1/*-
2 * Copyright (c) 2003-2005 Nate Lawson (SDG)
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 */
26
27#include <sys/cdefs.h>
28__FBSDID("$FreeBSD$");
29
30#include "opt_acpi.h"
31#include <sys/param.h>
32#include <sys/kernel.h>
33#include <sys/proc.h>
34#include <sys/sched.h>
35#include <sys/bus.h>
36#include <sys/cpu.h>
37#include <sys/power.h>
38#include <sys/malloc.h>
39#include <sys/module.h>
40#include <sys/sbuf.h>
41#include <sys/pcpu.h>
42
43#include <machine/bus.h>
44#include <machine/resource.h>
45#include <sys/rman.h>
46
47#include <contrib/dev/acpica/include/acpi.h>
48
49#include <dev/acpica/acpivar.h>
50
51#include "cpufreq_if.h"
52
53/*
54 * Support for ACPI processor performance states (Px) according to
55 * section 8.3.3 of the ACPI 2.0c specification.
56 */
57
58struct acpi_px {
59	uint32_t	 core_freq;
60	uint32_t	 power;
61	uint32_t	 trans_lat;
62	uint32_t	 bm_lat;
63	uint32_t	 ctrl_val;
64	uint32_t	 sts_val;
65};
66
67/* Offsets in struct cf_setting array for storing driver-specific values. */
68#define PX_SPEC_CONTROL	0
69#define PX_SPEC_STATUS	1
70
71#define MAX_PX_STATES	16
72
73struct acpi_perf_softc {
74	device_t	 dev;
75	ACPI_HANDLE	 handle;
76	struct resource	*perf_ctrl;	/* Set new performance state. */
77	int		 perf_ctrl_type; /* Resource type for perf_ctrl. */
78	struct resource	*perf_status;	/* Check that transition succeeded. */
79	int		 perf_sts_type;	/* Resource type for perf_status. */
80	struct acpi_px	*px_states;	/* ACPI perf states. */
81	uint32_t	 px_count;	/* Total number of perf states. */
82	uint32_t	 px_max_avail;	/* Lowest index state available. */
83	int		 px_curr_state;	/* Active state index. */
84	int		 px_rid;
85	int		 info_only;	/* Can we set new states? */
86};
87
88#define PX_GET_REG(reg) 				\
89	(bus_space_read_4(rman_get_bustag((reg)), 	\
90	    rman_get_bushandle((reg)), 0))
91#define PX_SET_REG(reg, val)				\
92	(bus_space_write_4(rman_get_bustag((reg)), 	\
93	    rman_get_bushandle((reg)), 0, (val)))
94
95#define ACPI_NOTIFY_PERF_STATES		0x80	/* _PSS changed. */
96
97static void	acpi_perf_identify(driver_t *driver, device_t parent);
98static int	acpi_perf_probe(device_t dev);
99static int	acpi_perf_attach(device_t dev);
100static int	acpi_perf_detach(device_t dev);
101static int	acpi_perf_evaluate(device_t dev);
102static int	acpi_px_to_set(device_t dev, struct acpi_px *px,
103		    struct cf_setting *set);
104static void	acpi_px_available(struct acpi_perf_softc *sc);
105static void	acpi_px_startup(void *arg);
106static void	acpi_px_notify(ACPI_HANDLE h, UINT32 notify, void *context);
107static int	acpi_px_settings(device_t dev, struct cf_setting *sets,
108		    int *count);
109static int	acpi_px_set(device_t dev, const struct cf_setting *set);
110static int	acpi_px_get(device_t dev, struct cf_setting *set);
111static int	acpi_px_type(device_t dev, int *type);
112
113static device_method_t acpi_perf_methods[] = {
114	/* Device interface */
115	DEVMETHOD(device_identify,	acpi_perf_identify),
116	DEVMETHOD(device_probe,		acpi_perf_probe),
117	DEVMETHOD(device_attach,	acpi_perf_attach),
118	DEVMETHOD(device_detach,	acpi_perf_detach),
119
120	/* cpufreq interface */
121	DEVMETHOD(cpufreq_drv_set,	acpi_px_set),
122	DEVMETHOD(cpufreq_drv_get,	acpi_px_get),
123	DEVMETHOD(cpufreq_drv_type,	acpi_px_type),
124	DEVMETHOD(cpufreq_drv_settings,	acpi_px_settings),
125
126	DEVMETHOD_END
127};
128
129static driver_t acpi_perf_driver = {
130	"acpi_perf",
131	acpi_perf_methods,
132	sizeof(struct acpi_perf_softc),
133};
134
135static devclass_t acpi_perf_devclass;
136DRIVER_MODULE(acpi_perf, cpu, acpi_perf_driver, acpi_perf_devclass, 0, 0);
137MODULE_DEPEND(acpi_perf, acpi, 1, 1, 1);
138
139static MALLOC_DEFINE(M_ACPIPERF, "acpi_perf", "ACPI Performance states");
140
141static void
142acpi_perf_identify(driver_t *driver, device_t parent)
143{
144	ACPI_HANDLE handle;
145	device_t dev;
146
147	/* Make sure we're not being doubly invoked. */
148	if (device_find_child(parent, "acpi_perf", -1) != NULL)
149		return;
150
151	/* Get the handle for the Processor object and check for perf states. */
152	handle = acpi_get_handle(parent);
153	if (handle == NULL)
154		return;
155	if (ACPI_FAILURE(AcpiEvaluateObject(handle, "_PSS", NULL, NULL)))
156		return;
157
158	/*
159	 * Add a child to every CPU that has the right methods.  In future
160	 * versions of the ACPI spec, CPUs can have different settings.
161	 * We probe this child now so that other devices that depend
162	 * on it (i.e., for info about supported states) will see it.
163	 */
164	if ((dev = BUS_ADD_CHILD(parent, 0, "acpi_perf", -1)) != NULL)
165		device_probe_and_attach(dev);
166	else
167		device_printf(parent, "add acpi_perf child failed\n");
168}
169
170static int
171acpi_perf_probe(device_t dev)
172{
173	ACPI_HANDLE handle;
174	ACPI_OBJECT *pkg;
175	struct resource *res;
176	ACPI_BUFFER buf;
177	int error, rid, type;
178
179	if (resource_disabled("acpi_perf", 0))
180		return (ENXIO);
181
182	/*
183	 * Check the performance state registers.  If they are of type
184	 * "functional fixed hardware", we attach quietly since we will
185	 * only be providing information on settings to other drivers.
186	 */
187	error = ENXIO;
188	handle = acpi_get_handle(dev);
189	buf.Pointer = NULL;
190	buf.Length = ACPI_ALLOCATE_BUFFER;
191	if (ACPI_FAILURE(AcpiEvaluateObject(handle, "_PCT", NULL, &buf)))
192		return (error);
193	pkg = (ACPI_OBJECT *)buf.Pointer;
194	if (ACPI_PKG_VALID(pkg, 2)) {
195		rid = 0;
196		error = acpi_PkgGas(dev, pkg, 0, &type, &rid, &res, 0);
197		switch (error) {
198		case 0:
199			bus_release_resource(dev, type, rid, res);
200			bus_delete_resource(dev, type, rid);
201			device_set_desc(dev, "ACPI CPU Frequency Control");
202			break;
203		case EOPNOTSUPP:
204			device_quiet(dev);
205			error = 0;
206			break;
207		}
208	}
209	AcpiOsFree(buf.Pointer);
210
211	return (error);
212}
213
214static int
215acpi_perf_attach(device_t dev)
216{
217	struct acpi_perf_softc *sc;
218
219	sc = device_get_softc(dev);
220	sc->dev = dev;
221	sc->handle = acpi_get_handle(dev);
222	sc->px_max_avail = 0;
223	sc->px_curr_state = CPUFREQ_VAL_UNKNOWN;
224	if (acpi_perf_evaluate(dev) != 0)
225		return (ENXIO);
226	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_px_startup, NULL);
227	if (!sc->info_only)
228		cpufreq_register(dev);
229
230	return (0);
231}
232
233static int
234acpi_perf_detach(device_t dev)
235{
236	/* TODO: teardown registers, remove notify handler. */
237	return (ENXIO);
238}
239
240/* Probe and setup any valid performance states (Px). */
241static int
242acpi_perf_evaluate(device_t dev)
243{
244	struct acpi_perf_softc *sc;
245	ACPI_BUFFER buf;
246	ACPI_OBJECT *pkg, *res;
247	ACPI_STATUS status;
248	int count, error, i, j;
249	static int once = 1;
250	uint32_t *p;
251
252	/* Get the control values and parameters for each state. */
253	error = ENXIO;
254	sc = device_get_softc(dev);
255	buf.Pointer = NULL;
256	buf.Length = ACPI_ALLOCATE_BUFFER;
257	status = AcpiEvaluateObject(sc->handle, "_PSS", NULL, &buf);
258	if (ACPI_FAILURE(status))
259		return (ENXIO);
260
261	pkg = (ACPI_OBJECT *)buf.Pointer;
262	if (!ACPI_PKG_VALID(pkg, 1)) {
263		device_printf(dev, "invalid top level _PSS package\n");
264		goto out;
265	}
266	sc->px_count = pkg->Package.Count;
267
268	sc->px_states = malloc(sc->px_count * sizeof(struct acpi_px),
269	    M_ACPIPERF, M_WAITOK | M_ZERO);
270
271	/*
272	 * Each state is a package of {CoreFreq, Power, TransitionLatency,
273	 * BusMasterLatency, ControlVal, StatusVal}, sorted from highest
274	 * performance to lowest.
275	 */
276	count = 0;
277	for (i = 0; i < sc->px_count; i++) {
278		res = &pkg->Package.Elements[i];
279		if (!ACPI_PKG_VALID(res, 6)) {
280			if (once) {
281				once = 0;
282				device_printf(dev, "invalid _PSS package\n");
283			}
284			continue;
285		}
286
287		/* Parse the rest of the package into the struct. */
288		p = &sc->px_states[count].core_freq;
289		for (j = 0; j < 6; j++, p++)
290			acpi_PkgInt32(res, j, p);
291
292		/*
293		 * Check for some impossible frequencies that some systems
294		 * use to indicate they don't actually support this Px state.
295		 */
296		if (sc->px_states[count].core_freq == 0 ||
297		    sc->px_states[count].core_freq == 9999 ||
298		    sc->px_states[count].core_freq == 0x9999 ||
299		    sc->px_states[count].core_freq >= 0xffff)
300			continue;
301
302		/* Check for duplicate entries */
303		if (count > 0 &&
304		    sc->px_states[count - 1].core_freq ==
305			sc->px_states[count].core_freq)
306			continue;
307
308		count++;
309	}
310	sc->px_count = count;
311
312	/* No valid Px state found so give up. */
313	if (count == 0)
314		goto out;
315	AcpiOsFree(buf.Pointer);
316
317	/* Get the control and status registers (one of each). */
318	buf.Pointer = NULL;
319	buf.Length = ACPI_ALLOCATE_BUFFER;
320	status = AcpiEvaluateObject(sc->handle, "_PCT", NULL, &buf);
321	if (ACPI_FAILURE(status))
322		goto out;
323
324	/* Check the package of two registers, each a Buffer in GAS format. */
325	pkg = (ACPI_OBJECT *)buf.Pointer;
326	if (!ACPI_PKG_VALID(pkg, 2)) {
327		device_printf(dev, "invalid perf register package\n");
328		goto out;
329	}
330
331	error = acpi_PkgGas(sc->dev, pkg, 0, &sc->perf_ctrl_type, &sc->px_rid,
332	    &sc->perf_ctrl, 0);
333	if (error) {
334		/*
335		 * If the register is of type FFixedHW, we can only return
336		 * info, we can't get or set new settings.
337		 */
338		if (error == EOPNOTSUPP) {
339			sc->info_only = TRUE;
340			error = 0;
341		} else
342			device_printf(dev, "failed in PERF_CTL attach\n");
343		goto out;
344	}
345	sc->px_rid++;
346
347	error = acpi_PkgGas(sc->dev, pkg, 1, &sc->perf_sts_type, &sc->px_rid,
348	    &sc->perf_status, 0);
349	if (error) {
350		if (error == EOPNOTSUPP) {
351			sc->info_only = TRUE;
352			error = 0;
353		} else
354			device_printf(dev, "failed in PERF_STATUS attach\n");
355		goto out;
356	}
357	sc->px_rid++;
358
359	/* Get our current limit and register for notifies. */
360	acpi_px_available(sc);
361	AcpiInstallNotifyHandler(sc->handle, ACPI_DEVICE_NOTIFY,
362	    acpi_px_notify, sc);
363	error = 0;
364
365out:
366	if (error) {
367		if (sc->px_states) {
368			free(sc->px_states, M_ACPIPERF);
369			sc->px_states = NULL;
370		}
371		if (sc->perf_ctrl) {
372			bus_release_resource(sc->dev, sc->perf_ctrl_type, 0,
373			    sc->perf_ctrl);
374			bus_delete_resource(sc->dev, sc->perf_ctrl_type, 0);
375			sc->perf_ctrl = NULL;
376		}
377		if (sc->perf_status) {
378			bus_release_resource(sc->dev, sc->perf_sts_type, 1,
379			    sc->perf_status);
380			bus_delete_resource(sc->dev, sc->perf_sts_type, 1);
381			sc->perf_status = NULL;
382		}
383		sc->px_rid = 0;
384		sc->px_count = 0;
385	}
386	if (buf.Pointer)
387		AcpiOsFree(buf.Pointer);
388	return (error);
389}
390
391static void
392acpi_px_startup(void *arg)
393{
394
395	/* Signal to the platform that we are taking over CPU control. */
396	if (AcpiGbl_FADT.PstateControl == 0)
397		return;
398	ACPI_LOCK(acpi);
399	AcpiOsWritePort(AcpiGbl_FADT.SmiCommand, AcpiGbl_FADT.PstateControl, 8);
400	ACPI_UNLOCK(acpi);
401}
402
403static void
404acpi_px_notify(ACPI_HANDLE h, UINT32 notify, void *context)
405{
406	struct acpi_perf_softc *sc;
407
408	sc = context;
409	if (notify != ACPI_NOTIFY_PERF_STATES)
410		return;
411
412	acpi_px_available(sc);
413
414	/* TODO: Implement notification when frequency changes. */
415}
416
417/*
418 * Find the highest currently-supported performance state.
419 * This can be called at runtime (e.g., due to a docking event) at
420 * the request of a Notify on the processor object.
421 */
422static void
423acpi_px_available(struct acpi_perf_softc *sc)
424{
425	ACPI_STATUS status;
426	struct cf_setting set;
427
428	status = acpi_GetInteger(sc->handle, "_PPC", &sc->px_max_avail);
429
430	/* If the old state is too high, set current state to the new max. */
431	if (ACPI_SUCCESS(status)) {
432		if (sc->px_curr_state != CPUFREQ_VAL_UNKNOWN &&
433		    sc->px_curr_state > sc->px_max_avail) {
434			acpi_px_to_set(sc->dev,
435			    &sc->px_states[sc->px_max_avail], &set);
436			acpi_px_set(sc->dev, &set);
437		}
438	} else
439		sc->px_max_avail = 0;
440}
441
442static int
443acpi_px_to_set(device_t dev, struct acpi_px *px, struct cf_setting *set)
444{
445
446	if (px == NULL || set == NULL)
447		return (EINVAL);
448
449	set->freq = px->core_freq;
450	set->power = px->power;
451	/* XXX Include BM latency too? */
452	set->lat = px->trans_lat;
453	set->volts = CPUFREQ_VAL_UNKNOWN;
454	set->dev = dev;
455	set->spec[PX_SPEC_CONTROL] = px->ctrl_val;
456	set->spec[PX_SPEC_STATUS] = px->sts_val;
457
458	return (0);
459}
460
461static int
462acpi_px_settings(device_t dev, struct cf_setting *sets, int *count)
463{
464	struct acpi_perf_softc *sc;
465	int x, y;
466
467	sc = device_get_softc(dev);
468	if (sets == NULL || count == NULL)
469		return (EINVAL);
470	if (*count < sc->px_count - sc->px_max_avail)
471		return (E2BIG);
472
473	/* Return a list of settings that are currently valid. */
474	y = 0;
475	for (x = sc->px_max_avail; x < sc->px_count; x++, y++)
476		acpi_px_to_set(dev, &sc->px_states[x], &sets[y]);
477	*count = sc->px_count - sc->px_max_avail;
478
479	return (0);
480}
481
482static int
483acpi_px_set(device_t dev, const struct cf_setting *set)
484{
485	struct acpi_perf_softc *sc;
486	int i, status, sts_val, tries;
487
488	if (set == NULL)
489		return (EINVAL);
490	sc = device_get_softc(dev);
491
492	/* If we can't set new states, return immediately. */
493	if (sc->info_only)
494		return (ENXIO);
495
496	/* Look up appropriate state, based on frequency. */
497	for (i = sc->px_max_avail; i < sc->px_count; i++) {
498		if (CPUFREQ_CMP(set->freq, sc->px_states[i].core_freq))
499			break;
500	}
501	if (i == sc->px_count)
502		return (EINVAL);
503
504	/* Write the appropriate value to the register. */
505	PX_SET_REG(sc->perf_ctrl, sc->px_states[i].ctrl_val);
506
507	/*
508	 * Try for up to 10 ms to verify the desired state was selected.
509	 * This is longer than the standard says (1 ms) but in some modes,
510	 * systems may take longer to respond.
511	 */
512	sts_val = sc->px_states[i].sts_val;
513	for (tries = 0; tries < 1000; tries++) {
514		status = PX_GET_REG(sc->perf_status);
515
516		/*
517		 * If we match the status or the desired status is 8 bits
518		 * and matches the relevant bits, assume we succeeded.  It
519		 * appears some systems (IBM R32) expect byte-wide access
520		 * even though the standard says the register is 32-bit.
521		 */
522		if (status == sts_val ||
523		    ((sts_val & ~0xff) == 0 && (status & 0xff) == sts_val))
524			break;
525		DELAY(10);
526	}
527	if (tries == 1000) {
528		device_printf(dev, "Px transition to %d failed\n",
529		    sc->px_states[i].core_freq);
530		return (ENXIO);
531	}
532	sc->px_curr_state = i;
533
534	return (0);
535}
536
537static int
538acpi_px_get(device_t dev, struct cf_setting *set)
539{
540	struct acpi_perf_softc *sc;
541	uint64_t rate;
542	int i;
543	struct pcpu *pc;
544
545	if (set == NULL)
546		return (EINVAL);
547	sc = device_get_softc(dev);
548
549	/* If we can't get new states, return immediately. */
550	if (sc->info_only)
551		return (ENXIO);
552
553	/* If we've set the rate before, use the cached value. */
554	if (sc->px_curr_state != CPUFREQ_VAL_UNKNOWN) {
555		acpi_px_to_set(dev, &sc->px_states[sc->px_curr_state], set);
556		return (0);
557	}
558
559	/* Otherwise, estimate and try to match against our settings. */
560	pc = cpu_get_pcpu(dev);
561	if (pc == NULL)
562		return (ENXIO);
563	cpu_est_clockrate(pc->pc_cpuid, &rate);
564	rate /= 1000000;
565	for (i = 0; i < sc->px_count; i++) {
566		if (CPUFREQ_CMP(sc->px_states[i].core_freq, rate)) {
567			sc->px_curr_state = i;
568			acpi_px_to_set(dev, &sc->px_states[i], set);
569			break;
570		}
571	}
572
573	/* No match, give up. */
574	if (i == sc->px_count) {
575		sc->px_curr_state = CPUFREQ_VAL_UNKNOWN;
576		set->freq = CPUFREQ_VAL_UNKNOWN;
577	}
578
579	return (0);
580}
581
582static int
583acpi_px_type(device_t dev, int *type)
584{
585	struct acpi_perf_softc *sc;
586
587	if (type == NULL)
588		return (EINVAL);
589	sc = device_get_softc(dev);
590
591	*type = CPUFREQ_TYPE_ABSOLUTE;
592	if (sc->info_only)
593		*type |= CPUFREQ_FLAG_INFO_ONLY;
594	return (0);
595}
596