(device_get_parent(dev)), dev,
655			    &t->irq))) {
656				device_printf(dev,
657				    "Can't allocate interrupt for t%d: %d\n",
658				    i, j);
659			}
660		}
661#endif
662		else if (dvectors & t->vectors) {
663			t->irq = ffs(dvectors & t->vectors) - 1;
664			dvectors &= ~(1 << t->irq);
665		}
666		if (t->irq >= 0) {
667			t->intr_rid = hpet_find_irq_rid(dev, t->irq, t->irq);
668			t->intr_res = bus_alloc_resource(dev, SYS_RES_IRQ,
669			    &t->intr_rid, t->irq, t->irq, 1, RF_ACTIVE);
670			if (t->intr_res == NULL) {
671				t->irq = -1;
672				device_printf(dev,
673				    "Can't map interrupt for t%d.\n", i);
674			} else if (bus_setup_intr(dev, t->intr_res,
675			    INTR_TYPE_CLK, hpet_intr_single, NULL, t,
676			    &t->intr_handle) != 0) {
677				t->irq = -1;
678				device_printf(dev,
679				    "Can't setup interrupt for t%d.\n", i);
680			} else {
681				bus_describe_intr(dev, t->intr_res,
682				    t->intr_handle, "t%d", i);
683				num_msi++;
684			}
685		}
686		if (t->irq < 0 && (cvectors & t->vectors) != 0) {
687			cvectors &= t->vectors;
688			sc->useirq |= (1 << i);
689		}
690	}
691	if (sc->legacy_route && sc->t[0].irq < 0 && sc->t[1].irq < 0)
692		sc->legacy_route = 0;
693	if (sc->legacy_route)
694		hpet_enable(sc);
695	/* Group timers for per-CPU operation. */
696	num_percpu_et = min(num_msi / mp_ncpus, sc->per_cpu);
697	num_percpu_t = num_percpu_et * mp_ncpus;
698	pcpu_master = 0;
699	cur_cpu = CPU_FIRST();
700	for (i = 0; i < num_timers; i++) {
701		t = &sc->t[i];
702		if (t->irq >= 0 && num_percpu_t > 0) {
703			if (cur_cpu == CPU_FIRST())
704				pcpu_master = i;
705			t->pcpu_cpu = cur_cpu;
706			t->pcpu_master = pcpu_master;
707			sc->t[pcpu_master].
708			    pcpu_slaves[cur_cpu] = i;
709			bus_bind_intr(dev, t->intr_res, cur_cpu);
710			cur_cpu = CPU_NEXT(cur_cpu);
711			num_percpu_t--;
712		} else if (t->irq >= 0)
713			bus_bind_intr(dev, t->intr_res, CPU_FIRST());
714	}
715	bus_write_4(sc->mem_res, HPET_ISR, 0xffffffff);
716	sc->irq = -1;
717	/* If at least one timer needs legacy IRQ - set it up. */
718	if (sc->useirq) {
719		j = i = fls(cvectors) - 1;
720		while (j > 0 && (cvectors & (1 << (j - 1))) != 0)
721			j--;
722		sc->intr_rid = hpet_find_irq_rid(dev, j, i);
723		sc->intr_res = bus_alloc_resource(dev, SYS_RES_IRQ,
724		    &sc->intr_rid, j, i, 1, RF_SHAREABLE | RF_ACTIVE);
725		if (sc->intr_res == NULL)
726			device_printf(dev, "Can't map interrupt.\n");
727		else if (bus_setup_intr(dev, sc->intr_res, INTR_TYPE_CLK,
728		    hpet_intr, NULL, sc, &sc->intr_handle) != 0) {
729			device_printf(dev, "Can't setup interrupt.\n");
730		} else {
731			sc->irq = rman_get_start(sc->intr_res);
732			/* Bind IRQ to BSP to avoid live migration. */
733			bus_bind_intr(dev, sc->intr_res, CPU_FIRST());
734		}
735	}
736	/* Program and announce event timers. */
737	for (i = 0; i < num_timers; i++) {
738		t = &sc->t[i];
739		t->caps &= ~(HPET_TCNF_FSB_EN | HPET_TCNF_INT_ROUTE);
740		t->caps &= ~(HPET_TCNF_VAL_SET | HPET_TCNF_INT_ENB);
741		t->caps &= ~(HPET_TCNF_INT_TYPE);
742		t->caps |= HPET_TCNF_32MODE;
743		if (t->irq >= 0 && sc->legacy_route && i < 2) {
744			/* Legacy route doesn't need more configuration. */
745		} else
746#ifdef DEV_APIC
747		if ((t->caps & HPET_TCAP_FSB_INT_DEL) && t->irq >= 0) {
748			uint64_t addr;
749			uint32_t data;
750
751			if (PCIB_MAP_MSI(
752			    device_get_parent(device_get_parent(dev)), dev,
753			    t->irq, &addr, &data) == 0) {
754				bus_write_4(sc->mem_res,
755				    HPET_TIMER_FSB_ADDR(i), addr);
756				bus_write_4(sc->mem_res,
757				    HPET_TIMER_FSB_VAL(i), data);
758				t->caps |= HPET_TCNF_FSB_EN;
759			} else
760				t->irq = -2;
761		} else
762#endif
763		if (t->irq >= 0)
764			t->caps |= (t->irq << 9);
765		else if (sc->irq >= 0 && (t->vectors & (1 << sc->irq)))
766			t->caps |= (sc->irq << 9) | HPET_TCNF_INT_TYPE;
767		bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(i), t->caps);
768		/* Skip event timers without set up IRQ. */
769		if (t->irq < 0 &&
770		    (sc->irq < 0 || (t->vectors & (1 << sc->irq)) == 0))
771			continue;
772		/* Announce the reset. */
773		if (maxhpetet == 0)
774			t->et.et_name = "HPET";
775		else {
776			sprintf(t->name, "HPET%d", maxhpetet);
777			t->et.et_name = t->name;
778		}
779		t->et.et_flags = ET_FLAGS_PERIODIC | ET_FLAGS_ONESHOT;
780		t->et.et_quality = 450;
781		if (t->pcpu_master >= 0) {
782			t->et.et_flags |= ET_FLAGS_PERCPU;
783			t->et.et_quality += 100;
784		} else if (mp_ncpus >= 8)
785			t->et.et_quality -= 100;
786		if ((t->caps & HPET_TCAP_PER_INT) == 0)
787			t->et.et_quality -= 10;
788		t->et.et_frequency = sc->freq;
789		t->et.et_min_period =
790		    ((uint64_t)(HPET_MIN_CYCLES * 2) << 32) / sc->freq;
791		t->et.et_max_period = (0xfffffffeLLU << 32) / sc->freq;
792		t->et.et_start = hpet_start;
793		t->et.et_stop = hpet_stop;
794		t->et.et_priv = &sc->t[i];
795		if (t->pcpu_master < 0 || t->pcpu_master == i) {
796			et_register(&t->et);
797			maxhpetet++;
798		}
799	}
800	acpi_GetInteger(sc->handle, "_UID", &sc->acpi_uid);
801
802	make_dev_args_init(&mda);
803	mda.mda_devsw = &hpet_cdevsw;
804	mda.mda_uid = UID_ROOT;
805	mda.mda_gid = GID_WHEEL;
806	mda.mda_mode = 0644;
807	mda.mda_si_drv1 = sc;
808	error = make_dev_s(&mda, &sc->pdev, "hpet%d", device_get_unit(dev));
809	if (error == 0) {
810		sc->mmap_allow = 1;
811		TUNABLE_INT_FETCH("hw.acpi.hpet.mmap_allow",
812		    &sc->mmap_allow);
813		sc->mmap_allow_write = 0;
814		TUNABLE_INT_FETCH("hw.acpi.hpet.mmap_allow_write",
815		    &sc->mmap_allow_write);
816		SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
817		    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
818		    OID_AUTO, "mmap_allow",
819		    CTLFLAG_RW, &sc->mmap_allow, 0,
820		    "Allow userland to memory map HPET");
821		SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
822		    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
823		    OID_AUTO, "mmap_allow_write",
824		    CTLFLAG_RW, &sc->mmap_allow_write, 0,
825		    "Allow userland write to the HPET register space");
826	} else {
827		device_printf(dev, "could not create /dev/hpet%d, error %d\n",
828		    device_get_unit(dev), error);
829	}
830
831	return (0);
832}
833
834static int
835hpet_detach(device_t dev)
836{
837	ACPI_FUNCTION_TRACE((char *)(uintptr_t) __func__);
838
839	/* XXX Without a tc_remove() function, we can't detach. */
840	return (EBUSY);
841}
842
843static int
844hpet_suspend(device_t dev)
845{
846//	struct hpet_softc *sc;
847
848	/*
849	 * Disable the timer during suspend.  The timer will not lose
850	 * its state in S1 or S2, but we are required to disable
851	 * it.
852	 */
853//	sc = device_get_softc(dev);
854//	hpet_disable(sc);
855
856	return (0);
857}
858
859static int
860hpet_resume(device_t dev)
861{
862	struct hpet_softc *sc;
863	struct hpet_timer *t;
864	int i;
865
866	/* Re-enable the timer after a resume to keep the clock advancing. */
867	sc = device_get_softc(dev);
868	hpet_enable(sc);
869	/* Restart event timers that were running on suspend. */
870	for (i = 0; i < sc->num_timers; i++) {
871		t = &sc->t[i];
872#ifdef DEV_APIC
873		if (t->irq >= 0 && (sc->legacy_route == 0 || i >= 2)) {
874			uint64_t addr;
875			uint32_t data;
876
877			if (PCIB_MAP_MSI(
878			    device_get_parent(device_get_parent(dev)), dev,
879			    t->irq, &addr, &data) == 0) {
880				bus_write_4(sc->mem_res,
881				    HPET_TIMER_FSB_ADDR(i), addr);
882				bus_write_4(sc->mem_res,
883				    HPET_TIMER_FSB_VAL(i), data);
884			}
885		}
886#endif
887		if (t->mode == TIMER_STOPPED)
888			continue;
889		t->next = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
890		if (t->mode == TIMER_PERIODIC &&
891		    (t->caps & HPET_TCAP_PER_INT) != 0) {
892			t->caps |= HPET_TCNF_TYPE;
893			t->next += t->div;
894			bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num),
895			    t->caps | HPET_TCNF_VAL_SET);
896			bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
897			    t->next);
898			bus_read_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num));
899			bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
900			    t->div);
901		} else {
902			t->next += sc->freq / 1024;
903			bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
904			    t->next);
905		}
906		bus_write_4(sc->mem_res, HPET_ISR, 1 << t->num);
907		bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num), t->caps);
908	}
909	return (0);
910}
911
912/* Print some basic latency/rate information to assist in debugging. */
913static void
914hpet_test(struct hpet_softc *sc)
915{
916	int i;
917	uint32_t u1, u2;
918	struct bintime b0, b1, b2;
919	struct timespec ts;
920
921	binuptime(&b0);
922	binuptime(&b0);
923	binuptime(&b1);
924	u1 = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
925	for (i = 1; i < 1000; i++)
926		u2 = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
927	binuptime(&b2);
928	u2 = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
929
930	bintime_sub(&b2, &b1);
931	bintime_sub(&b1, &b0);
932	bintime_sub(&b2, &b1);
933	bintime2timespec(&b2, &ts);
934
935	device_printf(sc->dev, "%ld.%09ld: %u ... %u = %u\n",
936	    (long)ts.tv_sec, ts.tv_nsec, u1, u2, u2 - u1);
937
938	device_printf(sc->dev, "time per call: %ld ns\n", ts.tv_nsec / 1000);
939}
940
941#ifdef DEV_APIC
942static int
943hpet_remap_intr(device_t dev, device_t child, u_int irq)
944{
945	struct hpet_softc *sc = device_get_softc(dev);
946	struct hpet_timer *t;
947	uint64_t addr;
948	uint32_t data;
949	int error, i;
950
951	for (i = 0; i < sc->num_timers; i++) {
952		t = &sc->t[i];
953		if (t->irq != irq)
954			continue;
955		error = PCIB_MAP_MSI(
956		    device_get_parent(device_get_parent(dev)), dev,
957		    irq, &addr, &data);
958		if (error)
959			return (error);
960		hpet_disable(sc); /* Stop timer to avoid interrupt loss. */
961		bus_write_4(sc->mem_res, HPET_TIMER_FSB_ADDR(i), addr);
962		bus_write_4(sc->mem_res, HPET_TIMER_FSB_VAL(i), data);
963		hpet_enable(sc);
964		return (0);
965	}
966	return (ENOENT);
967}
968#endif
969
970static device_method_t hpet_methods[] = {
971	/* Device interface */
972	DEVMETHOD(device_identify, hpet_identify),
973	DEVMETHOD(device_probe, hpet_probe),
974	DEVMETHOD(device_attach, hpet_attach),
975	DEVMETHOD(device_detach, hpet_detach),
976	DEVMETHOD(device_suspend, hpet_suspend),
977	DEVMETHOD(device_resume, hpet_resume),
978
979#ifdef DEV_APIC
980	DEVMETHOD(bus_remap_intr, hpet_remap_intr),
981#endif
982
983	DEVMETHOD_END
984};
985
986static driver_t	hpet_driver = {
987	"hpet",
988	hpet_methods,
989	sizeof(struct hpet_softc),
990};
991
992DRIVER_MODULE(hpet, acpi, hpet_driver, hpet_devclass, 0, 0);
993MODULE_DEPEND(hpet, acpi, 1, 1, 1);
994