xref: /freebsd/sys/dev/acpica/acpi_hpet.c (revision ff0ba87247820afbdfdc1b307c803f7923d0e4d3)
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 #if defined(__amd64__)
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 
67 ACPI_SERIAL_DECL(hpet, "ACPI HPET support");
68 
69 static devclass_t hpet_devclass;
70 
71 /* ACPI CA debugging */
72 #define _COMPONENT	ACPI_TIMER
73 ACPI_MODULE_NAME("HPET")
74 
75 struct 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 
116 static d_open_t hpet_open;
117 static d_mmap_t hpet_mmap;
118 
119 static 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 
126 static u_int hpet_get_timecount(struct timecounter *tc);
127 static void hpet_test(struct hpet_softc *sc);
128 
129 static char *hpet_ids[] = { "PNP0103", NULL };
130 
131 /* Knob to disable acpi_hpet device */
132 bool acpi_hpet_disabled = false;
133 
134 static u_int
135 hpet_get_timecount(struct timecounter *tc)
136 {
137 	struct hpet_softc *sc;
138 
139 	sc = tc->tc_priv;
140 	return (bus_read_4(sc->mem_res, HPET_MAIN_COUNTER));
141 }
142 
143 static void
144 hpet_enable(struct hpet_softc *sc)
145 {
146 	uint32_t val;
147 
148 	val = bus_read_4(sc->mem_res, HPET_CONFIG);
149 	if (sc->legacy_route)
150 		val |= HPET_CNF_LEG_RT;
151 	else
152 		val &= ~HPET_CNF_LEG_RT;
153 	val |= HPET_CNF_ENABLE;
154 	bus_write_4(sc->mem_res, HPET_CONFIG, val);
155 }
156 
157 static void
158 hpet_disable(struct hpet_softc *sc)
159 {
160 	uint32_t val;
161 
162 	val = bus_read_4(sc->mem_res, HPET_CONFIG);
163 	val &= ~HPET_CNF_ENABLE;
164 	bus_write_4(sc->mem_res, HPET_CONFIG, val);
165 }
166 
167 static int
168 hpet_start(struct eventtimer *et, sbintime_t first, sbintime_t period)
169 {
170 	struct hpet_timer *mt = (struct hpet_timer *)et->et_priv;
171 	struct hpet_timer *t;
172 	struct hpet_softc *sc = mt->sc;
173 	uint32_t fdiv, now;
174 
175 	t = (mt->pcpu_master < 0) ? mt : &sc->t[mt->pcpu_slaves[curcpu]];
176 	if (period != 0) {
177 		t->mode = 1;
178 		t->div = (sc->freq * period) >> 32;
179 	} else {
180 		t->mode = 2;
181 		t->div = 0;
182 	}
183 	if (first != 0)
184 		fdiv = (sc->freq * first) >> 32;
185 	else
186 		fdiv = t->div;
187 	if (t->irq < 0)
188 		bus_write_4(sc->mem_res, HPET_ISR, 1 << t->num);
189 	t->caps |= HPET_TCNF_INT_ENB;
190 	now = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
191 restart:
192 	t->next = now + fdiv;
193 	if (t->mode == 1 && (t->caps & HPET_TCAP_PER_INT)) {
194 		t->caps |= HPET_TCNF_TYPE;
195 		bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num),
196 		    t->caps | HPET_TCNF_VAL_SET);
197 		bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
198 		    t->next);
199 		bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
200 		    t->div);
201 	} else {
202 		t->caps &= ~HPET_TCNF_TYPE;
203 		bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num),
204 		    t->caps);
205 		bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
206 		    t->next);
207 	}
208 	now = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
209 	if ((int32_t)(now - t->next + HPET_MIN_CYCLES) >= 0) {
210 		fdiv *= 2;
211 		goto restart;
212 	}
213 	return (0);
214 }
215 
216 static int
217 hpet_stop(struct eventtimer *et)
218 {
219 	struct hpet_timer *mt = (struct hpet_timer *)et->et_priv;
220 	struct hpet_timer *t;
221 	struct hpet_softc *sc = mt->sc;
222 
223 	t = (mt->pcpu_master < 0) ? mt : &sc->t[mt->pcpu_slaves[curcpu]];
224 	t->mode = 0;
225 	t->caps &= ~(HPET_TCNF_INT_ENB | HPET_TCNF_TYPE);
226 	bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num), t->caps);
227 	return (0);
228 }
229 
230 static int
231 hpet_intr_single(void *arg)
232 {
233 	struct hpet_timer *t = (struct hpet_timer *)arg;
234 	struct hpet_timer *mt;
235 	struct hpet_softc *sc = t->sc;
236 	uint32_t now;
237 
238 	if (t->mode == 0)
239 		return (FILTER_STRAY);
240 	/* Check that per-CPU timer interrupt reached right CPU. */
241 	if (t->pcpu_cpu >= 0 && t->pcpu_cpu != curcpu) {
242 		if ((++t->pcpu_misrouted) % 32 == 0) {
243 			printf("HPET interrupt routed to the wrong CPU"
244 			    " (timer %d CPU %d -> %d)!\n",
245 			    t->num, t->pcpu_cpu, curcpu);
246 		}
247 
248 		/*
249 		 * Reload timer, hoping that next time may be more lucky
250 		 * (system will manage proper interrupt binding).
251 		 */
252 		if ((t->mode == 1 && (t->caps & HPET_TCAP_PER_INT) == 0) ||
253 		    t->mode == 2) {
254 			t->next = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER) +
255 			    sc->freq / 8;
256 			bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
257 			    t->next);
258 		}
259 		return (FILTER_HANDLED);
260 	}
261 	if (t->mode == 1 &&
262 	    (t->caps & HPET_TCAP_PER_INT) == 0) {
263 		t->next += t->div;
264 		now = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
265 		if ((int32_t)((now + t->div / 2) - t->next) > 0)
266 			t->next = now + t->div / 2;
267 		bus_write_4(sc->mem_res,
268 		    HPET_TIMER_COMPARATOR(t->num), t->next);
269 	} else if (t->mode == 2)
270 		t->mode = 0;
271 	mt = (t->pcpu_master < 0) ? t : &sc->t[t->pcpu_master];
272 	if (mt->et.et_active)
273 		mt->et.et_event_cb(&mt->et, mt->et.et_arg);
274 	return (FILTER_HANDLED);
275 }
276 
277 static int
278 hpet_intr(void *arg)
279 {
280 	struct hpet_softc *sc = (struct hpet_softc *)arg;
281 	int i;
282 	uint32_t val;
283 
284 	val = bus_read_4(sc->mem_res, HPET_ISR);
285 	if (val) {
286 		bus_write_4(sc->mem_res, HPET_ISR, val);
287 		val &= sc->useirq;
288 		for (i = 0; i < sc->num_timers; i++) {
289 			if ((val & (1 << i)) == 0)
290 				continue;
291 			hpet_intr_single(&sc->t[i]);
292 		}
293 		return (FILTER_HANDLED);
294 	}
295 	return (FILTER_STRAY);
296 }
297 
298 static ACPI_STATUS
299 hpet_find(ACPI_HANDLE handle, UINT32 level, void *context,
300     void **status)
301 {
302 	char 		**ids;
303 	uint32_t	id = (uint32_t)(uintptr_t)context;
304 	uint32_t	uid = 0;
305 
306 	for (ids = hpet_ids; *ids != NULL; ids++) {
307 		if (acpi_MatchHid(handle, *ids))
308 		        break;
309 	}
310 	if (*ids == NULL)
311 		return (AE_OK);
312 	if (ACPI_FAILURE(acpi_GetInteger(handle, "_UID", &uid)) ||
313 	    id == uid)
314 		*status = acpi_get_device(handle);
315 	return (AE_OK);
316 }
317 
318 /*
319  * Find an existing IRQ resource that matches the requested IRQ range
320  * and return its RID.  If one is not found, use a new RID.
321  */
322 static int
323 hpet_find_irq_rid(device_t dev, u_long start, u_long end)
324 {
325 	u_long irq;
326 	int error, rid;
327 
328 	for (rid = 0;; rid++) {
329 		error = bus_get_resource(dev, SYS_RES_IRQ, rid, &irq, NULL);
330 		if (error != 0 || (start <= irq && irq <= end))
331 			return (rid);
332 	}
333 }
334 
335 static int
336 hpet_open(struct cdev *cdev, int oflags, int devtype, struct thread *td)
337 {
338 	struct hpet_softc *sc;
339 
340 	sc = cdev->si_drv1;
341 	if (!sc->mmap_allow)
342 		return (EPERM);
343 	else
344 		return (0);
345 }
346 
347 static int
348 hpet_mmap(struct cdev *cdev, vm_ooffset_t offset, vm_paddr_t *paddr,
349     int nprot, vm_memattr_t *memattr)
350 {
351 	struct hpet_softc *sc;
352 
353 	sc = cdev->si_drv1;
354 	if (offset > rman_get_size(sc->mem_res))
355 		return (EINVAL);
356 	if (!sc->mmap_allow_write && (nprot & PROT_WRITE))
357 		return (EPERM);
358 	*paddr = rman_get_start(sc->mem_res) + offset;
359 	*memattr = VM_MEMATTR_UNCACHEABLE;
360 
361 	return (0);
362 }
363 
364 /* Discover the HPET via the ACPI table of the same name. */
365 static void
366 hpet_identify(driver_t *driver, device_t parent)
367 {
368 	ACPI_TABLE_HPET *hpet;
369 	ACPI_STATUS	status;
370 	device_t	child;
371 	int		i;
372 
373 	/* Only one HPET device can be added. */
374 	if (devclass_get_device(hpet_devclass, 0))
375 		return;
376 	for (i = 1; ; i++) {
377 		/* Search for HPET table. */
378 		status = AcpiGetTable(ACPI_SIG_HPET, i, (ACPI_TABLE_HEADER **)&hpet);
379 		if (ACPI_FAILURE(status))
380 			return;
381 		/* Search for HPET device with same ID. */
382 		child = NULL;
383 		AcpiWalkNamespace(ACPI_TYPE_DEVICE, ACPI_ROOT_OBJECT,
384 		    100, hpet_find, NULL, (void *)(uintptr_t)hpet->Sequence,
385 		    (void *)&child);
386 		/* If found - let it be probed in normal way. */
387 		if (child) {
388 			if (bus_get_resource(child, SYS_RES_MEMORY, 0,
389 			    NULL, NULL) != 0)
390 				bus_set_resource(child, SYS_RES_MEMORY, 0,
391 				    hpet->Address.Address, HPET_MEM_WIDTH);
392 			continue;
393 		}
394 		/* If not - create it from table info. */
395 		child = BUS_ADD_CHILD(parent, 2, "hpet", 0);
396 		if (child == NULL) {
397 			printf("%s: can't add child\n", __func__);
398 			continue;
399 		}
400 		bus_set_resource(child, SYS_RES_MEMORY, 0, hpet->Address.Address,
401 		    HPET_MEM_WIDTH);
402 	}
403 }
404 
405 static int
406 hpet_probe(device_t dev)
407 {
408 	ACPI_FUNCTION_TRACE((char *)(uintptr_t) __func__);
409 
410 	if (acpi_disabled("hpet") || acpi_hpet_disabled)
411 		return (ENXIO);
412 	if (acpi_get_handle(dev) != NULL &&
413 	    ACPI_ID_PROBE(device_get_parent(dev), dev, hpet_ids) == NULL)
414 		return (ENXIO);
415 
416 	device_set_desc(dev, "High Precision Event Timer");
417 	return (0);
418 }
419 
420 static int
421 hpet_attach(device_t dev)
422 {
423 	struct hpet_softc *sc;
424 	struct hpet_timer *t;
425 	int i, j, num_msi, num_timers, num_percpu_et, num_percpu_t, cur_cpu;
426 	int pcpu_master;
427 	static int maxhpetet = 0;
428 	uint32_t val, val2, cvectors, dvectors;
429 	uint16_t vendor, rev;
430 
431 	ACPI_FUNCTION_TRACE((char *)(uintptr_t) __func__);
432 
433 	sc = device_get_softc(dev);
434 	sc->dev = dev;
435 	sc->handle = acpi_get_handle(dev);
436 
437 	sc->mem_rid = 0;
438 	sc->mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &sc->mem_rid,
439 	    RF_ACTIVE);
440 	if (sc->mem_res == NULL)
441 		return (ENOMEM);
442 
443 	/* Validate that we can access the whole region. */
444 	if (rman_get_size(sc->mem_res) < HPET_MEM_WIDTH) {
445 		device_printf(dev, "memory region width %ld too small\n",
446 		    rman_get_size(sc->mem_res));
447 		bus_free_resource(dev, SYS_RES_MEMORY, sc->mem_res);
448 		return (ENXIO);
449 	}
450 
451 	/* Be sure timer is enabled. */
452 	hpet_enable(sc);
453 
454 	/* Read basic statistics about the timer. */
455 	val = bus_read_4(sc->mem_res, HPET_PERIOD);
456 	if (val == 0) {
457 		device_printf(dev, "invalid period\n");
458 		hpet_disable(sc);
459 		bus_free_resource(dev, SYS_RES_MEMORY, sc->mem_res);
460 		return (ENXIO);
461 	}
462 
463 	sc->freq = (1000000000000000LL + val / 2) / val;
464 	sc->caps = bus_read_4(sc->mem_res, HPET_CAPABILITIES);
465 	vendor = (sc->caps & HPET_CAP_VENDOR_ID) >> 16;
466 	rev = sc->caps & HPET_CAP_REV_ID;
467 	num_timers = 1 + ((sc->caps & HPET_CAP_NUM_TIM) >> 8);
468 	/*
469 	 * ATI/AMD violates IA-PC HPET (High Precision Event Timers)
470 	 * Specification and provides an off by one number
471 	 * of timers/comparators.
472 	 * Additionally, they use unregistered value in VENDOR_ID field.
473 	 */
474 	if (vendor == HPET_VENDID_AMD && rev < 0x10 && num_timers > 0)
475 		num_timers--;
476 	sc->num_timers = num_timers;
477 	if (bootverbose) {
478 		device_printf(dev,
479 		    "vendor 0x%x, rev 0x%x, %jdHz%s, %d timers,%s\n",
480 		    vendor, rev, sc->freq,
481 		    (sc->caps & HPET_CAP_COUNT_SIZE) ? " 64bit" : "",
482 		    num_timers,
483 		    (sc->caps & HPET_CAP_LEG_RT) ? " legacy route" : "");
484 	}
485 	for (i = 0; i < num_timers; i++) {
486 		t = &sc->t[i];
487 		t->sc = sc;
488 		t->num = i;
489 		t->mode = 0;
490 		t->intr_rid = -1;
491 		t->irq = -1;
492 		t->pcpu_cpu = -1;
493 		t->pcpu_misrouted = 0;
494 		t->pcpu_master = -1;
495 		t->caps = bus_read_4(sc->mem_res, HPET_TIMER_CAP_CNF(i));
496 		t->vectors = bus_read_4(sc->mem_res, HPET_TIMER_CAP_CNF(i) + 4);
497 		if (bootverbose) {
498 			device_printf(dev,
499 			    " t%d: irqs 0x%08x (%d)%s%s%s\n", i,
500 			    t->vectors, (t->caps & HPET_TCNF_INT_ROUTE) >> 9,
501 			    (t->caps & HPET_TCAP_FSB_INT_DEL) ? ", MSI" : "",
502 			    (t->caps & HPET_TCAP_SIZE) ? ", 64bit" : "",
503 			    (t->caps & HPET_TCAP_PER_INT) ? ", periodic" : "");
504 		}
505 	}
506 	if (testenv("debug.acpi.hpet_test"))
507 		hpet_test(sc);
508 	/*
509 	 * Don't attach if the timer never increments.  Since the spec
510 	 * requires it to be at least 10 MHz, it has to change in 1 us.
511 	 */
512 	val = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
513 	DELAY(1);
514 	val2 = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
515 	if (val == val2) {
516 		device_printf(dev, "HPET never increments, disabling\n");
517 		hpet_disable(sc);
518 		bus_free_resource(dev, SYS_RES_MEMORY, sc->mem_res);
519 		return (ENXIO);
520 	}
521 	/* Announce first HPET as timecounter. */
522 	if (device_get_unit(dev) == 0) {
523 		sc->tc.tc_get_timecount = hpet_get_timecount,
524 		sc->tc.tc_counter_mask = ~0u,
525 		sc->tc.tc_name = "HPET",
526 		sc->tc.tc_quality = 950,
527 		sc->tc.tc_frequency = sc->freq;
528 		sc->tc.tc_priv = sc;
529 		tc_init(&sc->tc);
530 	}
531 	/* If not disabled - setup and announce event timers. */
532 	if (resource_int_value(device_get_name(dev), device_get_unit(dev),
533 	     "clock", &i) == 0 && i == 0)
534 	        return (0);
535 
536 	/* Check whether we can and want legacy routing. */
537 	sc->legacy_route = 0;
538 	resource_int_value(device_get_name(dev), device_get_unit(dev),
539 	     "legacy_route", &sc->legacy_route);
540 	if ((sc->caps & HPET_CAP_LEG_RT) == 0)
541 		sc->legacy_route = 0;
542 	if (sc->legacy_route) {
543 		sc->t[0].vectors = 0;
544 		sc->t[1].vectors = 0;
545 	}
546 
547 	/* Check what IRQs we want use. */
548 	/* By default allow any PCI IRQs. */
549 	sc->allowed_irqs = 0xffff0000;
550 	/*
551 	 * HPETs in AMD chipsets before SB800 have problems with IRQs >= 16
552 	 * Lower are also not always working for different reasons.
553 	 * SB800 fixed it, but seems do not implements level triggering
554 	 * properly, that makes it very unreliable - it freezes after any
555 	 * interrupt loss. Avoid legacy IRQs for AMD.
556 	 */
557 	if (vendor == HPET_VENDID_AMD || vendor == HPET_VENDID_AMD2)
558 		sc->allowed_irqs = 0x00000000;
559 	/*
560 	 * NVidia MCP5x chipsets have number of unexplained interrupt
561 	 * problems. For some reason, using HPET interrupts breaks HDA sound.
562 	 */
563 	if (vendor == HPET_VENDID_NVIDIA && rev <= 0x01)
564 		sc->allowed_irqs = 0x00000000;
565 	/*
566 	 * ServerWorks HT1000 reported to have problems with IRQs >= 16.
567 	 * Lower IRQs are working, but allowed mask is not set correctly.
568 	 * Legacy_route mode works fine.
569 	 */
570 	if (vendor == HPET_VENDID_SW && rev <= 0x01)
571 		sc->allowed_irqs = 0x00000000;
572 	/*
573 	 * Neither QEMU nor VirtualBox report supported IRQs correctly.
574 	 * The only way to use HPET there is to specify IRQs manually
575 	 * and/or use legacy_route. Legacy_route mode works on both.
576 	 */
577 	if (vm_guest)
578 		sc->allowed_irqs = 0x00000000;
579 	/* Let user override. */
580 	resource_int_value(device_get_name(dev), device_get_unit(dev),
581 	     "allowed_irqs", &sc->allowed_irqs);
582 
583 	/* Get how much per-CPU timers we should try to provide. */
584 	sc->per_cpu = 1;
585 	resource_int_value(device_get_name(dev), device_get_unit(dev),
586 	     "per_cpu", &sc->per_cpu);
587 
588 	num_msi = 0;
589 	sc->useirq = 0;
590 	/* Find IRQ vectors for all timers. */
591 	cvectors = sc->allowed_irqs & 0xffff0000;
592 	dvectors = sc->allowed_irqs & 0x0000ffff;
593 	if (sc->legacy_route)
594 		dvectors &= 0x0000fefe;
595 	for (i = 0; i < num_timers; i++) {
596 		t = &sc->t[i];
597 		if (sc->legacy_route && i < 2)
598 			t->irq = (i == 0) ? 0 : 8;
599 #ifdef DEV_APIC
600 		else if (t->caps & HPET_TCAP_FSB_INT_DEL) {
601 			if ((j = PCIB_ALLOC_MSIX(
602 			    device_get_parent(device_get_parent(dev)), dev,
603 			    &t->irq))) {
604 				device_printf(dev,
605 				    "Can't allocate interrupt for t%d: %d\n",
606 				    i, j);
607 			}
608 		}
609 #endif
610 		else if (dvectors & t->vectors) {
611 			t->irq = ffs(dvectors & t->vectors) - 1;
612 			dvectors &= ~(1 << t->irq);
613 		}
614 		if (t->irq >= 0) {
615 			t->intr_rid = hpet_find_irq_rid(dev, t->irq, t->irq);
616 			t->intr_res = bus_alloc_resource(dev, SYS_RES_IRQ,
617 			    &t->intr_rid, t->irq, t->irq, 1, RF_ACTIVE);
618 			if (t->intr_res == NULL) {
619 				t->irq = -1;
620 				device_printf(dev,
621 				    "Can't map interrupt for t%d.\n", i);
622 			} else if (bus_setup_intr(dev, t->intr_res,
623 			    INTR_TYPE_CLK, hpet_intr_single, NULL, t,
624 			    &t->intr_handle) != 0) {
625 				t->irq = -1;
626 				device_printf(dev,
627 				    "Can't setup interrupt for t%d.\n", i);
628 			} else {
629 				bus_describe_intr(dev, t->intr_res,
630 				    t->intr_handle, "t%d", i);
631 				num_msi++;
632 			}
633 		}
634 		if (t->irq < 0 && (cvectors & t->vectors) != 0) {
635 			cvectors &= t->vectors;
636 			sc->useirq |= (1 << i);
637 		}
638 	}
639 	if (sc->legacy_route && sc->t[0].irq < 0 && sc->t[1].irq < 0)
640 		sc->legacy_route = 0;
641 	if (sc->legacy_route)
642 		hpet_enable(sc);
643 	/* Group timers for per-CPU operation. */
644 	num_percpu_et = min(num_msi / mp_ncpus, sc->per_cpu);
645 	num_percpu_t = num_percpu_et * mp_ncpus;
646 	pcpu_master = 0;
647 	cur_cpu = CPU_FIRST();
648 	for (i = 0; i < num_timers; i++) {
649 		t = &sc->t[i];
650 		if (t->irq >= 0 && num_percpu_t > 0) {
651 			if (cur_cpu == CPU_FIRST())
652 				pcpu_master = i;
653 			t->pcpu_cpu = cur_cpu;
654 			t->pcpu_master = pcpu_master;
655 			sc->t[pcpu_master].
656 			    pcpu_slaves[cur_cpu] = i;
657 			bus_bind_intr(dev, t->intr_res, cur_cpu);
658 			cur_cpu = CPU_NEXT(cur_cpu);
659 			num_percpu_t--;
660 		} else if (t->irq >= 0)
661 			bus_bind_intr(dev, t->intr_res, CPU_FIRST());
662 	}
663 	bus_write_4(sc->mem_res, HPET_ISR, 0xffffffff);
664 	sc->irq = -1;
665 	/* If at least one timer needs legacy IRQ - set it up. */
666 	if (sc->useirq) {
667 		j = i = fls(cvectors) - 1;
668 		while (j > 0 && (cvectors & (1 << (j - 1))) != 0)
669 			j--;
670 		sc->intr_rid = hpet_find_irq_rid(dev, j, i);
671 		sc->intr_res = bus_alloc_resource(dev, SYS_RES_IRQ,
672 		    &sc->intr_rid, j, i, 1, RF_SHAREABLE | RF_ACTIVE);
673 		if (sc->intr_res == NULL)
674 			device_printf(dev, "Can't map interrupt.\n");
675 		else if (bus_setup_intr(dev, sc->intr_res, INTR_TYPE_CLK,
676 		    hpet_intr, NULL, sc, &sc->intr_handle) != 0) {
677 			device_printf(dev, "Can't setup interrupt.\n");
678 		} else {
679 			sc->irq = rman_get_start(sc->intr_res);
680 			/* Bind IRQ to BSP to avoid live migration. */
681 			bus_bind_intr(dev, sc->intr_res, CPU_FIRST());
682 		}
683 	}
684 	/* Program and announce event timers. */
685 	for (i = 0; i < num_timers; i++) {
686 		t = &sc->t[i];
687 		t->caps &= ~(HPET_TCNF_FSB_EN | HPET_TCNF_INT_ROUTE);
688 		t->caps &= ~(HPET_TCNF_VAL_SET | HPET_TCNF_INT_ENB);
689 		t->caps &= ~(HPET_TCNF_INT_TYPE);
690 		t->caps |= HPET_TCNF_32MODE;
691 		if (t->irq >= 0 && sc->legacy_route && i < 2) {
692 			/* Legacy route doesn't need more configuration. */
693 		} else
694 #ifdef DEV_APIC
695 		if ((t->caps & HPET_TCAP_FSB_INT_DEL) && t->irq >= 0) {
696 			uint64_t addr;
697 			uint32_t data;
698 
699 			if (PCIB_MAP_MSI(
700 			    device_get_parent(device_get_parent(dev)), dev,
701 			    t->irq, &addr, &data) == 0) {
702 				bus_write_4(sc->mem_res,
703 				    HPET_TIMER_FSB_ADDR(i), addr);
704 				bus_write_4(sc->mem_res,
705 				    HPET_TIMER_FSB_VAL(i), data);
706 				t->caps |= HPET_TCNF_FSB_EN;
707 			} else
708 				t->irq = -2;
709 		} else
710 #endif
711 		if (t->irq >= 0)
712 			t->caps |= (t->irq << 9);
713 		else if (sc->irq >= 0 && (t->vectors & (1 << sc->irq)))
714 			t->caps |= (sc->irq << 9) | HPET_TCNF_INT_TYPE;
715 		bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(i), t->caps);
716 		/* Skip event timers without set up IRQ. */
717 		if (t->irq < 0 &&
718 		    (sc->irq < 0 || (t->vectors & (1 << sc->irq)) == 0))
719 			continue;
720 		/* Announce the reset. */
721 		if (maxhpetet == 0)
722 			t->et.et_name = "HPET";
723 		else {
724 			sprintf(t->name, "HPET%d", maxhpetet);
725 			t->et.et_name = t->name;
726 		}
727 		t->et.et_flags = ET_FLAGS_PERIODIC | ET_FLAGS_ONESHOT;
728 		t->et.et_quality = 450;
729 		if (t->pcpu_master >= 0) {
730 			t->et.et_flags |= ET_FLAGS_PERCPU;
731 			t->et.et_quality += 100;
732 		} else if (mp_ncpus >= 8)
733 			t->et.et_quality -= 100;
734 		if ((t->caps & HPET_TCAP_PER_INT) == 0)
735 			t->et.et_quality -= 10;
736 		t->et.et_frequency = sc->freq;
737 		t->et.et_min_period =
738 		    ((uint64_t)(HPET_MIN_CYCLES * 2) << 32) / sc->freq;
739 		t->et.et_max_period = (0xfffffffeLLU << 32) / sc->freq;
740 		t->et.et_start = hpet_start;
741 		t->et.et_stop = hpet_stop;
742 		t->et.et_priv = &sc->t[i];
743 		if (t->pcpu_master < 0 || t->pcpu_master == i) {
744 			et_register(&t->et);
745 			maxhpetet++;
746 		}
747 	}
748 
749 	sc->pdev = make_dev(&hpet_cdevsw, 0, UID_ROOT, GID_WHEEL,
750 	    0600, "hpet%d", device_get_unit(dev));
751 	if (sc->pdev) {
752 		sc->pdev->si_drv1 = sc;
753 		sc->mmap_allow = 1;
754 		TUNABLE_INT_FETCH("hw.acpi.hpet.mmap_allow",
755 		    &sc->mmap_allow);
756 		sc->mmap_allow_write = 1;
757 		TUNABLE_INT_FETCH("hw.acpi.hpet.mmap_allow_write",
758 		    &sc->mmap_allow_write);
759 		SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
760 		    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
761 		    OID_AUTO, "mmap_allow",
762 		    CTLFLAG_RW, &sc->mmap_allow, 0,
763 		    "Allow userland to memory map HPET");
764 		SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
765 		    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
766 		    OID_AUTO, "mmap_allow_write",
767 		    CTLFLAG_RW, &sc->mmap_allow_write, 0,
768 		    "Allow userland write to the HPET register space");
769 	} else
770 		device_printf(dev, "could not create /dev/hpet%d\n",
771 		    device_get_unit(dev));
772 
773 	return (0);
774 }
775 
776 static int
777 hpet_detach(device_t dev)
778 {
779 	ACPI_FUNCTION_TRACE((char *)(uintptr_t) __func__);
780 
781 	/* XXX Without a tc_remove() function, we can't detach. */
782 	return (EBUSY);
783 }
784 
785 static int
786 hpet_suspend(device_t dev)
787 {
788 //	struct hpet_softc *sc;
789 
790 	/*
791 	 * Disable the timer during suspend.  The timer will not lose
792 	 * its state in S1 or S2, but we are required to disable
793 	 * it.
794 	 */
795 //	sc = device_get_softc(dev);
796 //	hpet_disable(sc);
797 
798 	return (0);
799 }
800 
801 static int
802 hpet_resume(device_t dev)
803 {
804 	struct hpet_softc *sc;
805 	struct hpet_timer *t;
806 	int i;
807 
808 	/* Re-enable the timer after a resume to keep the clock advancing. */
809 	sc = device_get_softc(dev);
810 	hpet_enable(sc);
811 	/* Restart event timers that were running on suspend. */
812 	for (i = 0; i < sc->num_timers; i++) {
813 		t = &sc->t[i];
814 #ifdef DEV_APIC
815 		if (t->irq >= 0 && (sc->legacy_route == 0 || i >= 2)) {
816 			uint64_t addr;
817 			uint32_t data;
818 
819 			if (PCIB_MAP_MSI(
820 			    device_get_parent(device_get_parent(dev)), dev,
821 			    t->irq, &addr, &data) == 0) {
822 				bus_write_4(sc->mem_res,
823 				    HPET_TIMER_FSB_ADDR(i), addr);
824 				bus_write_4(sc->mem_res,
825 				    HPET_TIMER_FSB_VAL(i), data);
826 			}
827 		}
828 #endif
829 		if (t->mode == 0)
830 			continue;
831 		t->next = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
832 		if (t->mode == 1 && (t->caps & HPET_TCAP_PER_INT)) {
833 			t->caps |= HPET_TCNF_TYPE;
834 			t->next += t->div;
835 			bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num),
836 			    t->caps | HPET_TCNF_VAL_SET);
837 			bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
838 			    t->next);
839 			bus_read_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num));
840 			bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
841 			    t->div);
842 		} else {
843 			t->next += sc->freq / 1024;
844 			bus_write_4(sc->mem_res, HPET_TIMER_COMPARATOR(t->num),
845 			    t->next);
846 		}
847 		bus_write_4(sc->mem_res, HPET_ISR, 1 << t->num);
848 		bus_write_4(sc->mem_res, HPET_TIMER_CAP_CNF(t->num), t->caps);
849 	}
850 	return (0);
851 }
852 
853 /* Print some basic latency/rate information to assist in debugging. */
854 static void
855 hpet_test(struct hpet_softc *sc)
856 {
857 	int i;
858 	uint32_t u1, u2;
859 	struct bintime b0, b1, b2;
860 	struct timespec ts;
861 
862 	binuptime(&b0);
863 	binuptime(&b0);
864 	binuptime(&b1);
865 	u1 = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
866 	for (i = 1; i < 1000; i++)
867 		u2 = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
868 	binuptime(&b2);
869 	u2 = bus_read_4(sc->mem_res, HPET_MAIN_COUNTER);
870 
871 	bintime_sub(&b2, &b1);
872 	bintime_sub(&b1, &b0);
873 	bintime_sub(&b2, &b1);
874 	bintime2timespec(&b2, &ts);
875 
876 	device_printf(sc->dev, "%ld.%09ld: %u ... %u = %u\n",
877 	    (long)ts.tv_sec, ts.tv_nsec, u1, u2, u2 - u1);
878 
879 	device_printf(sc->dev, "time per call: %ld ns\n", ts.tv_nsec / 1000);
880 }
881 
882 #ifdef DEV_APIC
883 static int
884 hpet_remap_intr(device_t dev, device_t child, u_int irq)
885 {
886 	struct hpet_softc *sc = device_get_softc(dev);
887 	struct hpet_timer *t;
888 	uint64_t addr;
889 	uint32_t data;
890 	int error, i;
891 
892 	for (i = 0; i < sc->num_timers; i++) {
893 		t = &sc->t[i];
894 		if (t->irq != irq)
895 			continue;
896 		error = PCIB_MAP_MSI(
897 		    device_get_parent(device_get_parent(dev)), dev,
898 		    irq, &addr, &data);
899 		if (error)
900 			return (error);
901 		hpet_disable(sc); /* Stop timer to avoid interrupt loss. */
902 		bus_write_4(sc->mem_res, HPET_TIMER_FSB_ADDR(i), addr);
903 		bus_write_4(sc->mem_res, HPET_TIMER_FSB_VAL(i), data);
904 		hpet_enable(sc);
905 		return (0);
906 	}
907 	return (ENOENT);
908 }
909 #endif
910 
911 static device_method_t hpet_methods[] = {
912 	/* Device interface */
913 	DEVMETHOD(device_identify, hpet_identify),
914 	DEVMETHOD(device_probe, hpet_probe),
915 	DEVMETHOD(device_attach, hpet_attach),
916 	DEVMETHOD(device_detach, hpet_detach),
917 	DEVMETHOD(device_suspend, hpet_suspend),
918 	DEVMETHOD(device_resume, hpet_resume),
919 
920 #ifdef DEV_APIC
921 	DEVMETHOD(bus_remap_intr, hpet_remap_intr),
922 #endif
923 
924 	DEVMETHOD_END
925 };
926 
927 static driver_t	hpet_driver = {
928 	"hpet",
929 	hpet_methods,
930 	sizeof(struct hpet_softc),
931 };
932 
933 DRIVER_MODULE(hpet, acpi, hpet_driver, hpet_devclass, 0, 0);
934 MODULE_DEPEND(hpet, acpi, 1, 1, 1);
935